WINSTEPS MINISTEP Rasch-Model Computer Programs User's Guide

Below you will find brief information for WINSTEPS, MINISTEP. These are Rasch-model computer programs and tools for analyzing rating scale data, dichotomous data, and more. They perform item recoding, item deletion, scoring key implementation, and partial credit analysis. Useful for combining tests, comparing samples, and paired comparisons.
PDF
Document
A User's Guide to
WINSTEPS
MINISTEP
Rasch-Model
Computer Programs
John M. Linacre
www.winsteps.com
A User's Guide to
WINSTEPS
MINISTEP
Rasch-Model
Computer Programs
Program Manual by
John M. Linacre
P. O. Box 811322
Chicago IL 60681-1322
Tel. &amp; FAX (312) 264-2352
winsteps.com
&copy; Copyright 1991-2006, John M. Linacre
Permission to copy is granted.
ISBN 0-941938-03-4
Winsteps&reg; is a registered trademark
&reg;
Table of Contents
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
54.
55.
56.
57.
Control Variable Index .......................................................................................................................................1
Control Variable Index by function .....................................................................................................................3
Control variables from Specification menu ........................................................................................................6
Output Table Index ............................................................................................................................................8
Output Graphs and Plots Index........................................................................................................................11
Output Files Index ............................................................................................................................................12
Rasch analysis and WINSTEPS ......................................................................................................................12
References.......................................................................................................................................................14
About the Users' guide ................................................................................................................................14
Getting further help ......................................................................................................................................15
What is supplied ..........................................................................................................................................15
Installation instructions for WINSTEPS .......................................................................................................15
Starting WINSTEPS in Windows .................................................................................................................16
Using Winsteps under Windows..................................................................................................................18
Stopping WINSTEPS...................................................................................................................................19
Uninstalling WINSTEPS ..............................................................................................................................20
Menu bar......................................................................................................................................................20
Batch menu..................................................................................................................................................20
Data Setup menu.........................................................................................................................................21
Diagnosis menu ...........................................................................................................................................21
Edit menu.....................................................................................................................................................21
File menu .....................................................................................................................................................22
Graphs menu ...............................................................................................................................................23
Help menu ...................................................................................................................................................24
Output Files menu .......................................................................................................................................25
Output Tables menu ....................................................................................................................................26
Plots menu...................................................................................................................................................27
SAS/SPSS menu .........................................................................................................................................28
Specification menu ......................................................................................................................................28
Control and data file setup window..............................................................................................................29
Reading in a pre-existent control file ...........................................................................................................29
Data display .................................................................................................................................................30
Item labels ...................................................................................................................................................31
Category labels............................................................................................................................................31
Do-it-yourself control and data file construction ..........................................................................................32
Control file and template.txt.........................................................................................................................35
Data file........................................................................................................................................................36
Data from Excel and other spreadsheets ....................................................................................................37
Data from SAS files .....................................................................................................................................37
Data from SPSS files ...................................................................................................................................39
Data from STATA files .................................................................................................................................41
Data file with other delimiters ......................................................................................................................43
Example 0: Rating scale data: The Liking for Science data........................................................................43
Example 1: Dichotomous data: Simple control file with data included ........................................................45
Example 2: Control and anchor files............................................................................................................47
Example 3: Item recoding and item deletion ...............................................................................................48
Example 4: Selective item recoding ............................................................................................................48
Example 5: Scoring key for items, also CAT responses .............................................................................49
Example 6: Keys in data record FORMAT ..................................................................................................50
Example 7: A partial credit analysis.............................................................................................................51
Example 8: Items with various rating scale models.....................................................................................52
Example 9: Grouping and modeling items ..................................................................................................52
Example 10: Combining tests with common items......................................................................................53
Example 11: Item responses two characters wide ......................................................................................56
Example 12: Comparing high and low samples with rating scales .............................................................56
Example 13: Paired comparisons as the basis for measurement ...............................................................59
ii
58.
59.
60.
61.
62.
63.
64.
65.
66.
67.
68.
69.
70.
71.
72.
73.
74.
75.
76.
77.
78.
79.
80.
81.
82.
83.
84.
85.
86.
87.
88.
89.
90.
91.
92.
93.
94.
95.
96.
97.
98.
99.
100.
101.
102.
103.
104.
105.
106.
107.
108.
109.
110.
111.
112.
113.
114.
115.
116.
Example 14: Multiple rescorings, response structures and widths .............................................................60
Example 15: Figure skating: Multidimensionality, DIF or Bias ....................................................................64
Example 16: Disordered categories - thresholds and person anchoring ....................................................65
Example 17: Rack, stack, principal components.........................................................................................68
ALPHANUM alphabetic numbering ............................................................................................................69
ASCII output only ASCII characters ...........................................................................................................69
ASYMPTOTE item upper and lower asymptotes .......................................................................................70
BATCH Batch mode analysis .....................................................................................................................71
BYITEM display graphs for items ...............................................................................................................74
CATREF reference category for Table 2 ....................................................................................................74
CFILE scored category label file.................................................................................................................75
CHART graphical plots in Tables 10, 13-15 ...............................................................................................76
CLFILE codes label file...............................................................................................................................77
CODES valid data codes ............................................................................................................................78
CONVERGE select convergence criteria ....................................................................................................80
CSV comma-separated values in output files ............................................................................................81
CURVES probability curves for Table 21 and Table 2 ...............................................................................82
CUTHI cut off responses with high probability of success .........................................................................82
CUTLO cut off responses with low probability of success .........................................................................83
DATA name of data file ..............................................................................................................................83
DELIMITER data field delimiters ................................................................................................................83
DIF columns within person label for Table 30 .............................................................................................85
DISCRIMINATION item discrimination ........................................................................................................86
DISFILE category/distractor/option count file .............................................................................................87
DISTRACTOR output option counts in Tables 10, 13-15...........................................................................87
DPF columns within item label for Table 31 ................................................................................................88
EDFILE edit data file...................................................................................................................................88
END LABELS or END NAMES....................................................................................................................89
EQFILE code equivalences .........................................................................................................................89
EXTRSC extreme score correction for extreme measures ........................................................................89
FITHIGH higher bar in charts ......................................................................................................................90
FITI item misfit criterion ..............................................................................................................................91
FITLOW lower bar in charts.........................................................................................................................91
FITP person misfit criterion.........................................................................................................................91
FORMAT reformat data ..............................................................................................................................91
FORMFD the form feed character ..............................................................................................................97
FRANGE half-range of fit statistics on plots ...............................................................................................97
GRFILE probability curve coordinate output file.........................................................................................98
GROUPS or ISGROUPS assigns items to rating scale groupings .............................................................98
GRPFROM location of ISGROUPS..........................................................................................................101
GUFILE (G0ZONE, G1ZONE) Guttmanized response file ......................................................................102
HEADER display or suppress subtable headings ....................................................................................102
HIADJ correction for top rating scale categories ......................................................................................103
HLINES heading lines in output files ........................................................................................................103
IAFILE item anchor file .............................................................................................................................103
IANCHQU anchor items interactively .......................................................................................................105
ICORFILE item residual correlation file ....................................................................................................105
IDELETE item one-line item deletion .......................................................................................................106
IDELQU delete items interactively............................................................................................................106
IDFILE item deletion file ...........................................................................................................................107
IDROPEXTREME drop items with extreme scores...................................................................................109
IFILE item output file.................................................................................................................................109
ILFILE item label file .................................................................................................................................111
IMAP item label on item maps Tables 1, 12.............................................................................................111
INUMB label items by sequence numbers ...............................................................................................111
IPMATRIX response-level matrix ..............................................................................................................112
IREFER identifying items for recoding .....................................................................................................114
ISELECT item selection criterion..............................................................................................................115
ISFILE item structure output file ...............................................................................................................116
iii
117.
118.
119.
120.
121.
122.
123.
124.
125.
126.
127.
128.
129.
130.
131.
132.
133.
134.
135.
136.
137.
138.
139.
140.
141.
142.
143.
144.
145.
146.
147.
148.
149.
150.
151.
152.
153.
154.
155.
156.
157.
158.
159.
160.
161.
162.
163.
164.
165.
166.
167.
168.
169.
170.
171.
172.
173.
174.
175.
ISORT column within item name for alphabetical sort in Table 15 ...........................................................118
ISUBTOTAL columns within item label for subtotals in Table 27.............................................................119
ITEM title for item labels ...........................................................................................................................120
ITEM1 column number of first response ..................................................................................................120
ITLEN maximum length of item label .......................................................................................................121
IVALUEx recoding of data ........................................................................................................................121
IWEIGHT item (variable) weighting ..........................................................................................................122
KEYn scoring key .....................................................................................................................................124
KEYFROM location of KEYn ....................................................................................................................125
KEYSCR reassign scoring keys ...............................................................................................................125
LCONV logit change at convergence .......................................................................................................126
LINLEN length of printed lines in Tables 7, 10-16, 22..............................................................................127
LOCAL locally restandardize fit statistics .................................................................................................127
LOGFILE accumulates control files ...........................................................................................................127
LOWADJ correction for bottom rating scale categories ...........................................................................128
MAKEKEY construct MCQ key.................................................................................................................128
MATRIX correlation output format ............................................................................................................128
MAXPAGE the maximum number of lines per page ................................................................................129
MFORMS reformat input data ...................................................................................................................129
MHSLICE Mantel-Haenszel slice width .....................................................................................................132
MISSCORE scoring of missing data codes ..............................................................................................134
MJMLE maximum number of JMLE iterations..........................................................................................135
MNSQ show mean-square or standardized fit statistics ..........................................................................135
MODELS assigns model types to items ...................................................................................................136
MODFROM location of MODELS .............................................................................................................137
MPROX maximum number of PROX iterations........................................................................................138
MRANGE half-range of measures on plots ..............................................................................................138
NAME1 first column of person label .........................................................................................................139
NAMLEN length of person label ...............................................................................................................139
NAMLMP name length on map for Tables 1, 12, 16 ................................................................................140
NEWSCORE recoding values ..................................................................................................................140
NI number of items ...................................................................................................................................141
NORMAL normal distribution for standardizing fit ....................................................................................142
OSORT option/distractor sort ...................................................................................................................142
OUTFIT sort misfits on infit or outfit..........................................................................................................142
PAFILE person anchor file........................................................................................................................143
PAIRED correction for paired comparison data .......................................................................................144
PANCHQU anchor persons interactively..................................................................................................144
PCORFIL person residual correlation file .................................................................................................144
PDELETE person one-line item deletion ...................................................................................................145
PDELQU delete persons interactively ......................................................................................................145
PDFILE person deletion file......................................................................................................................146
PDROPEXTREME drop persons with extreme scores .............................................................................147
PERSON title for person labels ................................................................................................................147
PFILE person output file ...........................................................................................................................147
PMAP person label on person map: Tables 1, 16....................................................................................149
PRCOMP residual type for principal components analyses in Tables 23, 24 ..........................................149
PSELECT person selection criterion ........................................................................................................149
PSORT column within person label for alphabetical sort in Table 19 ......................................................150
PSUBTOTAL columns within person label for subtotals in Table 28 .......................................................151
PTBIS compute point-biserial correlation coefficients ..............................................................................152
PVALUE proportion correct or average rating...........................................................................................153
PWEIGHT person (case) weighting..........................................................................................................153
QUOTED quote-marks around labels........................................................................................................154
RCONV score residual at convergence ...................................................................................................155
REALSE inflate S.E. for misfit ..................................................................................................................155
RESCORE response recoding .................................................................................................................155
RESFROM location of RESCORE ...........................................................................................................157
RFILE scored response file ......................................................................................................................157
iv
176.
177.
178.
179.
180.
181.
182.
183.
184.
185.
186.
187.
188.
189.
190.
191.
192.
193.
194.
195.
196.
197.
198.
199.
200.
201.
202.
203.
204.
205.
206.
207.
208.
209.
210.
211.
212.
213.
214.
215.
216.
217.
218.
219.
220.
221.
222.
223.
224.
225.
226.
227.
228.
229.
230.
231.
232.
233.
234.
SAFILE item structure anchor file.............................................................................................................158
SAITEM item numbers in SAFILE with one grouping................................................................................163
SANCHQ anchor category structure interactively ....................................................................................163
SCOREFILE person score file..................................................................................................................164
SDELQU delete category structure interactively ......................................................................................165
SDFILE category structure deletion file ....................................................................................................165
SFILE category structure output file .........................................................................................................166
SIFILE simulated data file.........................................................................................................................167
SPFILE supplementary control file ...........................................................................................................168
STBIAS correct for estimation bias...........................................................................................................168
STKEEP keep non-observed intermediate categories in structure ..........................................................169
STEPT3 include category structure summary in Table 3 or 21................................................................170
T1I# number of items summarized by &quot;#&quot; symbol in Table 1 ...................................................................170
T1P# number of persons summarized by &quot;#&quot; symbol in Table 1 .............................................................170
TABLES output tables ..............................................................................................................................170
TARGET estimate using information-weighting .......................................................................................171
TFILE input file listing tables to be output ................................................................................................171
TITLE title for output listing.......................................................................................................................172
TOTALSCORE show total scores with extreme observations ..................................................................173
UANCHOR anchor values supplied in user-scaled units .........................................................................173
UASCALE the anchor user-scale value of 1 logit.....................................................................................174
UCOUNT number of unexpected responses: Tables 6, 10.......................................................................174
UDECIMALS number of decimal places reported ....................................................................................174
UIMEAN the mean or center of the item difficulties..................................................................................174
UPMEAN the mean or center of the person abilities................................................................................175
USCALE the user-scaled value of 1 logit .................................................................................................175
W300 Output files in Winsteps 3.00 format ...............................................................................................176
WHEXACT Wilson-Hilferty exact normalization ........................................................................................176
XFILE analyzed response file...................................................................................................................177
XMLE consistent, almost unbiased, estimation........................................................................................178
XWIDE columns per response .................................................................................................................179
@Field name for location in label .............................................................................................................179
&amp;END end of control variables ..................................................................................................................180
&amp;INST start of control instructions .............................................................................................................180
The Iteration Screen ..................................................................................................................................180
Table heading ............................................................................................................................................182
Table 1.0, 1.2, 1.3, 1.10, 1.12 Distribution maps......................................................................................182
Table 1.1, 1.4 Distribution maps................................................................................................................184
Table 2 Multiple-choice distractor plot .......................................................................................................186
Table 2 Most probable, expected, cumulative, structure, average measures..........................................188
Table 3.1, 27.3, 28.3 Summaries of persons and items............................................................................191
Table 3.2 Summary of rating scale category structure..............................................................................194
Table 4.1, 5.1, 8.1, 9.1 Fit plots .................................................................................................................198
Table 5.2, 9.2 Fit plots ...............................................................................................................................199
Table 6.1, 10.1 Person and item statistics ................................................................................................199
Table 6.2, 10.2, 13.2, 14.2, 15.2, 17.2, 18.2, 19.2, 25.2 Person and item statistics .................................202
Table 6.5, 10.5 Most unexpected observations.........................................................................................203
Table 6.4, 10.4 Most misfitting response strings .......................................................................................203
Table 6.6, 10.6 Most unexpected response list.........................................................................................203
Table 7.1, 11.1 Misfitting responses..........................................................................................................204
Table 7.2, 17.3, 18.3, 19.3 Diagnostic KeyForms for Persons..................................................................204
Table 10.3, 13.3, 14.3, 15.3, 25.3 Item option/distracter frequencies .......................................................205
Table 12.2, 12.12 Item distribution maps ..................................................................................................206
Table 12.5 Item map with expected score zones ......................................................................................207
Table 12.6 Item map with 50% cumulative probabilities ...........................................................................208
Table 13.1, 14.1, 15.1, 25.1, 26.1 Item statistics.......................................................................................209
Table 16.3 Person distribution map...........................................................................................................212
Table 17.1, 18.1, 19.1 Person statistics ....................................................................................................213
Table 20 Complete score-to-measure table on test of all items................................................................215
v
235.
236.
237.
238.
239.
240.
241.
242.
243.
244.
245.
246.
247.
248.
249.
250.
251.
252.
253.
254.
255.
256.
257.
258.
259.
260.
261.
262.
263.
264.
265.
266.
267.
268.
269.
270.
271.
272.
273.
274.
275.
276.
277.
278.
279.
280.
281.
282.
283.
284.
285.
286.
287.
288.
289.
290.
291.
292.
293.
Table 20.3 Complete score-to-calibration table for tests based on whole sample....................................217
Table 21 Probability curves .......................................................................................................................218
Table 22.1 Sorted observed data matrix (scalogram) ...............................................................................219
Table 22.2 Guttman scalogram of zoned responses.................................................................................220
Table 22.3 Guttman scalogram of original codes......................................................................................220
Table 23.0, 24.0 Variance components scree plot for items or persons ...................................................221
Table 23.2, 24.2 Principal components plots of item or person loadings..................................................222
Table 23.3, 24.3 Principal components analysis/contrast of residuals......................................................224
Table 23.4, 24.4 Person or item contrast ..................................................................................................225
Table 23.99, 24.99 Largest residual correlations for items or persons .....................................................226
Table 27.1, 28.1 Subtotal summaries on one line .....................................................................................226
Table 27.2, 28.2 Subtotal measure histograms.........................................................................................227
Table 29 Empirical ICCs and option frequencies ......................................................................................228
Table 30 Differential item functioning DIF bias analysis............................................................................229
Table 31 Differential person functioning DPF bias/interaction analysis ....................................................231
Table 32 Control specifications .................................................................................................................233
Table 33 DIF/DPF/DGF interactions by class ...........................................................................................233
Table 34 Columnar statistical comparison and scatterplot........................................................................234
Table 0 The title page ................................................................................................................................235
Table 0.1 Analysis identification ................................................................................................................236
Table 0.2 Convergence report...................................................................................................................236
Table 0.3 Control file .................................................................................................................................237
Probability category curves .......................................................................................................................238
Empirical category curves .........................................................................................................................240
Expected score ICC...................................................................................................................................240
Empirical ICC.............................................................................................................................................240
Empirical randomness ...............................................................................................................................241
Cumulative probabilities ............................................................................................................................241
Item information function ...........................................................................................................................242
Category information function....................................................................................................................242
Conditional probability curves....................................................................................................................242
Test characteristic curve............................................................................................................................243
Test information function ...........................................................................................................................243
Test randomness .......................................................................................................................................243
Multiple item ICCs......................................................................................................................................244
Compare statistics .....................................................................................................................................244
Bubble charts.............................................................................................................................................246
Keyform plot...............................................................................................................................................248
DIF Plot......................................................................................................................................................249
DPF Plot ....................................................................................................................................................250
DIF-DPF Plot .............................................................................................................................................252
Advice to novice analysts ..........................................................................................................................253
Anchored estimation ..................................................................................................................................254
Average measures, distractors and rating scales .....................................................................................254
Automating file selection............................................................................................................................255
Batch mode example: Score Tables .........................................................................................................255
Biserial correlation .....................................................................................................................................256
Category boundaries and thresholds.........................................................................................................257
Column and classification selection and transformation ...........................................................................258
Comparing estimates with other Rasch software ......................................................................................260
Connection ambiguities .............................................................................................................................260
Convergence considerations .....................................................................................................................261
Decimal and percentage data....................................................................................................................263
Dependency and unidimensionality...........................................................................................................264
Dichotomous mean-square fit statistics.....................................................................................................266
DIF - DPF - bias - interactions concepts ...................................................................................................266
DIF item characteristic curves ...................................................................................................................269
Dimensionality: contrasts &amp; variances.......................................................................................................272
Dimensionality: when is a test multidimensional? .....................................................................................277
vi
294.
295.
296.
297.
298.
299.
300.
301.
302.
303.
304.
305.
306.
307.
308.
309.
310.
311.
312.
313.
314.
315.
316.
317.
318.
319.
320.
321.
322.
323.
324.
325.
326.
327.
328.
329.
330.
331.
332.
333.
334.
335.
336.
337.
338.
339.
340.
341.
342.
343.
344.
345.
346.
347.
348.
349.
350.
351.
352.
Disjoint strings of responses......................................................................................................................278
Disordered rating categories .....................................................................................................................278
Displacement measures ............................................................................................................................279
Edit Initial Settings (Winsteps.ini) file.........................................................................................................281
Equating and linking tests..........................................................................................................................283
Estimation bias correction - warnings........................................................................................................287
Estimation methods: JMLE, PROX, XMLE................................................................................................289
Exact Match: OBS% and EXP%................................................................................................................293
Exporting Tables to EXCEL.......................................................................................................................294
Extra Specifications prompt.......................................................................................................................294
Extreme scores: what happens .................................................................................................................295
Global fit statistics......................................................................................................................................295
GradeMap interface ...................................................................................................................................296
Guttman patterns .......................................................................................................................................297
Half-rounding .............................................................................................................................................298
How big an analysis can I do?...................................................................................................................299
How long will an analysis take?.................................................................................................................299
Information - item and test.........................................................................................................................299
Item difficulty: definition .............................................................................................................................300
Item discrimination or slope estimation .....................................................................................................300
Iterations - PROX &amp; JMLE.........................................................................................................................302
Local Dependence.....................................................................................................................................304
Logit and probit ..........................................................................................................................................304
Misfit diagnosis: infit outfit mean-square standardized..............................................................................306
Missing data...............................................................................................................................................310
Mixed and Saltus models ..........................................................................................................................310
Multiple t-tests............................................................................................................................................311
Null or unobserved categories...................................................................................................................312
One observation per respondent ...............................................................................................................312
Order of elements in Control file................................................................................................................314
Partial Credit model ...................................................................................................................................314
Plausible values.........................................................................................................................................315
Plotting with EXCEL ..................................................................................................................................315
Poisson counts ..........................................................................................................................................316
Point-measure correlation .........................................................................................................................317
Polytomous mean-square fit statistics .......................................................................................................317
Quality-control misfit selection criteria .......................................................................................................318
Rank order data .........................................................................................................................................319
Rectangular copying ..................................................................................................................................319
Reliability and separation of measures .....................................................................................................320
Right-click functions...................................................................................................................................321
Rules for assigning values to control variables (key-words) .....................................................................322
Shortcut Keys ............................................................................................................................................324
Specifying how data are to be recoded .....................................................................................................324
Standard errors: model and real................................................................................................................324
Starting Winsteps from the DOS prompt ...................................................................................................325
Subtest scoring ..........................................................................................................................................326
Transposing the data matrix ......................................................................................................................326
Unobserved and dropped categories ........................................................................................................327
User-friendly rescaling ...............................................................................................................................328
Using a word processor or text editor........................................................................................................330
Weighting items and persons ....................................................................................................................330
Winsteps: history and steps.......................................................................................................................331
Wordpad or text editor ...............................................................................................................................332
Data vanishes ............................................................................................................................................333
Display too big ...........................................................................................................................................333
File misread problems ...............................................................................................................................333
If Winsteps does not work .........................................................................................................................335
Initialization fails.........................................................................................................................................335
vii
353.
Not enough disk space ..............................................................................................................................335
354.
Not enough memory ..................................................................................................................................336
355.
Plotting problems .......................................................................................................................................336
356.
Tab-delimited file problems .......................................................................................................................336
357.
Winsteps problems and error codes..........................................................................................................336
358.
Winsteps SPSS error codes ......................................................................................................................337
Control Variable Index ............................................................................................................................................338
viii
2.
Control Variable Index
Commonly-used Control Variables
CFILE=
scored category label file
CLFILE= codes label file
CODES= valid &amp; missing data codes
CSV=
comma-separated values in output files
DATA=
name of data file
EDFILE= edit data file
GROUPS= assigns items to rating scale or partial credit groupings (same as ISGROUPS=)
IAFILE=
item anchor file
IDFILE=
item deletion file
IFILE=
item output file
IREFER= identifying items for recoding
ISELECT= item selection criterion
ISGROUPS= assigns items to rating scale or partial credit groupings (same as GROUPS=)
ISORT=
sort column in item label
ISUBTOTAL= subtotal items by specified columns
ITEM=
title for item labels
ITEM1=
column number of first response
IVALUEx= recoding for items
KEYn=
scoring key
MRANGE= half-range of measures on plots
NAME1= first column of person label
NAMLEN= length of person label
NEWSCORE= recoding values
NI=
number of items
PAFILE= person anchor file
PDFILE= person deletion file
PERSON= title for person labels
PFILE=
person output file
PSELECT= person selection criterion
PSORT= sort column in person label
PSUBTOTAL= subtotal persons by specified columns
RESCORE=
response recoding
SAFILE= structure anchor file
SFILE=
structure output file
STKEEP= keep non-observed categories in structure
TITLE=
title for output listing
UANCHOR=
anchor values in user-scaled units
UASCALE= anchor user-scale value of 1 logit
UDECIMALS= number of decimal places reported
UIMEAN= reported user-set mean of item measures
UMEAN= reported user-set mean of item measures
UPMEAN= reported user-set mean of person measures
USCALE= reported user-scaled value for 1 logit
XFILE=
analyzed response file
XWIDE=
columns per response
Special-purpose Control Variables
@FIELD= user-defined field locations
ALPHANUM= alphabetic numbering to exceed 9 with XWIDE=1
ASCII=
output only ASCII characters
ASYMPTOTE= report estimates of upper and lower asymptotes
BATCH= set batch mode
BYITEM= show empirical curves for items
CATREF= reference category: Table 2
1
CHART= graphical plots of measures
CONVERGE= select convergence criteria
CURVES= probability curve selection: Tables 21, 2
CUTHI=
cut off responses with high probability
CUTLO= cut off responses with low probability
DELIMITER= delimiter of input response data fields
DIF=
person label columns for DIF
DISCRIM= display item discrimination estimates
DISFILE= category/distractor/response option count file
DISTRT= output category/distractor/option counts
DPF=
item label columns for DPF
EQFILE= code equivalences
EXTRSCORE= extreme score adjustment
FITHIGH= lower bar in charts
FITI=
item misfit criterion
FITLOW= lower bar in charts
FITP=
person misfit criterion
FORMAT= reformat data
FORMFD= form-feed character
FRANGE= half-range of fit statistics on plots
G0ZONE= percent of 0's within 0-zone among which all 1's are turned to 0's
G1ZONE= percent of 1's within 1-zone among which all 0's are turned to 1's
GRFILE= graphing file for probability curves
GRPFROM=
location of ISGROUPS=
GUFILE= Guttmanized file of responses
HEADER= display or suppress Sub-Table Headings after the first
HIADJ=
correction for top categories
HLINES= heading lines in output files
IANCHQU= anchor items interactively
ICORFILE= inter-item residual correlations
IDELETE= one-line item deletion list
IDELQU= delete items interactively
IDROPEXTR= remove items with extreme scores
ILFILE=
item label file
IMAP=
item label for item maps
INUMB=
label items by sequence number
IPMATRIX= response matrix (Output Files menu only)
ISFILE=
item-structure output file
ITLEN=
maximum length of item label
IWEIGHT= item (variable) weighting
KEYFORM=
KeyForm skeleton for Excel plot
KEYFROM=
location of KEYn=
KEYSCR= reassign scoring keys
LCONV= logit change at convergence
LINLEN= length of printed lines
LOCAL=
locally restandardize fit statistics
LOGFILE= accumulates control files
LOWADJ= correction for bottom categories
MAKEKEY=
construct an MCQ scoring key
MAXPAGE=
maximum number of lines per page
MFORMS= reformatting input records &amp; multiple forms
MHSLICE= size of Mantel or Mantel-Haenszel slice
MISSCORE= scored value of missing data: not your missing-data code
MJMLE= maximum number of JMLE iterations
MNSQ=
show mean-square or t standardized fit
MODELS= assigns model types to items
MODFROM= location of MODELS=
MPROX= maximum number of PROX iterations
MUCON= maximum number of JMLE iterations
2
NAMLMP= label length on maps
NORMAL= normal distribution for standardizing fit
OSORT= category/option/distractor sorted by score, count or measure
OUTFIT= sort misfits on outfit or infit
PAIRED= correction for paired comparison data
anchor persons interactively
PANCHQU=
PCORFIL= inter-person residual correlations
PDELETE= one-line person deletion list
PDELQU= delete persons interactively
PDROPEXTR= remove persons with extreme scores
PMAP=
person label for person maps
PRCOMP= residual type for principal components/contrast analysis
PTBISERIAL= report point-biserial correlations
PVALUE= report item p-values
PWEIGHT= person (case) weighting
QUOTED= quote-marks around labels
RCONV= score residual at convergence
REALSE= inflate S.E. of measures for misfit
RESFRM= location of RESCORE=
RFILE=
scored response output file
SAITEM= multiple ISGROUP= format in SFILE= &amp; SAFILE=
SANCHQU=
anchor structure interactively
SCOREFILE= score-to-measure file
SDELQU= delete structure interactively
SDFILE= structure deletion file
SEPARATOR= delimiter of input response data fields
SIFILE=
simulated data file
SPFILE= supplemental control file
STBIAS= correct for estimation bias
STEPT3= structure summary in Table 3 or 21
T1I#=
items per # in Table 1
T1P#=
persons per # in Table 1
TABLES= output table selection
TARGET= information-weighted estimation
TFILE=
input file of table numbers to be output
TOTALSCORE=
show total observed score and count
UCOUNT= most unexpected responses in Tables 6 and 10.
WHEXACT=
Wilson-Hilferty exact normalization
W300=
produce IFILE= and PFILE= in 3.00 format
XMLE=
consistent, almost unbiased, estimation (experimental)
3.
Control Variable Index by function
Data file layout:
DATA=
name of data file
DELIMITER= delimiter of input response data fields
FORMAT= reformat data
ILFILE=
item label file
ITEM1=
column number of first response
ITLEN=
maximum length of item label
INUMB=
label items by sequence number
MFORMS= reformatting input records &amp; multiple forms
NAME1= first column of person label
NAMLEN= length of person label
NI=
number of items
SEPARATOR= delimiter of input response data fields
XWIDE=
columns per response
@FIELD= user-defined field locations
3
Data selection and recoding:
ALPHANUM= alphabetic numbering to exceed 9 with XWIDE=1
CODES= valid &amp; missing data codes
CUTHI=
cut off responses with high probability
CUTLO= cut off responses with low probability
EDFILE= edit data file
IREFER= identifying items for recoding
IVALUEx= recoding for items
IWEIGHT= item (variable) weighting
KEYn=
scoring key
KEYFROM=
location of KEYn=
KEYSCR= reassign scoring keys
MAKEKEY=
construct an MCQ scoring key
MISSCORE= scored value of missing data: not your missing-data code
NEWSCORE= recoding values
PWEIGHT= person (case) weighting
RESCORE=
response recoding
RESFRM= location of RESCORE=
Items: deleting, anchoring and selecting:
IAFILE=
item anchor file
IANCHQU= anchor items interactively
IDELETE= one-line item deletion list
IDELQU= delete items interactively
IDFILE=
item deletion file
IDROPEXTR= remove items with extreme scores
ISELECT= item selection criterion
Person: deleting, anchoring and selecting:
PAFILE= person anchor file
PANCHQU=
anchor persons interactively
PDELETE= one-line person deletion list
PDELQU= delete persons interactively
PDFILE= person deletion file
PDROPEXTR= remove persons with extreme scores
PSELECT= person selection criterion
Rating scales, partial credit items and polytomour response structures:
GROUPS= assigns items to rating scale or partial credit groupings (same as ISGROUPS=)
GRPFROM=
location of ISGROUPS=
ISGROUPS= assigns items to rating scale or partial credit groupings (same as GROUPS=)
MODELS= assigns model types to items
MODFROM= location of MODELS=
STKEEP= keep non-observed categories in structure
Category structure: anchoring, labeling, deleting:
CFILE=
scored category label file
CLFILE= codes label file
SAFILE= structure anchor file
SAITEM= multiple ISGROUP= format in SFILE= &amp; SAFILE=
SANCHQU=
anchor structure interactively
SDELQU= delete structure interactively
SDFILE= structure deletion file
Measure origin, anchoring and user-scaling:
UANCHOR=
anchor values in user-scaled units
UASCALE= anchor user-scaled value for 1 logit
UDECIMALS= number of decimal places reported
UIMEAN= reported user-set mean of item measures
4
UMEAN= reported user-set mean of item measures
UPMEAN= reported user-set mean of person measures
USCALE= reported user-scaled value for 1 logit
Output table selection and format:
output only ASCII characters
ASCII=
FORMFD= form-feed character
HEADER= display or suppress Sub-Table Headings after the first
ITEM=
title for item labels
LINLEN= length of printed lines
MAXPAGE=
maximum number of lines per page
PERSON= title for person labels
TABLES= output table selection
TFILE=
input file of table numbers to be output
TITLE=
title for output listing
TOTALSCORE=
show total observed score and count
Output tables, files and graphs: specific controls
ASYMPTOTE= report estimates of upper and lower asymptotes
BYITEM= show empirical curves for items
CATREF= reference category: Table 2
CHART= graphical plots of measures
CURVES= probability curve selection: Tables 21, 2
DIF=
person label columns for DIF
DISCRIM= display item discrimination estimates
DISTRT= output category/distractor/option counts
DPF=
item label columns for DPF
EQFILE= code equivalences
FITHIGH= lower bar in charts
FITI=
item misfit criterion
FITLOW= lower bar in charts
FITP=
person misfit criterion
FRANGE= half-range of fit statistics on plots
HIADJ=
correction for top categories
IMAP=
item label for item maps
ISORT=
sort column in item label
ISUBTOTAL= subtotal items by specified columns
LOWADJ= correction for bottom categories
MNSQ=
show mean-square or t standardized fit
MRANGE= half-range of measures on plots
NAMLMP= label length on maps
OSORT= category/option/distractor sorted by score, count or measure
OUTFIT= sort misfits on outfit or infit
PMAP=
person label for person maps
PSORT= sort column in person label
PSUBTOTAL= subtotal persons by specified columns
PVALUE= report item p-values
STEPT3= structure summary in Table 3 or 21
T1I#=
items per # in Table 1
T1P#=
persons per # in Table 1
UCOUNT= most unexpected responses in Tables 6 and 10.
Output files:
DISFILE= category/distractor/response option count file
GRFILE= graphing file for probability curves
GUFILE= Guttmanized file of responses
ICORFILE= inter-item residual correlations
IFILE=
item output file
IPMATRIX= response matrix (Output Files menu only)
5
ISFILE=
item-structure output file
KEYFORM=
KeyForm skeleton for Excel plot
LOCAL=
locally restandardize fit statistics
LOGFILE= accumulates control files
PCORFIL= inter-person residual correlations
person output file
PFILE=
RFILE=
scored response output file
SFILE=
structure output file
SCOREFILE= score-to-measure file
SIFILE=
simulated data file
XFILE=
analyzed response file
Output file format control:
CSV=
comma-separated values in output files
G0ZONE= percent of 0's within 0-zone among which all 1's are turned to 0's
G1ZONE= percent of 1's within 1-zone among which all 0's are turned to 1's
HLINES= heading lines in output files
QUOTED= quote-marks around labels
W300=
produce IFILE= and PFILE= in 3.00 format
Estimation, operation and convergence control:
CONVERGE= select convergence criteria
EXTRSCORE= extreme score adjustment
LCONV= logit change at convergence
LOCAL=
locally restandardize fit statistics
MJMLE= maximum number of JMLE iterations
MPROX= maximum number of PROX iterations
MUCON= maximum number of JMLE iterations
NORMAL= normal distribution for standardizing fit
PAIRED= correction for paired comparison data
PRCOMP= residual type for principal components/contrast analysis
PTBISERIAL= compute point-biserial correlations
RCONV= score residual at convergence
REALSE= inflate S.E. of measures for misfit
STBIAS= correct for estimation bias
TARGET= information-weighted estimation
WHEXACT=
Wilson-Hilferty exact normalization
XMLE=
consistent, almost unbiased, estimation (experimental)
Program operation:
BATCH= set batch mode
MAKEKEY=
construct an MCQ scoring key
SPFILE= supplemental control file
&amp;END
end of control variable list
&amp;INST
start of control variable list (ignored)
END LABELS end of item label list
4.
Control variables from Specification menu
These control variables can be changed using the Specification pull-down menu after measures have
been estimated.
They do not alter the estimates from the main analysis. They only change how it is reported.
@FIELD= user-defined field locations
ASCII=
output only ASCII characters
ASYMPTOTE= report estimates of upper and lower asymptotes
BYITEM= show empirical curves for items
CATREF= reference category: Table 2
6
CFILE=
scored category label file (file name only, blank deletes category labels)
CHART= graphical plots of measures
CLFILE= codes label file (file name only, blank deletes code labels)
CSV=
comma-separated values in output files
CURVES= probability curve selection: Tables 21, 2
DISCRIM= display item discrimination estimates
DISTRT= output category/distractor/option counts
FITHIGH= lower bar in charts
FITI=
item misfit criterion
FITLOW= lower bar in charts
FITP=
person misfit criterion
FORMFD= form-feed character
FRANGE= half-range of fit statistics on plots
G0ZONE= percent of 0's within 0-zone among which all 1's are turned to 0's
G1ZONE= percent of 1's within 1-zone among which all 0's are turned to 1's
HEADER= display or suppress Sub-Table Headings after the first
HIADJ=
correction for top categories
HLINES= heading lines in output files
IDELETE= one-line item deletion list
IDFILE=
item deletion file (file name only: blank resets temporary item deletions)
ILFILE=
item label file (file name only: blank not allowed)
IMAP=
item label for item maps
ISELECT= item selection criterion
ISORT=
sort column in item label
ITEM=
title for item labels
LINLEN= length of printed lines
LOWADJ= correction for bottom categories
MAXPAGE=
maximum number of lines per page
MHSLICE= size of Mantel or Mantel-Haenszel slice
MNSQ=
show mean-square or t standardized fit
MRANGE= half-range of measures on plots
NAMLMP= label length on maps
OSORT= category/option/distractor sorted by score, count or measure
OUTFIT= sort misfits on outfit or infit
PDELETE= one-line person deletion list (blank resets temporary person deletions)
PDFILE= person deletion file (file name only: blank resets temporary person deletions)
PERSON= title for person labels
PMAP=
person label for person maps
PSELECT= person selection criterion
PSORT= sort column in person label
PVALUE= report item p-values
PWEIGHT= person (case) weighting
QUOTED= quote-marks around output labels
STEPT3= structure summary in Table 3 or 21
T1I#=
items per # in Table 1
T1P#=
persons per # in Table 1
TITLE=
title for output listing
TOTALSCORE=
show total observed score and count
UDECIMALS= number of decimal places reported
UIMEAN= reported user-set mean of item measures
UMEAN= reported user-set mean of item measures
UPMEAN= reported user-set mean of person measures
USCALE= reported user-scaled value for 1 logit
W300=
produce IFILE= and PFILE= in 3.00 format
Control variables that can be set with other pull-down menus
DIF=
person label columns for DIF (from Output Tables menu)
DISFILE= category/distractor/response option count file (from Output Files menu)
DPF=
item label columns for DPF (from Output Tables menu)
7
GRFILE= graphing file for probability curves (from Output Files menu)
GUFILE= Guttmanized file of responses (from Output Files menu)
ICORFILE= inter-item residual correlations (from Output Files menu)
IFILE=
item output file (from Output Files menu)
IPMATRIX= response matrix (from Output Files menu)
item-structure output file (from Output Files menu)
ISFILE=
ISUBTOTAL= subtotal items by specified columns (from Output Tables menu)
KEYFORM=
KeyForm skeleton for Excel plot (from Plots menu)
PCORFIL= inter-person residual correlations (from Output Files menu)
PFILE=
person output file (from Output Files menu)
PSUBTOTAL= subtotal persons by specified columns (from Output Tables menu)
scored response output file (from Output Files menu)
RFILE=
SCOREFILE= score-to-measure file (from Output Files menu)
SFILE=
structure output file (from Output Files menu)
SIFILE=
simulated data file (from Output Files menu)
XFILE=
analyzed response file (from Output Files menu)
5.
Output Table Index
Table Description
1 Maps of person and item measures. Show Rasch measures.
1.0 One page map with names.
1.1 Map of distributions - persons and items
1.2 Item labels with person distribution (squeezed onto one page)
1.3 Person labels with item distribution (squeezed onto one page)
1.4 Rating scale or partial credit map of distributions: persons with items at high, mean, low
1.10 One page map with person names by measure, item names by easiness.
1.12 Item labels, by easiness, with person distribution (squeezed onto one page)
2
Measures and responses plots. Response categories for each item, listed in measure order, plotted
against person measures, shown as modal categories, expected values and cumulative probabilities.
Table 2 for multiple-choice items.
By observed categories
2.6 Observed average measures of persons (empirical averages)
By scored categories (illustrated by an observed category code for each score)
2.1 Modal categories (most probable)
2.2 Mean categories (average or expected: Rasch-half-point thresholds)
2.3 Median categories (cumulative probabilities: Rasch-Thurstone thresholds)
2.4 Structure calibrations (Rasch model parameters: rating scale, partial credit, &quot;restricted&quot;, &quot;unrestricted&quot;:
Rasch-Andrich thresholds)
2.5 Observed average measures of persons (empirical averages)
2.7 Expected average measures of persons
By category scores (if category scores differ from category codes in the data):
2.11 Modal categories (most probable)
2.12 Mean categories (average or expected: Rasch-half-point thresholds)
2.13 Median categories (cumulative probabilities: Rasch-Thurstone thresholds)
2.14 Structure calibrations (Rasch model parameters: rating scale, partial credit, &quot;restricted&quot;,
&quot;unrestricted&quot;)
2.15 Observed average measures of persons (empirical averages)
2.16 Observed average measures of persons (empirical averages)
2.17 Expected average measures of persons
3 Summary statistics. Person, item, and category measures and fit statistics.
3.1 Summaries of person and items: means, S.D.s, separation, reliability.
3.2 Summary of rating categories and probability curves.
8
4 Person infit plot. Person infits plotted against person measures.
4.1 Person infit vs. person measure plot.
5 Person outfit plot. Person outfits plotted against person measures.
5.1 Person outfit vs. person measure plot.
5.2 Person infit vs. person outfit plot.
6 Person statistics - fit order. Misfitting person list. Scalogram of unexpected responses.
6.1 Table of person measures in descending order of misfit. (Specify FITP=0 to list all persons)
6.2 Chart of person measures, infit mean-squares and outfit mean-squares. (Chart=Yes)
6.3 (Not produced for persons)
6.4 Scalogram of most misfitting person response strings.
6.5 Scalogram of most unexpected responses.
6.6 Most unexpected responses list.
7 Misfitting Persons. Lists response details for persons with t standardized fit greater than FITP=.
7.1 Response strings for most misfitting persons.
7.2-... KeyForms of responses of misfitting persons.
8 Item infit plot. Item infits plotted against item calibrations.
8.1 Item infit vs. item measure plot.
9 Item outfit plot. Item outfits plotted against item calibrations.
9.1 Item outfit vs. item measure plot.
9.2 Item infit vs. item outfit plot
10 Item statistics - fit order. Misfitting item list with option counts. Scalogram of unexpected responses.
10.1 Table of item measures in descending order of misfit. (Specify FITI=0 to list all persons)
10.2 Chart of item measures, infit mean-squares and outfit mean-squares. (Chart=Yes)
10.3 Item response-structure categories/options/distractors: counts and average measures.
(Distractors=Yes)
10.4 Scalogram of most misfitting item response strings.
10.5 Scalogram of most unexpected responses.
10.6 Most unexpected responses list.
11 Misfitting Items. Response details for items with t standardized fit greater than FITI=.
11.1 Response strings for most misfitting items.
12 Item distribution map. Horizontal histogram of item distribution with abbreviated item names.
12.2 Item labels with person distribution (same as 1.2)
12.5 Item labels with expected score zones
12.6 Item labels with 50% cumulative probabilities
12.12 Item labels, by easiness, with person distribution (same as 1.12)
13 Item statistics - measure order list and graph with category/option/distractor counts.
13.1 Table of items in descending measure order.
13.2 Chart of item measures, infit mean-squares and outfit mean-squares. (Chart=Yes)
13.3 Item response-structure categories/options/distractors: counts and average measures.
(Distractors=Yes)
14 Item statistics - entry order list and graph with category/option/distractor counts.
14.1 Table of items in entry number (sequence) order.
14.2 Chart of item measures, infit mean-squares and outfit mean-squares. (Chart=Yes)
14.3 Item response-structure categories/options/distractors: counts and average measures.
(Distractors=Yes)
15 Item statistics - alphabetical order list and graph with category/option/distractor counts.
15.1 Table of item measures in alphabetical order by label. (Specify ISORT= for sort column)
15.2 Chart of item measures, infit mean-squares and outfit mean-squares. (Chart=Yes)
9
15.3 Item response-structure categories/options/distractors: counts and average measures.
(Distractors=Yes)
16 Person distribution map. Horizontal histogram of person distribution, with abbreviated person-ids.
16.3 Person labels with item distribution (same as 1.3)
17 Person statistics - measure order list and chart.
17.1 Table of persons in descending measure order.
17.2 Chart of person measures, infit mean-squares and outfit mean-squares. (Chart=Yes)
17.3-... KeyForms of responses of persons.
18 Person statistics - entry order list and chart.
18.1 Table of persons in entry number (sequence) order.
18.2 Chart of person measures, infit mean-squares and outfit mean-squares. (Chart=Yes)
18.3-... KeyForms of responses of persons.
19 Person statistics - alphabetical order list and chart.
19.1 Table of person measures in alphabetical order by label. (Specify PSORT= for sort column)
19.2 Chart of person measures, infit mean-squares and outfit mean-squares. (Chart=Yes)
19.3-... KeyForms of responses of persons.
20 Measures for all scores on a test of all calibrated items, with percentiles.
20.1 Table of person measures for every score on complete test. (Specify ISELECT= for subtests).
20.2 Table of measures for every score, with sample percentiles and norm-referenced measures.
20.3 Table of item difficulty measures (calibrations) for every score (p-value) by complete sample.
21
Category probability curves. Category probabilities plotted against the difference between person and
item measures, then the expected score and cumulative probability and expected score ogives.
21.1 Category probability curves (modes, structure calibrations).
21.2 Expected score ogive (means. model Item Characteristic Curve).
21.3 Cumulative category probability curves (medians, shows 50% cumulative probabilities).
22 Sorted observations. Data sorted by person and item measures into scalogram patterns.
22.1 Guttman scalogram of sorted scored responses.
22.2 Guttman scalogram showing out-of-place responses.
22.3 Guttman scalogram showing original responses.
23 Item principal components/contrasts. Identifies structure in response residuals (BIGSTEPS Table: 10.3)
23.0 Scree plot of variance components.
23.2 Plot of loadings on first contrast in residuals vs. item measures.
23.3 Items in contrast loading order.
23.4 Persons exhibiting contrast.
23.5 Items in measure order.
23.6 Items in entry order.
23.7 etc. Subsequent contrasts.
23.99 Tables of items with highly correlated residuals. (Reported as last subtable in Table 23)
24 Person principal components/contrasts. Identifies structure in residuals (not in BIGSTEPS)
24.0 Scree plot of variance components.
24.2 Plot of loadings on first contrast in residuals vs. person measures.
24.3 Persons in contrast loading order.
24.4 Items exhibiting contrast.
24.5 Persons in measure order.
24.6 Persons in entry order.
24.7 etc. Subsequent contrasts.
24.99 Tables of persons with highly correlated residuals. (Reported as last subtable in Table 24)
25 Item statistics - displacement order list and graph with category/option/distractor counts.
25.1 Table of items in descending displacement order.
10
25.2 Chart of item measures, infit mean-squares and outfit mean-squares. (Chart=Yes)
25.3 Item response-structure categories/options/distractors: counts and average measures.
(Distractors=Yes)
26 Item statistics - correlation order list and graph with category/option/distractor counts.
26.1 Table of items in ascending correlation order (Point-biserial, if PTBIS=Yes, else Point-measure).
26.2 Chart of item measures, infit mean-squares and outfit mean-squares.
26.3 Item response-structure categories/options/distractors: counts and average measures.
27 Item subtotals.
27.1 Measure sub-totals, controlled by ISUBTOT=
27.2 Measure sub-totals histograms, controlled by ISUBTOT=
27.3 Measure sub-totals summary statistics, controlled by ISUBTOT=
28 Person subtotals.
28.1 Measure sub-totals, controlled by PSUBTOT=
28.2 Measure sub-totals histograms, controlled by PSUBTOT=
28.3 Measure sub-totals summary statistics, controlled by PSUBTOT=
29 Empirical item character curves and response frequency plots.
29.1 Empirical and model ICCs (see also Graph Menu)
29.2 Use of response categories by measure
30 Differential Item Function across Person classifications
30.1 DIF report (paired), controlled by DIF=
30.2 DIF report (measure list: person class within item)
30.3 DIF report (measure list: item within person class)
31 Differential Person Function across Item classifications
31.1 DPF report (paired), controlled by DPF=
31.2 DPF report (measure list: item class within person)
31.3 DPF report (measure list: person within item class)
32
Control Variable Listing of the current settings of all Winsteps control variables - appears on the Output
Files pull-down menu.
33 Differential Item Classification vs. Person Classification interactions/biases
33.1 DIF report (paired person classifications on each item classification), controlled by DIF= and DPF=
33.2 DIF report (measure list of item classification differential difficulties)
33.3 DPF report (paired item classifications on each person classification)
33.4 DPF report (measure list of person classification differential abilities)
0
6.
Control Variables and Convergence report. Lists the control variables and shows estimation
convergence. (Only appears at end of Output Report File).
0.0 Title page
0.1 Analysis identification
0.2 Convergence table
0.3 Control file
Output Graphs and Plots Index
Graphs - from the Graphs pull-down menu
Category Probability Curves
Expected Score ICC (Item Characteristic Curve)
Cumulative Probabilities
Information Function
Category Information
Conditional Probability Curves
Test Characteristic Curve
11
Test Information
Plots - from the Plots pull-down menu
Compare statistics generates an Excel-based cross-plot of the values in Table 34.
Bubble chart generates a Bond &amp; Fox-style bubble chart.
Keyform plot - Horizontal generates a horizontal keyform layout.
Keyform plot - Vertical generates a vertical keyformlayout.
DIF plot - the DIF values in Table 30.
DPF plot - the DPF values in Table 31.
DIF+DPF plot - the DIF+DPF values in Table 33.
7.
Output Files Index
These can be accessed from the Output Files pull-down menu
Output Files - in control file or pull-down menu
DISFILE= category/distractor/response option count file
GRFILE= graphing file for probability curves
GUFILE= Guttmanized response file
ICORFILE= inter-item residual correlations
IFILE=
item output file (use for anchoring)
ISFILE=
item-structure output file (do not use for anchoring)
PFILE=
person output file (useful or anchoring)
PCORFIL= inter-person residual correlations
RFILE=
scored response output file
SCOREFILE= score-to-measure file
SFILE=
structure output file (use for anchoring)
SIFILE=
simulated data file
XFILE=
analyzed response file
Output files - only from pull-down menu
Control variable file control variable listing (same as Table 32)
IPMATRIX= matrix of response-level data
TRANSPOSE= transposed (rows-columns) control file
GradeMap GradeMap model specification and student data files
To control output file formatting in the control file:
CSV=
fixed fields, tab-delimited or comma-separated values in output files
HLINES= heading lines written to output files
8.
Rasch analysis and WINSTEPS
Winsteps is Windows-based software which assists with many applications of the Rasch model, particularly in the
areas of educational testing, attitude surveys and rating scale analysis. There is more information at:
www.winsteps.com
Rasch analysis is a method for obtaining objective, fundamental, linear measures (qualified by standard errors
and quality-control fit statistics) from stochastic observations of ordered category responses. Georg Rasch, a
Danish mathematician, formulated this approach in 1953 to analyze responses to a series of reading tests
(Rasch G, Probabilistic Models for Some Intelligence and Attainment Tests, Chicago: MESA Press, 1992, with
instructive Foreword and Afterword by B.D. Wright). Rasch is pronounced like the English word rash in Danish,
and like the English sounds raa-sch in German. The German pronunciation, raa-sch, is used to avoid
misunderstandings.
The person and item total raw scores are used to estimate linear measures. Under Rasch model conditions, these
measures are item-free (item-distribution-free) and person-free (person-distribution-free). So that the measures
are statistically equivalent for the items regardless of which persons (from the same population) are analyzed, and
for the items regardless of which items (from the same population) are analyzed. Analysis of the data at the
response-level indicates to what extent these ideals are realized within any particular data set.
12
The Rasch models implemented in Winsteps include the Georg Rasch dichotomous, Andrich &quot;rating scale&quot;,
Masters &quot;partial credit&quot;, Bradley-Terry &quot;paired comparison&quot;, Glas &quot;success model&quot;, Linacre &quot;failure model&quot; and
most combinations of these models. Other models such as binomial trials and Poisson can also be analyzed by
anchoring (fixing) the response structure to accord with the response model. (If you have a particular need,
please let us know as Winsteps is continually being enhanced.)
The estimation method is JMLE, &quot;Joint Maximum Likelihood Estimation&quot;, with initial starting values provided by
PROX, &quot;Normal Approximation Algorithm&quot;.
The Rasch Family of Models
The necessary and sufficient transformation of ordered qualitative observations into linear measures is a Rasch
model. Rasch models are logit-linear models, which can also be expressed as log-linear models. Typical Rasch
models operationalized with Winsteps are:
The dichotomous model:
loge(Pni1 / Pni0 ) = Bn - Di
The polytomous &quot;Rating Scale&quot; model:
log(Pnij/ Pni(j-1) ) = Bn - Di - Fj
The polytomous &quot;Partial Credit&quot; model:
log(Pnij/ Pni(j-1) ) = Bn - Di - Fij = Bn - Dij
The polytomous &quot;Grouped response-structure&quot; model:
log(Pnij/ Pni(j-1) ) = Bn - Dig - Fgj
where
Pnij is the probability that person n encountering item i is observed in category j,
Bn is the &quot;ability&quot; measure of person n,
Di is the &quot;difficulty&quot; measure of item i, the point where the highest and lowest categories of the item are equally
probable.
Fj is the &quot;calibration&quot; measure of category j relative to category j-1, the point where categories j-1 and j are
equally probable relative to the measure of the item.
Also models with the form of &quot;Continuation Ratio&quot; models, such as the &quot;Success&quot; model and the &quot;Failure&quot;
model.
For methods of estimation, see RSA, pp. 72-77.
Work-flow with Winsteps
Control + Data file or Control file and Data file(s)
↓
User-interaction →
↑
Winsteps ←
↓
Anchor Files
↑
Report Output File + Output Tables + Graphs + Output Files
↓
Word Processor, Spreadsheet, Statistical Package
↓
Actions
WINSTEPS is designed to construct Rasch measurement from the responses of a set of persons to a set of
items. Responses may be recorded as letters or integers and each recorded response may be of one or two
characters. Alphanumeric characters, not designated as legitimate responses, are treated as missing data. This
causes these observations, but not the corresponding persons or items, to be omitted from the analysis. The
13
responses to an item may be dichotomous (&quot;right&quot;/&quot;wrong&quot;, &quot;yes&quot;/&quot;no&quot;), or may be on a rating scale (&quot;good&quot;/
&quot;better&quot;/&quot;best&quot;, &quot;disagree&quot;/&quot;neutral&quot;/&quot;agree&quot;), or may have &quot;partial credit&quot; or other hierarchical structures. The items
may all be grouped together as sharing the one response structure, or may be grouped into subsets of one or
more items which share the same response structure.
WINSTEPS begins with a central estimate for each person measure, item calibration and response-structure
calibration, unless pre-determined, &quot;anchor&quot; values are provided by the analyst. An iterative version of the PROX
algorithm is used reach a rough convergence to the observed data pattern. The JMLE method is then iterated to
obtain more exact estimates, standard errors and fit statistics.
Output consists of a variety of useful plots, graphs and tables suitable for import into written reports. The statistics
can also be written to data files for import into other software. Measures are reported in Logits (log-odds units)
unless user-rescaled. Fit statistics are reported as mean-square residuals, which have approximate chi-square
distributions. These are also reported t standardized, N(0,1).
9.
References
Please cite the current Winsteps computer program as:
Linacre, J. M. (2006) WINSTEPS Rasch measurement computer program. Chicago: Winsteps.com
• RSA means Wright B.D. &amp; Masters G.N. Rating Scale Analysis, Chicago: MESA Press, 1982, especially p. 100:
www.rasch.org/books.htm
• BTD means Wright B.D. &amp; Stone M.H. Best Test Design, Chicago: MESA Press, 1979:
www.rasch.org/books.htm
Other recommended sources:
• Rasch Measurement Transactions: www.rasch.org/rmt/
•Journal of Applied Measurement: www.jampress.org
•&quot;Applying the Rasch Model: Fundamental Measurement in the Human Sciences&quot;, by Trevor G. Bond &amp; Christine
M. Fox, 2001. Mahwah NJ: Lawrence Erlbaum Assoc. 0-8058-3476-1. Authors' website. Using Winsteps: Bond &amp;
Fox examples
&quot;Introduction to Rasch Measurement&quot;, Everett V. Smith, Jr. &amp; Richard M. Smith (Eds.) JAM Press, 2004
www.jampress.org
10.
About the Users' guide
You don't need to know about every WINSTEPS option in order to use the program successfully. Glance through
the examples and find one similar to yours. Adapt the example to match your requirements. Then &quot;fine tune&quot; your
analysis as you become familiar with further options.
Most of this Guide is in proportionately-spaced type.
When it is important to be precise about blanks or
spaces, or about
column alignment,
fixed-space type is used.
When it is important to show everything that appears on a long line, small type is used.
Suggestions that we have found helpful are shown like this in italics.
Please cite the current Winsteps computer program as:
Linacre, J. M. (2006) WINSTEPS Rasch measurement computer program. Chicago: Winsteps.com
We acknowledge the kind permission granted by Chris Hanscom of Veign for the use of their Jeweled Style
Command Button.
An Ode from a user:
WINSTEPS
14
This is a program that's much alive,
With scores and maps, and curves Ogive,
Persons and items to separate,
Giving results that do relate.
Although we love what WINSTEPS does,
Some problems make us Yelp!
But if we give John Michael a buzz,
He is always there to help!
Jim Houston, Nov. 30, 2001
11.
Getting further help
Common installation problems are solved at: www.winsteps.com/problems.htm
WINSTEPS is a powerful weapon in the struggle to wrest meaning from the chaos of empirical data. As you
become skilled in using WINSTEPS, you will find that it helps you to conceptualize what you are measuring, and
to diagnose measurement aberrations. The Special Topics section of this User's Guide contains a wealth of
information and advice.
Rasch Measurement Transactions, contains instructive articles on the fundamentals of Rasch analysis as well as
the latest ideas in theory and practice. There are other useful books and journals, including: Journal of Applied
Measurement, Trevor Bond &amp; Christine Fox: &quot;Applying the Rasch Model&quot;, Lawrence Erlbaum Assoc.
You may also find that you can use a more personal word of advice on occasion. The author of WINSTEPS, Mike
Linacre, is happy to answer e-mailed questions to do with the operation of WINSTEPS or the nature of Rasch
analysis. More prolonged consultations can also be arranged.
12.
What is supplied
WINSTEPS is supplied in three forms:
1) MinistepInstall.exe
To install MINISTEP, the student/evaluation version of Winsteps.
2) WinstepsInstall.exe
To install WINSTEPS under Windows
or 3) WinstepsPasswordInstall.exe To install WINSTEPS under Windows with password-protected installation.
These create directory, C:\WINSTEPS, and install in it WINSTEPS or MINISTEP.
Sample control and data (.TXT) files are also installed in c:\WINSTEPS\EXAMPLES to help you get started:
KCT.TXT is the Knox Cube Test data (BTD p.31 - see Section 1.1) The results in BTD were obtained with
more approximate algorithms and do not agree exactly with WINSTEPS results.
SF.TXT is the Liking For Science data (RSA p.18)
There are many more EXAMPLE files described later in this manual.
13.
Installation instructions for WINSTEPS
Under Windows XP/98/NT/ME/2000/...:
Run WinstepsInstall.exe from the downloaded file or from the CD-ROM.
If program hangs during &quot;Constructing Winsteps.ini ...&quot; then see Initialization Fails
To Run WINSTEPS:
Click on Start button (or use desktop Winsteps icon)
Point to Programs
Point to WINSTEPS
Click on WINSTEPS icon
15
Type SF.TXT
Press Enter key
Press Enter key
Press Enter key
Winsteps will run. Examine its output with the pull-down menus, particularly Output Tables
Additional Notes:
a) WINSTEPS is usually installed in the C:\WINSTEPS directory.
b) When WINSTEPS ends, pull down the &quot;Edit&quot; menu to edit control and output files.
c) All information on the screen is also in the Report output file.
d) Files in the C:\TEMP directory and with the suffix .TMP may be deleted.
Macintosh Computer: A Winsteps users reports: I've successfully installed Winsteps on a PowerMac G3 running
Virtual PC and OS 9.2. It seems to work fine. Surprisingly fast in fact, considering the Mac is emulating a
Windows environment.
14.
Starting WINSTEPS in Windows
A typical analysis requires two components: control information and data. These can be in separate computer
files or can be combined in one file. The results of the analysis are written to output files.
1) Double-click the Winsteps Icon on the Desktop.
To change the Winsteps data directory,
right-click on the icon
highlight and click on properties
click the shortcut tab
change the &quot;Start in&quot; path to the path to the desired directory.
or 2)
Click on Start button
Point to Programs
Point to WINSTEPS
Click on WINSTEPS icon
or 3)
Drag your control file onto your Winsteps icon.
Setup procedure: takes you to the Control and data file setup window
Instructions only: takes you to the do-it-yourself instructions
No: WINSTEPS asks you for the names of your input and report output files. There are example files already in
the Winsteps\examples directory.
Don't ask again: makes &quot;No&quot; the standard option here. You can reset this using Edit Initial Settings
If you need help to set up a new control file, go to control and data file set-up.
To select your control file:
Winsteps asks: Please enter name of WINSTEPS control file:
16
(a) You can type it in ...
Please enter name of WINSTEPS control file: KCT.TXT(Enter)
or (b) Click on the Files pull-down menu, &quot;Open File&quot;, to get a file dialog box.
or (c) Click-on a file name from the bottom of the Files menu list.
or (d) Press your Enter key
This displays a standard Windows open dialog box - try it.
You can also edit files directly from the file dialog box by right-clicking on them and selecting &quot;Edit&quot;.
Example Analysis:
Control file name? (e.g., kct.txt). Press Enter for Dialog Box: Example0.txt(Enter)
Please enter name of report output file: (Enter)
If you only press Enter, a temporary output file will be used.
If you type in a file name, that file will be created and used.
If you want the file dialog box, use 'Open File' on the File pull-down menu.
Extra specifications? (e.g., MJMLE=1), or press Enter: (Enter)
Usually you have none, so merely press Enter.
WINSTEPS will now construct measures (i.e., analyze) the Liking for Science data from RSA.
Use the Edit pull-down menu to simplify inspection and editing of the control, input and output files. This is done
with WordPad or your own text editor.
17
At the end of the run, using the Output Tables pull-down menu you can request further output Tables to be
produced interactively and displayed on the screen. If you want to save any of these, use the Save As option
when they are displayed. If you omit to save them, they are written as &quot;ws.txt&quot; files, and can be recovered
from the Recycle Bin.
At the end of the run, you can also use the Output Files pull down menu, to write out person and item measures
to computer-readable PFILE= and IFILE= files.
If the Edit menu and Output Tables menu don't work properly, then see Changing your Word Processor setting.
15.
Using Winsteps under Windows
Winsteps provides a familiar &quot;pull-down&quot; user interface, intended to provide the user with maximum speed and
flexibility. There are three main ways that you direct Winsteps:
(a) You respond to prompts on the screen.
There are two frequent ones:
Name of control file:
This is the &quot;DOS-text with line breaks&quot; or ASCII file where your control specifications reside.
You can press Enter to browse for it.
Report output file name:
Press Enter for a temporary output file, or
Type in an output file name or use the pull-down file menu
Extra specifications (or press Enter):
Press Enter!
This is used for making just-in-time changes to your control file instructions, for this run only.
18
(b) You use the pull-down menus.
A frequent choice is the first choice on both the File and Edit menus:
Edit Control File=
Select this to edit your Winsteps control file.
(c) You use the WordPad text editor.
All editing and display of Winsteps output is done using text files and Wordpad or your own text editor.
This gives you great flexibility to:
modify control and anchor files
view, copy and paste output into Word (or other) files
16.
Stopping WINSTEPS
The WINSTEPS program ceases execution when
1) The program stops itself:
The estimation procedure has reached an acceptable level of convergence and all pre-specified output has been
produced. This happen when:
a) The estimates are within the convergence criteria (LCONV= and RCONV= as controlled by CONVERGE=)
b) The maximum number of iterations has been reached (MPROX= and then MJMLE=)
To instruct WINSTEPS to run indefinitely (up to 2,000,000,000 iterations), set
MJMLE=0
LCONV=0
RCONV=0
CONVERGE=F
c) The estimates are not improving. This can occur when the limits of the computational precision of your
computer have been reached.
2) You stop the iterative process:
a) If you press Ctrl with F (or use File menu) during PROX iterations:
PROX iteration will cease as soon extreme scores have been identified and point-biserial correlations have been
calculated. JMLE iterations then start.
b) If you press Ctrl with F during JMLE iterations:
JMLE iteration will cease at the end of this iteration. Fit statistics will then be calculated and output tables
written to disk.
c) If you press Ctrl with F during the output phase:
Output will cease at the end of the current output operation.
Acknowledgment of your Ctrl with F instruction is shown by the replacement of = by # in the horizontal bar drawn
across you screen which indicates progress through the current phase of analysis.
3) You cancel WINSTEPS execution immediately:
For WINSTEPS:
From the File menu, choose Exit.
No more analysis or output is performed.
When Winsteps exits ...
It deletes all temporary files it has created and releases memory. You may have output Tables, files or graphs
open on your screen. Winsteps asks if you want these closed.
Yes: close all open windows. If some windows have been modified, but not saved, you may be asked if you want
19
to save those.
No: leave all windows as they are, but close the Winsteps analysis window.
To make Yes or No the automatic standard, click &quot;and from now on&quot;. This choice may be reset in Edit Initial
Settings
17.
Uninstalling WINSTEPS
Depending on the installation procedure:
(a) Select &quot;Uninstall&quot; from the Programs menu
or
(b) Go to &quot;Settings&quot;, &quot;Control Panel&quot;, &quot;Add/Remove Programs&quot; and double-click on &quot;WINSTEPS Uninstall&quot;
(c) Delete WINSTEPS directory, and delete the WINSTEPS entries from &quot;Windows\Start Menu\Programs&quot;
(d) Use a Windows clean-up utility to tidy up loose ends.
(e) Delete files in C:\TEMP and C:\WINDOWS\TEMP (or your Windows temporary file) and files ending .....ws.txt
18.
Menu bar
Winsteps has a useful set of pull-down menus:
File
overall control of the analysis/
Edit
display and editing of input and output files and tables.
Diagnosis Tables for understanding, evaluating and improving your measurement system.
Output Tables produces all output Tables produced by Winsteps.
Output Filesproduces output primarily intended for input into other software.
Batch
facilitates running Winsteps in batch mode
Help
displays Help file
Specification
allows entry of specifications after the analysis, one at a time, in the form of specification=value.
Plots
uses Excel to display and compare analyses.
SAS/SPSS reformat SAS .sas7bdat and SPSS .sav files into Winsteps control and data files.
Graphs Menu bit-mapped graphics for test, item and category display.
Data Setup provides a immediate means for setting up control and data files
19.
Batch menu
This facilitates running Winsteps in batch mode.
Running Winsteps in Batch mode Summary instructions for running Winsteps in batch mode.
Help
Displays help information for batch mode.
Edit
Edit batch file
Run
Run batch file: done by right-clicking on batch file name (.bat or .cmd), then clicking on open on the
right-click menu.
20
20.
Data Setup menu
This is described at Control and data file setup.
21.
Diagnosis menu
The Diagnosis pull-down menu suggests a step-by-step procedure for investigating the results of your analysis.
A. Item Polarity
check that all items are aligned in the same direction on the latent variable, same as Table
26. Check that all items have positive correlations. Use IREFER= and IVALUE= to point all items in the same
direction, or KEY1= to correct a multiple-choice key error.
B. Empirical Item-Category Measures check that all categories for all items are aligned in the same direction,
same as Table 2.6. For multiple-choice items, see Table 2 for MCQ. Check that correct answers, and higher
category values corresponding to &quot;more&quot; of the variable, are to the right.
C. Category Function check that all categorization functioned as intended, same as Table 3.2. Check that the
&quot;average measures&quot; for the categories advance, and that no category is especially noisy. Use IREFER= and
IVALUE= to collapse or remove discordant categories. use ISGROUPS= to identify category functioning. If more
details are required, look at the option/distractor analysis of the Item Tables.
D. Dimensionality check that all items share the same dimension, same as Table 23. This identifies substructures, &quot;secondary dimensions&quot;, in the data by performing a principal components/contrast decomposition of
the observation residuals. If there are large sub-structures, then it may be wiser to divide the data into two
measurement instruments.
E. Item Misfit
check that items cooperate to measure, same as Table 10. Are there misbehaving items?
Look for large mean-squares, and also for contradictory use of responses in the option/distractor listing.
F. Construct KeyMap check that the item hierarchy is as intended (construct validity), same as Table 2. This
locates items, response categories and your sample in one picture. Does your item measure hierarchy make
sense? What is the typical person in your sample saying?
G. Separation
check that the items discriminate different levels of person performance (&quot;test&quot; reliability),
same as Table 3.1. Also that persons are able to discriminate differences in item calibration.
22.
Edit menu
Display and editing of input and output files and tables.
21
Edit Control File=
Display and edit the current control file. Alters this analysis if no computation has been
done, otherwise the next analysis done with this control file.
Edit Report Output File= display and edit the report output file written during the main analysis phase. This
contains Table 0 and output specified with TABLES= and TFILE=.
Edit/create new control file from= ....\template.txt template.txt is a generic control file which can be edited and
saved under another name to give you a flying start at setting up your own analysis. There is a control and
data file setup procedure. It is easier!
Edit/create file with wordpad launches WordPad or your own text editor.
Undo
Cut
Copy
Paste
Delete
undo most recent change to the Output screen
copy characters from an output screen line to the Windows clipboard and delete them from the screen
copy characters from an output screen line to the Windows clipboard
paste characters from the Windows clipboard to a screen line
delete character from a screen line
for more substantial editing, save the screen using the File pull-down menu.
Edit initial settings
Edit Table ...
menu
Edit ... File
23.
change standard files and settings in Winsteps.ini
display and edit a Table produced from the Diagnosis, Output Tables or other pull-down
display and edit a file produced from the Output Files or other pull-down menu
File menu
This menu launches and terminates Winsteps analysis.
Edit Control File=
Edit the current control file. Alters this analysis if no computation has been done,
otherwise the next analysis done with this control file.
Exit, then Restart &quot;WINSTEPS ...&quot; Stop and then restart this analysis, usually after editing the control file.
22
Restart &quot;WINSTEPS ...&quot;
Restart this analysis, leaving the current one running.
Open File
Select control file for this analysis
Start another Winsteps
Launch a new copy of Winsteps. More than one copy of Winsteps can run at the
same time.
Exit
Exit from Winsteps immediately
Finish iterating
Finish the current phase as quickly as possible.
Close open output windows close any open windows for output tables, files or plots.
Enter
Acts as the Enter key
Save
Save As...
Print
Save the information displayed on the processing screen to disk.
Save the screen output to a named disk file.
Print the screen output
Location of the EXCEL program (if installed on your computer): can be change in Edit
Excel=
Initial Settings
SPSS=
Location of the SPSS program (if installed on your computer): can be change in Edit
Initial Settings
C:\WINSTEPS\examples\exam1.txt
24.
Previously used Winsteps control files, select these to analyze them
Graphs menu
Winsteps produces bit-mapped images, using the Graphs menu. Winsteps produces character-based graphs in
Table 21
Initially, select which type of curves you want to see. You can look at the others later without going back to this
menu. Graphs are plotted relative to the central difficulty of each item or response structure. Model-based curves
(such as probability and information functions) are the same for all items which share the same model definition in
ISGROUPS=. Empirical curves differ across items.
Category Probability Curves: model-based probability of observing each category of the response structure at
each point on the latent variable (relative to the item difficulty)
Empirical Category Curves: data-based relative frequencies of categories in each interval along the latent
variable
Expected Score ICC shows the model-based Item Characteristic Curve (or Item Response Function IRF) for the
item or response structure. This is controlled BYITEM= or the last two entries in this menu.
Empirical ICC shows the data-based empirical curve.
Empirical randomness shows the observed randomness (mean-square fit) in each interval on the variable with
logarithmic scaling. The model expection is 1.0
Cumulative Probabilities plot the model-based sum of category probabilities. The category median boundaries
are the points at which the probability is .5. Click on a line to obtain the category accumulation.
Item Information Function plots the model-based Fisher statistical information for the item. This is also the model
variance of the responses, see RSA p. 100.
Category Information plots the model-based item information partitioned according to the probability of observing
the category. Click on a line to obtain the category number.
23
Conditional Probability Curves show the model-based relationship between probabilities of adjacent categories.
These follow dichotomous logistic ogives. Click on a line to obtain the category pairing.
Test Characteristic Curve is the model-based test score-to-measure characteristic curve.
Test Information Function plots the model-based test information function, the sum of the item information
functions.
Test randomness shows the observed randomness (mean-square fit) in each interval on the variable with
logarithmic scaling. The model expection is 1.0
Multiple Item ICCs supports the display of several model and empirical ICCs simultaneously.
Only on the Graphs menu:
Display by item shows these curves for individual items, also controlled by BYITEM=. Model-based output is the
same for all items with the same ISGROUPS= designation.
Display by scale group for each ISGROUPS= code, a set of curves is shown. An example item number is also
shown - all other items in the grouping are included in the one set of grouping plots. Also controlled by BYITEM=.
25.
Help menu
Index displays the index of control variables
Contents displays the Table of Contents of the Help file.
About shows version number and compile date. Please mention these when reporting problems.
www.winsteps.com takes you to our website.
Bongo is the Winsteps &quot;Adjutant's Call&quot; - play this when you are summoning the data in preparation for
constructing measures!
Scaling calculator is designed to help you linearly rescale your measures in the way most meaningful for your
audience:
Under Current measure: enter two measures from your current analysis.
Under Desired measure: enter the values with which you want them to be reported.
Under Decimals: enter the number of decimal places for the measure Tables, Udecimals=.
Press Compute New to calculate the revised values of Uimean= and Uscale=.
The current values of Uimean= and Uscale= are displayed and also the revised New values. The New values can
be altered if you wish.
24
Press Specify New to action the New values. Or the values can be copied (Ctrl+c) and pasted into your Winsteps
control file.
26.
Output Files menu
It produces output primarily intended for input into other software. These files are referenced in the Output Files
Index
Most of these files can be written in several formats, so a dialog box is shown:
Display the Output File with:
Text Editor:
this is usually WordPad or your own text editor.
Excel:
the file is automatically input into Excel. Tab-delimited format is recommended.
SPSS:
SPSS is launched, if available: if this malfunctions, check that the path to SPSS is correct with Edit
Initial Settings.
Don't display: the file is written to disk, but no further action is taken.
File format:
Text: space-separated: fixed field: this is usually easiest to look at.
Text: tab-delimited:
columns are separated by Tab characters, which EXCEL expects.
Text: comma-separated fields:
columns are separated by commas or their international equivalents.
Labels in &quot;quotation marks&quot; place non-numeric values within quotation marks
SPSS: .sav format:
this can be input directly into SPSS (compatible with SPSS 6.0 and later). To see its
25
contents as plain text, use the SPSS menu. If this format is not available: Write the file in Excel (tabseparated). Then with variable names in row 1 and the data beginning in row 2, save the Excel data as an
Excel spreadsheet (not a workbook) with the extension .xls. Use the SPSS Open command to import this
into SPSS, and ask it to use the first row as variable names.
Column Headings:
Include column headings convenient when looking at the file, or reading into EXCEL
No column headings
convenient when the file is to be processed directly by other software.
File status::
Permanent file:
Temporary file:
27.
the file will not be deleted when Winsteps terminates. A file name is requested.
the file will be deleted when Winsteps terminates. &quot;Save as&quot; to make the file permanent.
Output Tables menu
Output Tables are listed in the Output Table Index. They are written into temporary files if selected from the
Output Tables menu. Output Tables are written into the Report Output File if specified using TABLES= or TFILE=
in the control file. Table 0 is always written to the Report Output File.
3.2 Rating (partial credit) response-structure and most Tables shown
Click on the Table to write it to a file and show it on the screen. Here is &quot;3.2 Rating response-structure Structure&quot;.
It is written into temporary file 03-859ws.txt. &quot;03&quot; refers to Table number 3. &quot;859&quot; is a unique number for this
analysis. &quot;ws.txt&quot; means &quot;Winsteps text file&quot;.
TABLE 3.2 LIKING FOR SCIENCE (Wright &amp; Masters p. ZOU859ws.txt Oct 9 10:54 2002
INPUT: 76 PUPILS, 25 ACTS MEASURED: 75 PUPILS, 12 ACTS, 3 CATS
WINSTEPS 3.36
-------------------------------------------------------------------------------SUMMARY OF CATEGORY STRUCTURE. Model=&quot;R&quot;
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ|| MEASURE | MEASURE|
|-------------------+------------+------------++---------+--------+
| 0
0
667 33| -1.30 -1.30|
.96
.95|| NONE
|( -2.04)| 00 dislike
| 1
1
757 37| -.08 -.09|
.90
.78||
-.82 |
.00 | 01 neutral
| 2
2
609 30| 1.40 1.41| 1.09 1.33||
.82 |( 2.04)| 02 like
+-----------------------------------------------------------------AVERAGE MEASURE is mean of measures in category.
Tables 27, 28, 30, 31, 33
These all allow the user to change the relevant control command on execution. ISUBTOTAL= controls the subtotal segments for Table 27 with a selection command so you are asked to confirm or change this value, before
the Table is produced.
26
Request Subtables
Any Table (except Table 0) can be displayed using this command. It also accepts the special fields available with
TFILE=
Weight Selection. See weighting. When IWEIGHT= or PWEIGHT= are used in estimation, reports can be
adjusted to reflect those weights or not. Weights of zero are useful for pilot items, variant items or persons with
unusual characteristics. These can be reported exclusively or excluded from reports.
28.
Plots menu
Here is the Plots menu.
Plotting problems? - these are usually due to the Winsteps-Excel interface. See
www.winsteps.com/problems.htm
Compare statistics - enables you to draw scatterplots of Winsteps statistics within or between analyses. It also
produces the tabular output of Table 34.
Bubble chart generates a Bond &amp; Fox-style bubble chart.
Keyform Plot - Horizontal generates a horizontal keyform layout.
Keyform Plot - Vertical generates a vertical keyform layout.
Plot 30 - DIF plots the DIF values in Table 30.
Plot 31 - DPF plots the DPF values in Table 31.
Plot 33 - DIF &amp; DPF plots the DIF+DPF values in Table 33.
27
29.
SAS/SPSS menu
SAS: This is described under Data from SAS files
SPSS: This is described under Data from SPSS files
30.
Specification menu
This allows entry of some specifications after the analysis, one at a time, in the form of specification=value. Click
on &quot;OK&quot; to action the specification and return to the standard screen. &quot;OK and again&quot; to action the specification
and redisplay this entry box.
Some specifications can be entered after the analysis has completed. They do not change the analysis but do
alter the output. They are useful for making selections (e.g., PSELECT= and ISELECT=), setting output Table
control values (e.g., MRANGE=) and changing user-scaling (e.g., USCALE=10).
Specifications with &quot;file name only&quot;:
CFILE=
scored category label file (file name only, blank deletes labels)
CLFILE= codes label file (file name only, blank deletes labels)
IDFILE=
item deletion file (file name only: blank resets temporary deletions)
ILFILE=
item label file (file name only: blank not allowed)
PDFILE= person deletion file (file name only: blank resets temporary deletions)
do not support &quot;*&quot;.
So, instead of
CLFILE=*
1 strongly disagree
2 disagree
3 agree
4 strongly agree
*
use the Edit menu, &quot;Create/Edit with Wordpad&quot;,
then, in Wordpad, type
1 strongly disagree
2 disagree
3 agree
4 strongly agree
save as &quot;clfile.txt&quot;
and in the Specification box, enter:
CLFILE = clfile.txt
28
31.
Control and data file setup window
This interface simplifies setting up Winsteps control and data files. It can be used for entering data with matching
specifications or for constructing the specifications that match existing data.
Select &quot;Setup Procedure&quot; on Winsteps startup, or use the Setup pull-down menu.
If you are already in Winsteps, then on the menu bar:
This displays the Control File Set-Up screen:
A multiple-choice test key, KEY1=, can be specified, if desired.
Items can be clustered into similar response-structure groupings using ISGROUPS=, using a one character code
for each grouping.
Use the Files menu to read in pre-existing control or data files. Uses the boxes and the data grid to enter new
control and data information. Use the Files or Winsteps menu to Save what you have done.
After performing the set-up, save the file and return to Winsteps using the Winsteps pull-down menu.
32.
Reading in a pre-existent control file
Reading in a pre-existing control file is easy. You can add or change control specifications, and add or change the
data.
From the Setup screen, use the File pull-down menu:
29
Select your desired pre-existing control or data file:
This fills in the Setup screen:
The control specifications and data are those in the Control and Data file. Control values are filled in as far as
possible. The data are filled in with one row per data record and one character per data column.
To see the item labels (between &amp;End and END LABELS in the Control file) either drag the column wider or click
on &quot;Item Labels Enter/Edit&quot;
33.
Data display
This grid displays the data file with one character per column.
During data entry, more columns are automatically added to the right, as needed.
Double-click the extreme row or column for an extra row or column.
Click on &quot;Refresh Data Display&quot; if the display does not show the current specification settings.
30
Press shift + left-click on the Person or Item No. row to dynamically select person label or item response columns.
34.
Item labels
Output reports and displays are much more useful and informative when the items are identified with short, clear
identifying labels.
These are usually entered in the specification control file after &amp;END and before END LABELS. There is one item
identifying label per line, so there should be as many lines of item identification as there are items on the test or
instrument.
In the Setup routine, they are entered in a special screen.
35.
Category labels
Category labels describe the categories in a response structure, levels in a partial credit item, or such like.
Categories are identified in the data codes (CODES=). If there are different categories for different items, then use
item grouping (ISGROUPS=) to identifier clusters of items which share the same category structures. Both of
these can be entered on the main set-up screen.
Example grouping &quot; &quot; means that this is the standard common grouping.
Double-click on the bottom line for another blank line.
31
36.
Do-it-yourself control and data file construction
There is a control and data file setup procedure. It is even easier!
Here is a step-by-step guide to setting up and running Winsteps. It is a little tricky the first time, but you'll soon find
it's a breeze!
The first stage is to set up your data in a rectangular data file in &quot;MS-DOS text with line breaks&quot; format.
1. Obtain your data
You'll need to be a little organized. Think of your data as a wall of pigeon-holes.
(a) Each column corresponds to one item, probe, prompt, task, agent, ....
For each column, you will need an item name or label. Make these short, only one or two words long. Make a list
of these in a document file. Put the label of the first item on the first line, etc.
Put END LABELS on the line after the last item.
Your list will look like this:
Eating
Dressing
....
Walking
Stair climbing
END LABELS
You can use WordPad or your own text editor or pull-down the Winsteps &quot;Edit&quot; menu, and select &quot;Create/Edit
file with WordPad&quot;
(b) Each row of pigeon-holes corresponds to one person, subject, case, object, ...
You will need some useful identifying codes for each row such as age, gender, demographics, diagnosis, timepoint. Winsteps doesn't require these. but its is much more useful when they appear. Give each of these
identifiers one or two letter codes, e.g., F=Female, M=Male, and give each identifier a column of pigeon-holes.
(c) The Data must be carefully lined up.
It is simpler if each data point, observation, response, rating can be squeezed into one character - numeric or
alphabetic.
Now create the data file. It will look like something this:
M
F
F
M
29
27
32
31
B
W
H
W
001
002
003
004
210212110200102
122121010201020
222210102112100
002010021000210
or, less conveniently,
M29B001210212110200102
F27W002122121010201020
32
F32H003222210102112100
M31W004002010021000210
After the END LABELS line, or in a separate file,
on each line enter the person identifying codes. Line them up so that each column has a meaning. This is easier if
you set the font to Courier.
Then enter the responses, starting with the first item and continuing to the last.
Do not place spaces or tabs between the responses.
If the lines start to wrap-around, reduce the font size, or increase the page size.
Excel, SPSS, SAS, ACCESS
Your data may already be entered in a spread-sheet, statistics program or database.
&quot;Copy and Paste&quot;, Save As, Export or Print to disk the data from that program into &quot;DOS-text with line breaks&quot;
or ASCII file.
If the program puts in extra blanks or separators (e.g., Tabs or commas), remove them with a &quot;global replace&quot; in
your text editor or word processor.
To replace a Tab with nothing, highlight the space where a Tab is. Then Ctrl+c to copy. Global replace. Ctrl+V put
a Tab into &quot;From&quot;. Put nothing in &quot;To&quot;. Action Global Replace.
In Excel, reduce the column width to one column, then
&quot;Save As&quot; Formatted Text (Spaced delimited) (*.prn)
In SPSS, see SPSS pull-down menu.
2. Set up your Winsteps Control and Data file
(a) Edit Template.txt
Pull-down the Winsteps &quot;Edit&quot; menu, select &quot;Create new control file from= ....\Template.txt&quot;
The Template.txt will be displayed on your screen by WordPad or your own text editor.
(b) Template.txt is a Winsteps Control and Data file
Find the three sections:
top: down to &amp;END are Control Specifications
we will edit this in a moment
middle: between &amp;END and END LABELS are the Item Labels
copy and paste the item labels from your list into this area.
one item label per line, in the order of the labels.
bottom: below END LABELS are the data
33
copy and paste the person labels and responses into this area.
one person per line (row).
(c) Edit the Control Specifications
Find the line &quot;Title=&quot;
Replace Put your page heading here with your own page heading.
Look at a data line, and count across:
In which column does the person identifying label start, the first position of the person name?
This is the Name1= value e.g., if it is column 4, then Name1=-4
How long is the person identifying label, the name length?
This is the Namlen= value e.g., if it is 10 columns, then Namlen=10
In which column is the response to the first item?
This is the Item1= value e.g., if the first response is in column 12, then Item1=12
How many items are there, the number of items?
e.g., if the number of items is 50, then NI=50
This is the NI= value
What are the valid item response codes in the data file?
This is the Codes= value
e.g., if the codes are 1,2,3,4, then Codes=1234
If your codes are not numeric, then you will need to rescore them.
See Data recoding
This is usually enough for your first Winsteps run.
(d) &quot;Save As&quot; the Template.txt file with your own file name.
Winsteps accepts any valid file name.
3. Run Winsteps
To the prompt:
Control file name? (e.g., KCT.txt). Press Enter for Dialog Box:
Press the Enter key
Select your control file from the dialog box
Press the Enter key
Report output file name (or press Enter for temporary file):
Press the Enter key
Extra specifications (or press Enter):
Press the Enter key
4. Your analysis commences
5. Your analysis concludes.
If there is an error message:
select &quot;Edit Control File=&quot; from the Winsteps Edit menu
correct the control file
save it
select &quot;Exit, then Restart Winsteps&quot; from the Winsteps File menu
34
If &quot;Measures constructed&quot; use the Output Tables pull-down menus to look at the Output Tables
here is the list of output tables.
6. Exit Winsteps using the X in the top right corner.
37.
Control file and template.txt
The control file tells what analysis you want to do. The template file, TEMPLATE.TXT. gives you an outline to start
from. The easiest way to start is to look at one of the examples in the next section of this manual, or on the
program disk. The control file contains control variables. These are listed in the index of this manual. Only two
control variables must have values assigned for every analysis: NI= and ITEM1=. Almost all others can be left at
their automatic standard values, which means that you can defer learning how to use most of the control variables
until you know you need to use them.
When in doubt, don't specify control variables, then they keep their standard values.
Here is a version of TEMPLATE.TXT. Copy and paste this, if your TEMPLATE.TXT is corrupted.
; this is a WINSTEPS specification control file template.
; Save it with your own name, e.g., control.txt
; a semi-colon means a comment: remove semi-colons as needed.
&amp;INST ; optional
TITLE = &quot;Put your page heading here&quot;
;Input
NAME1
NAMLEN
ITEM1
NI
XWIDE
PERSON
ITEM
Data Format
= 1
= 30
= ?
= ??
= 1
= Person
= Item
; DATA =
;
;
;
;
;
;
;
column of start of person information
maximum length of person information
column of first item-level response
number of items = test length
number of columns per response
Persons are called ...
Items are called ...
; data after control specifications
; For rescoring
;
0
1
2
3
4
5
6
7
;
1234567890123456789012345678901234567890123456789012345678901234567890
35
;ISGROUPS=0
; specify that each item has its own response structure (partial credit)
;IREFER=AABBCC....
; Data Scoring
CODES = &quot;01&quot;
;IVALUEA= &quot;01&quot;
;IVALUEB= &quot;10&quot;
;IVALUEC= &quot; &quot;
;
; valid response codes
for rescoring for item type A
for rescoring for item type B
for rescoring for item type C
Codes in IREFER with no IVALUE are not changed
;
;
;
;
CLFILE = *
; label the categories in Table 3
0 Strongly Disagree ; 0 in the data means &quot;Strongly Disagree&quot;
1 Strongly Agree
; 1 in the data means &quot;Strongly Agree&quot;
*
;NEWSCORE = &quot;10&quot;
;
KEY1 =
; use to rescore all items
; key for MCQ items
; XWIDE = 2
; for all codes 00 to 99
; CODES = &quot;000102030405060708091011121314151617181920212223242526272829+
;
+303132333435363738394041424344454647484950515253545556575859+
;
+606162636465666768697071727374757677787980818283848586878889+
;
+90919293949596979899&quot;
; codes reversed, in case needed
; NEWSCR= &quot;999897969594939291908988878685848382818079787776757473727170+
;
+696867666564636261605958575655545352515049484746454443424140+
;
+393837363534333231302928272625242322212019181716151413121110+
;
+09080706050403020100&quot;
; MISSCORE = -1
;User Scaling
UMEAN = 50
USCALE = 10
UDECIM = 1
MRANGE = 50
; -1 = missing data treated as not administered
;
;
;
;
user-set item mean - standard is 0.00
user-scaled measure units - standard is 1.00
reported decimal places - standard is 2
half-range on maps - standard is 0 (auto-scaled)
&amp;END
;Put item labels here for NI= lines
END LABELS
;Put data here - and delete this comment to prevent it being processed as a data line.
38.
Data file
If your data file is small, it is easiest merely to have it at the end of your control file. If your data is extensive, keep
it in a separate data file.
Your data file is expected to contain a record for each person containing a person-id field and a string of
responses to some items. Your data can be placed either at the end of your control file or in a separate disk file.
WINSTEPS reads up to 30 columns of person-id information as standard. Normally the person-id is assumed to
end when the response data begin or when the end of your data record is reached. However, an explicit length of
up to 300 characters can be given using the NAMLEN= control variable.
By the term &quot;response&quot; is meant a data value which can be a category label or value, score on an item or a
multiple-choice option code. The responses can be one or two characters wide. Every record must contain
responses (or missing data codes) to the same items. The response (or missing data code) for a particular item
must be in the same position in the same format in every record. If every person was not administered every item
then mark the missing responses blank or make them some otherwise unused code, so that the alignment of item
responses from record to record is maintained.
36
A table of valid responses is entered using the CODES= character string. Any other response found in your data
is treated as missing. By using the CODES=, KEYn=, NEWSCORE= and IVALUE= options, virtually any type of
response, e.g. &quot;01&quot;, &quot;1234&quot;, &quot; 1 2 3 4&quot;, &quot;abcd&quot;, &quot; a b c d&quot;, can be scored and analyzed. Missing responses are
usually ignored, but the MISSCORE= control variable allows such responses to be treated as, say, &quot;wrong&quot;.
When writing a file from SPSS, the syntax is:
FORMATS ITEM1 ITEM2 ITEM3 (F1). i.e., FORMATS varlist (format) [varlist..]
The procedure is FORMATS and then the variable list. Enclosed in parentheses is the format type. F signifies
numeric while 1 signifies the width. (F2) would signify a numeric with a width of 2 columns for XWIDE=2. See
pages 216 and 217 of the SPSS Reference Guide (1990). See also the SPSS pull-down menu.
39.
Data from Excel and other spreadsheets
It is easy to copy data from an Excel spreadsheet into a Winsteps data file.
(i) Organize your data.
Transform all item responses into columns one or two columns wide, e.g., &quot;1&quot; or &quot;23&quot;
Transform all demographics into columns, one column wide, e.g., &quot;M&quot; and &quot;F&quot; for male and female.
(ii) Organize your Excel spread sheet.
Put all item responses (one item per column) into one block to the left of your spreadsheet.
Put all person identifiers (one item per column) into one block, immediately to the right of the last item column.
(iii) Organize your column widths.
Make all item column widths the same (usually one or two columns).
Person identifier widths can match the identifiers, but these are best at one column wide.
(iv) Replace missing data with &quot;*&quot; or &quot;.&quot; or &quot;-&quot;
Global replace nothing in a cell with a convenient clear missing data indicator, which is not a number.
(v) Use the Excel format function to inset leading zeroes etc.
Select the item columns, then
Format - Cells - Custom
and enter 0 for 1 character wide columns, 00 for 2 character-wide columns, etc.
(vi) Select all cells.
(vii) Copy into clipboard (Ctrl+C), or write to a tab-delimited file
or write to a &quot;Formatted Text (space delimited) (*.prn)&quot; file
(viii) Open WordPad or your own text editor.
Paste (Ctrl+V) or open the tab-delimited file.
(ix) Removing tabs
Highlight a tab (area between two columns)
Copy (Ctrl+C)
Replace all tabs: (Ctrl+V) tab if necessary
with nothing.
(x) The file should now look like a standard Winsteps rectangular data file.
Save as a text file.
40.
Data from SAS files
There are two approaches to analyzing SAS data with Winsteps:
1. Use the Winsteps SAS/SPSS menu: SAS option:
37
This may require you to install Microsoft MDAC and SAS OLE DB Local Provider freeware, see SAS Conversion
Problems
Clicking &quot;Construct Winsteps file from SAS file&quot; displays:
Select SAS file: choose the SAS .sas7bdat file that you want to convert to a Winsteps control and data file.
If this fails, you may not have installed the free SAS interface software or your version of Windows may not fully
support OLE. See www.winsteps.com/problems.htm
This displays in the text box the details of the SAS file:
;
SAS dataset name: sastest
;
Number of Cases: 200
; Number of SAS Variables: 6
;
;
;
;
;
;
;
Move SAS variables under &quot;! Person Label Variables&quot;
and &quot;! Item Response Variables&quot;
SAS variables can be placed in both sections.
Numeric item variables are truncated to integers.
Constant fields may be specified with &quot; &quot;
XWIDE= is set according to the biggest item response value.
&quot;Create...&quot; when completed
! SAS File (do not delete this line): C:\WINSTEPS\examples\sastest.sas7bdat
! Person Label Variables. (Do not delete this line)
! Item Response Variables. (Do not delete this line)
!Other SAS Variables (ignored)
;Variable
Format
SN
; double-precision floating-point
item 1 ; double-precision floating-point
item 2 ; double-precision floating-point
ita
; double-precision floating-point
itb
; double-precision floating-point
itc
; double-precision floating-point
value
value
value
value
value
value
Cut-and-paste the SAS variables you want as Winsteps person and item variables:
! SAS File (do not delete this line): C:\WINSTEPS\examples\sastest.sas7bdat
! Person Label Variables. (Do not delete this line)
SN
; double-precision floating-point value
38
! Item Response Variables.
item 1 ; double-precision
item 2 ; double-precision
ita
; double-precision
itb
; double-precision
itc
; double-precision
(Do not delete
floating-point
floating-point
floating-point
floating-point
floating-point
this line)
value
value
value
value
value
!Other SAS Variables (ignored)
;Variable
Format
Save and Exit: save the text box and exit from the SAS conversion. The contents of the text box will
automatically redisplay if SAS conversion is requested from the same Winsteps run.
Permanent Output File: A Winsteps control and data file is created with a file name you select.
Temporary Output File: A Winsteps control and data file is created with a temporary name which will
automatically be deleted when the conversion screen is closed.
Display Output File: display the converted Winsteps control and data file, temporary or permanent. This can be
edited and saved or &quot;saved as&quot;.
&amp;INST
Title= &quot;C:\WINSTEPS\examples\sastest.sas7bdat&quot;
ITEM1= 1 ; Starting column of item responses
NI= 5 ; Number of items
; SAS Cases processed = 200
; datum: 0 count: 450
; datum: 1 count: 550
XWIDE = 1 ; this matches the biggest data value observed
CODES= 01 ; matches the data
NAME1 = 7 ; Starting column for person label in data record
; Person Label variables: columns in label: columns in line
@SN = 1E3 ; SN 1-3 7-9
@sascase = 5E7 ; case 5-7 11-13
NAMLEN = 7 ; Length of person label
&amp;END ; Item labels follow: columns in label
item 1 ; Item 1 : 1-1
item 2 ; Item 2 : 2-2
ita ; Item 3 : 3-3
itb ; Item 4 : 4-4
itc ; Item 5 : 5-5
END NAMES
01000 70
1
11101 121
2
01101 86
3
.....
Launch Winsteps: launch a new copy of Winsteps using the permanent or temporary control and data file.
2. SAS provides an environment within which Winsteps can run. See Kazuaki Uekawa's instructions,
www.estat.us/id2.html.
Sample instructions:
/*type the location where winsteps is installed*/
%let win= C:\WINSTEPS\winsteps;
....
option xwait xsync;
/*This run uses the whole sample*/
x &quot;start &amp;win &amp;WD&amp;scale..con &amp;WD&amp;scale._whole.out ifile=&amp;WD&amp;scale._whole.ifile
pfile=&amp;WD&amp;scale._whole.pfile &quot;;
....
/*item files produced by winsteps are now read by SAS*/
.....
41.
Data from SPSS files
Winsteps control and data files can easily be constructed from SPSS .sav files (compatible with SPSS 6.0 and
39
later). This can be done using the SPSS pull-down menu. Winsteps uses an interface routine provided by SPSS
which should work for all recent SPSS versions.
There are two approaches:
(a) Make the Winsteps control and data files directly. This option also allows inspection of SPSS variable
definitions.
(b) Convert the SPSS file into an EXCEL file for manipulation when SPSS is not available. This option can be
used for examining the contents of any SPSS .sav file.
1. Select SPSS file and variables. This displays the format of the SPSS variables. This is a utility function, and
can be used inspecing any SPSS .sav file.
Choose the variables you want in the person labels and the item response strings. Copy and paste the wanted
variables under:
! Person Label Variables. (Do not delete this line)
and/or
! Item Response Variables. (Do not delete this line)
Constant fields can be added using &quot; &quot; or ' '
then click Save. Not &quot;Save as&quot;
Example of display:
; Move SPSS variables under &quot;!Person Label Variables&quot;
; and &quot;!Item Response Variables&quot;
; SPSS variables can be placed in both sections.
;
Numeric variables are truncated to integers.
;
XWIDE= is set according to the biggest item response value.
; &quot;Save&quot; when completed
! SPSS File (do not delete this line): C:\WINDOWS\Desktop\older driver.sav
;
Number of Cases: 144
;
Number of SPSS Variables: 55
! Person Label Variables. (Do not delete this line)
INITIALS ; A8
Pasted in
SUMMARY
; F8.2
&quot;A&quot;
; a constant field containing &quot;A&quot; used to identify these records when this data file
is analyzed with other data files using DATA= file1 + file2 + ...
! Item Response Variables. (Do not delete this line)
&quot;.&quot;
; a constant of . used to indicate missing data for the CLOCK0 item in other data
files.
CLOCK1
; F1.0
closed circle Pasted in
CLOCK2
; F1.0
numbers in correct positions
!Other SPSS Variables (ignored)
;Variable
Format
Label
IDNUMBER ; F6.0
INITIALS ; A8
SUMMARY
; F8.2
IFNOT
; F8.2
if not complete
CLOCK1
; F1.0
closed circle
CLOCK2
; F1.0
numbers in correct positions
2. Edit SPSS variable selection
40
This permits you to change your variable selection.
3. Construct Winsteps file from SPSS file
This uses your variable selection to construct a Winsteps control and data file.
If &quot;No SPSS variables selected&quot;, then back to step 1, and be sure to &quot;Save&quot;, not &quot;Save as&quot;
Your SPSS file has been converted into a temporary .txt file. &quot;Save as&quot; your own permanent .txt file.
Example of Winsteps control and data file
; Save this file as your Winsteps control and data file
Title=&quot;C:\WINDOWS\Desktop\older driver.sav&quot;
ITEM1=1
NI=2
XWIDE = 3
CODES =
; Please supply your values here
; SPSS Cases processed = 144
NAME1 = 8
; Person Label variables
; INITIALS 8-10
; SUMMARY 11-13
NAMLEN = 6
&amp;END
CLOCK3
; Item 1
CLOCK4
; Item 2
END NAMES
0 1
kah6
1
1 1
b-f2
2
1 1
mbh1
3
....
Note: that SPSS variables are truncated to integers in the range 0-254 for Winsteps items. They are
converted with decimal places for person variables.
4. Convert SPSS file to EXCEL file
The selected SPSS file is converted into a Tab-delimited file and EXCEL is launched. EXCEL automatically reads
the SPSS variables into columns. &quot;Save as&quot; to keep this worksheet.
5. Convert SPSS file to EXCEL-compatible Tab-separated file
The selected SPSS file is converted into a Tab-delimited file. Copy and paste this file, or &quot;Save as&quot;.
This is a utility-function for converting SPSS files for any purpose.
42.
Data from STATA files
Fred Wolfe has provided a Stata module which produces WINSTEPS control and data files. It is at
ideas.uqam.ca/ideas/data/Softwares/bocbocodeS423302.html and has an accompanying help file.
Here is what this file looked like on 12-25-2001:
*! raschcvt version 1.2.1 fw 8/1/00 Prepares data file and control file for Winsteps
*! 5/26/01 update to version 7. turns off log output, 1.2.2 (12/20/01) corrects
*! decimal point formatting error and removes &quot;
= 40&quot;.
program define raschcvt
version 7.0
syntax varlist, outfile(string) id(varlist) max(integer) [min(integer 0)] [Xwide(integer 2)]
tokenize `varlist'
set more off
preserve
confirm variable `id'
di in gr &quot;Building Rasch files&quot;
local itemno = 0
local counter = 0
while &quot;`1'&quot; != &quot;`id'&quot; {
41
local itemno = `itemno' + 1
mac shift
}
capture log close
qui log using `outfile'.con, replace
di &quot;;The id variable is &quot; &quot;`id'&quot;
di &quot;;There are `itemno' items&quot;
di &quot;;Items and lengths&quot;
tokenize `varlist'
while &quot;`1'&quot; != &quot;&quot; {
capture assert `1' == int(`1')
if _rc != 0 {
di &quot;non-integer value found&quot;
exit _rc
}
local counter = `counter' + 1
qui su `1'
local f = length(&quot;`r(max)'&quot;)
di &quot;;&quot; &quot;`1'&quot; &quot; &quot; `f' &quot; &quot; `g'
if `counter' &lt;= `itemno' {
local f = `xwide'
}
format `1' %0`f'.0f
qui replace `1' = 99 if `1' == . &amp; `f' == 2
qui replace `1' = 999 if `1' == . &amp; `f' == 3
macro shift
}
order `varlist'
qui outfile `varlist' using `outfile'.dat, nolabel wide replace
di &quot;;`outfile' has been written to disc&quot;
di in gr &quot;;Start control file below&quot;
di
di &quot;TITLE=&quot;
di &quot;DATA=`outfile'.dat&quot;
di &quot;ITEM1=1&quot;
di &quot;NI=&quot; &quot;`itemno'&quot;
di &quot;NAME1=&quot; `itemno' +1
di &quot;DELIMITER = SPACE&quot;
di &quot;XWIDE=`xwide'&quot;
di &quot;CODES=&quot;_c
if `xwide' == 1{
for num `min' / `max', noheader: di X _c
}
if `xwide' == 2{
for num `min' / `max', noheader: if X == 0{di &quot;00&quot;_c} \ if X &gt;0 &amp; X &lt;10{di &quot;0&quot;X _c} \ if X &gt;9 &amp; X
&lt; 99 {di X _c}
}
if `xwide' == 3{
for num `min' / `max', noheader: if X == 0{di &quot;000&quot;_c} \ if X &gt;0 &amp; X &lt;10{di &quot;00&quot;X _c} \ if X &gt;9 &amp;
X &lt; 99 {di &quot;0&quot;X _c} \ if X &gt;99 {di X _c}
}
di
di &quot;MAXPAG=60&quot;
di &quot;PRCOMP=S&quot;
di &quot;MJMLE=0&quot;
di &quot;LCONV= .001
di &quot;RCONV= .1&quot;
di &quot;ISGROUPS=&quot;
di &quot;HLINES=Y&quot;
di &quot;;PSELECT= ?????1*&quot;
di &quot;;TABLES=11110110011111000001111&quot;
di &quot;;ISFILE=`outfile'.isf&quot;
di &quot;;IFILE=`outfile'.IFL&quot;
di &quot;;PFILE=`outfile'.PFL&quot;
di &quot;;XFILE=`outfile'.XFL&quot;
di
di &quot;&amp;END&quot;
di
tokenize `varlist'
while &quot;`1'&quot; != &quot;`id'&quot; {
local lbl:variable label `1'
42
di &quot;`lbl'&quot;
macro shift
}
di &quot;END LABELS&quot;
di
tokenize `varlist'
while &quot;`1'&quot; != &quot;`id'&quot; {
di &quot;`1'&quot;
macro shift
}
di &quot;END NAMES&quot;
qui log close
restore
end
43.
Data file with other delimiters
It is often convenient to organize your data with delimiters, such as commas, semi-colons or spaces, rather than
in fixed column positions. However, often the delimiter (a Tab, space or comma) only takes one column position.
In which case, it may be easier to include it in the CODES= or use MFORMS= or FORMAT=. See also
DELIMITER=.
44.
Example 0: Rating scale data: The Liking for Science data
Rather than attempting to construct a control file from scratch, it is usually easier to find one of these examples
that is similar to your problem, and modify it.
Control and data file, Example0.txt, for the Liking for Science data (see RSA) contains the responses of 75
children to 25 rating-scale items. The responses are 0-dislike, 1-neutral, 2-like. To analyze these data, start
WINSTEPS, then:
Control file name?:
&quot;Files&quot; pull-down menu
&quot;Control file name&quot;
&quot;Examples&quot; folder
Example0.txt
Open
Report output file: (Enter)
Extra specifications:(Enter)
press Enter for a temporary file.
no extra specifications at present.
; This is file &quot;example0.txt&quot; - &quot;;&quot; starts a comment
TITLE= 'LIKING FOR SCIENCE (Wright &amp; Masters p.18)'
NI=
25 ; 25 items
ITEM1= 1
; responses start in column 1 of the data
NAME1= 30 ; person-label starts in column 30 of the data
ITEM=
ACT ; items are called &quot;activities&quot;
PERSON= KID ; persons are called &quot;kids&quot;
CODES= 012 ; valid response codes (ratings) are 0, 1, 2
CLFILE= *
; label the response categories
0 Dislike
; names of the response categories
1 Neutral
2 Like
*
; &quot;*&quot; means the end of a list
&amp;END
; this ends the control specifications
; These are brief descriptions of the 25 items
WATCH BIRDS
READ BOOKS ON ANIMALS
READ BOOKS ON PLANTS
WATCH GRASS CHANGE
FIND BOTTLES AND CANS
LOOK UP STRANGE ANIMAL OR PLANT
WATCH ANIMAL MOVE
LOOK IN SIDEWALK CRACKS
LEARN WEED NAMES
43
LISTEN TO BIRD SING
FIND WHERE ANIMAL LIVES
GO TO MUSEUM
GROW GARDEN
LOOK AT PICTURES OF PLANTS
READ ANIMAL STORIES
MAKE A MAP
WATCH WHAT ANIMALS EAT
GO ON PICNIC
GO TO ZOO
WATCH BUGS
WATCH BIRD MAKE NEST
FIND OUT WHAT ANIMALS EAT
WATCH A RAT
FIND OUT WHAT FLOWERS LIVE ON
TALK W/FRIENDS ABOUT PLANTS
END NAMES ;this follows the item names: - the data follow:
1211102012222021122021020
ROSSNER, MARC DANIEL
2222222222222222222222222
ROSSNER, LAWRENCE F.
2211011012222122122121111
ROSSNER, TOBY G.
1010010122122111122021111
ROSSNER, MICHAEL T.
1010101001111001122111110
ROSSNER, REBECCA A.
1011211011121010122101210
ROSSNER, TR CAT
2220022202222222222222222
WRIGHT, BENJAMIN
2210021022222020122022022
LAMBERT, MD., ROSS W.
0110100112122001121120211
SCHULZ, MATTHEW
2100010122221121122011021
HSIEH, DANIEL SEB
2220011102222220222022022
HSIEH, PAUL FRED
0100220101210011021000101
LIEBERMAN, DANIEL
1211000102121121122111000
LIEBERMAN, BENJAMIN
2110020212022100022000120
HWA, NANCY MARIE
1111111111111111112211111
DYSON, STEPHIE NINA
2221121122222222222122022
BUFF, MARGE BABY
2222022222222222222222122
SCHATTNER, GAIL
2222021022222221222122022
ERNST, RICHARD MAX
2221011002222211122020022
FONTANILLA, HAMES
1100010002122212122020022
ANGUIANO, ROB
1111011112122111122021011
EISEN, NORM L.
1211111112222121122121121
HOGAN, KATHLEEN
2211112112222121222122122
VROOM, JEFF
1221012102222110122021022
TOZER, AMY ELIZABETH
2221120222222221222122022
SEILER, KAREN
1111111111111111111111111
NEIMAN, RAYMOND
1110011101122111022120221
DENNY, DON
2211102002122121022012011
ALLEN, PETER
1000101001110000022110200
LANDMAN, ALAN
1000000002122000022012100
NORDGREN, JAN SWEDE
2211011112221011122121111
SABILE, JACK
1210010011222110121022021
ROSSNER, JACK
2212021212222220222022021
ROSSNER, BESS
2222122222222222222222122
PASTER, RUTH
1110211011122111122011111
RINZLER, JAMES
1110201001122111122011111
AMIRAULT, ZIPPY
1210010012122111122012021
AIREHEAD, JOHN
2211021112222121122222121
MOOSE, BULLWINKLE
2221022112222122222122022
SQURREL, ROCKY J.
2221011022222221222122022
BADENOV, BORIS
2222222222222222222222122
FATALE, NATASHA
2121011112122221122011021
LEADER, FEARLESS
2220112022222121222022021
MAN, SPIDER
1210021012212211022011021
CIANCI, BUDDY
2221022222222222222222022
MCLOUGHLIN, BILLY
2200020002222000222122010
MULLER, JEFF
1110220112212111022010000
VAN DAM, ANDY
2221022222222221222122022
CHAZELLE, BERNIE
1100110101110112111111210
BAUDET, GERARD
2222121222222222222122122
DOEPPNER, TOM
44
1210021022222221122021022
1111111111111111122111111
0100101010110011021010000
1111021121122112122021011
1010010011122112222111110
2221012012222222222122022
2220022022222222222022022
1221010122222221222021021
2221021122222221222022222
1221111122222121122121021
2111010022122111122011022
2220000022222220222020000
1211010122222122222122010
2210022112222121222122011
1211211002222012122121200
1201010012222120222022021
1211001001212121222121012
1200010102120120022001010
1210010122222122222021021
2222022012222122222022021
2201222001220022220222201
1001000100011010021200201
2012010122210201102202022
2220022002222222222022012
1200110112122221022020010
REISS, STEVE
STULTZ, NEWELL
SABOL, ANDREW
BABBOO, BABOO
DRISKEL, MICHAEL
EASTWOOD, CLINT
CLAPP, CHARLIE
CLAPP, DOCENT
CLAPP, LB
SQUILLY, MICHAEL
SQUILLY, BAY OF
BEISER, ED
BECKER, SELWIN
CORLEONE, VITO
CORLEONE, MICHAEL
PINHEAD, ZIPPY
MALAPROP, MRS.
BOND, JAMES
BLOFELD, VILLAIN
KENT, CLARK
STOLLER, DAVE
JACKSON, SOLOMON
SANDBERG, RYNE
PATRIARCA, RAY
PAULING, LINUS
When analysis completes, use the &quot;Output Tables&quot;pull-down menu to examine the measures and quality-control
fit statistics. Start with Table 1, which gives you a quick summary. Here are the Science Activity items in
measure order, Table 13. The hierachy of item labels, in the right-hand column, defines the construct that has
been measured. &quot;Go on picnic&quot; is easiest to like. &quot;Find bottles&quot; is hardest to like.
ACT STATISTICS: MEASURE ORDER
+-------------------------------------------------------------------------------------------------+
|ENTRY
RAW
MODEL|
INFIT | OUTFIT |PTMEA|
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.| ACT
|
|------------------------------------+----------+----------+-----+--------------------------------|
|
5
35
74
2.42
.22|2.30
5.6|3.62
7.3| .05| FIND BOTTLES AND CANS
|
|
23
40
74
2.18
.21|2.41
6.3|4.11
9.0| .00| WATCH A RAT
|
|
20
48
74
1.83
.20|1.33
2.0|1.82
3.7| .42| WATCH BUGS
|
|
4
50
74
1.75
.20| .89
-.7| .91
-.4| .60| WATCH GRASS CHANGE
|
|
8
52
74
1.67
.20|1.10
.7|1.21
1.2| .51| LOOK IN SIDEWALK CRACKS
|
|
7
67
74
1.10
.19| .97
-.1|1.01
.1| .59| WATCH ANIMAL MOVE
|
|
9
78
74
.71
.19|1.18
1.3|1.17
1.0| .53| LEARN WEED NAMES
|
|
16
81
74
.60
.19| .97
-.2| .95
-.3| .51| MAKE A MAP
|
|
25
83
74
.53
.19| .80 -1.5| .74 -1.6| .66| TALK W/FRIENDS ABOUT PLANTS
|
|
3
86
74
.42
.19| .57 -3.5| .54 -3.0| .72| READ BOOKS ON PLANTS
|
|
14
86
74
.42
.19| .82 -1.3| .75 -1.4| .62| LOOK AT PICTURES OF PLANTS
|
|
6
89
74
.31
.19| .81 -1.4| .76 -1.4| .61| LOOK UP STRANGE ANIMAL OR PLANT|
|
17
93
74
.16
.19| .65 -2.7| .59 -2.4| .70| WATCH WHAT ANIMALS EAT
|
|
22
95
74
.08
.19| .83 -1.2| .74 -1.4| .63| FIND OUT WHAT ANIMALS EAT
|
|
24
105
74
-.31
.20| .90
-.6| .79
-.9| .60| FIND OUT WHAT FLOWERS LIVE ON |
|
1
107
74
-.40
.21| .55 -3.5| .49 -2.5| .64| WATCH BIRDS
|
|
15
109
74
-.48
.21| .78 -1.5| .64 -1.6| .61| READ ANIMAL STORIES
|
|
2
114
74
-.71
.22| .93
-.4| .72 -1.0| .58| READ BOOKS ON ANIMALS
|
|
21
117
74
-.85
.22| .84
-.9| .65 -1.3| .58| WATCH BIRD MAKE NEST
|
|
11
119
74
-.96
.23| .63 -2.4| .49 -1.9| .59| FIND WHERE ANIMAL LIVES
|
|
13
125
74
-1.29
.25|1.22
1.1| .94
.0| .47| GROW GARDEN
|
|
10
128
74
-1.49
.26| .78 -1.1| .57 -1.1| .50| LISTEN TO BIRD SING
|
|
12
135
74
-2.04
.31| .70 -1.2| .51 -1.0| .45| GO TO MUSEUM
|
|
19
139
74
-2.48
.36|1.08
.4|1.10
.4| .30| GO TO ZOO
|
|
18
143
74
-3.15
.47|1.50
1.2|1.23
.5| .14| GO ON PICNIC
|
|------------------------------------+----------+----------+-----+--------------------------------|
| MEAN
93.0
74.0
.00
.23|1.02
-.2|1.08
.0|
|
|
| S.D.
30.9
.0
1.41
.06| .45
2.3| .87
2.8|
|
|
+-------------------------------------------------------------------------------------------------+
45.
Example 1: Dichotomous data: Simple control file with data included
A control file, EXAM1.TXT, for an analysis of the Knox Cube Test (see BTD) a test containing 18 items, each item
45
is already scored dichotomously as 0,1. The person-id data begins in column 1 and the item string begins in
column 11. No items will be deleted, recoded, or anchored. The number of data lines is counted to determine
how many children took the test. use the pull-down &quot;Diagnosis&quot; and &quot;Output Table&quot; menus to see the output
tables. For an explanation of the output obtained, see later in the manual. Run this example with:
Control file: EXAM1.TXT
Report output file: (Enter)
Extra specifications: (Enter)
; This file is EXAM1.TXT - (&quot;;&quot; starts a comment)
TITLE='KNOX CUBE TEST' ; Report title
NAME1=1
; First column of person label in data file
ITEM1=11
; First column of responses in data file
NI=18
; Number of items
CODES=01
; Valid response codes in the data file
CLFILE=*
; Labels the observations
0 Wrong
; 0 in data is &quot;wrong&quot;
1 Right
; 1 in data is &quot;right&quot;
*
; &quot;*&quot; is the end of a list
PERSON=KID
; Person title: KID means &quot;child&quot;
ITEM=TAP
; Item title: TAP means &quot;tapping pattern&quot;
&amp;END
; Item labels for 18 items follow
1-4
; tapping pattern of first item: cube 1 then cube 4 are tapped.
2-3
1-2-4
1-3-4
2-1-4
3-4-1
1-4-3-2
1-4-2-3
1-3-2-4
2-4-3-1
1-3-1-2-4
1-3-2-4-3
1-4-3-2-4
1-4-2-3-4-1
1-3-2-4-1-3
1-4-2-3-1-4
1-4-3-1-2-4
4-1-3-4-2-1-4
; last tapping pattern: 7 actions to remember!
END NAMES
; END NAMES or END LABELS must come at end of list
Richard M 111111100000000000
; Here are the 35 person response strings
Tracie F 111111111100000000
Walter M 111111111001000000
Blaise M 111100101000000000
Ron
M 111111111100000000
William M 111111111100000000
Susan
F 111111111111101000
Linda
F 111111111100000000
Kim
F 111111111100000000
Carol
F 111111111110000000
Pete
M 111011111000000000
Brenda F 111110101100000000
Mike
M 111110011111000000
Zula
F 111111111110000000
Frank
M 111111111111100000
Dorothy F 111111111010000000
Rod
M 111101111100000000
Britton F 111111111100100000
Janet
F 111111111000000000
David
M 111111111100100000
Thomas M 111111111110100000
Betty
F 111111111111000000
Bert
M 111111111100110000
Rick
M 111111111110100110
; best performance
46
Don
Barbara
Adam
Audrey
Anne
Lisa
James
Joe
Martha
Elsie
Helen
M
F
M
F
F
F
M
M
F
F
F
111011000000000000
111111111100000000
111111100000000000
111111111010000000
111111001110010000
111111111000000000
111111111100000000
111111111110000000
111100100100000000
111111111101010000
111000000000000000
46.
; worst performance:
last data line - just stop!
Example 2: Control and anchor files
A control file, EXAM2.TXT, for the analysis of a test containing 18 items, each item already scored dichotomously
as 0,1. The person-id data begins in column 1 and the item-response string begins in column 11. The standard
tables will be appear in the printout. There is user scaling. Items 2, 4, 6 and 8 are anchored at 400, 450, 550 and
600 units respectively, supplied in file EXAM2IAF.TXT Your data is in file EXAM2DAT.TXT:
; This file is EXAM2.TXT
TITLE='KNOX CUBE TEST - ANCHORED'
; the title for output
NI=18
; the number of items
ITEM1=11
; position of first response in data record
NAME1=1
; first column of person-id in data record
PERSON=KID
ITEM=TAP
DATA=EXAM2DAT.TXT
; name of data file
IAFILE=EXAM2IAF.TXT
; this is item anchor (input) file: it is the IFILE= of an
earlier analysis
CONVERGE=L
; use only Logits for convergence criterion
LCONV=.005
; converged when biggest change is too small to show on any
report.
; What follows is equivalent to the IAFILE= above
; IAFILE=*
;2 400
;4 450
;6 550
;8 600
; *
UIMEAN=500
USCALE=100
UDECIM=0
&amp;END
1-4
.
4-1-3-4-2-1-4
END NAMES
;
;
;
;
; item anchor file list
item 2 anchored at 400 units
item 4 anchored at 450 units
item 6 anchored at 550 units
item 8 anchored at 600 units
; user scaling - item mean
; user scaling - 1 logit = 100 user units
; print measures without decimals
; item labels, starting with the first item
; End of this file
The anchoring information is contained in file EXAM2IAF.TXT and contains the following lines, starting in column
1:
2 400
; item 2 anchored at 400 units:
; if logits are user-rescaled, then anchor values are also expected to be userrescaled.
; for logit anchor values, specify UANCHOR=no
4 450 ; item 4 anchored at 450 units
6 550 ; item 6 anchored at 550 units
8 600 ; item 8 anchored at 600 units
Item calibration files, IFILE=, from prior runs can be used as item anchor files, IAFILE=, of later runs.
Your data is in the separate file, EXAM2DAT.TXT, with person-id starting in column 1, and item responses
starting in column 11:
47
Richard
Tracie
.
Elsie
Helen
M 111111100000000000
F 111111111100000000
F 111111111101010000
F 111000000000000000
47.
End of this file
Example 3: Item recoding and item deletion
The test has 25 items, specified in EXAM3.TXT. The item response string starts in column 12. Person-id's start
in column 1 (the standard value). Original item codes are &quot;0&quot;, &quot;1&quot;, &quot;2&quot; and &quot;X&quot;. All items are to be recoded and
the original-to-new-code assignments will be 0 0, 1 2, 2 1 and X 3. Items 5, 8, and 20 through 25 are to be
deleted from the analysis, and are specified in the control. The misfit criterion for person or item behavior is 3.0.
Tables 1, 2, 3, 4, 5, 6, 7, 8, 9, 11, 13, 15, 17, 19, 20 and 21 are to appear in your report output file
EXAM3OUT.TXT. Sequence numbers are used as item names. Data are in file EXAM3DAT.TXT.
; This file is EXAM3.TXT
TITLE=&quot;A test with 25 recoded items&quot;
; informative title
NI=25
; 25 items
ITEM1=12
; item responses start in column 12
CODES =012X
; valid codes
NEWSCORE=0213
; corresponding response score
RESCORE=2
; specifies rescore all items
TABLES=1111111110101010101110
; selected tables to go in output file
FITP=3.0
; person misfit cut-off for reporting
FITI=3.0
; item misfit cut-off for reporting
DATA=EXAM3DAT.TXT
; name of data file
IDFILE=*
; list of items to delete: or use IDFILE=file name
5
; delete item 5
8
; delete item 8
20-25
; delete items 20 through 25
*
; end of list
INUMB=Y
; use item sequence numbers as names
&amp;END
; end of control specifications
The data is in file EXAM3DAT.TXT
101F20FJDP
102M20PFNP
.
175F FBDP
176F23FEDP
21XX2XXXXX111X1200001X2X
X2210122211222211222120X2
48.
Example 4: Selective item recoding
1X00X00000200012X02220100
21121022012002121 2202000 person id's contain demographic information
The test has 18 items, specified in file EXAM4.TXT. The response string starts in column 1. Person-id's start in
column 41. Original codes are 0,1 in data file EXAM4DAT.TXT. Items 2, 3, and 4 are to be recoded as 1,0. All
tables are to appear in report file EXAM4OUT.TXT, in a standardized form.
TITLE=&quot;An example of selective item recoding&quot;
; page title
NI=18
; 18 items
ITEM1=1
; item responses start in column 1
NAME1=41
; person-id starts in column 41
NAMLEN=9
; person-id has 9 characters: Richard M
PSELECT= ????????M
; select M in position 9 of person name
NAMLMP=7
; 7 characters to appear on maps: &quot;Richard&quot;
CODES=01
; the observed response codes
CLFILE=*
; the code label list
0 Incorrect
; the observed response codes
1 Correct
*
; end of list
IREFER=ABBBAAAAAAAAAAAAAA ; can be letters or numbers
48
IVALUEA=01
IVALUEB=10
; A-type items keep original scoring (this line can be omitted)
; B-type items have reverse scoring
DATA=EXAM4DAT.TXT
; name of data file
MRANGE=5
; plots and maps have half-range of 5 logits
LINLEN=0
; allow maximum line length
MAXPAG=0
; no maximum page length
TABLES=1111111111111111111111111111111111 ; output all Tables to the Output file
&amp;END
1-4
R2-3
R1-2-4
2-1-4
.
4-1-3-4-2-1-4
END NAMES
; R reminds us item coding was reversed
The data file, EXAM4DAT.TXT, is
100011100000000000
100011111100000000
.
100100000000000000
49.
Richard M
Tracie F
Helen
F
Example 5: Scoring key for items, also CAT responses
A multiple choice adaptive test, in file EXAM5.TXT with responses &quot;a&quot;, &quot;b&quot;, &quot;c&quot;, &quot;d&quot; and a scoring key for 69 items.
Your data are in the control file. This was administered as a CAT test, then the response file formatted into a &quot;flat&quot;
file with one row per person and one column per item.
; This file is EXAM5.TXT
TITLE=&quot;An MCQ Test&quot; ; the title
NI=69
; 69 items
ITEM1=10
; response string starts in column 10
NAME1=1
; person-id starts in column 1
CODES=abcd
; valid response codes
MISSCORE=-1
; standard scoring of missing data, means that blanks are ignored
KEY1 = dcbbbbadbdcacacddabadbaaaccbddddcaadccccdbdcccbbdbcccbdcddbacaccbcddb
; scoring key of correct answers
ITEM=TOPIC
; items are topics
PERSON=STDNT
; respondents are students
NAMLMP=2
; first 2 characters on maps, e.g., nl
PFILE=EXAM5PF.TXT
; write out person measures
CSV=Y
; separate values by commas in PFILE=
HLINES=N
; write out no heading lines in PFILE=
; Many spreadsheets and statistics programs expect a file of numbers separated by commas.
; Use IFILE= or PFILE= with CSV=Y and HLINES=N.
MJMLE=0
; allow as many JMLE iterations as necessary
EXTRSC=0.5
; most conservative (central) extreme measures wanted
XFILE=EXAM5XF.TXT
; write out individual response residual file
&amp;END
nl01 Month
nl02 Sign
.
sb02 newspaper
sb03 newspaper
END NAMES
IM CAT
a
dcacc ccabbcaa
NM KAT
b badad accaaba aa c dd
ab
c
.
NH RIC ddb b dbdcbcaadba ba acd bad db
c
d
cc
IL HOL
a
a da d
d ccbddd bcd dc ca
49
50.
Example 6: Keys in data record FORMAT
Do not use FORMAT= unless there is no other way. It is tricky to set up correctly. MFORMS= is much easier.
A test of 165 multiple-choice items with multiple data lines per data record. The scoring key is formatted in the
same way as the data lines:
; This file is EXAM6.TXT
TITLE='Demonstration of KEY1 record'
title
FORMAT=(1X,10A,T23,50A,/,T23,50A,/,T23,50A,/,T23,15A)
;
;
;
;
;
;
;
;
;
;
;
;
;
;
;
;
The first character is ignored,
then 10 characters in first record are person id,
then starting in column 23,
50 columns in first 3 records,
and 15 responses in fourth record.
Using MFORMS= to
MFORMS=*
Data= filename
L=4
P1-10=2
I1-50=23
I51-100=2:23
I101-150=3:23
I151-165=4:23
*
reformat the same data record
;
;
;
;
;
;
;
;
;
put the input data in a separate file
4 data input lines per output record
person label characters 1-10 start in column 2 of line 1 of input data
responses to items 1-50 start in column 23 of line 1 of input data
responses to items 51-100 start in column 23 of line 2 of input data
responses to items 101-150 start in column 23 of line 3 of input data
responses to items 151-165 start in column 23 of line 4 of input data
end of MFORMS=
Note this does not reformat the Keyfrm=, so use KEY1=
; In the reformatted record
NAME1=1
ITEM1=11
NI=165
CODES=&quot;ABCD &quot;
MISSCORE=0
KEYFRM=1
RFILE=exam6rf.txt
PTBIS=YES
TFILE=*
1.0
3
6
10
20
*
;
;
;
;
;
;
;
;
;
;
;
;
;
Person-id starts in column 1
Item responses start in column 11 of reformatted record
There are 165 items
The raw responses are ABCD and BLANK.
Put character strings in &quot; &quot; if blanks are to be included.
Blanks and invalid codes scored wrong=0
There is a KEY1 record after &amp;END which is formatted
exactly like a data record specifying the correct responses.
this shows effect of reformatting and scoring
Raw score point-biserial
List of desired tables
Subtable 0 of Table 1
Table 3
; end of list
&amp;END
; KEY1= formatted like your data follows:
Key 1 Record
CDABCDBDABCADCBDBCADBABDDCDABCBABDCACBADACBADBAACD
after &amp;END
CCBDACABDADCBDCABBCACDBAABCDADCDCADBCABCDCADABACDA
in FORMAT= format
BADCDBADCBADCDBACBADBCAADBCBBDCBACDBACBADCDADBACDB
before item names
ABDACDCDBADBCAB
A1
First item name
A2
.
A164
A165
END NAMES
090111000102
10001 BDABADCDACCDCCADBCBDBCDDACADDCACCCBCCADBDABADCAADD
ABDDDCABDADCBDACDBCACADABCDCCDCBDBCCABBCDCADDCDCDA
BDCCDBABCDCDDDCBADCACBDCBDBACBCBCADBABAADCDCBABAAC
DCBCCACABCDDCBC
090111000202
10002 BDCDCDCDADCBCCBDBDCABCBDACDABCAABCAACBBBACAADDAACA
ACBCACBBDADCBDCBBBCDCCDACCBCADCACCAACDBCCDADDBACDA
BADCDCBDBDCDCCBACCCBBAABDBCDBCCBAADBABBADBDDABDCAA
50
DCDBCDCDBADBCCB
.
090111008402
10084
090111008502
10085
CDABCDADDDCDDDDCBDCCBCCDACDBBCACDBCCCBDDACBADCAACD
ACBDCCDBDADCCBCDDBBDCABACCBDBDCBCCACCBACDCADABACDA
BABBDCADBDDBCDADDDCDDBCABCBDCCCAACDBACBDDBDBCCAACB
DBACDBCDBADDCBC
CDABADADABCADCDDBDADBBCBACDABCCABACCCDAAACBADAAACD
ACBCDCBBDADCDDCADBCCCDBADDBBBDCACAABCBDDDCADABACDA
BADABBADBBADCADACDABBAACACAABDCBACDBADBACCDBACBADA
BCABCBCDBADDCCC
or, using continuation lines for the key:
CODES=&quot;ABCD &quot;
; The raw responses are ABCD and BLANK
;KEYFRM=
; omit this, not needed
KEY1 = CDABCDBDABCADCBDBCADBABDDCDABCBABDCACBADACBADBAACD+
+CCBDACABDADCBDCABBCACDBAABCDADCDCADBCABCDCADABACDA+
+BADCDBADCBADCDBACBADBCAADBCBBDCBACDBACBADCDADBACDB+
+ABDACDCDBADBCAB
; &quot;+&quot; are continuation characters
51.
Example 7: A partial credit analysis
A 30 item MCQ Arithmetic Test is to be analyzed in which credit is given for partial solutions to the questions.
Each item is conceptualized to have its own response structure, as in the Masters' Partial Credit model.
Estimation for the Partial Credit model is described in RSA, p. 87.
In this example, item 1 has 3 scoring levels. &quot;C&quot; is correct, worth 2 points. &quot;A&quot; and &quot;D&quot; are partially correct, worth 1
point. &quot;B&quot; is incorrect, worth 0 points. CODES= identifies all possible valid responses. In this example, KEY1=
identifies responses worth 2 points. KEY2= and KEY3= identify reponses worth 1 point. The values of KEY1=,
KEY2= and KEY3= are set by KEYSCR=. So for item 1, KEY1=C..., KEY2=A..., and KEY3=D. Response B is not
in a KEY= and so is scored 0. Here, invalid responses are treated as not-administered. If invalid responses are to
be treated as &quot;wrong&quot;, specify MISSCORE=0.
; This file is EXAM7.TXT
TITLE=&quot;A Partial Credit Analysis&quot; page heading
NAME1=1
; Person-id starts in column 1
ITEM1=23
; Item responses start in column 23
NI=30
; There are 30 items
CODES=ABCD
; Scores entered as A through D
KEY1=CDABCDBDABCADCBDBCADBABDDCDABC
; Fully correct
KEY2=ABCDABCCBADDABACABBBACCCCDABAB
; Partially correct
KEY3=DABABAAACC*CCABAC*************
; Some partially correct
; if no matching response, use a character not in CODES=, e.g., *
; the keys are matched in sequence, &quot;B&quot; for item 15 matches Key1=, and
Key3= is ignored
KEYSCR=211
; KEY1 fully correct (2 points),
; KEY2, KEY3 partially correct (1 point each)
ISGROUPS=0
; Each item is its own grouping, i.e., the Partial Credit model
MODELS=R
; Each item has its own Andrich rating scale
STKEEP=Y
; Keep all intermediate categories in analysis, even if never observed
CURVES=111
; Print all 3 item curves in Tables 2 and 21
CATREF=2
; Use category 2 for ordering in Table 2
&amp;END
.7
A1
1.1 A2
.
3.1 A29
2.7 A30
END NAMES
090111000102
10001 BDABADCDACCDCCADBCBDBCDDACADDC
090111000202
10002 BDCDCDCDADCBCCBDBDCABCBDACDABC
51
.
090111005302
090111005402
52.
10053
10054
BDABCDDDCCCDDCDBACBABABCMCDBDC
BDADCDCCDACBCCABBDADBBCBBDDDDC
Example 8: Items with various rating scale models
A 4 item test in which each item has a 9-level scoring rubric. We suspect that there are not really 9 levels of
competence. After several WINSTEPS analyses, we choose to recode the rating scale or partial credit items by
collapsing categories in order to increase measurement effectiveness (separation) and increase parameter
stability. An objection to collapsing is that it violates the Rasch model. This is only true if the uncollapsed data
strictly accord with the model. In fact, the collapsed data may fit the Rasch model better than than the
uncollapsed. We have to compare the collapsed and uncollapsed analyses to decide.
; This file is EXAM8.TXT
TITLE=&quot;Success and Failure Items&quot;
NAME1=6
ITEM1=1
NI=4
; 4 items
ISGROUPS=1234
; one item per grouping, same as ISGROUPS=0
IREFER=ABCD
; the 4 items are to be recoded differently. Item 1 is type &quot;A&quot;, etc.
CODES=123456789
; the codes in the data file
IVALUEA=333456666
; the recoding for A-type items in IREFER=, i.e., Item 1
IVALUEB=333455555
IVALUEC=333444444
IVALUED=444456777
&amp;END
Maze
Passengers
Blocks
Egg race
END NAMES
5536 M Richard
4545 F Tracie
4345 M Walter
3454 M Blaise
4435 M Ron
....
Table 14.3 shows the recoding:
ITEMS CATEGORY/OPTION/Distractor FREQUENCIES: ENTRY ORDER
+------------------------------------------------------------------------+
|ENTRY
DATA SCORE |
DATA
|
USED
AVERAGE OUTF|
|
|NUMBER CODE VALUE | COUNT
% | COUNT
%
MEASURE MNSQ| ITEM
|
|--------------------+------------+--------------------------+-----------|
|
1
2
3 |
1
3 |
1
3
-2.39
.3|Maze
|
|
3
3 |
7 22 |
6 19
-1.57
.7|
|
|
4
4 |
10 31 |
10 32
-.54
.5|
|
|
5
5 |
10 31 |
10 32
1.23
.6|
|
|
6
6 |
3
9 |
3 10
2.42
.7|
|
|
7
6 |
1
3 |
1
3
2.78
.5|
|
53.
Example 9: Grouping and modeling items
A 20 item test. Items 1, 2 are dichotomous items, coded &quot;Y&quot;, &quot;N&quot;. Item 3 is a &quot;Partial Credit&quot; item. Items 4-10 are
all ratings on one Andrich scale or test-level partial-credit scale (Grouping 1), and items 11-20 are all ratings on
another Andrich scale or test-level partial-credit scale (Grouping 2). These are grouped with ISGROUPS=.
Winsteps discovers from the data what the item structure is. Items 3-20 have response codes &quot;A&quot;, &quot;B&quot;, &quot;C&quot;, &quot;D&quot;,
&quot;E&quot; or &quot;a&quot;, &quot;b&quot;, &quot;c&quot;, &quot;d&quot;, &quot;e&quot;.
; This file is EXAM9.TXT
TITLE=&quot;Grouping and Modeling&quot;
ITEM1=11
; Item responses start in column 11
52
NI=20
; There are 20 items
ISGROUPS=DD011111112222222222 ; The groupings in item order
IREFER=DDAAAAAAAAAAAAAAAAAA
; for recoding
CODES = 10ABCDEabcde
; Response codes to all items
IVALUED= 10**********
; for Items 1 &amp; 2
IVALUEA= **1234512345
; for Items 3-20
DATA=EXAM9DAT.TXT
; Location of data file
IWEIGHT=*
; Item weighting file list
3 2.5
; Item 3 has weight of 2.5
*
&amp;END
RD Prompt 1 ; Item id's remind us of MODELS= and ISGROUPS=
RD Prompt 2 ; MODELS= and ISGROUPS= are shown in item measure Tables 10, 13, 14, 15
S0 Logic
R1 Grammar 1
R1 Grammar 2
R1 Grammar 3
R1 Grammar 4
R1 Grammar 5
R1 Grammar 6
R1 Grammar 7
R2 Meaning 1
R2 Meaning 2
R2 Meaning 3
R2 Meaning 4
R2 Meaning 5
R2 Meaning 6
R2 Meaning 7
R2 Meaning 8
R2 Meaning 9
R2 Meaning 10
END NAMES
The data is in file EXAM9DAT.TXT:
Richard
Tracie
.
James
Joe
54.
M 00bCDCDddCDddddCDccE
F 00BcBABBccbBbbBbBBBb
M 00ccaBbabBAcbacbaBbb
M 10BdBBBBccBccbbccbcC
Example 10: Combining tests with common items
This uses MFORMS=, but it can also be done, more awkwardly, with FORMAT=
Test A, in file EXAM10A.TXT, and TEST B, in EXAM10B.TXT, are both 20 item tests. They have 5 items in
common, but the distractors are not necessarily in the same order. The responses must be scored on an
individual test basis. Also the validity of each test is to be examined separately. Then one combined analysis is
wanted to equate the tests and obtain bankable item measures. For each file of original test responses, the
person information is in columns 1-11, the item responses in 41-60.
The combined data file specified in EXAM10C.TXT, is to be in RFILE= format. It contains
Person information of 11 characters: Columns 1-30 (always), but only 1-11 are active.
Item responses to 35 items: Columns 31-64
The identification of the common items is:
Test
Item Number (=Location in item string)
Bank: 1 2 3 4 5 6-20
21-35
A: 3 1 7 8 9 2,4-6,10-20
B: 4 5 6 2 11
1,3,7-10,12-20
53
I. From Test A, make a response (RFILE=) file..
; This file is EXAM10A.TXT
TITLE=&quot;Analysis of Test A&quot;
RFILE=EXAM10AR.TXT ; The constructed response file for Test A
NI=20
; 20 items
ITEM1=41
; Items start in column 41 of data record
NAME1=1
; Start of person label
NAMELEN=11
; Length of person label
CODES=&quot;ABCD# &quot;
; Beware of blanks and # meaning wrong!
; Blanks are included in CODES=, but they are scored incorrect, because
never keyed correct
KEY1=CCBDACABDADCBDCABBCA
; the MCQ key
&amp;END
BANK 2
TEST A 1
; first item name
BANK 6
TEST A 2
BANK 1
TEST A 3
.
BANK 20 TEST A 20
END NAMES
Person 01 A
BDABCDBDDACDBCACBDBA
.
Person 12 A
BADCACADCDABDDDCBACA
The RFILE= file, EXAM10AR.TXT, is:
0
1
2
3
4
5
12345678901234567890123456789012345678901234567890
Person 01 A
00000000110010001001
.........
Person 12 A
00001110000001001011
II. From Test B, make a response (RFILE=) file.
; This file is EXAM10B.TXT
TITLE=&quot;Analysis of Test B&quot;
RFILE=EXAM10BR.TXT ; The constructed response file for Test B
NI=20
ITEM1=41
; Items start in column 26 of reformatted record
NAME1=1
; Start of person label
NAMELEN=11
; Length of person label
CODES=&quot;ABCD# &quot;
; Beware of blanks meaning wrong!
KEY1=CDABCDBDABCADCBDBCAD
; Key in data record format
&amp;END
BANK 21 TEST B 1
BANK 4
TEST B 2
BANK 22 TEST B 3
.
BANK 35 TEST B 20
END NAMES
Person 01 B
BDABDDCDBBCCCCDAACBC
.
Person 12 B
BADABBADCBADBDBBBBBB
The RFILE= file, EXAM10BR.TXT, is:
Person 01 B
...
Person 12 B
01110101011001000100
00000001010000101000
III. Analyze Test A's and Test B's RFILE='s together:
; This file is EXAM10C.TXT
TITLE=&quot;Analysis of Tests A &amp; B (already scored)&quot;
NI=35
; 35 items in total
54
ITEM1=31
CODES=01
; reformatted data record
; scored right-wrong.
; Blanks ignored as &quot;not administered&quot;
MFORMS=*
; multiple data files in different formats
DATA=EXAM10AR.txt ; first file
L=1
; one line per data record
P1-11 = 1
; person id starts in column 1 of input file
I1=33
; item 1 is Test A's item 3 in Test A's column 33
I2=31
; item 2 is Test A's item 1 in column 31
I3-5=37
; items 3-5 are Test A's items 7-9 starting in column 37
I6=32
I7-9=34
I10-20=40
#
DATA=EXAM10BR.txt ; second data file
L=1
P1-11 = 1
; person id starts in column 1 of input file
I1-3 = 34
; items 1-3 are Test B's items 4-6 starting in Test B's column 34
I4 = 32
; item 4 is Test B's item 2 in column 32
I5 = 41
; items 5 in Test B's item 11 in column 41
I21 = 31
I22 = 33
I23-26 = 37
I27-35 = 42
*
&amp;END
BANK 1
TEST A 3 B 4
...
BANK 35 TEST B 20
END NAMES
The combined data file (which can be accessed from the Edit pull-down menu) is:
Person
...
Person
Person
...
Person
01 A
00001000010010001001
12 A
01 B
00100001100001001011
10111
010101001000100
12 B
00000
000101000101000
After running EXAM10C, I want to see two ICCs: One for test A and another Test B. How do I do this?
This graph is not produced directly by Winsteps, but can be produced in Excel.
After the Exam10C analysis,
55
use the &quot;Specification&quot; pull-down menu to delete items not in Test A: Use IDELETE=21-35
Display the Test Characteristic Curve
Select &quot;Copy data to clipboard&quot;.
Paste into an Excel worksheet
Use the &quot;Specification&quot; pull-down menu to reinstate all items: IDELETE=+1-35
Use the &quot;Specification&quot; pull-down menu to delete items not in Test B: Use IDELETE=6-20
Display the Test Characteristic Curve
Select &quot;Copy data to clipboard&quot;.
Paste into an Excel worksheet
In Excel, scatterplot the pasted columns.
55.
Example 11: Item responses two characters wide
The &quot;Liking for Science&quot; data (see RSA) is in file EXAM11.TXT. Each observation is on a rating scale where 0
means &quot;dislike&quot;, 1 means &quot;don't care/don't know&quot; and 2 means &quot;like&quot;. The data has been recorded in two columns
as 00, 01 and 02. XWIDE= is used.
; This file is EXAM11.TXT
TITLE='LIKING FOR SCIENCE (Wright &amp; Masters p.18)'
XWIDE=2
; Responses are 2 columns wide
CODES=000102
; Codes are 00 01 and 02
CLFILE=*
; Category label filelist
00 Dislike
; Category 00 in data file means &quot;Dislike&quot;
01 Neutral
02 Like
*
; End of category list
ITEM1=1
; Items start in column 1
NI=25
; 25 Items
NAME1=51
; Person id starts in column 51
NAMLMP=20
; Show 20 characters of id on maps
TABLES=111111111111111111111111111111 ; All Tables
CURVES=111
; Print all curves in Tables 2 and 21
IFILE = EXAM11IF.TXT
; Output item measure file
PFILE = EXAM11PF.TXT
; Output person measure file
SFILE = EXAM11SF.TXT
; Output structure calibration file
RFILE = EXAM11RF.TXT
; Output reformatted response file
XFILE = EXAM11XF.TXT
; Output observation and residual file
UIMEAN = 455
; User scaling: mean 455
USCALE = 94
; 94 user units per logit
LINLEN = 0
; Print with minimum of split lines
MAXPAG = 0
; Print with no page breaks in long tables
&amp;END
WATCH BIRDS
READ BOOKS ON ANIMALS
.
FIND OUT WHAT FLOWERS LIVE ON
TALK W/FRIENDS ABOUT PLANTS
END NAMES
01020101010002000102020202000201010202000201000200ROSSNER, MARC DANIEL
02020202020202020202020202020202020202020202020202ROSSNER, LAWRENCE F.
.
02020200000202000002020202020202020202000202000102PATRIARCA, RAY
01020000010100010102010202020201000202000200000100PAULING, LINUS
BLANK RECORD
56.
Example 12: Comparing high and low samples with rating scales
Rasch estimates are constructed to be as sample independent as is statistically possible, but you must still take
care to maintain comparability of measures across analyses. For instance, if a rating scale or partial credit
structure is used, and a high-low measure split is made, then the low rating scale (or partial credit) categories may
56
not appear in the data for the high measure sample and vice versa. To compare item calibrations for the two
samples requires the response structure to be calibrated on both samples together, and then the response
structure calibrations to be anchored for each sample separately. Comparison of patient measures from separate
analyses requires both the response structure calibrations and the item calibrations to share anchor calibrations.
35 arthritis patients have been through rehabilitation therapy. Their admission to therapy and discharge from
therapy measures are to be compared. They have been rated on the 13 mobility items of the Functional
Independence Measure (FIM™). Each item has seven levels. At admission, the patients could not perform at the
higher levels. At discharge, all patients had surpassed the lower levels (Data courtesy of C.V. Granger &amp; B.
Hamilton, ADS). A generic control file is in EXAM12.TXT. The admission ratings are in EXAM12LO.TXT and the
discharge ratings in EXAM12HI.TXT. Three analyses are performed: 1) joint analysis of the admission (low) and
discharge (high) data to obtain response structure calibrations, 2 &amp; 3) separate runs for the admission (low) and
discharge (high) data to obtain item calibrations. For a more complex situation, see Example 17.
; This common control file is EXAM12.TXT
TITLE='GENERIC ARTHRITIS FIM CONTROL FILE'
ITEM1=7
; Responses start in column 7
NI=13
; 13 mobility items
CODES=1234567
; 7 level rating scale
CLFILE=*
; Defines the rating scale
1 0% Independent
2 25% Independent
3 50% Independent
4 75% Independent
5 Supervision
6 Device
7 Independent
*
&amp;END
A. EATING
B. GROOMING
C. BATHING
D. UPPER BODY DRESSING
E. LOWER BODY DRESSING
F. TOILETING
G. BLADDER
H. BOWEL
I. BED TRANSFER
J. TOILET TRANSFER
K. TUB, SHOWER
L. WALK/WHEELCHAIR
M. STAIRS
57
END NAMES
The admission data is in file EXAM12LO.TXT:
21101
21170
|
22618
22693
5523133322121
4433443345454
Patient number in cols 1-5, ratings in 7-19
4433255542141
3524233421111
The discharge data is in file EXAM12HI.TXT:
21101
21170
|
22618
22693
5734366655453
6466677777676
Ratings generally higher than at admission
7667656666565
7776677676677
The batch file to run this is (see BATCH=), under Windows XP, 2000, EXAM12.CMD:
REM COMPARISON OF ITEM CALIBRATIONS FOR HIGH AND LOW SAMPLES
START /WAIT ..\WINSTEPS EXAM12.TXT EXAM12OU.TXT DATA=EXAM12LO.TXT+EXAM12HI.TXT
TITLE=ADMIT+DISCHARGE SFILE=EXAM12SF.TXT BATCH=Y
START /WAIT ..\WINSTEPS EXAM12.TXT EXAM12LU.TXT DATA=EXAM12LOF.TXT TITLE=ADMIT
SAFILE=EXAM12SF.TXT IFILE=EXAM12LI.TXT BATCH=Y
START /WAIT ..\WINSTEPS EXAM12.TXT EXAM12HU.TXT DATA=EXAM12HIF.TXT TITLE=DISCHARGE
SAFILE=EXAM12SF.TXT IFILE=EXAM12HI.TXT BATCH=Y
under WINDOWS-95 or -98, EXAM12BAT.BAT:
REM COMPARISON OF ITEM CALIBRATIONS FOR HIGH AND LOW SAMPLES
START /w ..\WINSTEPS EXAM12.TXT EXAM12OU.TXT DATA=EXAM12LO.TXT+EXAM12HI.TXT
TITLE=ADMIT+DISCHARGE SFILE=EXAM12SF.TXT BATCH=Y
START /w ..\WINSTEPS EXAM12.TXT EXAM12LU.TXT DATA=EXAM12LO.TXT TITLE=ADMIT
SAFILE=EXAM12SF.TXT IFILE=EXAM12LIF.TXT BATCH=Y
START /w ..\WINSTEPS EXAM12.TXT EXAM12HU.TXT DATA=EXAM12HI.TXT TITLE=DISCHARGE
SAFILE=EXAM12SF.TXT IFILE=EXAM12HIF.TXT BATCH=Y
Under WINDOWS-NT (early versions), EXAM12NT.BAT:
REM COMPARISON OF ITEM CALIBRATIONS FOR HIGH AND LOW SAMPLES
..\WINSTEPS EXAM12.TXT EXAM12OU.TXT DATA=EXAM12LO.TXT+EXAM12HI.TXT TITLE=ADMIT&amp;DISCHARGE
SFILE=EXAM12SF.TXT BATCH=Y
..\WINSTEPS EXAM12.TXT EXAM12LU.TXT DATA=EXAM12LO.TXT TITLE=ADMIT SAFILE=EXAM12SF.TXT
IFILE=EXAM12LIF.TXT BATCH=Y
..\WINSTEPS EXAM12.TXT EXAM12HU.TXT DATA=EXAM12HI.TXT TITLE=DISCHARGE SAFILE=EXAM12SF.TXT
IFILE=EXAM12HIF.TXT BATCH=Y
To run this, select &quot;Run batch file&quot; from &quot;Batch&quot; pull-down menu, and right-click on &quot;Exam12bat.bat&quot; or
&quot;Exam12cmd.cmd&quot; in the dialog box, then left-click on &quot;open&quot;.
The shared structure calibration anchor file is EXAM12SF.TXT:
;
;
;
;
structure measure FILE FOR
ADMIT&amp;DISCHARGE
May 23 13:56 1993
CATEGORY structure measure
1
.00
2 -2.11
3 -1.61
4 -1.25
5
.06
6
1.92
7
2.99
58
The item calibrations measures for admission and discharge are written into IFILE= files, with comma-separated
values (CSV=Y), so that they can easily be imported into a spreadsheet.
57.
Example 13: Paired comparisons as the basis for measurement
Paired comparisons can be modeled directly with the Facets computer program. For WINSTEPS a dummy facet
of &quot;occasion&quot; must be introduced. On each occasion (in this example, each column), there is a winner '1', a loser
'0', or a draw 'D' recorded for the two players. In column 1 of the response data in this example, Browne (1)
defeated Mariotti (0). In column 2, Browne (D) drew with Tatai (D). Specifying PAIRED=YES adjusts the
measures for the statistical bias introduced by this stratagem. Each player receives a measure and fit statistics.
Occasion measures are the average of the two players participating. Misfitting occasions are unexpected
outcomes. Point-biserial correlations have little meaning. Check the occasion summary statistics in Table 3 to
verify that all occasions have the same raw score.
; This common control file is EXAM13.TXT
TITLE = 'Chess Matches at the Venice Tournament, 1971'
Name1 = 1
; Player's name
Item1 = 11
; First match results
PERSON = PLAYER
ITEM
= MATCH
; Example of paired comparison
CODES = 0D1
; 0 = loss, D = draw (non-numeric), 1 = win
; if you wish to just consider won-loss, and ignore the draws, omit the following line:
NEWSCORE = 012
; 0 = loss, 1 = draw, 2 = win
CLFILE=*
0 Loss
D Draw
1 Win
*
NI
= 66
; 66 matches (columns) in total
PAIRED = YES
; specify the paired comparison adjustment
INUMBER = YES
; number the matches in the Output
&amp;END
Browne
1D 0 1
1
1
1
D
D
1
1
Mariotti 0 1 D 0
1
1
1
D
1
D
1
Tatai
D0 0 1
D
D
1
1
1
1
D
Hort
1D1
D
D
1
D
D
D
1
0
Kavalek
010D
D
D
1
D
1
1
D
Damjanovic
00DDD
D
D
D
1
D
1
Gligoric
00D0DD
D
1
1
1
0
Radulov
000D0DD
D
1
D
1
Bobotsov
DD0DDD0D
0
0
1
Cosulich
D00D00001
1
1
Westerinen
0D000D0D10
1
Zichichi
00D1D010000
Part of the output is:
PLAYER STATISTICS: MEASURE ORDER
+---------------------------------------------------------------------+
|ENTRY
RAW
|
INFIT | OUTFIT |
|
|NUMBR SCORE COUNT MEASURE ERROR|MNSQ ZSTD|MNSQ ZSTD| PLAYER
|
|-----------------------------------+----------+----------+-----------|
|
1
17
11
1.09
.35|1.10
.2|1.02
.1| Browne
|
|
2
15
11
.68
.32|1.02
.0| .96
-.1| Mariotti |
|
3
14
11
.50
.31| .86
-.4| .83
-.5| Tatai
|
|
4
14
11
.50
.31|1.34
.9|1.54
1.3| Hort
|
|
5
13
11
.33
.31| .81
-.6| .80
-.6| Kavalek
|
|
6
11
11
.00
.30| .35 -2.8| .37 -2.6| Damjanovic|
|
7
10
11
-.17
.30| .90
-.3| .91
-.3| Gligoric |
|
8
9
11
-.34
.31| .52 -1.8| .52 -1.7| Radulov
|
|
9
8
11
-.51
.31|1.00
.0|1.00
.0| Bobotsov |
59
|
10
8
11
-.51
.31|1.18
.5|1.15
.4| Cosulich |
|
11
7
11
-.69
.32| .95
-.1| .89
-.3| Westerinen|
|
12
6
11
-.88
.33|1.86
1.8|1.90
1.7| Zichichi |
|-----------------------------------+----------+----------+-----------|
| MEAN
11.
11.
.00
.32| .99
-.2| .99
-.2|
|
+---------------------------------------------------------------------+
58.
Example 14: Multiple rescorings, response structures and widths
Introductory example: A test has 30 multiple choice question (keyed A, B, C, D) and 5 essay questions, each has
its own rating scale definition (scored 0, 1, or 2), i.e., they accord with the partial credit model. ISGROUPS= is
used to identify the partial credit items. IREFER= and IVALUE= are used for scoring the questions
; This control file is EXAM14.TXT
TITLE = &quot;AN EXAMPLE OF MULTIPLE ITEM TYPES&quot;
; THIS PART DEFINES THE ANALYTICAL MODEL
; MCQ ARE ONE GROUPING, WHICH WE WILL CALL &quot;M&quot; - PARTIAL CREDIT ITEMS ARE GROUPED AS &quot;0&quot;
; &quot;0&quot; MEANS EACH ITEM IS IN A GROUPING BY ITSELF, i.e., partial credit
ISGROUPS = MMMMMMMMMMMMMMMMMMMMMMMMMMMMMM00000
; THIS PART DEFINES THE RESPONSE SCORING
; The MCQ key is: ABCAAACBAAABABABBAAABABBAAABAB
IREFER = ABCAAACBAAABABABBAAABABBAAABAB00000 ; REFERENCE THE MCQ ITEMS BY THEIR SCORING KEY
CODES = ABCD012; ALL VALID CODES IN THE DATA FILE
IVALUE0= 0000012; THESE ARE THE ESSAY ITEMS
IVALUEA= 1000000 ; MCQ ITEMS WITH A AS THE CORRECT ANSWER
IVALUEB= 0100000 ; B IS CORRECT
IVALUEC= 0010000 ; C IS CORRECT
IVALUED= 0001000 ; D IS CORRECT
MISSING-VALUES-SCORED = 0 ; SO ALL 9'S (OR ANYTHING ELSE NOT IN CODES=) ARE SCORED 0
ITEM1 = 1
; START OF ITEM STRING
NI
= 35
; NUMBER OF ITEM RESPONSES
NAME1 = 37
; START OF PERSON IDENTIFICATION
&amp;END
; ITEM LABELS HERE
END LABELS
ACAACBAA9DC9AB9C9DCC9ABBA9AACB00102 1
; DATA ARE HERE
A9D9AAAB9DA999AA9BBD999999B9AA12100
A9C9AACD9AB9AB9B99D9DB9ACBD9AB10120
A9C9AACB9AD99999CA9ABD9999999900120
A9C9AAAB9AA99999CD9ABB9999999910120
A9C9ADCBBAA9CDABD9DACCABBA9ABC21212
B9D9AAAB9CB99999DB9ABC9999999900100
A9D9BAAB9AA99999BA9ABB9999999900101
A9D9BACB9DC99999CA9DBB9999999921201
A9A9AACD9BA99999AD9ABB9999999910120
A9C9AACB9AA99999DC9ABD9999999900120
A9C9AAAB9CA99999BA9CBB9999999921201
A9C9CACB9AC99999CB9ABD9999999910120
A9D9AACB9AA99999AA9ABD999999D900000
A9C9DDCB9AA99999CB9ABD99999C9D21201
A9C9AABB9AD9999CD9ABC9999ABDAB11110
A9CAACB9ABBC9ADBB9ABDABBA9ADCB00120
CBBAB9CAAC9BBBC9BCBACDD9ADDCAB10120
A9C9BAAB9AD99CB9BBBA9ABDACDD9A00120
C9D9BDDB9BBBB9ACBADBC9AADBBCBC21201
A9C9AABB9DABA9ABDDABCABBA9AACB00120
A9D9BDCB9DCAC9DBBADBBACBA9ABAC00001
A9C9BACB9DADA9DBBDABBACBA9ABBB21201
A9D9BACC9AADC9DBBAABBADBC9ABCB10012
A9D9BAAB9ABCC9ABBDABBACBB9ABBB21201
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Example 14: A test comprises multiple-choice questions (MCQ), true-false, and two different response structure
formats. These are combined in one data file. The MCQ scoring key, and also the True-False scoring key are
used as item cluster references in the IREFER= and IVALUE= specifications to simplify key checking.
EXAM14DT.TXT data format is:
60
Cols 1-5
Cols 6-10
Cols 11-15
correct
Cols 16-25
Cols 26-29
Person id.
5 MCQ items (A,B,C,D, # = missing, wrong. Some items may be miskeyed
5 True-False items, responses (S, N). For some of these S=&quot;True&quot; is correct, for others &quot;N&quot;=False is
10 Rating scale items (N=0,P=1,M=2,S=3). Some items may be reversed.
2 Evaluation items (0-12). - (See exam14b.txt) - second analysis only.
First analysis: all responses one column wide
The control file, EXAM14.TXT, is:
; This file is EXAM14.TXT
TITLE=&quot;Multiple Response Formats, Same Response Width&quot;
DATA=exam14dt.txt
;EXAM14DT.TXT data format is
;Cols 1-5 Person id.
;Cols 6-10 5 MCQ items (A,B,C,D, # = missing, wrong)
; Some items may be miskeyed
;Cols 11-15 5 True-False items, responses (S, N)
; For some of these S=&quot;True&quot; is correct, for others &quot;N&quot;=False is correct
;Cols 16-25 10 Rating scale items (N=0,P=1,M=2,S=3)
; Some items may be reversed.
;Cols 26-29 2 Evaluation items (0-12). - (See exam14b.con)
NAME1=1
ITEM1=6
NI=20
; THESE CODES LINE UP WITH THE ITEM COLUMNS
;
0
1
2
;
12345678901234567890
; TO DEFINE RESPONSE STRUCTURE CLUSTERS
ISGROUPS = 11111222223333333333
; TO DEFINE RESCORING CLUSTERS
IREFER = BACDCSNSNSRRRRRRRRRR
; IREFER = X MATCHES IVALUEX=
; IVALUE?= MATCHES WITH CODES=
CODES
= ABCD#SNPM ; CODES IN ORIGINAL DATA FILE
IVALUEA = 10000**** ; MCQ RESPONSE A IS CORRECT
IVALUEB = 01000**** ; MCQ RESPONSE B IS CORRECT
IVALUEC = 00100**** ; MCQ RESPONSE C IS CORRECT
IVALUED = 00010**** ; MCQ RESPONSE D IS CORRECT
IVALUES = *****10** ; &quot;S&quot; IS THE CORRECT ANSWER
IVALUEN = *****01** ; &quot;N&quot; IS THE CORRECT ANSWER
IVALUER = *****3012 ; &quot;NPMS&quot; RATING SCALE
STKEEP=YES ; KEEP UNUSED INTERMEDIATE CATEGORIES IN RATING SCALES
INUMBER=YES ; NO ITEM INFORMATION AVAILABLE
&amp;END
Second analysis: responses one and two columns wide
Including the last two items with long numeric response structures and 2-column format, the control file becomes
EXAM14B.TXT. Since some responses are in 2 column format, XWIDE=2. FORMAT= is used to transform all
responses into XWIDE=2 format.
EXAM14B.TXT is:
&amp;INST
61
TITLE=&quot;Multiple Response Formats, Same Response Width&quot;
DATA=EXAM14DT.TXT
;EXAM14DT.TXT data format is
;Cols 1-5 Person id.
;Cols 6-10 5 MCQ items (A,B,C,D, # = missing, wrong)
; Some items may be miskeyed
;Cols 11-15 5 True-False items, responses (S, N)
; For some of these S=&quot;True&quot; is correct, for others &quot;N&quot;=False is correct
;Cols 16-25 10 Rating scale items (N=0,P=1,M=2,S=3)
; Some items may be reversed.
;Cols 26-29 2 Evaluation items (0-12)
NAME1=1
ITEM1=6
NI=22
;20-1 COLUMN + 2 2-COLUMN ITEMS
XWIDE=2 ;XWIDE FOR WIDEST FIELD
FORMAT=(5A1, 20A1,2A2) ; PERSON LABEL &amp; FIRST 20 ITEMS 1 CHARACTER COLUMNS
;
; LAST 2 ITEMS ARE 2 CHARACTER COLUMNS
; THESE ARE SET UP FOR XWIDE=2
; FOR RESPONSE STRUCTURE DEFINITIONS
; TO DEFINE RATING SCALE CLUSTERS
ISGROUPS = 1111122222333333333344
IREFER = BACDCSNSNSRRRRRRRRRREE
; IREFER = X MATCHES IVALUEX=
; IVALUE?= MATCHES WITH CODES=
; CODES ARE 2 CHARACTERS WIDES
CODES
= &quot;A B C D # S N P M 121110 9 8 7 6 5 4 3
IVALUEA = &quot;1 0 0 0 0 * * * * * * * * * * * * * *
IVALUEB = &quot;0 1 0 0 0 * * * * * * * * * * * * * *
IVALUEC = &quot;0 0 1 0 0 * * * * * * * * * * * * * *
IVALUED = &quot;0 0 0 1 0 * * * * * * * * * * * * * *
IVALUES = &quot;* * * * * 1 0 * * * * * * * * * * * *
IVALUEN = &quot;* * * * * 0 1 * * * * * * * * * * * *
IVALUER = &quot;* * * * * 3 0 1 2 * * * * * * * * * *
IVALUEE = &quot;* * * * * * * * * 121110 9 8 7 6 5 4 3
STKEEP=YES ; KEEP UNUSED INTERMEDIATE CATEGORIES
INUMBER=YES ; NO ITEM INFORMATION AVAILABLE
&amp;END
2 1 0&quot;; CODES IN DATA
* * *&quot;; MCQ A IS CORRECT
* * *&quot;; MCQ B IS CORRECT
* * *&quot;; MCQ C IS CORRECT
* * *&quot;; MCQ D IS CORRECT
* * *&quot;; &quot;S&quot; IS CORRECT
* * *&quot;; &quot;N&quot; IS CORRECT
* * *&quot;; &quot;NPMS&quot; RATING SCALE
2 1 0&quot;; 0-12 RATING SCALE
IN RATING SCALES
This can also be done with MFORMS=
EXAM14C.TXT is:
; This file is EXAM14C.TXT
&amp;INST
TITLE=&quot;Multiple Response Formats, Same Response Width&quot;
;EXAM14DT.TXT data format is
;Cols 1-5 Person id.
;Cols 6-10 5 MCQ items (A,B,C,D, # = missing, wrong)
; Some items may be miskeyed
;Cols 11-15 5 True-False items, responses (S, N)
; For some of these S=&quot;True&quot; is correct, for others &quot;N&quot;=False is correct
;Cols 16-25 10 Rating scale items (N=0,P=1,M=2,S=3)
; Some items may be reversed.
;Cols 26-29 2 Evaluation items (0-12)
; Reformatted data record is:
;Cols 1-5 Person id
;Cols 6-7 Item 1 = original Col. 6
62
;Cols 8-9 Item 2 = original Col. 7
;.....
;Cols 48-49 Item 22
NAME1=1
ITEM1=6
NI=22
XWIDE=2
;20-1 COLUMN + 2 2-COLUMN ITEMS
;XWIDE FOR WIDEST FIELD
mforms=*
data=exam14dt.txt ; the name of an input data file
L = 1
; the are 2 lines in input data file for each data record
P1-5 = 1
; person label characters 1 through 5 start in column 1
; in the following &quot;C&quot; is used because &quot;I&quot; uses XWIDE=2
C6 = 6
; original item 1 in column 6 goes in column 6
C8 = 7
; original item 2 in column 7 goes in column 8
C10 = 8
C12 = 9
C14 = 10
C16 = 11
C18 = 12
C20 = 13
C22 = 14
C24 = 15
C26 = 16
C28 = 17
C30 = 18
C32 = 19
C34 = 20
C36 = 21
C38 = 22
C40 = 23
C42 = 24
C44 = 25
; original item 20 in column 25 goes in column 44-45
I21-22 = 26
; two-character items 21 and 22 start in column 26
*
; end of mforms= command
; THESE ARE SET UP FOR XWIDE=2 FOR RESPONSE STRUCTURE DEFINITIONS
; TO DEFINE RATING SCALE CLUSTERS
ISGROUPS = 1111122222333333333344
IREFER = BACDCSNSNSRRRRRRRRRREE
; IREFER = X MATCHES IVALUEX=
; IVALUE?= MATCHES WITH CODES=
; CODES ARE 2 CHARACTERS WIDES
CODES
= &quot;A B C D # S N P M 121110 9 8 7 6 5 4 3
IVALUEA = &quot;1 0 0 0 0 * * * * * * * * * * * * * *
IVALUEB = &quot;0 1 0 0 0 * * * * * * * * * * * * * *
IVALUEC = &quot;0 0 1 0 0 * * * * * * * * * * * * * *
IVALUED = &quot;0 0 0 1 0 * * * * * * * * * * * * * *
IVALUES = &quot;* * * * * 1 0 * * * * * * * * * * * *
IVALUEN = &quot;* * * * * 0 1 * * * * * * * * * * * *
IVALUER = &quot;* * * * * 3 0 1 2 * * * * * * * * * *
IVALUEE = &quot;* * * * * * * * * 121110 9 8 7 6 5 4 3
STKEEP=YES ; KEEP UNUSED INTERMEDIATE CATEGORIES
INUMBER=YES ; NO ITEM INFORMATION AVAILABLE
&amp;END
63
2 1 0&quot;; CODES IN DATA
* * *&quot;; MCQ A IS CORRECT
* * *&quot;; MCQ B IS CORRECT
* * *&quot;; MCQ C IS CORRECT
* * *&quot;; MCQ D IS CORRECT
* * *&quot;; &quot;S&quot; IS CORRECT
* * *&quot;; &quot;N&quot; IS CORRECT
* * *&quot;; &quot;NPMS&quot; RATING SCALE
2 1 0&quot;; 0-12 RATING SCALE
IN RATING SCALES
59.
Example 15: Figure skating: Multidimensionality, DIF or Bias
The Pairs Skating competition at the 2002 Winter Olympics in Salt Lake City was contentious. It resulted in the
awarding of Gold Medals to both a Russian and a Canadian pair, after the French judge admitted to awarding
biased scores. Multidimensionality, differential item functioning, and item bias are all manifestations of disparate
subdimensions within the data. In judged competitions, judge behavior can introduce unwanted subdimensions.
For this analysis, each pair is allowed to have a different skill level, i.e., different measure, on each skill of each
performance. The judges are modeled to maintain their leniencies across all performances. The control file and
data are in exam15.txt.
; This control file is EXAM15.TXT
Title = &quot;Pairs Skating: Winter Olympics, SLC 2002&quot;
Item
= Judge
Person = Pair
NI
= 9
; the judges
Item1 = 14
; the leading blank of the first rating
Xwide = 3
; Observations are 3 CHARACTERS WIDE for convenience
NAME1 = 1
; start of person identification
NAMELENGTH = 13
; 13 characters identifiers
; CODES NEXT LINE HAS ALL OBSERVED RATING SCORES
CODES= &quot; 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44+
+ 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60&quot;
STEPKEEP=YES
; maintain missing intermediate rating scores in the scoring structure
@order = 1-2
; order number at finish of competition in person label columns 1-2
DIF = @order
; judge &quot;DIF&quot; across skating pairs
tfile=*
30
; produce Table 30 for judge &quot;DIF&quot;
*
&amp;END
1 Rus ;Mrs. Marina SANAIA : RUSSIA
2 Chn ;Mr. Jiasheng YANG : CHINA
3 USA ;Mrs. Lucy BRENNAN : USA
4 Fra ;Miss Marie Reine LE GOUGNE : FRANCE
5 Pol ;Mrs. Anna SIEROCKA : POLAND
6 Can ;Mr. Benoit LAVOIE : CANADA
7 Ukr ;Mr. Vladislav PETUKHOV : UKRAINE
8 Ger ;Mrs. Sissy KRICK : GERMANY
9 Jap ;Mr. Hideo SUGITA : JAPAN
; Description of Person Identifiers
; Cols. Desciption
; 1-2 Order immediately after competition (@order)
; 4-5 Skaters' initials
; 7-9 Nationality
; 11
Program: S=Short F=Free
; 13
Skill: T=Technical Merit, A=Artistic Impression
END LABELS
1 BS-Rus S T 58 58 57 58 58 58 58 58 57 ; 1 BEREZHNAYA
1 BS-Rus S A 58 58 58 58 59 58 58 58 58 ; 2 BEREZHNAYA
1 BS-Rus F T 58 58 57 58 57 57 58 58 57 ; 3 BEREZHNAYA
1 BS-Rus F A 59 59 59 59 59 58 59 58 59 ; 4 BEREZHNAYA
2 SP-Can S T 57 57 56 57 58 58 57 58 56 ; 5 SALE Jamie
2 SP-Can S A 58 59 58 58 58 59 58 59 58 ; 6 SALE Jamie
2 SP-Can F T 58 59 58 58 58 59 58 59 58 ; 7 SALE Jamie
2 SP-Can F A 58 58 59 58 58 59 58 59 59 ; 8 SALE Jamie
3 SZ-Chn S T 57 58 56 57 57 57 56 57 56 ; 9 SHEN Xue /
.....
Elena / SIKHARULIDZE Anton
Elena / SIKHARULIDZE Anton
Elena / SIKHARULIDZE Anton
Elena / SIKHARULIDZE Anton
/ PELLETIER David : CAN
/ PELLETIER David : CAN
/ PELLETIER David : CAN
/ PELLETIER David : CAN
ZHAO Hongbo : CHN
:
:
:
:
RUS
RUS
RUS
RUS
From this data file, estimate judge severity. In my run this took 738 iterations, because the data are so thin, and
the rating scale is so long.
Here is some of the output of Table 30, for Judge DIF, i.e., Judge Bias by skater pair order number, @order =
$S1W2.
+-------------------------------------------------------------------------+
| Pair
DIF
DIF Pair
DIF
DIF
DIF
JOINT
Judge
|
| CLASS ADDED S.E. CLASS ADDED S.E. CONTRAST S.E.
t
Number Name |
64
|-------------------------------------------------------------------------|
| 13
-.93
.40 18
1.50
.39
-2.43
.56 -4.35
9 9 Jap |
| 14
-1.08
.36 18
1.50
.39
-2.58
.53 -4.83
9 9 Jap |
+-------------------------------------------------------------------------+
The most significant statistical bias is by the Japanese judge on skater pairs 13 and 14 vs. 18. These pairs are
low in the final order, and so of little interest.
Table 23, the principal components/contrast analysis of Judge residuals is more interesting. Note that Judge 4,
the French judge, is at the top with the largest contrast loading. The actual distortion in the measurement
framework is small, but crucial to the awarding of the Gold Medal!
.6
.5
C
O
N
T
R
A
S
T
.4
.3
.2
.1
1
.0
L
O
A
D
I
N
G
-.1
-.2
-.3
-.4
-.5
60.
STANDARDIZED RESIDUAL CONTRAST PLOT
-1
0
1
++--------------------------------+--------------------------------++
+
4
|
+
|
|
|
+
| 5
7
+
|
|
|
+
1
|
+
|
|
|
+
|
+
|
|
|
+
|
+
|
|
|
+
|
+
|
|
|
+---------------------------------|---------------------------------+
|
2
|
|
+
|
+
|
|
|
+
|
+
|
|
|
+
|
+
|
| 9
|
+
|
8
+
|
|
6
|
+
| 3
+
|
|
|
++--------------------------------+--------------------------------++
-1
0
1
Judge MEASURE
Example 16: Disordered categories - thresholds and person anchoring
This example illustrates two types of disordering in rating scale structures: category disordering and RaschAndrich threshold disordering. For further discussion, see disordering. It also illustrates anchor values in the
person labels and category labeling.
The control file and data are in exam16.txt.
; This control file is EXAM16.TXT
title = &quot;Attitude Survey illustrating two types of disordering&quot;
ni
= 3
; three items
item1 = 1
; item responses start in column 1
xwide = 2
; data codes are two columns wide
codes = &quot;1 2 3 &quot; ; valid two character codes
name1 = 9
; person label starts in column 9
namelength = 3
; length of person labels
; pafile = $s1w3 ; person anchor values in columns 1-3 of person labels
@panchors = 1-3 ; person anchor field is in columns 1 to 3 of person labels
pafile = @panchors ; person anchors in field @panchors
ISGROUPS = 0
clfile = *
1+1 Never
;
1+2 Sometimes
1+3 0ften
2+1 Car
;
2+2 Ship
;
2+3 Plane
;
3+1 No
3+2 Neutral
3+3 Yes
; allow each item to have its own rating scale (the partial credit model)
; category labels: item+category
item 1 category 1 is &quot;Never&quot;
; well-behaved rating scale
categories as ordered in frequency in 1930
now these categories are disordered
ship travel now rarer than planes
; very few in this narrow intermediate category
65
*
&amp;END
; item labels follow
Smooth advance everywhere - probability curves a &quot;range of hills&quot;
Disordered categories - disordered empirical average measures for categories
Low middle frequency - high discrimination - disordered Rasch-Andrich thresholds
END LABELS
; end of item labels, data follow ...
1 1 1
0.0 ; person anchor logits in person labels
1 1 1
0.1 ; advancing anchored person measures
2 3 1
0.2 ; but not advancing categories for item 2
2 3 1
0.3
2 2 2
0.4 ; only one observation of category 2 for item 3, but in the correct place =
high discrimination
2 2 3
0.5
2 2 3
0.6
3 2 3
0.7
3 2 3
0.8 ; last data record
On the Diagnosis Menu: Empirical Item-Category Measures:
TABLE 2.5 Attitude Survey
OBSERVED AVERAGE MEASURES FOR PERSONS
0
1
2
3
4
5
+-----+-----+-----+-----+-----| NUM
|
1
2
3
|
1
|
1
3 2
|
2
|
1 2 3
|
3
+-----+-----+-----+-----+-----| NUM
(BY OBSERVED CATEGORY)
ITEM
Smooth advance everywhere
Disordered categories
Low middle frequency - high discrimination
ITEM
On the Diagnosis Menu: Category Function:
TABLE 3.2 Attitude Survey
FOR GROUPING &quot;0&quot; ITEM NUMBER: 1
Smooth advance everywhere - probability curves a &quot;range of hills&quot;
ITEM ITEM DIFFICULTY MEASURE OF 2.00 ADDED TO MEASURES
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ||CALIBRATN| MEASURE|
|-------------------+------------+------------++---------+--------+
| 1
1
2 22| 1.00 1.62|
.49
.52|| NONE
|( -.29)| 1 1 Never
| 2
2
5 56| 2.00 2.00|
.00
.00||
-1.11 |
2.00 | 2 2 Sometimes &lt;- Ordered RaschAndrich thresholds
| 3
3
2 22| 3.00 2.38|
.49
.52||
1.11 |( 4.29)| 3 3 0ften &lt;- Ordered categories
+------------------------------------------------------------------
TABLE 3.3 Attitude Survey
FOR GROUPING &quot;0&quot; ITEM NUMBER: 2
Disordered categories - disordered empirical average measures for
66
categories
ITEM ITEM DIFFICULTY MEASURE OF 2.00 ADDED TO MEASURES
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ||CALIBRATN| MEASURE|
|-------------------+------------+------------++---------+--------+
| 1
1
2 22| 1.00 1.62|
.49
.52|| NONE
|( -.29)| 1 1 Car
| 2
2
5 56| 2.40 2.00| 1.01 1.01||
-1.11 |
2.00 | 2 2 Ship
thresholds
| 3
3
2 22| 2.00* 2.38| 1.28 1.23||
1.11 |( 4.29)| 3 3 Plane
+------------------------------------------------------------------
TABLE 3.4 Attitude Survey
FOR GROUPING &quot;0&quot; ITEM NUMBER: 3
thresholds
&lt;- Ordered Rasch-Andrich
&lt;- Disordered categories
Low middle frequency - high discrimination - disordered Rasch-Andrich
ITEM ITEM DIFFICULTY MEASURE OF 2.00 ADDED TO MEASURES
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ||CALIBRATN| MEASURE|
|-------------------+------------+------------++---------+--------+
| 1
1
4 44| 1.50 1.65|
.74
.64|| NONE
|(
.86)| 1 1 No
| 2
2
1 11| 2.00 2.00|
.00
.00||
1.21 |
2.00 | 2 2 Neutral
Andrich thresholds
| 3
3
4 44| 2.50 2.35|
.74
.64||
-1.21 |( 3.14)| 3 3 Yes
+------------------------------------------------------------------
67
&lt;- Disordered Rasch&lt;- Ordered categories
61.
Example 17: Rack, stack, principal components
Comparisons of measures at two time-points or in two situations can become complicated. Example 12 illustrates
a straightforward situation. It uses the 13 motor items of the FIM&reg;. This example uses the 18 items of the FIM, 13
motor and 5 cognitive, at two time points. There is a 7-level rating scale that is intended to operate in the same
way across all items.
In exam17s.txt, the data have been stacked. The items are modeled to maintain their difficulties across the two
time points, and the 52 patients are entered twice, once at admission to rehabilitation and once at discharge from
it, so there 104 data records. Changes in patient independence can be identified by cross-plotting the admission
and discharge measures for each patient, as in Example 12. This can be done by using the Plots menu to plot the
measures against themselves, and then, in EXCEL, pasting the discharge measures over the top of the
admission measures for the y-axis.
In exam17r.txt, the data have been racked. The persons are modeled to maintain their abilities across the two
time points, and the 18 FIM items are entered twice, once at admission to rehabilitation and once at discharge
from it, so there 36 items. Average changes in patient performance on individual items can be identified by crossplotting the admission and discharge measures for each item. This can be done by using the Plots menu to plot
the measures against themselves, and then, in EXCEL, pasting the discharge measures over the top of the
admission measures for the y-axis.
A further feature is the contrasts in the Principal Components Analysis of Residuals, Tables 23 and 24. The 13
motor items and 5 cognitive items are probing different aspects of patient independence. Do they function as one
variable in this sample? See Principal components/contrast. Patients also have different impairments. Are their
measures comparable? See Differential Item Function.
; This control
TITLE='Example
ITEM1=1
NI=18
NAME1=20
; variables in
@SETTING=$S1W1
@SEX=$S3W1
@IGC=$S5W2
13=Other
@ID=$S8W2
file is EXAM17S.TXT
17S: 18-item FIM control file: stacked admission then discharge'
; Responses start in column 7
; 13 motor items + 5 cognitive items
; start of person label
person label
; setting: A=Admission, D=Discharge
; Gender: 1=Male, 2=Female
; Impairment Group Code: 8=Orthopedic, 9=Cardiac, 10=Pulmonary, 11=Burns, 12=Congenital,
; Patient sequ. number
; variables in item label
@SUBTEST=$S1W1 ; subtest: M=Motor, C=Cognitive
@ILETTER=$S3W1 ; code letter of item
CODES=1234567 ; 7 level rating scale
CLFILE=*
; Defines the rating scale
1 0% Independent
2 25% Independent
3 50% Independent
4 75% Independent
5 Supervision
6 Device
7 Independent
*
&amp;END
M A. EATING
....
C R. MEMORY
END NAMES
334412312331123112 A 1 8 1
432322442223134444 A 2 8 2
....
; This control
TITLE='Example
ITEM1=1
NI=37
NAME1=39
file is EXAM17R.TXT
17: 18-item FIM control file: racked admission and discharge'
; Responses start in column 7
; 13 motor items + 5 cognitive items: admission and discharge
; start of person label
68
; variables in person label
@SEX=$S1W1
; Gender: 1=Male, 2=Female
@IGC=$S3W2
; Impairment Group Code: 8=Orthopedic, 9=Cardiac, 10=Pulmonary, 11=Burns, 12=Congenital,
13=Other
@ID=$S5W2
; Patient sequ. number
; variable in item label
@SETTING=$S1W1 ; setting: A=Admission, D=Discharge
@SUBTEST=$S3W1 ; subtest: M=Motor, C=Cognitive
@ILETTER=$S5W1 ; code letter of item
CODES=1234567 ; 7 level rating scale
CLFILE=*
; Defines the rating scale
1 0% Independent
2 25% Independent
3 50% Independent
4 75% Independent
5 Supervision
6 Device
7 Independent
*
&amp;END
A M A. EATING
....
A C R. MEMORY
Blank for ease of seeing
D M A. EATING
....
D C R. MEMORY
END NAMES
334412312331123112 554535546665345545 1 8
432322442223134444 777677777666567777 2 8
....
62.
1
2
ALPHANUM alphabetic numbering
Normally XWIDE=1 limits the numerical score range to 0-9, but with ALPHANUM= this can be extended much
further.
ALPHANUM= is followed by a string of characters that represent the cardinal numbers in order starting with 0, 1,
2, 3, ...
Example:
Represent the numbers 0-20 using XWIDE=1
ALPHANUM = 0123456789ABCDEFGHIJK ; then A=10, B=11, .., K=20
XWIDE = 1
CODES = 0123456789ABCDEFGHIJK
NI = 5
ITEM1 = 1
&amp;END
21BF3 Mary
; Mary's responses are 2, 1, 11, 15, 3
K432A Mark
; Mark's responses are 20, 4, 3, 2, 10
63.
ASCII output only ASCII characters
MS-DOS Tables include graphic characters which some printers can't print. These graphics characters can be
replaced by the ASCII characters | and -.
ASCII=N use graphics characters
ASCII=Y
replace graphics characters with ASCII characters (the standard).
Example:
ASCII=N produces what follows, or else accented letters, e.g., &agrave;&agrave;&agrave;:
OVERVIEW TABLES
ITEM CALIBRATIONS
1* PERSON AND ITEM DISTRIBUTION MAP 12* ITEM MAP BY NAME
69
ASCII=Y always produces:
+----------------------------------------------------------------------------+
|
OVERVIEW TABLES
|
ITEM CALIBRATIONS
|
|-------------------------------------+--------------------------------------|
| 1* PERSON AND ITEM DISTRIBUTION MAP | 12. ITEM MAP BY NAME
|
64.
ASYMPTOTE item upper and lower asymptotes
Persons responding to multiple-choice questions (MCQ) can exhibit guessing and carelessness. In the threeparameter IRT model (3-PL), guessing is parameterized as a lower asymptote to the item's logistic ogive of the
probability of a correct answer. In the four-parameter IRT model (4-PL), carelessness is parameterized as an
upper asymptote. Winsteps reports a first approximationto these parameter values, but does not use the
estimates to alter the Rasch measures. The literature suggests that when the lower asymptote is .10 or greater, it
is &quot;substantial&quot; (How Many IRT Parameters Does It Take to Model Psychopathology Items? Steven P. Reise,
Niels G. Waller, Psychological Methods, 2003, 8, 2, 164–184).
ASYMPTOTE=Y
report the values of the Upper and Lower asymptotes in the Item Tables and IFILE=
ASYMPTOTE=N
do not report values for the Upper and Lower asymptotes.
Example: Estimate the 4-PL IRT parameters for the Knox Cube Test data:
Run Exam1.txt
After the analysis completes, use the &quot;Specification&quot; pull-down menu:
Enter: DISCRIM = Yes
to report the Item Discrimination
Enter: ASYMP = Yes
to report the asymptotes
On the &quot;Output Tables&quot; menu, select an item table, e.g., Table 14.
+-------------------------------------------------------------------------------------------------+
|ENTRY
RAW
|
INFIT | OUTFIT |PTMEA|ESTIM| ASYMPTOTE |
|
|NUMBER SCORE COUNT MEASURE ERROR|MNSQ ZSTD|MNSQ ZSTD|CORR.|DISCR|LOWER UPPER| TAP
|
|------------------------------------+----------+----------+-----+-----+-----------+--------------|
|
4
32
34
-4.40
.81| .90
.0| .35
.8| .55| 1.09| .00 1.00| 1-3-4
|
|
5
31
34
-3.83
.70|1.04
.2| .52
.6| .55| 1.01| .07 1.00| 2-1-4
|
|
6
30
34
-3.38
.64|1.17
.6| .96
.6| .53| .87| .10 1.00| 3-4-1
|
|
7
31
34
-3.83
.70|1.33
.9|2.21
1.2| .40| .54| .09
.98| 1-4-3-2
|
Estimation
Item Response Theory (IRT) three-parameter and four-parameter (3-PL, 4-PL) models estimate lower-asymptote
parameters (&quot;guessability&quot;, &quot;pseudo-guessing&quot;) and upper-asymptote parameters (&quot;mistake-ability&quot;) and use
these estimates to modify the item difficulty and person ability estimates. Rasch measurement models
guessability and mistake-ability as misfit, and does not attempt to make adjustments for item difficulties and
personabilities. But initial approximations for the values of the asymptotes can be made, and output by Winsteps
with ASYMPTOTE=Yes.
A lower-asymptote model for dichotomies or polytomies is:
Tni = ci + (mi - ci) (Eni/mi)
where Tni is the expected observation for person n on item i, ci is the lower asymptote for item i, mi is the highest
category for item i (counting up from 0), and Eni is the Rasch expected value (without asymptotes). Rewriting:
ci = mi (Tni - Eni) / (mi - Eni)
This provides the basis for a model for estimating ci. Since we are concerned about the lower asymptote, let us
only consider Bni=Bn-Di&lt;B(Eni=0.5) and weight the observations, Xni, with Wni = Bni - B(Eni=0.5),
ci ≈ Σ(Wni mi (Xni - Eni)) / Σ(Wni (mi - Eni)) for Bni&lt;B(Eni=0.5)
Similarly, for di, the upper asymptote,
di ≈ Σ(Wni mi Xni) / Σ(Wni Eni)) for Bni&gt;B(Eni=mi-0.5)
But if the data are sparse in the asymptotic region, the estimates may not be good. This is a known problem in 3PL estimation, leading many analysts to impute, rather than estimate, asymptotic values.
Birnbaum A. (1968) Some latent trait models and their uses in inferring an examinee's ability. In F.M. Lord &amp; M.R.
Novick, Statistical theories of mental test scores (pp. 395-479). Reading, MA: Addison-Wesley.
Barton M.A. &amp; Lord F.M. (1981) An upper asymptote for the three-parameter logistic item-response model.
Princeton, N.J.: Educational Testing Service.
70
65.
BATCH Batch mode analysis
If you want Winsteps to close itself as performing an analysis and writing out any output specified in the control
file, e.g., by TABLES=, then specify BATCH=YES in the control file. You can launch batch files from the Batch
menu.
If you want Winsteps to run in &quot;background&quot; with the minimum user interaction, then specify BATCH=YES in the
Shortcut, DOS or Shell command which invokes Winsteps.
Running Winsteps in Batch mode: If this won't work for you, see Difficulty below.
Under Windows-2000, -XP and later Windows-NT
It is often useful to run multiple WINSTEPS tasks, one after the other, without keyboard intervention. This can be
accomplished by running WINSTEPS in CMD batch mode.
i) On the main WINSTEPS screen, click on the &quot;Batch&quot; menu item.
ii) On the pull-down menu, select &quot;Edit batch file&quot;.
iii) In the dialog box, select Winbatchcmd.cmd and click on &quot;Open&quot;
iv) The following batch file is available to edit:
echo This is the version for Windows-NT, 2000
echo This is a batch file to run WINSTEPS in batch mode
echo Edit the next lines and add more.
echo Format of lines is:
echo START /WAIT ..\WINSTEPS BATCH=YES Control-file Output-file
Extra=specifications
START /WAIT ..\WINSTEPS BATCH=YES EXAMPLE0.txt EXAMPLE0.OUT TABLES=111
START /WAIT ..\WINSTEPS BATCH=YES SF.txt SF.OUT TFILE=* 1 *
PERSON=CASE
START /WAIT ..\WINSTEPS BATCH=YES KCT.txt KCT.OUT TFILE=* 3 20 * MRANGE=4
These characters have special meanings in batch files: @ &amp; ^ ( )
v) The lines starting with &quot;echo&quot; are comments.
v) Lines starting &quot;START /WAIT WINSTEPS BATCH=YES&quot; execute WINSTEPS
vi) The format is START /WAIT WINSTEPS BATCH=YES control-file output-file extra-specifications
vii) Each new WINSTEPS line is an additional run of the WINSTEPS program
viii) Edit and save this file. You can save it with any name ending &quot;.cmd&quot;
ix) From the &quot;Batch&quot; pull-down menu, select &quot;Run batch file&quot;.
x) Right-click on the desired batch file
xi) In the right-click menu, left-click on &quot;open&quot;
x) The batch file will run - if nothing happens, the batch file is incorrect.
xi) Exit from the Winsteps dialog by clicking on &quot;Cancel&quot;.
xii) You can minimize the batch screen by clicking on the underline in the top right corner.
xiii) You can cancel the batch run by right clicking on the Batch icon in the Task bar, usually at the bottom of the
screen.
Under early versions of Windows (e.g. -95, -98) except early Windows-NT
It is often useful to run multiple WINSTEPS tasks, one after the other, without keyboard intervention. This can be
accomplished by running WINSTEPS in batch mode.
i) On the main WINSTEPS screen, click on the &quot;Batch&quot; menu item.
ii) On the pull-down menu, select &quot;Edit batch file&quot;.
iii) In the dialog box, select Winbatchbat.bat and click on &quot;Open&quot;
iv) The following batch file is available to edit:
echo
This is the version for WINDOWS-95, WINDOWS-98 and ME
71
echo
echo
echo
echo
START
START
START
This is a batch file to run WINSTEPS in batch mode
Edit the next lines and add more.
Format of lines is:
START /w ..\WINSTEPS BATCH=YES Control-file Output-file Extra=specifications
/w ..\WINSTEPS BATCH=YES EXAMPLE0.txt EXAMPLE0.OUT.txt TABLES=111
/w ..\WINSTEPS BATCH=YES SF.txt SF.OUT.txt TFILE=* 1 * PERSON=CASE
/w ..\WINSTEPS BATCH=YES KCT.txt KCT.OUT.txt TFILE=* 3 20 * MRANGE=4
These characters have special meanings in batch files: @ &amp; ^ ( )
v) The lines starting with &quot;echo&quot; are comments.
v) Lines starting &quot;START /w WINSTEPS BATCH=YES&quot; execute WINSTEPS
vi) The format is START /w WINSTEPS BATCH=YES control-file output-file extra-specifications
vii) Each new WINSTEPS line is an additional run of the WINSTEPS program
viii) Edit and save this file. You can save it with any name ending &quot;.bat&quot;
ix) From the &quot;Batch&quot; pull-down menu, select &quot;Run batch file&quot;.
x) Right-click on the desired batch file
xi) In the right-click menu, left-click on &quot;open&quot;
x) The batch file will run - if nothing happens, the batch file is incorrect.
xi) Exit from the Winsteps dialog by clicking on &quot;Cancel&quot;.
xii) You can minimize the batch screen by clicking on the underline in the top right corner.
xiii) You can cancel the batch run by right clicking on the Batch icon in the Task bar, usually at the bottom of the
screen.
Under early Windows-NT
It is often useful to run multiple WINSTEPS tasks, one after the other, without keyboard intervention. This can be
accomplished by running WINSTEPS in batch mode.
i) On the main WINSTEPS screen, click on the &quot;Batch&quot; menu item.
ii) On the pull-down menu, select &quot;Edit batch file&quot;.
iii) In the dialog box, select NTbatch.bat and click on &quot;Open&quot;
iv) The following batch file is available to edit:
echo This is for early versions of WINDOWS-NT
echo For later versions use *.cmd files
echo This is a batch file to run WINSTEPS in batch mode
echo Edit the next lines and add more.
echo Format of lines is:
echo WINSTEPS BATCH=YES Control-file Output-file Extra=specifications
..\WINSTEPS BATCH=YES EXAMPLE0.txt EXAMPLE0.OUT.txt Tables=111
..\WINSTEPS BATCH=YES SF.txt SF.OUT.txt TFILE=* 1 * PERSON=CASE
..\WINSTEPS BATCH=YES KCT.txt KCT.OUT.txt TFILE=* 3 20 * MRANGE=4
These characters have special meanings in batch files: @ &amp; ^ ( )
v) The lines starting with &quot;echo&quot; are comments.
v) Lines starting &quot;WINSTEPS BATCH=YES&quot; execute WINSTEPS
vi) The format is WINSTEPS BATCH=YES control-file output-file extra-specifications
vii) Each new WINSTEPS line is an additional run of the WINSTEPS program
viii) Edit and save this file. You can save it with any name ending &quot;.bat&quot;
ix) From the &quot;Batch&quot; pull-down menu, select &quot;Run batch file&quot;.
x) Right-click on the desired batch file
xi) In the right-click menu, left-click on &quot;open&quot;
x) The batch file will run - if nothing happens, the batch file is incorrect.
xi) Exit from the Winsteps dialog by clicking on &quot;Cancel&quot;.
xii) You can minimize the batch screen by clicking on the underline in the top right corner.
72
xiii) You can cancel the batch run by right clicking on the Batch icon in the Task bar, usually at the bottom of the
screen.
Example: I want to automatically run multiple DIF reports for the same set of data.
Since Winsteps can only perform one DIF analysis at a time in batch mode, you can use anchor files:
First line in batch file, produce measure files
Winsteps BATCH=YES infile outfile dif=$s1w1
ifile=ifile.txt pfile=pfile.txt sfile=sfile.txt
Later lines in batch file, use measure files as anchor files
Winsteps BATCH=YES infile outfile2 dif=$s2w1
30 *
Winsteps BATCH=YES infile outfile3 dif=$s3w1
30 *
ifile=iafile.txt pfile=pafile.txt safile=sfile.txt tfile=*
ifile=iafile.txt pfile=pafile.txt safile=sfile.txt tfile=*
A Windows-XP batch processor
Batch files under Windows XP are used to test out new features in Winsteps. Here is what is done:
a) Create a new subfolder of c:\winsteps, called c:\winsteps\test
b) Copy into folder &quot;test&quot; all the control and data files to be analyzed. For instance all the Winsteps example
control and data files, which are found in c:\winsteps\examples
c) Use Notepad to create a file in c:\winsteps\test to do the analysis. This file is &quot;saved as&quot; test.bat
This file contains, for instance:
start /w ..\winsteps batch=yes exam1.txt exam1.out DISC=YES TABLES=111
start /w ..\winsteps batch=yes exam9.txt exam9.out DISC=YES TABLES=111
start /w ..\winsteps batch=yes sf.txt sf.out DISC=YES TABLES=111
You can replace ..\winsteps with the pathname to your copy of winsteps.exe
d) double-click on test.bat in c:\winsteps\test to run this batch file.
e) Winsteps &quot;flashes&quot; on the task bar several times, and progress through the batch file is shown in a DOS-style
window.
e) the .out files are written into c:\winsteps\test
Difficulty running Batch or Command files?
Microsoft Windows is designed to run interactively, not in batch mode. Microsoft are not consistent with the way
they implement batch files in different versions of Windows. So our challenge is to discover a method of running
batch files that works for the version of Windows we happen to have. Since Windows is very bad at running batch
or command files. You need to validate your instructions one step at a time:
Common problems are solved at: www.winsteps.com/problems.htm
i)
Run Winsteps in standard mode from the DOS prompt.
ii) Have the full paths to everything in your batch or command file, e.g., called mybatch.cmd,
START /WAIT c:\winsteps\WINSTEPS BATCH=YES c:\winsteps\examples\example0.txt
c:\winsteps\examples\example0.out.txt
also have full paths to everything in your Winsteps control file, e.g.,
DATA = c:\c:\winsteps\examples\mydata.txt
Note: In this Batch command:
START /WAIT c:\winsteps\WINSTEPS BATCH=YES c:\winsteps\examples\controlfile.txt outputfile.txt
file &quot;outputfile.txt&quot; will be placed in directory &quot;c:\winsteps\examples\&quot;
iii) Windows &quot;Start&quot; menu. &quot;Run&quot;. Copy and paste the following line into the Windows Run box on the Windows
Start menu. Click OK:
c:\winsteps\WINSTEPS c:\winsteps\examples\example0.txt c:\winsteps\examples\example0.out.txt table=1
73
Does Winsteps start in the ordinary way? This tests the Windows command line interface.
iv) Windows &quot;Start&quot; menu. &quot;Run&quot;. Copy and paste the following line into the Run box. Click OK:
c:\winsteps\WINSTEPS BATCH=YES c:\winsteps\examples\exam15.txt c:\winsteps\examples\exam15.out.txt
table=1
Does the Winsteps icon appear on the Task bar and then disappear? This tests Winsteps backgroup
processing.
v) On your desktop, right-click, &quot;New&quot;, &quot;Text document&quot;. Double-click on icon. Paste in:
START /WAIT c:\winsteps\WINSTEPS c:\winsteps\examples\example0.txt c:\winsteps\examples\example0.out.txt
table=1
&quot;Save as&quot; Test.cmd. Double-click on Test.cmd
Does Winsteps run in the ordinary way? This test the Windows START function. If this fails, &quot;Save as&quot;
Test.bat instead of Test.cmd.
vi) On your desktop, right-click, &quot;New&quot;, &quot;Text document&quot;. Double-click on icon. Paste in:
START /WAIT c:\winsteps\WINSTEPS BATCH=YES c:\winsteps\examples\exam15.txt
c:\winsteps\examples\exam15.out.txt table=1
&quot;Save as&quot; Test2.cmd. Double-click on Test2.cmd (or &quot;Save as&quot; Test2.bat if that works better on your
computer.)
Does the Winsteps icon flash on the task bar line, and then disappear? Winsteps has run in background.
vii) Now build your own .cmd batch file, using lines like:
START /WAIT c:\winsteps\WINSTEPS BATCH=YES c:\winsteps\examples\example0.txt
c:\winsteps\examples\example0.out.txt
Running Winsteps within other Software
Automating the standard version of Winsteps is straightforward using the control instruction BATCH=YES.
Winsteps will run under Windows in background (as much as Windows permits).
Let's assume your software is written in Visual Basic (or any other programming, database or statistical language)
(a) write out a Winsteps control file as a .txt file
(b) write out a Winsteps data file as a .txt file
(c) &quot;shell&quot; out to
&quot;Winsteps BATCH=YES controlfile.txt outputfile.txt data=datafile.txt ifile=ifile.txt pfile=pfile.txt .....&quot;
(d) read in the ifile.txt, pfile.txt or whatever Winsteps output you need to process.
This is being done routinely by users of SAS.
66.
BYITEM display graphs for items
In the bit-mapped graphs produced by the Graphs pull-down menu, the empirical item characteristic curves can
be produced at the grouping level or the item level. When ISGROUPS=0, the item-level and grouping-level curves
are the same.
BYITEM = Yes
show empirical curves at the item level.
BYITEM = No
show empirical curves at the grouping level.
67.
CATREF reference category for Table 2
If a particular category corresponds to a criterion level of performance, choose that category for CATREF=.
74
Table 2, &quot;most probable responses/scores&quot;, maps the items vertically and the most probable responses, expected
scores, and 50% cumulative probabilities horizontally. Generally, the vertical ordering is item difficulty measure.
If, instead, a particular category is to be used as the reference for sorting, give its value as scored and recoded.
Special uses of CATREF= are:
CATREF=-3
for item entry order
CATREF=-2
for item measure order
CATREF=-1
for items measure order with ISGROUPS=
CATREF=0 for item measure order
CATREF=1...254 for item measure order based on this category.
Example 1: You have 4-point partial-credit items, entered in your data as A,B,C,D, and then scored as 1,2,3,4.
You wish to list them based on the challenge of category C, rescored as 3,
CODES =ABCD
original responses
NEWSCORE=1234 rescored values
RESCORE=2
rescore all
CATREF=3
Table 2 reference category
ISGROUPS=0
partial credit: one item per grouping
If, for an item, the category value &quot;3&quot; is eliminated from the analysis or is the bottom category, the nearest higher
category is used for that item.
Example 2: You have 6 3-category items in Grouping 1, and 8 4-category items in Grouping 2. You wish to list
them in Table 2.2 by measure within grouping, and then by measure overall.
CODES=1234
NI= 14
ISGROUPS= 11111122222222
TFILE=*
2.2 0 0 0 -1
-1 means CATREF=-1
2.2 0 0 0 0
last 0 means CATREF=0
*
68.
CFILE scored category label file
Rating (or partial credit) scale output is easier to understand when the categories are shown with their substantive
meanings. Use CFILE= to label categories using their scored values, i.e., after rescoring. Use CLFILE= to label
categories using their original codes, i.e., before any rescoring.
Labels for categories, after they have been scored, can be specified using CFILE= and a file name, or CFILE=*
and placing the labels in the control file. Each category number is listed (one per line), followed by its descriptive
label. If the observations have been rescored (NEWSCORE=) or keyed (KEYn=), then use the recoded
category value in the CFILE= specification. When there are different category labels for different ISGROUPS=
of items, specify an example item from the grouping, followed immediately by &quot;+&quot; and the category number.
Blanks or commas can be used a separators between category numbers and labels.
Example 1: Identify the three LFS categories, 0=Dislike, 1=Don't know, 2=Like.
CODES=012
CFILE=*
0 Dislike
1 Don't know
2 Like
*
The labels are shown in Table 3 as:
------------------------------------------------------CATEGORY OBSERVED AVGE INFIT OUTFIT STRUCTURE
LABEL
COUNT
MEASURE MNSQ MNSQ MEASURE
-------------------------------------------------------
75
0
378
-.87 1.08 1.19
NONE
Dislike
1
620
.13
.85
.69
-.85
Don't know
2
852
2.23 1.00 1.46
.85
Like
-------------------------------------------------------
Example 2: Items 1-10 (Grouping 1) are &quot;Strong Disagree, Disagree, Agree, Strongly Agree&quot;. Items 11-20
(Grouping 2) are &quot;Never, Sometimes, Often, Always&quot;.
NI=20
CODES=1234
ISGROUPS=11111111112222222222
CFILE=*
7+1 Strongly Disagree ; We could use any item number in Grouping 1, i.e., 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
7+2 Disagree
; item 7 has been chosen
7+3 Agree
7+4 Strong Agree
13+1 Never
; We could use any item number in Grouping 2, i.e., 11, 12, 13, 14, 15, 16, 17, 18, 19, 20
13+2 Sometimes ; item 13 has been chosen
13+3 Often
13+4 Always
*
Example 3: To enter CFILE= information on the DOS Prompt or Extra Specifications lines, using commas instead
of blanks as separators:
C:&gt;WINSTEPS SF.TXT SFO.TXT CFILE=* 1,Dislike 2,Don't-know 3,Like *
Example 4: Some items have one rating scale definition, but most items have another rating scale definition. But
each item is calibrated with its own structure: ISGROUPS=0
NI=20
CODES=1234
ISGROUPS=0
CFILE=*
1 Strongly Disagree This scale is used by most items
2 Disagree
3 Agree
4 Strong Agree
16+1 Never
16 is one item using the other scale
16+2 Sometimes
16+3 Often
16+4 Always
17+1 Never
17 is another item using the other scale
17+2 Sometimes
17+3 Often
17+4 Always
....
for all the other items using the other scale
*
Example 5: Several categories are collapsed into one category. The original codes are A-H. After rescoring there
is only a dichotomy: 0, 1.
NI=30
CODES =ABCDEFGH
NEWSCORE=00011110
CFILE=*
0 Fail
1 Pass
*
69.
Specify the categories as recoded
CHART graphical plots in Tables 10, 13-15
The measures and fit statistics in the Tables can be displayed as graphical plots in subtables 6.2, 10.2, 13.2,
14.2, 15.2, 17.2, 18.2, 19.2, 25.2, 26.2.
76
CHART=N Omit the graphical plots.
CHART=Y Include graphical plots
PUPIL FIT GRAPH: OUTFIT ORDER
+----------------------------------------------------------------------------------+
|ENTRY| MEASURE | INFIT MEAN-SQUARE | OUTFIT MEAN-SQUARE |
|
|NUMBR| + |0
0.7 1 1.3
2|0
0.7 1 1.3
2| PUPIL
|
|-----+---------+---------------------+---------------------+----------------------|
|
72|*
|
: . :
* |A
: . :
*| JACKSON, SOLOMON
|
|
47| *
|
: . *:
|J
: . :
*
| VAN DAM, ANDY
|
|
53|*
|
: *. :
|K
: . :
* | SABOL, ANDREW
|
|
32| *
|
*: . :
|w
* . :
| ROSSNER, JACK
|
|
21| *
| *
: . :
|a *
: . :
| EISEN, NORM L.
|
+----------------------------------------------------------------------------------+
The fit information is shown in graphical format to aid the eye in identifying patterns and outliers. The fit bars are
positioned by FITLOW= and FITHIGH=. They may also be repositioned using TFILE=.
70.
CLFILE codes label file
Rating (or partial credit) scale output is easier to understand when the categories are shown with their substantive
meanings. Use CFILE= to label categories using their scored values, i.e., after rescoring. Use CLFILE= to label
categories using their original codes, i.e., before any rescoring. Labels for the original categories in the data can
be specified using CLFILE= and a file name, or CLFILE=* and placing the labels in the control file. Each category
number is listed (one per line), followed by its descriptive label. Original category values are used. There are
several options:
XWIDE=2 ;observations are two columns wide
CODES = &quot; 0 1 299&quot;
; codes are 0, 1 , 2, 99
CLFILE=*
99 Strongly Agree ; original code of 99 has the label &quot;Strongly Agree&quot;
2 Agree
; original code of blank+2 (or 2+blank) has the label &quot;Agree&quot;
2+99 Heartily Agree ; for item 2, code 99 has the label &quot;Heartily Agree&quot;
3+0 Disagree
; for item 3, code 0 means &quot;Disagree&quot;
*
Example 1: Identify the three LFS categories, D=Dislike, N=Don't know, L=Like.
CODES
=DNL
NEWSCORE=012
CLFILE=*
D Dislike
N Neutral
L Like
*
The labels are shown in Table 3 as:
------------------------------------------------------CATEGORY OBSERVED AVGE INFIT OUTFIT STRUCTURE
LABEL
COUNT
MEASURE MNSQ MNSQ MEASURE
------------------------------------------------------0
378
-.87 1.08 1.19
NONE
Dislike
1
620
.13
.85
.69
-.85
Don't know
2
852
2.23 1.00 1.46
.85
Like
-------------------------------------------------------
Example 2: Items 1-10 (Grouping 1) are &quot;Strong Disagree, Disagree, Agree, Strongly Agree&quot;. Items 11-20
(Grouping 2) are &quot;Never, Sometimes, Often, Always&quot;.
NI=20
CODES
=ABCD
NEWSCORE=1234
77
ISGROUPS=11111111112222222222
CLFILE=*
7+A Strongly Disagree 7 is any item in Grouping 1
7+B Disagree
7+C Agree
7+D Strong Agree
13+A Never
13 is any item in Grouping 2
13+B Sometimes
13+C Often
13+D Always
*
Example 3: To enter CLFILE= information on the DOS Prompt or Extra Specifications lines, using commas
instead of blanks as separators:
C:&gt;WINSTEPS SF.TXT SFO.TXT CLFILE=* D,Dislike N,Don't-know L,Like *
Example 4: One grouping of items has a unique response format, but the other groupings all have the same
format. Here, each grouping has only one item, i.e., ISGROUPS=0
NI=20
CODES=1234
ISGROUPS=0
CLFILE=*
1 Strongly Disagree This rating scale is used by most items
2 Disagree
3 Agree
4 Strong Agree
16+1 Never
16 is the one item using this rating scale
16+2 Sometimes
16+3 Often
16+4 Always
*
Example 5: Several categories are collapsed into one category. The original codes are A-H. After rescoring there
is only a dichotomy: 0, 1.
NI=30
CODES =ABCDEFGH
NEWSCORE=00011110
CFILE=*
0 Fail
1 Pass
*
; or
CLFILE=*
A Fail
B Fail
C Fail
D Pass
E Pass
F Pass
G Pass
H Pass
71.
Specify the categories as recoded
CODES valid data codes
Says what characters to recognize as valid codes in your data file. If XWIDE=1 (the standard), use one
column/character per legitimate code. If XWIDE=2, use two columns/characters per valid code. Characters in
your data not included in CODES= are given the MISSCORE= value.
Example 1:
A test has four response choices. These are &quot;1&quot;, &quot;2&quot;, &quot;3&quot;, and &quot;4&quot;. All other codes in the data file
are to be treated as &quot;item not administered&quot;. Each response uses 1 column in your data file. Data look like:
78
134212342132.3212343221
XWIDE=1
CODES=1234
one character wide (the standard)
four valid 1-character response codes
Example 2: There are four response choices. Each response takes up 2 columns in your data file and has
leading 0's, so the codes are &quot;01&quot;, &quot;02&quot;, &quot;03&quot; and &quot;04&quot;. Data look like: 0302040103020104040301
XWIDE=2
CODES=01020304
two characters wide
four valid 2-character response codes
Example 3: There are four response choices entered on the file with leading blanks, so that codes are &quot; 1&quot;, &quot; 2&quot;, &quot;
3&quot;, and &quot; 4&quot;. Data look like: 3 2 4 2 1 3 2
XWIDE=2
CODES=&quot; 1 2 3 4&quot;
two characters wide
&quot; required: blanks in 2-character responses
Note: when XWIDE=2 or more, both CODES= and the data value are left-aligned before matching, so both &quot; 1&quot;
and &quot;1 &quot; in CODES= match both &quot; 1&quot; and &quot;1 &quot;in your data file.
Example 4: Your data is a mixture of both leading blanks and leading 0's in the code field, e.g. &quot;01&quot;, &quot; 1&quot;, &quot; 2&quot;, &quot;02&quot;
etc. The numerical value of a response is calculated, where possible, so that both &quot;01&quot; and &quot; 1&quot; are analyzed as
1.
Data look like: 02 1 20102 1 2 01
XWIDE=2
two characters wide
CODES=&quot; 1 2 3 401020304&quot;
two characters per response
Example 5: Your valid data are 1,2,3,4,5 and your missing data codes are 7,8,9 which you want reported
separately on the distractor tables.
CODES =
12345789
NEWSCORE = 12345XXX
; missing values scored with non-numeric values
Example 6: The valid responses to an attitude survey are &quot;a&quot;, &quot;b&quot;, &quot;c&quot; and &quot;d&quot;. These responses are to be
recoded &quot;1&quot;, &quot;2&quot;, &quot;3&quot; and &quot;4&quot;. Data look like: adbdabcd
CODES
=abcd four valid response codes
NEWSCORE=1234 new values for codes
RESCORE=2
rescore all items
Typically, &quot;abcd&quot; data implies a multiple choice test. Then KEY1= is used to specify the correct response. But, in
this example, &quot;abcd&quot; always mean &quot;1234&quot;, so that the RESCORE= and NEWSCORE= options are easier to use.
Example 7: Five items of 1-character width, &quot;abcd&quot;, then ten items of 2-character width &quot;AA&quot;, &quot;BB&quot;, &quot;CC&quot;, &quot;DD&quot;.
These are preceded by person-id of 30 characters. Data look like:
George Washington Carver III dabcdBBAACCAADDBBCCDDBBAA
FORMAT=(30A1,5A1,10A2)
Name 30 characters, 5 1-chars, 10 2-chars
XWIDE =2
all converted to 2 columns
CODES =&quot;a b c d AABBCCDD&quot; &quot;a&quot; becomes &quot;a &quot;
NEWSCORE=&quot;1 2 3 4 1 2 3 4 &quot;
response values
RESCORE=2
rescore all items
NAME1=1
name starts column 1 of reformatted record
ITEM1=31
items start in column 31
NI=15
15 items, all XWIDE=2
Example 8: Items are to rescored according to Type A and Type B. Other items to keep original scoring.
CODES = 1234
IREFER = AAAAAAAABBBBBBBBCCCCCCC ; 3 item types
IVALUEA = 1223
Recode Type A items
IVALUEB = 1123
Recode Type B items
IVALUEC = 1234
Recode Type * item. Can be omitted
Example 9: The valid responses are percentages in the range 00 to 99.
XWIDE = 2
two columns each percent
; uses continuation lines
CODES = 0001020304050607080910111213141516171819+
+2021222324252627282930313233343536373839+
+4041424324454647484950515253545556575859+
79
+6061626364656667686970717273747576777879+
+8081828384858687888990919293949596979899
Example 10:
Codes are in the range 0-254 (the maximum possible).
XWIDE=3
; 3 characters per response
CODES=&quot; 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23+
+ 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47+
+ 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71+
+ 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95+
+ 96 97 98 99100101102103104105106107108109110111112113114115116117118119+
+120121122123124125126127128129130131132133134135136137138139140141142143+
+144145146147148149150151152153154155156157158159160161162163164165166167+
+168169170171172173174175176177178179180181182183184185186187188189190191+
+192193194195196197198199200201202203204205206207208209210211212213214215+
+216217218219220221222223224225226227228229230231232233234235236237238239+
+240241242243244245246247248249250251252253254&quot;
72.
CONVERGE select convergence criteria
This selects which of LCONV= and RCONV= set the convergence criterion. See convergence considerations.
CONVERGE=L
LCONV= for &quot;Logit change size&quot; controls convergence.
Iteration stops when the biggest logit change is less or equal to LCONV=, or when the biggest logit
change size increases (divergence).
CONVERGE=R
RCONV= for &quot;Residual size&quot; controls convergence.
Iteration stops when the biggest residual score is less or equal to RCONV=, or when the biggest residual
size increases (divergence).
CONVERGE=E
Either LCONV= for &quot;Logit change size&quot; or RCONV= for &quot;Residual size&quot; controls convergence.
Iteration stops when the biggest logit change is less or equal to LCONV=, or when the biggest residual
score is less or equal to RCONV=, or when both the biggest logit change size increases and the biggest
residual size increases (divergence).
CONVERGE=B
Both LCONV= for &quot;Logit change size&quot; and RCONV= for &quot;Residual size&quot; controls
convergence.
Iteration stops when both the biggest logit change is less or equal to LCONV= and the biggest residual
score is less or equal to RCONV=, or when both the biggest logit change size increases and the biggest
residual size increases (divergence).
CONVERGE=F
Force both LCONV= for &quot;Logit change size&quot; and RCONV= for &quot;Residual size&quot; to control
convergence.
Iteration stops when both the biggest logit change is less or equal to LCONV= and the biggest residual
score is less or equal to RCONV=.
Example 1: We want to be take a conservative position about convergence, requiring both small logit changes
and small residual sizes when iteration ceases.
CONVERGE=Both
Example 2: We want to set the convergence criteria to match BIGSTEPS version 2.59
CONVERGE=B ; the rule was LCONV= and RCONV=
RCONV= 0.5 ; the BIGSTEPS standards or whatever value you used
LCONV= .01
Example 3: We want to set the convergence criteria to match Winsteps version 3.20
CONVERGE=E ; the rule was LCONV or RCONV
RCONV= 0.5 ; the 3.20 standards or whatever value you used
LCONV= .01
Example 4: We want the convergence criteria to match Winsteps version 2.85
80
CONVERGE= F; force both LCONV and RCONV to be met
RCONV= 0.5 ; the 2.85 standards or whatever value you used
LCONV= .01
You may also want:
WHEXACT=NO ; centralized Wilson-Hilfterty was the default
Example 5:
Question: With anchored analyses, iterations never stop!
|----------------------------------------------------------------------------|
|
JMLE
MAX SCORE
MAX LOGIT
LEAST CONVERGED
CATEGORY
STEP |
| ITERATION
RESIDUAL*
CHANGE
EXID
BYCASE
CAT
RESIDUAL
CHANGE |
|----------------------------------------------------------------------------|
|
1
-239.04
.5562
1993
392*
6
85.70
-.5960|
|
2
-105.65
-.1513
1993
392*
4
-28.92
.2745|
.....
|
18
-5.35
-.0027
2228
352*
3
2.35
.0146|
|
19
-5.16
.0029
2228
352*
3
2.31
.0106|
|
20
-5.05
.0025
2228
352*
3
2.28
.0055|
|
21
-5.00
.0010
2228
352*
3
2.26
.0075|
|
22
-4.99
-.0008
2228
352*
3
2.25
.0025|
.....
|
170
-5.00
-.0011
1377
352*
3
1.96
.0109|
|
171
-5.00
.0018
187
352*
3
1.96
.0019|
.....
The standard convergence criteria in Winsteps are preset for &quot;free&quot; analyses. With anchored analyses,
convergence is effectively reached when the logit estimates stop changing in a substantively meaningful way.
This has effectively happened by iteration 20. Note that the logit changes are less than .01 logits - i.e., even the
biggest change would make no difference to the printed output (which is usually reported to 2 decimal places)
To have the current Winsteps do this automatically, set
CONVERGE=L
LCONV=.005 ; set to stop at iteration 22 - to be on the safe side.
73.
CSV comma-separated values in output files
To facilitate importing the IFILE=, ISFILE=, PFILE=, SFILE= and XFILE= files into spreadsheet and database
programs, the fields can be separated by commas, and the character values placed inside &quot; &quot; marks.
CSV=N
Use fixed field length format (the standard)
CSV=Y or CSV=, Separate values by commas (or their international replacements) with character fields in &quot; &quot;
marks.
CSV=T
Separate values by tab characters with character fields in &quot; &quot; marks (convenient for EXCEL).
CSV=S
SPSS format.
Examples:
Fixed space:
; MATCH Chess Matches at the Venice Tournament, 1971 Feb 11 0:47 2004
;ENTRY MEASURE STTS COUNT
SCORE ERROR IN.MSQ IN.ZSTD OUT.MS OUT.ZSTD DISPL PTME WEIGHT
NAME
1
.87 1
2.0
2.0
.69
1.17
.47
1.17
.47
.01
1.00
1.00
I0001
DISCR G M
1.97 1 R
Tab-delimited:
&quot;MATCH Chess Matches at the Venice Tournament, 1971 Feb 11 0:47 2004&quot;
&quot;;&quot;
&quot;ENTRY&quot; &quot;MEASURE&quot;
&quot;STATUS&quot;
&quot;COUNT&quot; &quot;SCORE&quot; &quot;ERROR&quot; &quot;IN.MSQ&quot;
&quot;IN.ZSTD&quot;
&quot; &quot;
1
.87
1
2.0
2.0
.69
1.17
.47
1.17
.47 .01 1.00
&quot;1&quot;
&quot;R&quot;
&quot;I0001&quot;
Comma-separated:
&quot;MATCH Chess Matches at the Venice Tournament, 1971 Feb 11 0:47 2004&quot;
&quot;;&quot;,&quot;ENTRY&quot;,&quot;MEASURE&quot;,&quot;STATUS&quot;,&quot;COUNT&quot;,&quot;SCORE&quot;,&quot;ERROR&quot;,&quot;IN.MSQ&quot;,&quot;IN.ZSTD&quot;,....
81
.....
1.00
1.97
&quot; &quot;,1,.87,1,2.0,2.0,.69,1.17,.47,1.17,.47,.01,1.00,1.00,1.97,&quot;1&quot;,&quot;R&quot;,&quot;I0001&quot;
SPSS format: This is the SPSS .sav file format.
74.
CURVES probability curves for Table 21 and Table 2
Interpreting rating (or partial credit) scale structure is an art, and sometimes counter-intuitive. See the examples
in RSA.
Table 21 provides three curves for each rating (or partial credit) scale definition. Table 2 provides the equivalent
response locations for each item. The first curve shows the probability of response in each category. The second
curve shows the expected score ogive. The third curve shows the cumulative probability of response in each
category or below. The 50% cumulative probability medians are at the intersections of these curves with the .5
probability line. The control indicators of &quot;1&quot; and &quot;0&quot;, in the 1st, 2nd or 3rd position of the CURVES= variable,
select or omit output of the corresponding curve.
1?? Probability curves
?1? Expected score ogives, model inter characteristic curves
??1 Cumulative probability curves, showing .5 probability median category boundaries.
CURVES=000 indicates no curves are to be drawn - Table 21 will be skipped, unless STEPT3=N, in which
case only the structure summaries are output.
CURVES=111 draw all 3 curves in Table 21 (and 3 versions of Table 2)
CURVES=001 draw only the 3rd, cumulative probability score, curve.
75.
CUTHI cut off responses with high probability of success
Use this if careless responses are evident. CUTHI= cuts off the top left-hand corner of the Scalogram in Table
22.
Eliminates (cuts off) observations where examinee measure is CUTHI= logits or more (as user-rescaled by
USCALE=) higher than item measure, so the examinee has a high probability of success. Removing off-target
responses takes place after PROX has converged. After elimination, PROX is restarted, followed by JMLE
estimation and fit calculation using only the reduced set of responses. This may mean that the original scorebased ordering is changed.
Usually with CUTLO= and CUTHI=, misfitting items aren't deleted - but miskeys etc. must be corrected first.
Setting CUTLO= and CUTHI= is a compromise between fit and missing data. If you loose too much data, then
increase the values. If there is still considerable misfit or skewing of equating, then decrease the values.
Example: Eliminate responses where examinee measure is 3 or more logits higher than item measure, to
eliminate ther worst careless wrong responses:
CUTHI= 3
This produces a scalogram with eliminated responses blanked out:
27
RESPONSES SORTED BY MEASURE:
KID
TAP
111111111
123745698013245678
-----------------------15 111
11100000 observations for extreme scores remain
14 111
1110000000
28 111
111010000000
30 1111 1111000000000
111111100000000000
82
76.
CUTLO cut off responses with low probability of success
Use this if guessing or response sets are evident. CUTLO= cuts off the bottom right-hand corner of the
Scalogram in Table 22.
Eliminates (cuts off) observations where examinee measure is CUTLO= logits or more (user-rescaled by
USCALE=) lower than item measure, so that the examinee has a low probability of success. The elimination of
off-target responses takes place after PROX has converged. After elimination, PROX is restarted, followed by
JMLE estimation and point-measure and fit calculation using only the reduced set of responses. This may mean
that the original score-based ordering is changed.
Usually with CUTLO= and CUTHI=, misfitting items aren't deleted - but miskeys etc. must be corrected first.
Setting CUTLO= and CUTHI= is a compromise between fit and missing data. If you loose too much data, then
increase the values. If there is still considerable misfit or skewing of equating, then decrease the values.
Example: Disregard responses where examinees are faced with too great a challenge, and so might guess
wildly, i.e., where examinee measure is 2 or more logits lower than item measure:
CUTLO= 2
RESPONSES SORTED BY MEASURE:
KID
TAP
111111111
123745698013245678
-------------------------15 111010101011100000 observations for extreme scores remain
14 111011001110000000
28 1110101110
00000
30 11110111
00000
27 111111
00000
77.
DATA name of data file
Your data can be the last thing in the control file (which is convenient if you only have a small amount of data), but
if you have a large amount of data, you can place it in a separate file, and then use DATA= to say where it is.
FORMAT= reformats these records. MFORMS= enables multiple reformatting.
Example 1: Read the observations from file &quot;A:\PROJECT\RESPONSE.TXT&quot;.
DATA=A:\PROJECT\RESPONSE.TXT
Example 2: Read scanned MCQ data from file DATAFILE in the current directory.
DATA=DATAFILE
You may specify that several data files be analyzed together in one run, by listing their file names, separated by
&quot;+&quot; signs. The list, e.g., FILE1.TXT+MORE.TXT+YOURS.D, can be up to 200 characters long. The layout of all
data files must be identical.
Example 3: A math test has been scanned in three batches into files &quot;BATCH.1&quot;, &quot;BATCH.2&quot; and &quot;BATCH.3&quot;.
They are to be analyzed together.
DATA=BATCH.1+BATCH.2+BATCH.3
78.
DELIMITER data field delimiters
It is often convenient to organize your data with delimiters, such as commas, semi-colons or spaces, rather than
in fixed column positions. However, often the delimiter (a Tab, space or comma) only takes one column position.
In which case, it may be easier to include it in the CODES= or use MFORMS= or FORMAT=.
To check that your data file has decoded properly, look at RFILE=
To do this, specify the following command DELIMITER= value (or SEPARATOR= value). This value is the
separator.
83
Examples: DELIMITER= &quot; &quot;
fixed-field values
DELIMITER= &quot;,&quot;
comma-separated values. The , must be &quot;,&quot;
DELIMITER=BLANK
blank-separated values
or DELIMITER=SPACE
space-separated values
DELIMITER=TAB tab-separated values
DELIMITER=&quot;;&quot;
semi-colon separated values. The ; must be &quot;;&quot;, otherwise it is treated as a comment.
When decoding delimited values, leading and trailing blanks, and leading and trailing quotation marks, &quot; &quot; and ' ' in
each value field are ignored. Responses are left-aligned, and sized according to XWIDE=.
For NAME1= and ITEM1=, specify the value number in the data line, starting with 1 as the leftmost value.
FORMAT= does not apply to this data design.
Combine your person name and demographic information into one field that is to be referenced by NAME1=.
Example 1 of a data line:
; the following is ONE data line:
&quot;01&quot;; 02; &quot;01&quot;; &quot;01&quot;; &quot;01&quot;; 00; 02; 00; &quot;01&quot;; 02; 02; 02; 02; 00; 02; &quot;01&quot;; &quot;01&quot;; 02; 02 ; 00; 02;
&quot;01&quot;; 00; 02; 00; ROSSNER, MARC DANIEL
;which decodes as:
01020101010002000102020202000201010202000201000200ROSSNER, MARC DANIEL
ITEM1=1
; item responses start in first field
NI=25
; there are 25 responses, i.e., 25 response fields
NAME1=26
; the person name is in the 26th field
DELIMITER = &quot;;&quot; ; the field delimiters are semi-colons
XWIDE=2
; values are right-aligned, 2 characters wide.
CODES=000102 ; the valid codes.
NAMLEN=20
; override standard person name length of 30 characters.
Example 2 of a data line:
; the following is ONE data line:
ROSSNER - MARC DANIEL, &quot;01&quot;, 02 , &quot;01&quot;, &quot;01&quot;, &quot;01&quot;, 00, 02, 00, &quot;01&quot;, 02, 02, 02, 02, 00, 02, &quot;01&quot;,
&quot;01&quot;, 02, 02, 00, 02, &quot;01&quot;, 00, 02, 00
;which decodes as:
01020101010002000102020202000201010202000201000200ROSSNER - MARC DANIEL
ITEM1=2
; item responses start in second field
NI=25
; there are 25 responses, i.e., 25 response fields
NAME1=1
; the person name is in the 1st field
DELIMITER = &quot;,&quot; ; the field delimiters are commas (so no commas in names)
XWIDE=2
; values are right-aligned, 2 characters wide.
CODES=000102 ; the valid codes
NAMLEN=20
; override standard person name length of 30 characters.
Example: Here is the data file, &quot;Book1.txt&quot;
fred,1,0,1,0
george,0,1,0,1
Here is the control file:
name1=1
; first field
item1=2
; second field
ni=4
; 4 fields
data=book1.txt
84
codes=01
delimiter = &quot;,&quot;
&amp;END
looking
viewing
peeking
seaking
END LABELS
Here is the reformatted file from the Edit Pull-Down menu: View Delimiter File:
1010fred
0101george
79.
DIF columns within person label for Table 30
DIF= specifies the part of the person label which is to be used for classifying persons in order to identify
Differential Item Function (DIF) - uniform or non-uniform - using the column selection rules. See also DIF Table
and DIF and DPF considerations.
DIF= location is usually column number within person label field. DIF=1 means &quot;DIF selection character is first
character of person label.&quot;
Example 0: I have tab-separated data and my DIF indicator is in a separate field from the Person Iabel.
Solution: for the DIF analysis, specify the DIF field as the person label field using NAME1=, then $S1W1
Example 1: Columns 18-20 of the person label (in columns 118-120 of the data record) contain a district code:
NAME1=101
; person label starts in column 101
DIF = $S18W3
; district starts in column 18 of person label with a width of 3
or
@district = 18W3 ; district starts in column 18 of person label with a width of 3
DIF = @district
; DIF classifier
tfile=*
30
Table 30 for the DIF report (or use Output Tables menu)
*
Example 2: Columns 18-20 of the person label (in columns 118-120 of the data record) contain a district code.
Column 121 has a gender code. Two independent DIF analyses are needed:
NAME1=101
; person label starts in column 101
DIF = *
$S18W3
; start in person label column 18 with a width of 3 - district
$S21W1
; start in person label column 21 with a width of 1 - gender
*
tfile=*
30
Table 30 for the DIF report (or use Output Tables menu)
*
Example 3: An investigation of non-uniform DIF with high-low ability classification for the KCT data.
; action the following with the Specification pull-down menu
@SEX = $S9W1
; the sex of participants is in column 9 of the person label
DIF = @SEX + MA2 ; look for non-uniform DIF (gender + two ability strata)
PSUBTOT = @SEX + MA2 ; summary statistics by gender and ability strata
Tfile=*
; This is more easily actioned through the Output Tables Menu
30
; Table 30 - DIF report
28
; Table 28 - Person subtotals for DIF classifications
*
Table 30: DIF specification is: DIF=@SEX+MA2
+-----------------------------------------------------------------------------------------+
| KID
DIF
DIF
KID
DIF
DIF
DIF
JOINT
TAP
|
| CLASS MEASURE S.E. CLASS MEASURE S.E. CONTRAST S.E.
t d.f. Number Name
|
|-----------------------------------------------------------------------------------------|
85
| F1
-1.86
1.09
M1
-4.54
1.15
2.68
1.59
1.69
8
6 6= 3-4-1
|
Table 28: Subtotal specification is: PSUBTOTAL=@SEX+MA2
+---------------------------------------------------------+
| KID
MEAN
S.E. OBSERVED MEDIAN
REAL
|
| COUNT MEASURE
MEAN
S.D.
SEPARATION CODE |
|---------------------------------------------------------|
|
4
-2.08
.90
.89
.00
F1
| &lt;- Non-extreme
|
6
-2.82
.41
.91
-2.86
.32
M1
|
Example 4: With Example0.txt (the Liking for Science rating scale data) you want to see if any items were biased
against names starting with any letter of the alphabet, then:
run example0.txt
request the DIF Table (Table 30) from the Output Tables menu
specify: $S1W1
a DIF table is produced.
The equivalent DIF specification is: DIF=$S1W1
Positive DIF size is higher ACT difficulty measure
+------------------------------------------------------------------------------------+
| KID
DIF
DIF
KID
DIF
DIF
DIF
JOINT
ACT
|
| CLASS MEASURE S.E. CLASS MEASURE S.E. CONTRAST S.E.
t d.f. Number Name
|
|------------------------------------------------------------------------------------|
| R
-.06
.54 W
.89&gt; 2.05
-.95 2.12 -.45
8
1 WATCH BIRDS |
| R
-.06
.54 L
-.65
.75
.59
.92
.64 12
1 WATCH BIRDS |
| R
-.06
.54 S
-.42
.57
.36
.78
.46 18
1 WATCH BIRDS |
| R
-.06
.54 H
-1.63 1.13
1.57 1.25 1.26 11
1 WATCH BIRDS |
| R
-.06
.54 D
.12
.86
-.18 1.01 -.18 11
1 WATCH BIRDS |
.......
80.
DISCRIMINATION item discrimination
Rasch models assert that items exhibit the model-specified item discrimination. Empirically, however, item
discriminations vary. During the estimation phase of Winsteps, all item discriminations are asserted to be equal, of
value 1.0, and to fit the Rasch model. But empirical item discriminations never are exactly equal, so Winsteps can
also report an estimate of those discriminations post-hoc (as a type of fit statistic). The amount of the departure of
a discrimination from 1.0 is an indication of the degree to which that item misfits the Rasch model.
DISCRIM=NO Do not report an estimate of the empirical item discrimination.
DISCRIM=YES Report an estimate of the empirical item discrimination in the IFILE= and Tables 6.1, 10.1, etc.
An estimated discrimination of 1.0 accords with Rasch model expectations for an item of this difficulty. A value
greater than 1 means that the item discriminates between high and low performers more than expected for an
item of this difficulty. A value less than 1 means that the item discriminates between high and low performers less
than expected for an item of this difficulty. In general, the geometric mean of the estimated discriminations
approximates 1.0, the Rasch item discrimination.
Rasch analysis requires items which provide indication of relative performance along the latent variable. It is this
information which is used to construct measures. From a Rasch perspective, over-discriminating items are
tending to act like switches, not measuring devices. Under-discriminating items are tending neither to stratify nor
to measure.
Over-discrimination is thought to be beneficial in many raw-score and IRT item analyses. High discrimination
usually corresponds to low MNSQ values, and low discrimination with high MNSQ values. In Classical Test
Theory, Guttman Analysis and much of Item Response Theory, the ideal item acts like a switch. High performers
pass, low performers fail. This is perfect discrimination, and is ideal for sample stratification, but such an item
provides no information about the relative performance of low performers, or the relative performers of high
performers.
Winsteps reports an approximation to what the discrimination parameter value would have been in a 2-PL IRT
86
program, e.g., BILOG for MCQ, or PARSCALE for partial credit items. IRT programs artificially constrain
discrimination values in order to make them estimable, so Winsteps discrimination estimates tend to be wider
than 2-PL estimates. For the lower asymptote, see ASYMPTOTE=.
A Rasch-Andrich threshold discrimination is also reported, see Table 3.2.
With DISCRIM=YES,
+-----------------------------------------------------------------------------------------------+
|ENTRY
RAW
|
INFIT | OUTFIT |SCORE|ESTIM|
|
|NUMBER SCORE COUNT MEASURE ERROR|MNSQ ZSTD|MNSQ ZSTD|CORR.|DISCR| ACTS
|
|------------------------------------+----------+----------+-----+-----+------------------------|
|
23
40
74
2.19
.21|2.42
6.3|4.13
8.9| .00| .09| WATCH A RAT
|
|
17
93
74
.16
.19| .65 -2.7| .59 -2.5| .70| 1.20| WATCH WHAT ANIMALS EAT |
81.
DISFILE category/distractor/option count file
DISFILE=filename produces an output file containing the counts for each distractor or option or category of each
item. This file contains 1 heading lines (unless HLINES=N), followed by one line for each CODES= of each item
containing:
Columns:
Start
End Description
1 10 Item entry number
11 20 Response code in CODES= or *** = missing
21 30 Scored value of code. (MISSCORE= -1)
31 40 Count of response code in data set (excludes missing except for &quot;MISSING&quot; lines)
41 50 % of responses to this item
51 60 Count of response codes used for measurement (non-extreme persons and items)
61 70 % of responses to this item
71 80 Average person measure for these used responses
81 90 Standard Error of average person measure
91 100 Outfit Mean-square of responses in this category.
101 110 Point-biserial PTBIS=Y or point-measure correlation for this response (distractor, option)
112 - Item label
Since the DISFILE= has the same number of CODES= and MISSING entries for every item, the repeated fields
are filled out with &quot;0&quot; for any unobserved response codes.
When CSV=Y, commas separate the values with quotation marks around the &quot;Item label&quot;, response in CODES=,
and MISSING When CSV=T, the commas are replaced by tab characters.
Example: You wish to write a file on disk called &quot;DISCOUNT.TXT&quot; containing the item distractor counts from
table 14.3, for use in constructing your own tables:
DISFILE=DISCOUNT.TXT
ITEM
1
1
1
1
1
82.
CODE
***
A
B
C
D
SCORE
-1
1
0
2
1
ALL
0
2
35
16
1
ALL %
.00
3.70
64.81
29.63
1.85
USED
.0
2.0
35.0
16.0
1.0
USED % AVGE MEAS S.E. MEAS OUTF MNSQ
.00
.00
.00
.00
3.70
.18
.11
1.56
64.81
.39
.07
1.09
29.63
.59
.11
1.17
1.85
.36
.00
.78
PTMEA
.00
-.12
-.17
.24
-.03
LABEL
.7
A1
.7
A1
.7
A1
.7
A1
.7
A1
DISTRACTOR output option counts in Tables 10, 13-15
This variable controls the reporting of counts of option, distractor or category usage in Table 10.3 etc. The
standard is DISTRT=Y, if more than two values are specified in CODES=.
DISTRACTOR=N Omit the option or distractor information.
DISTRACTOR=Y Include counts, for each item, for each of the values in CODES=, and for the number of
responses counted as MISSCORE=
87
83.
DPF columns within item label for Table 31
DPF= specifies the part of the item label which is to be used for classifying items in order to identify Differential
Person Function (DPF) - uniform or non-uniform - using the column selection rules. See also DPF Table and DIF
and DPF considerations.
See ISUBTOTAL for format.
See DPF Table.
Example 1: Columns 3 and 4 of the item label (between &amp;END and END LABELS) contains content-area code:
DPF = $S3E4 ; start in column 3 and end in column 4 of item label
tfile=*
31
; Table 31 is DPF Table (or use Output Tables menu)
*
Example 2: Columns 3 of the item label contains a content code. Column 5-6 have a complexity code. Two
independent DIF analyses are needed:
DPF = *
$S3W1
; content analysis
$S5W2
; complexity
*
tfile=*
31
; Table 31 is DPF Table (or use Output Tables menu)
*
84.
EDFILE edit data file
This permits the replacement of data values in your data file with other values, without altering the data file. Data
values are in the original data file format, specified in CODES=. If specified as decimals, they are rounded to the
nearest integers.
Its format is:
person entry number
item entry number
replacement data value
Ranges are permitted: first-last.
Example 1: In your MCQ test, you wish to correct a data-entry error. Person 23 responded to item 17 with a D,
not whatever is in the data file.
EDFILE=*
23 17 D
*
; person 23, item 17, data value of D
Example 2: Person 43 failed to read the attitude survey instructions correctly for items 32-56. Mark these
missing.
43 32-56 &quot; &quot; ; person 43, items 32 to 56, blanks are missing data.
Example 3: Persons 47-84 are to be given a rating of 4 on item 16.
47-84 16 4
; persons 47 to 84, item 16, data value of 4
Example 4: Items 1-10 are all to be assigned a datum of 1 for the control subsample, persons 345-682.
345-682 1-10 1
; persons 345-682, items 1 to 10, data value 1.
Example 5: Missing data values are to be imputed with the values nearest to their expectations.
88
a. Produce PFILE=, IFILE= and SFILE= from the original data (with missing).
b. Use those as PAFILE=, IAFILE=, SAFILE= anchor files with a data set in which all the original non-missing
data are made missing, and vice-versa - it doesn't matter what non-missing value is used.
c. Produce XFILE= to obtain a list of the expected values of the originally missing data.
d. Use the EDFILE= command to impute those values back into the data file. It will round expected values to
the nearest integer, for us as a category value.
17 6 2.6
85.
; persons 17, item 6, expected value 2.6, imputed as category &quot;3&quot;.
END LABELS or END NAMES
The first section of a control file contains control variables, one per line, and ends with &amp;END. This is followed by
the second section of item labels, one per line, matching the items in the analysis. This sections ends with END
LABELS or END NAMES, which mean the same thing. The data can follow as a third section, or the data can be
in a separate file specified by the control variable DATA=.
TITLE = &quot;5 item test&quot;
ITEM1 = 1
NI = 5
.....
&amp;END
Addition
; label for item 1
Subtraction ; label for item 2
Multiplication ; label for item 3
Division
; label for item 4
Geometry ; label for item 5
END LABELS
.....
; data here
86.
EQFILE code equivalences
This specifies that different demographic or item-type codes are to be reported as one code. This is useful for
Tables 27, 28, 30, 31, 33 and Use EQFILE=filename or EQFILE=*, followed by a list, followed by a *. These
values can be overwritten from the equivalence boxes when invoking the Tables from the Output Tables menu.
The format is
@Field name = $S1W1
; user defined field name and location, see selection rules.
EQFILE=* ; start of list
@Field name ; field to be referred to
Base Code Code Code Code .....
; code list
Base Code Code Code Code .....
Base Code Code Code Code .....
@Field name ; field to be referred to
Base Code Code Code Code .....
; code list
Base Code Code Code Code .....
Base Code Code Code Code .....
* ; end of list
where @Field name is the name of field in the person or item label, such as
@GENDER = $S1W1 ; M or F
@STRAND = $S10W2 ; 01 to 99
Base is the demographic or item-type code to be reported. It need not be present in a label
Code is a demographic or item-type code to be included with the Base code.
87.
EXTRSC extreme score correction for extreme measures
EXTRSCORE= is the fractional score point value to subtract from perfect scores, and to add to zero scores, in
89
order to estimate finite values for extreme scores (formerly MMADJ=). Look at the location of the E's in the tails
of the test ogive in Table 20. If they look too far away, increase EXTRSC= by 0.1. If they look too bunched up,
reduce EXTRSC= by 0.1.
The measure corresponding to an extreme (perfect or zero) score is not estimable, but the measure
corresponding to a score a little than perfect, or a little more than zero, is estimable, and is often a useful measure
to report.
Rasch programs differ in the way they estimate measures for extreme scores. Adjustment to the value of
EXTRSC= can enable a close match to be made to the results produced by other programs.
There is no &quot;correct&quot; answer to the question: &quot;How large should EXTRSC= be?&quot; The most conservative value,
and that recommended by Joseph Berkson, is 0.5. Some work by John Tukey indicates that 0.167 is a
reasonable value. The smaller you set EXTRSC=, the further away measures corresponding to extreme scores
will be located from the other measures. The technique used here is Strategy 1 in
www.rasch.org/rmt/rmt122h.htm.
Treatment of Extreme Scores
Tables
Placed at extremes of map
1, 12, 16
Positioned by estimated measure 13, 17, 22
Positioned by other criteria
14, 15, 18, 19
Omitted
2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 20, 21
Output files
IFILE=, ISFILE=, PFILE=, RFILE=
SFILE=, XFILE=
Example 1: You wish to estimate conservative finite measures for extreme scores by subtracting 0.4 score points
from each perfect score and adding 0.4 score points to each zero person score.
EXTRSCORE=0.4
Example 2: With the standard value of EXTRSCORE=, this Table is produced:
+-------------------------------------------------------------------------+
|ENTRY
RAW
|
INFIT | OUTFIT |PTBIS|
|
|NUMBER SCORE COUNT MEASURE ERROR|MNSQ ZSTD|MNSQ ZSTD|CORR.| PERSON |
|------------------------------------+----------+----------+-----+--------|
|
46
60
20
7.23
1.88| MAXIMUM ESTIMATED MEASURE | XPQ003 |
|
94
62
21
5.83
1.12| .44
-.9| .08
-.6| .62| XPQ011 |
|
86
18
6
5.11
1.90| MAXIMUM ESTIMATED MEASURE | XPQ009 |
|
64
50
17
4.94
1.09| .53
-.7| .13
-.6| .60| XPQ006 |
Here, the 5.11 corresponds to a perfect score on 6 easier items. The 5.83 was obtained on 21 harder items
(perhaps including the 6 easier items.) To adjust the &quot;MAXIMUM ESTIMATED&quot; to higher measures, lower the
value of EXTRSCORE=, e.g., to EXTRSCORE=0.2
88.
FITHIGH higher bar in charts
Use FITHIGH= to position the higher acceptance bars in Tables like 6.2. Use FITLOW= to position the lower bar.
FITHIGH=0 cancels the instruction.
Example: We want the lower mean-square acceptance bar to be shown at 1.4
FITHIGH=1.4 ; show higher fit bar
CHART=YES ; produce Tables like 6.2
ACTS FIT GRAPH: MISFIT ORDER
+------------------------------------------------------------------------------------------------+
|ENTRY | MEASURE | INFIT MEAN-SQUARE | OUTFIT MEAN-SQUARE |
|
|NUMBER| + |0.0
1 1.4
2|0.0
1 1.4
2| ACTS
G |
|------+---------+---------------------+---------------------+-----------------------------------|
|
23|
*|:
.
:
*|:
.
:
*| WATCH A RAT
0 |
|
5|
*|:
.
:
*|:
.
:
*| FIND BOTTLES AND CANS
0 |
|
20|
* |:
. *:
|:
.
: *
| WATCH BUGS
0 |
|
18|*
|:
. * :
|:
.* :
| GO ON PICNIC
0 |
|
8|
* |:
.* :
|:
. * :
| LOOK IN SIDEWALK CRACKS
0 |
90
89.
FITI item misfit criterion
Specifies the minimum t standardized fit value at which items are selected for reporting as misfits. For Table 10,
the table of item calibrations in fit order, an item is omitted only if the absolute values of both t standardized fit
statistics are less than FITI=, both mean-square statistics are closer to 1 than (FITI=)/10, and the item pointbiserial correlation is positive.
For Table 11, the diagnosis of misfitting items, all items with a t standardized fit greater than FITI= are reported.
Selection is based on the OUTFIT statistic, unless you set OUTFIT=N in which case the INFIT statistic is used. If
MNSQ=YES, then selection is based on the mean-square value: 1 + FITI=/10.
Example 1: You wish to focus on grossly &quot;noisy&quot; items in Tables 10 and 11.
FITI=4 an extreme positive value
Example 2: You wish to include all items in Tables 10 and 11.
FITI= -10 a value more negative than any fit statistic
90.
FITLOW lower bar in charts
Use FITLOW= to position the lower acceptance bars in Tables like 6.2. Use FITHIGH= to position the higher bar.
FITLOW=0 cancels the instruction.
Example: We want the lower mean-square acceptance bar to be shown at 0.6
FITLOW=0.6 ; show lower fit bar
CHART=YES ; produce Tables like 6.2
PUPIL FIT GRAPH: MISFIT ORDER
+--------------------------------------------------------------------------------------------+
|ENTRY | MEASURE | INFIT MEAN-SQUARE | OUTFIT MEAN-SQUARE |
|
|NUMBER| + | .0 0.6 1
:2| .0 0.6 1
:2| PUPIL
|
|------+---------+---------------------+---------------------+-------------------------------|
|
46| *
|
:
. *
|T
:
.*
| MULLER, JEFF
|
|
26| *
|
:* .
|U
:
. *
| NEIMAN, RAYMOND
|
|
30| *
|
:
*
|V
:
. *
| NORDGREN, JAN SWEDE
|
|
23|
*
|
* :
.
|c * :
.
| VROOM, JEFF
|
|
22|
*
|
* :
.
|b *
:
.
| HOGAN, Kathleen
|
|
21| *
| *
:
.
|a *
:
.
| RISEN, NORM L.
|
91.
FITP person misfit criterion
Specifies the minimum t standardized fit value at which persons are selected for reporting as misfits. For Table 6,
person measures in fit order, a person is omitted only if the absolute values of both t standardized fit statistics are
less than FITP=, both mean-square statistics are closer to 1 than (FITP=)/10, and the person point-biserial
correlation is positive.
For Table 7, the diagnosis of misfitting persons, persons with a t standardized fit greater than FITP= are reported.
Selection is based on the OUTFIT statistic, unless you set OUTFIT=N in which case the INFIT statistic is used. If
MNSQ=YES, then selection is based on the mean-square value: 1 + FITP=/10.
Example 1: You wish to examine wildly guessing persons in Tables 6 and 7.
FITP= 3
an extreme positive value
Example 2: You wish to include all persons in Tables 6 and 7.
FITP= -10 a value more negative than any fit statistic
92.
FORMAT reformat data
Enables you to process awkwardly formatted data! But MFORMS= is easier
FORMAT= is rarely needed when there is one data line per person.
91
Place the data in a separate file, then the Winsteps screen file will show the first record before and after
FORMAT=
Control instructions to pick out every other character for 25 two-character responses, then a blank, and then the
person label:
XWIDE=1
data=datafile.txt
format=(T2,25(1A,1X),T90,1A,Tl1,30A)
This displays on the Winsteps screen:
Opening: datafile.txt
Input Data Record before FORMAT=:
1
2
3
4
5
6
7
1234567890123456789012345678901234567890123456789012345678901234567890
---------------------------------------------------------------------01xx 1x1 10002000102020202000201010202000201000200ROSSNER, MARC DANIEL
Input Data Record after FORMAT=:
1x11102012222021122021020 L
^I
^N^P
^I is Item1= column
^N is the last item according to NI=
^P is Name1= column
FORMAT= enables you to reformat one or more data record lines into one new line in which all the component
parts of the person information are in one person-id field, and all the responses are put together into one
continuous item-response string. A FORMAT= statement is required if
1) each person's responses take up several lines in your data file.
2) if the length of a single line in your data file is more than 10000 characters.
3) the person-id field or the item responses are not in one continuous string of characters.
4) you want to rearrange the order of your items in your data record, to pick out sub-tests, or to move a set of
connected forms into one complete matrix.
5) you only want to analyze the responses of every second, or nth, person.
FORMAT= contains up to 512 characters of reformatting instructions, contained within (..), which follow special
rules. Instructions are:
nA read in n characters starting with the current column, and then advance to the next column after them.
Processing starts from column 1 of the first line, so that 5A reads in 5 characters and advances to the sixth
column.
nX means skip over n columns. E.g. 5X means bypass this column and the next 4 columns.
Tc go to column c. T20 means get the next character from column 20.
T55 means &quot;tab&quot; to column 55, not &quot;tab&quot; passed 55 columns (which is TR55).
TLc go c columns to the left. TL20 means get the next character the column which is 20 columns to the left of the
current position.
TRc
go c columns to the right. TR20 means get the next character the column which is 20 columns to the right
of the current position.
/
means go to column 1 of the next line in your data file.
n(..)repeat the string of instructions within the () exactly n times.
,
a comma is used to separate the instructions.
92
Set XWIDE=2 and you can reformat your data from original 1 or 2 column entries. Your data will all be analyzed
as XWIDE=2. Then:
nA2 read in n pairs of characters starting with the current column into n 2-character fields of the formatted record.
(For responses with a width of 2 columns.)
A1 read in n 1-character columns, starting with the current column, into n 2-character fields of the formatted
record.
Always use nA1 for person-id information. Use nA1 for responses entered with a width of 1-character when there
are also 2-character responses to be analyzed. When responses in 1-character format are converted into 2character field format (compatible with XWIDE=2), the 1-character response is placed in the first, left, character
position of the 2-character field, and the second, right, character position of the field is left blank. For example,
the 1-character code of &quot;A&quot; becomes the 2-character field &quot;A &quot;. Valid 1-character responses of &quot;A&quot;, &quot;B&quot;, &quot;C&quot;, &quot;D&quot;
must be indicated by CODES=&quot;A B C D &quot; with a blank following each letter.
ITEM1= must be the column number of the first item response in the formatted record created by the
FORMAT= statement. NAME1= must be the column number of the first character of the person-id in the
formatted record.
Example 1: Each person's data record file is 80 characters long and takes up one line in your data file. The
person-id is in columns 61-80. The 56 item responses are in columns 5-60. Codes are &quot;A&quot;, &quot;B&quot;, &quot;C&quot;, &quot;D&quot;. No
FORMAT= is needed. Data look like:
xxxxDCBDABCADCDBACDADABDADCDADDCCDADDCAABCADCCBBDADCACDBBADCZarathrustra-Xerxes
Without FORMAT=
XWIDE=1
ITEM1=5
NI=56
NAME1=61
NAMLEN=20
CODES=ABCD
response width (the standard)
start of item responses
number of items
start of name
length of name
valid response codes
With FORMAT=
Reformatted record will look like:
DCBDABCADCDBACDADABDADCDADDCCDADDCAABCADCCBBDADCACDBBADCZarathrustra-Xerxes
XWIDE=1
response width (the standard)
FORMAT=(4X,56A,20A) skip unused characters
ITEM1=1
start of item responses
NI=56
number of items
NAME1=57
start of name
NAMLEN=20
length of name
CODES=ABCD
valid response codes
Example 2: Each data record is one line of 80 characters. The person-id is in columns 61-80. The 28 item
responses are in columns 5-60, each 2 characters wide. Codes are &quot; A&quot;, &quot; B&quot;, &quot; C&quot;, &quot; D&quot;. No FORMAT= is
necessary. Data look like:
xxxx C D B A C B C A A D D D D C D D C A C D C B A C C B A CZarathrustra-Xerxes
Without FORMAT=
XWIDE=2
response width
ITEM1=5
start of item responses
NI=28
number of items
NAME1=61
start of name
NAMLEN=20
length of name
CODES=&quot; A B C D&quot;
valid response codes
With FORMAT=
Columns of reformatted record:
1-2-3-4-5-6-7-8-9-0-1-2-3-4-5-6-7-8-9-0-1-2-3-4-5-6-7-8-90123456789012345678
C D B A C B C A A D D D D C D D C A C D C B A C C B A CZarathrustra-Xerxes
XWIDE=2
response width
FORMAT=(4X,28A2,20A1)
skip unused characters
93
ITEM1=1
start of item responses in formatted record
NI=28
number of items
NAME1=29
start of name in &quot;columns&quot;
NAMLEN=20
length of name
CODES=&quot; A B C D&quot;
valid response codes
Example 3: Each person's data record is 80 characters long and takes one line in your data file. Person-id is in
columns 61-80. 30 1-character item responses, &quot;A&quot;, &quot;B&quot;, &quot;C&quot; or &quot;D&quot;, are in columns 5-34, 13 2-character item
responses, &quot;01&quot;, &quot;02&quot; or &quot;99&quot;, are in 35-60.
xxxxDCBDABCADCDBACDADABDADCDADDCCA01990201019902010199020201Zarathrustra-Xerxes.
becomes on reformatting:
Columns:
1234567890123456789012345678901-2-3-4-5-6-7-8-9-0-1-2-3-45678901234567890123
DCBDABCADCDBACDADABDADCDADDCCA01990201019902010199020201Zarathrustra-Xerxes
XWIDE=2
analyzed response width
FORMAT=(4X,30A1,13A2,20A1) skip unused
ITEM1=1
start of item responses in formatted record
NI=43
number of items
NAME1=44
start of name
NAMLEN=20
length of name
CODES=&quot;A B C D 010299&quot;
valid responses
^ 1-character code followed by blank
Example 4: The person-id is 10 columns wide in columns 15-24 and the 50 1-column item responses, &quot;A&quot;, &quot;B&quot;,
&quot;C&quot;, &quot;D&quot;, are in columns 4000-4019, then in 4021-50. Data look like:
xxxxxxxxxxxxxxJohn-Smithxxxx....xxxDCBACDADABCADCBCDABDxBDCBDADCBDABDCDDADCDADBBDCDABB
becomes on reformatting:
John-SmithDCBACDADABCADCBCDABDBDCBDADCBDABDCDDADCDADBBDCDABB
FORMAT=(T15,10A,T4000,20A,1X,30A)
NAME1=1
start of person name in formatted record
NAMLEN=10
length of name (automatic)
ITEM1=11
start of items in formatted record
NI=50
50 item responses
CODES=ABCD
valid response codes
Example 5: There are five records or lines in your data file per person. There are 100 items. Items 1-20 are in
columns 25-44 of first record; items 21-40 are in columns 25-44 of second record, etc. The 10 character personid is in columns 51-60 of the last (fifth) record. Codes are &quot;A&quot;, &quot;B&quot;, &quot;C&quot;, &quot;D&quot;. Data look like:
xxxxxxxxxxxxxxxxxxxxxxxxACDBACDBACDCABACDACD
xxxxxxxxxxxxxxxxxxxxxxxxDABCDBACDBACDCABACDA
xxxxxxxxxxxxxxxxxxxxxxxxACDBACDBACDCABACDACD
xxxxxxxxxxxxxxxxxxxxxxxxDABCDBACDBACDCABACDA
xxxxxxxxxxxxxxxxxxxxxxxxABCDBACDBACDCABACDADxxxxxxMary-Jones
becomes:
ACDBACDBACDCABACDACDDABCDBACDBACDCABACDAACDBACDBACDCABACDACDDABCDBACDBACDCABACDAABCDBACDBACDC
ABACDADMary-Jones
FORMAT=(4(T25,20A,/),T25,20A,T51,10A)
ITEM1=1
start of item responses
NI=100
number of item responses
NAME1=101
start of person name in formatted record
NAMLEN=10
length of person name
CODES=ABCD
valid response codes
Example 6: There are three lines per person. In the first line from columns 31 to 50 are 10 item responses, each
2 columns wide. Person-id is in the second line in columns 5 to 17. The third line is to be skipped. Codes are
&quot;A &quot;, &quot;B &quot;, &quot;C &quot;, &quot;D &quot;. Data look like:
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx A C B D A D C B A Dxxxxxxxx
xxxxJoseph-Carlosxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
becomes:
94
Columns:
1-2-3-4-5-6-7-8-9-0-1234567890123
A C B D A D C B A DJoseph-Carlos
FORMAT=(T31,10A2,/,T5,13A1,/)
ITEM1=1
start of item responses
NI=10
number of items
XWIDE=2
2 columns per response
NAME1=11
starting &quot;A&quot; of person name
NAMLEN=13
length of person name
CODES='A B C D '
valid response codes
If the third line isn't skipped, format a redundant extra column in the skipped last line. Replace the first
control variable in this with:
FORMAT=(T31,10A2,/,T5,13A1,/,A1) last A1 unused
Example 7: Pseudo-random data selection
You have a file with 1,000 person records. This time you want to analyze every 10th record, beginning with the
3rd person in the file, i.e., skip two records, analyze one record, skip seven records, and so on. The data records
are 500 characters long.
XWIDE = 1
FORMAT = (/,/,500A,/,/,/,/,/,/,/)
or
XWIDE = 2
FORMAT = (/,/,100A2,300A1,/,/,/,/,/,/,/)
responses, 300 other columns
; 100 2-character
Example 8: Test A, in file EXAM10A.TXT, and TEST B, in EXAM10B.TXT, are both 20 item tests. They have 5
items in common, but the distractors are not necessarily in the same order. The responses must be scored on an
individual test basis. Also the validity of each test is to be examined separately. Then one combined analysis is
wanted to equate the tests and obtain bankable item difficulties. For each file of original test responses, the
person information is in columns 1-25, the item responses in 41-60.
The combined data file specified in EXAM10C.TXT, is to be in RFILE= format. It contains
Person information 30 characters (always)
Item responses
Columns 31-64
The identification of the common items is:
Test
Item Number (=Location in item string)
Both: 1 2 3 4 5 6-20
21-35
A: 3 1 7 8 9 2,4-6,10-20
B: 4 5 6 2 11 1,3,7-10,12-20
I. From Test A, make a response (RFILE=) file rearranging the items with FORMAT=.
; This file is EXAM10A.TXT
&amp;INST
TITLE=&quot;Analysis of Test A&quot;
RFILE=EXAM10AR.TXT ; The constructed response file for Test A
NI=20
FORMAT=(25A,T43,A,T41,A,T47,3A,T42,A,T44,3A,T50,11A)
ITEM1=26
; Items start in column 26 of reformatted record
CODES=ABCD#
; Beware of blanks meaning wrong!
; Use your editor to convert all &quot;wrong&quot; blanks into another code,
; e.g., #, so that they will be scored wrong and not ignored as missing.
KEYFRM=1
; Key in data record format
&amp;END
Key 1 Record
CCBDACABDADCBDCABBCA
BANK 1
TEST A 3
; first item name
.
BANK 20 TEST A 20
END NAMES
95
Person 01 A
.
Person 12 A
BDABCDBDDACDBCACBDBA
BADCACADCDABDDDCBACA
The RFILE= file, EXAM10AR.TXT, is:
Person 01 A
Person 02 A
.
Person 12 A
00001000010010001001
00000100001110100111
00100001100001001011
II. From Test B, make a response (RFILE=) file rearranging the items with FORMAT=. Responses unique to Test
A are filled with 15 blank responses to dummy items.
; This file is EXAM10B.TXT
&amp;INST
TITLE=&quot;Analysis of Test B&quot;
RFILE=EXAM10BR.TXT ; The constructed response file for Test B
NI=35
FORMAT=(25A,T44,3A,T42,A,T51,A,T100,15A,T41,A,T43,A,T47,4A,T52,9A)
; Blanks are imported from an unused part of the data record to the
right!
; T100 means &quot;go beyond the end of the data record&quot;
; 15A means &quot;get 15 blank spaces&quot;
ITEM1=26
; Items start in column 26 of reformatted record
CODES=ABCD#
; Beware of blanks meaning wrong!
KEYFRM=1
; Key in data record format
&amp;END
Key 1 Record
CDABCDBDABCADCBDBCAD
BANK 1
TEST B 4
.
BANK 5
TEST B 11
BANK 6
TEST A 2
.
BANK 20 TEST A 20
BANK 21 TEST B 1
.
BANK 35 TEST B 20
END NAMES
Person 01 B
BDABDDCDBBCCCCDAACBC
.
Person 12 B
BADABBADCBADBDBBBBBB
The RFILE= file, EXAM10BR.TXT, is:
Person
Person
.
Person
Person
01 B
02 B
10111
00000
010101001000100
010000000001000
11 B
12 B
00010
00000
001000000000100
000101000101000
III. Analyze Test A's and Test B's RFILE='s together:
; This file is EXAM10C.TXT
&amp;INST
TITLE=&quot;Analysis of Tests A &amp; B (already scored)&quot;
NI=35
ITEM1=31
; Items start in column 31 of RFILE=
CODES=01
; Blanks mean &quot;not in this test&quot;
DATA=EXAM10AR.TXT+EXAM10BR.TXT
; Combine data files
;
;
;
;
or, first, at the DOS prompt,
C:&gt; COPY EXAM10AR.TXT+EXAM10BR.TXT EXAM10AB.TXT(Enter)
then, in EXAM10C.TXT,
DATA=EXAM10AB.TXT
96
PFILE=EXAM10CP.TXT
IFILE=EXAM10CI.TXT
tfile=*
3
10
*
PRCOMP=S
&amp;END
BANK 1
TEST A 3 B
.
BANK 35 TEST B 20
END NAMES
;
;
;
;
;
Person measures for combined tests
Item calibrations for combined tests
List of desired tables
Table 3 for summary statistics
Table 10 for item structure
; Principal components/contrast analysis with standardized residuals
4
Shortening FORMAT= statements
If the required FORMAT= statement exceeds 512 characters, consider using this technique:
Relocate an entire item response string, but use an IDFILE= to delete the duplicate items, i.e., replace them by
blanks. E.g., for Test B, instead of
FORMAT=(25A, T44,3A,T42,A,T51,A, T100,15A, 41,A,T43,A,T47,4A,T52,9A)
NI=35
Put Test 2 as items 21-40 in columns 51 through 70:
FORMAT=(25A, T44,3A,T42,A,T51,A, T100,15A, T41,20A)
NI=40
Blank out (delete) the 5 duplicated items with an IDFILE= containing:
24-26
22
31
93.
FORMFD the form feed character
Do not change FORMFD= unless you have problems printing the tables or importing them into some other
program.
The form feed character indicates the start of a new page of print-out. The DOS standard is Ctrl+L (ASCII 12)
which is what represented by ^ (Shift+6). The DOS standard is understood by most word-processing software
and PC printers as the instruction to skip to the top of a new page, i.e., form feed. The ASA (FORTRAN) form
feed character is 1.
WordPad does not have a &quot;form feed&quot; or page advance feature. You must put extra blank lines in the
output files.
Example 1: You want your EPSON LQ-500 printer to form feed automatically at each new page of output. (You
have already set the printer to use compressed print, at 15 cpi, because output lines contain up to 132
characters):
FORMFD=^ (the standard)
Example 2: Your main-frame software understands a &quot;1&quot; in the first position of a line of print-out to indicate the
top of a new page:
FORMFD=1
94.
FRANGE half-range of fit statistics on plots
Specifies the t standardized fit Y-axis half-range, (i.e. range away from the origin), for the t standardized fit plots.
FRANGE= is in units of t standardized fit (i.e., expected mean = 0, standard deviation = 1).
Example:
You want the fit plots to display from -3 to +3 units of t standardized fit:
FRANGE=3
97
95.
GRFILE probability curve coordinate output file
If GRFILE=filename is specified, a file is output which contains a list of measures (x-axis coordinates) and
corresponding expected scores and category probabilities (y-axis coordinates) to enable you to use your own
plotting program to produce item-category plots like those in Table 21.
These are usually relative to the item difficulty, so
If you want the item information function relative to the item difficulty, use the GRFILE=
If you want the item information function relative to the latent variable, add the item difficulty to the GRFILE=
value.
The plotted range is at least MRANGE= away from its center.
This file contains:
1. An example item number from the response-structure grouping (I5) - see ISGROUPS=
2. The measure (F7.2) (user-rescaled by USCALE=)
3. Expected score (F7.2)
4. Statistical information (F7.2)
5. Probability of observing lowest category (F7.2)
6 etc. Probability of observing higher categories (F7.2).
If CSV=Y, values are separated by commas. When CSV=T, values are separated by tab characters.
Example: You wish to write a file on disk called &quot;MYDATA.GR&quot; containing x- and y-coordinates for plotting your
own category response curves.
GRFILE=MYDATA.GR
With CSV=Y
&quot;PROBABILITY CURVES FOR LIKING FOR SCIENCE (Wright &amp; Masters p.18) Jul
;ITEM,MEAS,SCOR,INFO,0,1,2
1,-3.00,.11,.10,.89,.10,.00
1,-2.94,.12,.11,.89,.11,.00
1,-2.88,.12,.11,.88,.12,.00
1,-2.82,.13,.12,.87,.12,.00
1,-2.76,.14,.12,.87,.13,.00
1,-2.70,.14,.13,.86,.14,.00
4 16:03 2000&quot;
With CSV=N (fixed spacing)
; PROBABILITY CURVES FOR LIKING FOR SCIENCE (Wright &amp; Masters p.18) Jul
; ITEM
MEAS
SCOR
INFO
0
1
2
1 -3.00
.11
.10
.89
.10
.00
1 -2.94
.12
.11
.89
.11
.00
1 -2.88
.12
.11
.88
.12
.00
1 -2.82
.13
.12
.87
.12
.00
1 -2.76
.14
.12
.87
.13
.00
96.
4 16:03 2000
GROUPS or ISGROUPS assigns items to rating scale groupings
Items in the same &quot;grouping&quot; share the same dichotomous, rating scale or partial credit response structure. For
tests comprising only dichotomous items, or for tests in which all items share the same rating (or partial credit)
scale definition, all items belong to one grouping, i.e., they accord with the simple dichotomous Rasch model or
the Andrich &quot;Rating Scale&quot; model. For tests using the &quot;Masters' Partial Credit&quot; model, each item comprises its
own grouping (dichotomous or polytomous). For tests in which some items share one polytomous responsestructure definition, and other items another response-structure definition, there can be two or more item
groupings. Groups are called &quot;blocks&quot; in the PARSCALE software.
98
where P is a probability, and the Rasch parameters are Bn, the ability of person, Dgi, the difficulty of item i of
grouping g, and Fgj, the Rasch-Andrich threshold between categories j-1 and j of grouping g. If there is only one
grouping, this is the Andrich &quot;rating scale&quot; model. If each item forms a grouping of its own, i.e., g=i, this is the
Masters' &quot;partial credit' model. When several items share the rating scale, then this could be called an itemgrouping-level Andrich rating-scale model, or an item-grouping-level Masters' partial-credit model. They are the
same thing.
ISGROUPS= also acts as IREFER=, when IVALUE= is specified, but IREFER= is omitted.
ISGROUPS= has three forms: ISGROUPS=1101110 and ISGROUPS=* list * and ISGROUPS=*filename
ISGROUPS=&quot; &quot;
(standard if only one model specified with MODELS=)
All items belong to one grouping. This is sometimes called &quot;Andrich's Rating Scale Model&quot;
ISGROUPS=0
(standard if MODELS= specifies multiple models)
Each item has a grouping of its own, so that a different response structure is defined for each item, as in the
&quot;Masters' Partial Credit model&quot;. This is also used for rank order data.
ISGROUPS= some combination of numbers and letters: 0's, 1's, 2's, A's, B's, etc. (&quot;a&quot; is the same as &quot;A&quot;), also #,
@, !,&amp; etc.
Use only one letter or number per grouping, regardless of the XWIDE= setting.
Items are assigned to the grouping label whose location in the ISGROUPS= string matches the item
sequence number. Each item assigned to grouping 0 is allocated to a &quot;partial credit&quot; grouping by itself. Items in
groupings labeled &quot;1&quot;, &quot;2&quot;, &quot;A&quot;, etc. share the response-structure definition with other items in the same labeled
grouping.
Valid one-character ISGROUPS= codes include: !#$%&amp;./123456789&lt;&gt;@ABCDEFGHIJKLMNOPQRSTUVWXYZ^_|~
A-Z are the same as a-z.
For the ISGROUPS= specification, &quot;0&quot; has the special meaning: &quot;this is a one item grouping&quot; - and can be used
for every 1 item grouping.
Characters with ASCII codes from 129-255 can also be used, but display peculiarly:
&Icirc;&Iuml;&ETH;&Ntilde;&Ograve;&Oacute;&Ocirc;&Otilde;&Ouml;&times;&Oslash;&Ugrave;&Uacute;&Ucirc;&Uuml;&Yacute;&THORN;&szlig;&agrave;&aacute;&acirc;&atilde; etc.
When XWIDE=2 or more, then
either (a) Use one character per XWIDE= and blanks,
NI=8
XWIDE=2
ISGROUPS=' 1 0 1 0 1 0 1 1'
or (b) Use one character per item with no blanks
NI=8
XWIDE=2
ISGROUPS='10101011'
ISGROUPS=*
item number grouping code
item number-item number grouping code
*
Each line has an item number, e.g., 23, or an item number range, e.g., 24-35, followed by a space and then a
grouping code, e.g., 3. The items can be out of order. If an item is listed twice, the last entry dominates.
ISGROUPS=*file name
99
This has a file with the format of the ISGROUPS=* list.
Particularly with ISGROUPS=0, some extreme categories may only be observed for persons extreme scores. To
reinstate them into the measurement analysis, see Extreme Categories: Rescuing the Dropped.
Groupings vs. Separate Analyses
ISGROUPS= is very flexible for analyzing together different item response structures in one analysis. Suppose
that an attitude survey has 20 Yes-No items , followed by 20 5-category Likert (Strongly disagree - disagree neutral - agree- strongly agree) items, followed by 20 3-category frequency (never - sometimes - often) items.
When possible, we analyze these together using ISGROUPS=. But sometimes the measurement characteristics
are too different. When this happens, the fit statistics stratify by item type: so that, say, all the Yes-No items
overfit, and all the Frequency items underfit. Then analyze the test in 3 pieces, and equate them together usually into the measurement framework of the response structure that is easiest to explain. In this case, the
Yes-No framework, because probabilistic interpretation of polytomous logits is always difficult to explain or
perform.
The &quot;equation&quot; would be done by cross-plotting the person measures for different item types, and getting the
slope and intercept of the conversion from that. Drop out of the &quot;equation&quot; any noticeably off-trend-line measures.
These are person exhibiting differential performance on the different item types.
Example 1: Responses to all items are on the same 4-point rating scale, an Andrich &quot;Rating Scale&quot; model,
ISGROUPS=&quot; &quot;
Example 2: An arithmetic test in which partial credit is given for an intermediate level of success on some items.
There is no reason to assert that the intermediate level is equivalent for all items. 0=No success, 1=Intermediate
success (or complete success on items with no intermediate level), 2=Complete success on intermediate level
items.
CODES=012 valid codes
ISGROUPS=0 each item has own response structure, i.e., Masters' Partial Credit model
or
ISGROUPS=*
1 0 ; item 1 is in Grouping 0, no other items mentioned, so all assumed to be in Grouping 0
*
Example 3: An attitude survey consists of four questions on a 0,1,2 rating scale (grouping 1), an Andrich &quot;Rating
Scale&quot; model, followed by three 0,1 items (grouping 2), an other Andrich &quot;Rating Scale&quot; model, and ends with
one 0,1,2,3,4,5 question (grouped by itself, 0), a Masters' &quot;Partial Credit&quot; model.
NI=8
number of items
CODES=012345
valid codes for all items
ISGROUPS=11112220
the item groupings
or
ISGROUPS=*
1-4 1
5-7 2
8 0
; this line is optional, 0 is the standard.
*
When XWIDE=2, use two columns for each ISGROUPS= code. Each ISGROUPS= code must be one character,
a letter or number, specified once in the two columns, e.g. &quot; 1&quot; or &quot;1 &quot; mean &quot;1&quot;, and &quot; 0&quot; or &quot;0 &quot; mean &quot;0&quot;.
Example 4: You wish most items on the &quot;Liking for Science&quot; Survey to share the same rating scale, an Andrich
&quot;Rating Scale&quot; model, (in Grouping A). Items about birds (1, 10, 21) are to share a separate response structure,
another Andrich &quot;Rating Scale&quot; model, (in Grouping B). Items 5 (cans) and 18 (picnic) each has its own
response structure, i.e., the &quot;Masters' Partial Credit&quot; model, (Grouping 0).
NI=25
number of items
XWIDE=2
CODES=000102
valid codes for all items
ISGROUPS=' B A A A 0 A A A A B A A A A A A A 0 A A B A A A A'
100
item groupings - use only one letter/number codes.
or
ISGROUPS=* ; XWIDE=2 is not a worry here, but use one letter/number codes.
1-25
A ; most items in grouping A
1 B ; item 1 transferred to grouping B
10 B
21 B
5 0 ; grouping 0 means item is by itself
18 0
*
Example 5: Four separate tests of patient performance (some on a 4-point rating scale, some on a 5-point rating
scale) are to be Rasch-analyzed. All 500 persons were given all 4 tests. I analyzed each separately, to get an
idea of good-fitting and bad-fitting items, etc. Now, I'd like to run all 4 tests together using a partial credit model.
There is no problem running all four tests together. Put them all in one file, or use MFORMS=. If you intend
every item of every test to have its own rating scale (i.e., a strict partial-credit model), use ISGROUPS=0. But if
you intend items on test 1 to share the same rating scale, similarly test 2 etc. (i.e., a test-level partial-credit
model), then specify ISGROUPS=111111112222233334444.... matching the grouping number indicators to
the count of items in each test.
Example 6: Items are to be rescored in 3 different ways, and then the items are to be divided into 4 rating scale
structures.
ISGROUPS=11112223312444
; 4 RATING SCALE GROUPINGS
IREFER =AAAABBBCCABBBB
; 3 RECODINGS
CODES
=01234 ; ORIGINAL CODES IN DATA
IVALUEA =01234 ; ORIGINAL CODING MAINTAINED - THIS LINE CAN BE OMITTED
IVALUEB =43210 ; CODING IS REVERSED
IVALUEC =*112* ; DICHOTOMIZED WITH EXTREME CATEGORIES MARKED MISSING
Example 7: A five-item test.
Item 1 Dichotomy already scored 0-1 ; let's call this a &quot;D&quot; (for dichotomy) group item
Item 2 Dichotomy already scored 0-1 ; this is another &quot;D&quot; (for dichotomy) group item. Under the Rasch model,
all dichotomies have the same response structure.
Item 3 Partial credit polytomy already scored 0-1-2 ; this is an &quot;0&quot; type item. &quot;0&quot; means &quot;this item has its own
response structure&quot;.
Item 4 Rated polytomy already scored 1-2-3-4 ; let's call this an &quot;R&quot; group items
Item 5 Rated polytomy already scored 1-2-3-4 with the same rating scale as item 4, so this is another &quot;R&quot; group
item,
CODES = 01234
; this is all possible valid codes in the data
ISGROUPS = DD0RR ; Winsteps detects from the data which are the responses for each item-group and what
they mean.
97.
GRPFROM location of ISGROUPS
Only use this if you have too many items to put conveniently on one line of the ISGROUPS= control variable.
Instructs where to find the ISGROUPS= information.
GRPFROM=N
ISGROUPS= is a control variable before &amp;END (the standard).
GRPFROM=Y
ISGROUPS= information follows just after &amp;END, but before the item names. It is formatted exactly like a
data record. It is helpful to enter &quot;ISGROUPS=&quot;, for reference, where the person name would go.
Example: An attitude survey of 10 items with 3 rating scale definitions. Items 1 through 3 on Rating Scale 1,
items 4 through 6 on Rating Scale 2 and items 7 through 10 on Rating Scale 3. The ISGROUPS= information is
formatted like a data record and entered after &amp;END and before the item names. The responses are in columns
1-10, and the person-id in column 11 onwards.
NAME1=11
start of person-id
ITEM1=1
start of responses
101
NI=10
number of items
CODES=12345
valid responses
GRPFRM=Y
ISGROUPS= formatted like data
&amp;END
1112223333
ISGROUPS= information
Item name 1
item names
|
Item name 10
END NAMES
2213243223 John Smith
first data record
|
98.
GUFILE (G0ZONE, G1ZONE) Guttmanized response file
This writes out the response file edited to more closely match an ideal Guttman scalogram. It is in a format close
to the original data file, with items and person in entry order.
Outlying 1's are converted to 0's according to G0ZONE=
Outlying 0's are converted to 1's according to G1ZONE=
This removes unlikely 1's in the G0ZONE (e.g., lucky guesses)
and unlikely 0's in the G1ZONE (e.g. careless mistakes)
It is also useful for imputing theories about item hierarchies.
G0ZONE= sets the % of observed 0's, starting from the &quot;difficult&quot; side of the Guttman scalogram, among which all
1's are turned to 0's. (The item hierarchy is constructed with the current data, but can be enforced
through anchoring.) Standard value is 50.
G1ZONE= sets the % of observed 1's, starting from the &quot;easy&quot; side of the Guttman scalogram, among which all
0's are turned to 1's. Standard value is 50.
Example: GUFILE= guttman.txt
G0ZONE = 20%
G1ZONE = 40%
Original data (Guttman ordered)
11100110011001001010
becomes
11111110011001001000
The file format matches the input data file if both are in fixed-field format.
When GUFILE= is written with CSV=Y, comma-separated or CSV=T, tab-separated, the item responses precede
the person label.
Example: KCT.txt Guttmanized with fixed field format:
Richard M 111111100000000000
Tracie F 111111111100000000
Walter M 111111111001000000
KCT.txt Guttmanized with comma-separated, CSV=Y, format:
1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,Richard M
1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,Tracie F
1,1,1,1,1,1,1,1,1,0,0,1,0,0,0,0,0,0,Walter M
99.
HEADER display or suppress subtable headings
Subtables are usually displayed with two heading lines, showing Table number, Title, date, etc.
102
To display these (the standard), specify HEADER=YES.
To suppress these, specify HEADER=NO.
Example: In Table 7.2, the person misfits. Heading lines are not wanted between persons.
HEADER=NO
100.
HIADJ correction for top rating scale categories
The Rasch model models the measure corresponding to a top rating (or partial credit) scale category as infinite.
This is difficult to think about and impossible to plot. Consequently, graphically in Table 2.2 and numerically in
Table 3.1 a measure is reported corresponding to a top category. This is the measure corresponding to an
imaginary rating HIADJ= rating points below the top category. The corresponding instruction for the bottom
category is LOWADJ=.
Example: The standard spread in Table 2.2 is based on HIADJ=0.25. You wish the top category number to be
printed more to the right, further away from the other categories.
HIADJ=0.1
101.
HLINES heading lines in output files
To facilitate importing the IFILE=, PFILE=, SFILE= and XFILE= files into spreadsheet and database programs, the
heading lines can be omitted from the output files.
HLINES=Y
lines.
HLINES=N
Example:
0
1
2
Include heading lines in the output files (the standard)
In IFILE= and PFILE=, specifying HLINES=Y also puts &quot;;&quot; at the start of missing, deleted and extreme
Omit heading lines.
I want a tab-separated score-to-measure file, without the column headings:
SCOREFILE=mysc.txt
HLINES=NO
CSV=TAB
-6.46
-5.14
-4.22
1.83
1.08
.86
.28
.81
1.29
217
278
321
85
50
40
1
0
1
2.9
.0
2.9
1
1
2
2.9
2.9
5.7
1
3
4
with column headings, HLINES=YES, the standard:
&quot;;&quot;
&quot;KID&quot; &quot;SCORE FILE FOR&quot;
&quot;; TABLE OF SAMPLE NORMS (500/100) AND FREQUENCIES CORRESPONDING TO COMPLETE TEST&quot;
&quot;;SCORE&quot;
&quot;MEASURE&quot;
&quot;S.E.&quot; &quot;INFO&quot; &quot;NORMED&quot;
&quot;S.E.&quot; &quot;FREQUENCY&quot; &quot;%&quot; &quot;CUM.FREQ.&quot;
&quot;PERCENTILE&quot;
0 -6.46 1.83
.28 217 85 1 2.9 1 2.9 1
102.
&quot;%&quot;
IAFILE item anchor file
The IFILE= from one analysis can be used unedited as the item anchor file, IAFILE=, of another.
The item parameter values (deltas) can be anchored (fixed) using IAFILE=. Anchoring facilitates equating test
forms and building item banks. The items common to two test forms, or in the item bank and also in the current
form, can be anchored at their other form or bank calibrations. Then the measures constructed from the current
data will be equated to the measures of the other form or bank. Other measures are estimated in the frame of
reference defined by the anchor values.
In order to anchor items, a data file must be created of the following form:
1. Use one line per item-to-be-anchored.
2. Type the sequence number of the item in the current analysis, a blank, and the measure-value at which to
anchor the item (in logits if USCALE=1, or in your user-rescaled units otherwise).
103
Further values in each line are ignored. An IFILE= works well as an IAFILE=.
Anything after &quot;;&quot; is treated as a comment.
IAFILE = filename
Item anchor information is in a file containing lines of format
item entry number
anchor value
item entry number
anchor value
IAFILE=*
Item anchor information is in the control file in the format
IAFILE=*
item entry number anchor value
item entry number anchor value
*
IAFILE=$SnnEnn or IAFILE=$SnnWnn or @Field
Item anchor information is in the item labels using the column selection rules. Blanks or non-numeric values
indicate no anchor value.
Example 1: The third item is to be anchored at 1.5 logits, and the fourth at 2.3 logits.
1. Create a file named, say, &quot;ANC.FIL&quot;
2. Enter the line &quot;3 1.5&quot; into this file, which means &quot;item 3 in this test is to be fixed at 1.5 logits&quot;.
3. Enter a second line &quot;4 2.3&quot; into this file, which means &quot;item 4 in this test is to be fixed at 2.3 logits&quot;.
3. Specify, in the control file,
IAFILE=ANC.FIL
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
or place directly in the control file:
IAFILE=*
3 1.5
4 2.3
*
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
or in with the item labels:
IAFILE=$S10W4
; location of anchor value in item label
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
&amp;END
Zoo
House
1.5
Garden 2.3
Park
; item label and anchor value
END LABELS
To check: &quot;A&quot; after the measure means &quot;anchored&quot;
+----------------------------------------------------------------------------------------+
|ENTRY
RAW
|
INFIT | OUTFIT |PTMEA|
|
|
|NUMBER SCORE COUNT MEASURE ERROR|MNSQ ZSTD|MNSQ ZSTD|CORR.|DISPLACE| ITEMS
|
|------------------------------------+----------+----------+-----+--------+--------------|
|
3
32
35
1.5A
.05| .80
-.3| .32
.6| .53|
.40| House
|
Example 2: The calibrations from one run are to be used to anchor subsequent runs. The items have the same
numbers in both runs. This is convenient for generating tables not previously requested.
1. Perform the calibration run, say,
C:&gt; WINSTEPS SF.TXT SOMEO.TXT IFILE=ANCHORS.SF TABLES=111
104
2. Perform the anchored runs, say,
C:&gt; WINSTEPS SF.TXT MOREO.TXT IAFILE=ANCHORS.SF TABLES=0001111
C:&gt; WINSTEPS SF.TXT CURVESO.TXT IAFILE=ANCHORS.SF CURVES=111
Example 3: Score-to-measure Table 20 is to be produced from known item and rating scale structure difficulties.
Specify:
IAFILE=
; the item anchor file
SAFILE= ; the structure/step anchor file (if not dichotomies)
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
STBIAS=NO
; anchor values do not need estimation bias correction.
The data file comprises two dummy data records, so that every item has a non extreme score, e.g.,
For dichotomies:
Record 1: 10101010101
Record 2: 01010101010
For a rating scale from 1 to 5:
Record 1: 15151515151
Record 2: 51515151515
103.
IANCHQU anchor items interactively
Items to be anchored can be entered interactively by setting IANCHQ=Y. If you specify this, you are asked if you
want to anchor any items. If you respond &quot;yes&quot;, it will ask if you want to read these anchored items from a file; if
you answer &quot;yes&quot; it will ask for the file name and process that file in the same manner as if IAFILE= had been
specified. If you answer &quot;no&quot;, you will be asked to enter the sequence number of each item to be anchored, one at
a time, along with its logit (or user-rescaled by USCALE=, UMEAN=) value. When you are finished, enter a zero.
Example: You are doing a number of analyses, anchoring a few, but different, items each analysis. You don't
want to create a lot of small anchor files, but rather just enter the numbers at the terminal, so specify:
IANCHQ=Y
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
You want to anchor items 4 and 8.
WINSTEPS asks you:
DO YOU WANT TO ANCHOR ANY ITEMS?
respond YES(Enter)
DO YOU WISH TO READ THE ANCHORED ITEMS FROM A FILE?
respond NO(Enter)
INPUT ITEM TO ANCHOR (0 TO END):
respond 4(Enter) (the first item to be anchored)
INPUT VALUE AT WHICH TO ANCHOR ITEM:
respond 1.45(Enter) (the first anchor value)
INPUT ITEM TO ANCHOR (0 TO END): 8(Enter)
INPUT VALUE AT WHICH TO ANCHOR ITEM:-0.23(Enter)
INPUT ITEM TO ANCHOR (0 TO END): 0(Enter) (to end anchoring)
104.
ICORFILE item residual correlation file
This writes out the Table of inter-item correlations which is the basis of the principal components analysis of
residuals. Missing data: for these Winsteps substitutes their expectations when possible. For residuals and
standardized residuals, these are 0. Persons with extreme scores (minimum possible or maximum possible):
Winsteps drops these from the correlation computation.The reason for these choices is to make the principal
components analysis of residuals as meaningful as possible.
ICORFILE= file name
Example 1: Write out the Table of inter-item residual correlations. ICORFILE=file.txt - Then file.txt contains, for
105
SF.txt,
Item
1
1
.....
2
2
.....
Item
2
3
Correlation
-.04
.05
3
4
-.04
-.20
Example 2: When ICORFILE= is selected on the Output Files menu or MATRIX=YES, the Data Format: Matrix
option can be selected:
This produces:
1.0000
-.0451
.0447
......
105.
-.0451
1.0000
-.0448
......
.0447
-.0448
1.0000
......
.0095
-.2024
-.0437
......
.....
.....
.....
.....
IDELETE item one-line item deletion
A one-line list of items to be deleted or reinstated can be conveniently specified with IDELETE=. This is designed
to be used in the post-analysis Specification pull-down menu box.
The formats are:
IDELETE= 3
; an entry number: delete item 3
IDELETE= 6 1
; delete items 6 and 1
IDELETE= 2-5
; delete items 2, 3, 4, 5
IDELETE= +3-10 ; delete all items, then reinstate items 3 to 10.
IDELETE= 4-20 +8 ; delete items 4-20 then reinstate item 8
IDELETE= 3,7,4,10 ; delete items 3, 7, 4, 10. Commas, blanks and tabs are separators. At the &quot;Extra information&quot;
prompt, use commas.
IDELETE= (blank) ; resets temporary item deletions
Example 1: After an analysis is completed, delete all items except for one subtest in order to produce a score-tomeasure Table for the subtest.
In the Specifications pull-down box:
IDELETE = +11-26 ; the subtest is items 11-26
Screen displays: CURRENTLY REPORTABLE ITEMS = 16
In the Output Tables menu (or SCOREFILE=)
Table 20. Measures for all possible scores on items 11-26.
Example 2: 9 common items. 3 items on Form A. 4 items on Form B. Score-to-measure tables for the Forms.
For Form A: in the Specifications pull-down box:
IDELETE = 13-16 ; deletes Form B items
In the Output Tables menu:
Table 20. Measures for all possible scores on items in Form A.
For Form B: in the Specifications pull-down box:
IDELETE= ; to reset all deletions
then
IDELETE = 10-12 ; deletes Form A items
In the Output Tables menu:
Table 20. Measures for all possible scores on items in Form B.
106.
IDELQU delete items interactively
Use this if you have one or two items to delete or will be running repeatedly with different deletion and selection
patterns, otherwise use IDFILE=.
106
If your system is interactive, items to be deleted or selected can be entered interactively by setting IDELQU=Y. If
you specify this, you will be asked if you want to delete any items. If you respond &quot;yes&quot;, it will ask if you want to
read these deleted items from a file; if you answer &quot;yes&quot; it will ask for the file name and process that file in the
same manner as if IDFILE= had been specified. If you answer &quot;no&quot;, you will be asked to enter the sequence
number or numbers of items to be deleted or selected one line at a time, following the rules specified for IDFILE=.
When you are finished, enter a zero.
Example: You are doing a number of analyses, deleting a few, but different, items each analysis. You don't
want to create a lot of small delete files, but rather just enter the numbers directly into the program using:
NI=60
ITEM1=30
IDELQU=Y
&amp;END
You want to delete items 23 and 50 through 59.
WINSTEPS asks you:
DO YOU WANT TO DELETE ANY ITEMS?
respond YES(Enter)
DO YOU WISH TO READ THE DELETED ITEMS FROM A FILE?
respond NO(Enter)
INPUT ITEM TO DELETE (0 TO END):
respond 23(Enter) (the first item to be deleted)
INPUT ITEM TO DELETE (0 TO END): 50-59(Enter)
INPUT ITEM TO DELETE (0 TO END): 0(Enter) (to end deletion)
If you make a mistake, it is simple to start again, reinstate all items with
INPUT ITEM TO DELETE (0 TO END): +1-999(Enter)
where 999 is the length of your test or more, and start selection again.
107.
IDFILE item deletion file
Deletion or selection of items from a test for an analysis, but without removing the responses from your data file,
is easily accomplished by creating a file in which each line contains the sequence number or numbers of items to
be deleted or selected. Specify this file by means of the control variable, IDFILE=, or enter the deletion list in the
control file using IDFILE=*. Your control file should include item labels for all items, including the ones you are
deleting.
a) Delete an item: enter the item number. E.g., to delete item 5, enter
5
b) Delete a range of items: enter the starting and ending item number on the same line separated by a blank or
dash. E.g., to delete items 13 through 24
13-24
or
13 24
c) Select an item for analysis: enter a plus sign then the number.
E.g., to select item 19 from a previously deleted range
+19
d) Select a range of items for analysis: enter a plus sign, the starting number, a blank or dash, then the ending
number. E.g., to select items 17 through 22
+17-22
or
+17 22
e) If a + selection is the first entry in the deletion file, then all items are deleted before the first selection is
undertaken, so that the items analyzed will be limited to those selected, e.g,
if +10-20 is the only line in the item deletion file for a 250 item test, it means
107
1-250
; delete all 250 items
+10-20 ; reinstate items 10 through 20.
f) You may specify an item deletion list in your control with
IDFILE=*
(List)
*
e.g.,
IDFILE=*
17
; delete item 17
2
; delete item 2
*
Example 1: You wish to delete the fifth and tenth through seventeenth items from an analysis, but then keep item
fourteen.
1. Create a file named, say, ITEM.DEL
2. Enter into the file, the lines:
5
10-17
+14
3. Specify, in the control file,
NI=50
ITEM1=63
IDFILE=ITEM.DEL
TABLES=1110111
&amp;END
or, specify in the control file,
NI=50
ITEM1=63
IDFILE=*
5
10-17
+14
*
TABLES=1110111
&amp;END
Example 2: The analyst wants to delete the most misfitting items reported in Table 10.
1. Set up a standard control file.
2. Specify
IDFILE=*
*
3. Copy the target portion of Table 10.
4. Paste it between the &quot;*&quot;
5. Delete characters before the entry numbers.
6. Type ; after the entry numbers to make further numbers into comments.
TITLE = 'Example of item deletion list from Table 10'
IDFILE = *
Delete the border character before the entry number
; ENTRY
RAW
;
NUM SCORE
5; 2
8; 2
4; 3
9; 4
20 ; 1
24 ; 6
COUNT
4
4
4
4
4
4
INFIT
OUTFIT
MEASURE ERROR MNSQ ZSTD MNSQ ZSTD PTBIS ACTS
.00
1.03 1.48
1.8 1.50
1.8 A-.83 FIND BOTTLES AND CANS
.00
1.03 1.40
1.6 1.43
1.6 B-.71 LOOK IN SIDEWALK CRACKS
.00
.62 1.33
.7 1.49
.9 C-.21 WATCH GRASS CHANGE
.00
.74 1.51
.8 1.57
.9 D-.59 LEARN WEED NAMES
.00
1.03 1.12
.5 1.14
.6 E-.05 WATCH BUGS
.30
1.03 1.15
.6 1.13
.5 F-.15 FIND OUT WHAT FLOWERS LIVE ON
Enter the ; to make details to right of entry numbers into comments
108
G
0
0
0
0
0
0
*
Example 3: The analyst want to delete item 4 and items 18 to 23 on the DOS control (or Extra Specifications) line:
Extra specifications? IDFILE=* 4 18-23 * (Enter)
or
C:&gt;WINSTEPS CONTROL.FIL OUTPUT.FIL IDFILE=* 4 18-23 *
108.
IDROPEXTREME drop items with extreme scores
Unanchored items with extreme (zero or perfect, minimum possible or maximum possible) scores provide no
information for estimating person measures, but they are reported and included in summary statistics. To remove
them:
IDROPEXTREME = No ; do not drop extreme items (standard)
IDROPEXTREME = Yes or All ; drop items with zero and perfect scores
IDROPEXTREME = Zero or Low or Bottom or Minimum ; drop items with zero or minimum-possible scores
IDROPEXTREME = Perfect or High or Top or Maximum ; drop items with perfect or maximum-possible scores
Example: The instrument contains items asking about very rare conditions (scored &quot;0&quot; - not observed). These
are skewing the survey summary statistics:
IDROPEXTREME = Minimum ; items about conditions never observed in the sample are dropped.
109.
IFILE item output file
IFILE=filename produces an output file containing the information for each item. This file contains 4 heading lines
(unless HLINES=N), followed by one line for each item containing:
Columns:
Start
End Format Description
1 1 A1 Blank or &quot;;&quot; if HLINES=Y and there are no responses or deleted or extreme (status =0,-1, -2, -3)
2 6 I5 1. The item sequence number (ENTRY)
7 14 F8.2
2. Item's calibration (user-rescaled by UMEAN=, USCALE=, UDECIM) (MEASURE)
15 17 I3 3. The item's status (STATUS)
2 = Anchored (fixed) calibration
1 = Estimated calibration
0 = Extreme minimum (estimated using EXTRSC=)
-1 = Extreme maximum (estimated using EXTRSC=)
-2 = No responses available for calibration (or all responses in the same category with
ISGROUPS=0).
-3 = Deleted by user
18 25 F8.1
4. The number of responses used in calibrating (COUNT) or the observed count (TOTAL=Y)
26 34 F8.1
5. The raw score used in calibrating (SCORE) or the observed score (TOTAL=Y)
35 41 F7.2
6. Item calibration's standard error (user-rescaled by USCALE=, UDECIM=) (ERROR)
42 48 F7.2
7. Item infit: mean square infit (IN.MSQ)
49 55 F7.2
8. Item infit: t standardized (ZSTD), locally t standardized (ZEMP) or log-scaled (LOG)
56 62 F7.2
9. Item outfit: mean square outfit (OUT.MS)
63 69 F7.2
10. Item outfit: standardized (ZSTD), locally t standardized (ZEMP) or log-scaled (LOG)
70 76 F7.2
11. Item displacement (user-rescaled by USCALE=, UDECIM=) (DISPLACE)
77 83 F7.2
12. Item by test-score correlation: point-biserial (PTBS) or point-measure (PTME)
84 90 F7.2
13. Item weight (WEIGHT)
91 96 F6.1
14. Observed percent of observations matching prediction (OBSMA)
97 102 F6.1
15. Expected percent of observations matching prediction (EXPMA)
103 109 F7.2
16. Item discrimination
If ASYMPTOTE=Yes:
110 115 F6.2
17. Item lower asymptote (LOWER)
109
116 121 F6.2
18. Item upper asymptote(UPPER)
... add 12 to the next column locations
If PVALUE=Yes:
110 115 F6.2
19. Item p-value or average rating (PVALUE)
... add 6 to the next column locations
110 110 1X Blank
111 111 A1 20. Grouping to which item belongs (G)
112 112 1X Blank
113 113 A1 21. Model used for analysis (R=Rating, S=Success, F=Failure) (M)
114 114 1X Blank
115 144+ A30+ 22. Item name (NAME)
The format descriptors are:
In = Integer field width n columns
Fn.m = Numeric field, n columns wide including n-m-1 integral places, a decimal point and m decimal places
An = Alphabetic field, n columns wide
nX = n blank columns.
When CSV=Y, commas separate the values, which are squeezed together without spaces between. Quotation
marks surround the &quot;Item name&quot;, e.g., 1,2,3,4,&quot;Name&quot;. When CSV=T, the commas are replaced by tab
characters.
Example: You wish to write a file on disk called &quot;ITEM.CAL&quot; containing the item statistics for use in updating
your item bank, with values separated by commas:
IFILE=ITEM.CAL
CSV=Y
When W300=Yes, then this is produced in Winsteps 3.00, 1/1/2000, format:
Columns:
Start
End Format Description
1 1 A1 Blank or &quot;;&quot; if HLINES=Y and there are no responses or deleted (status = -2, -3)
2 6 I5 1. The item sequence number (ENTRY)
7 14 F8.2
2. Item's calibration (user-rescaled by UMEAN=, USCALE=, UDECIM) (MEASURE)
15 17 I3 3. The item's status (STATUS)
2 = Anchored (fixed) calibration
1 = Estimated calibration
0 = Extreme minimum (estimated using EXTRSC=)
-1 = Extreme maximum (estimated using EXTRSC=)
-2 = No responses available for calibration
-3 = Deleted by user
18 23 I6 4. The number of responses used in calibrating (COUNT) or the observed count (TOTAL=Y)
24 30 I6 5. The raw score used in calibrating (SCORE) or the observed score (TOTAL=Y)
31 37 F7.2
6. Item calibration's standard error (user-rescaled by USCALE=, UDECIM=) (ERROR)
38 44 F7.2
7. Item mean square infit (IN.MSQ)
45 51 F7.2
8. Item infit: t standardized (ZSTD), locally t standardized (ZEMP) or log-scaled (LOG)
52 58 F7.2
9. Item mean square outfit (OUT.MS)
59 65 F7.2
10. Item outfit: t standardized (ZSTD), locally t standardized (ZEMP) or log-scaled (LOG)
66 72 F7.2
11. Item displacement (user-rescaled by USCALE=, UDECIM=) (DISPLACE)
73 79 F7.2
12. Item by test-score correlation: point-biserial (PTBS) or point-measure (PTME)
80 80 1X 15. Blank
81 81 A1 16. Grouping to which item belongs (G)
82 82 1X 17. Blank
83 83 A1 18. Model used for analysis (R=Rating, S=Success, F=Failure) (M)
84 84 1X 19. Blank
85 132+ A30+ 18. Item name (NAME)
110
Example of standard IFILE=
; ITEM Knox Cube Test (Best Test Design p.31) Nov 10 15:40 2005
;ENTRY MEASURE STTS COUNT
SCORE ERROR IN.MSQ IN.ZSTD OUT.MS OUT.ZSTD DISPL PTME WEIGHT OBSMA EXPMA
DISCR G M NAME
;
1
-6.74 -1
34.0
34.0
1.87
1.00
.00
1.00
.00
.00
.00
1.00
.0
.0
1.00 1 R 1= 1-4
4
-4.54 1
34.0
32.0
.82
.92
-.16
.35
-.24
.00
.55
1.00 94.1 94.1
1.07 1 R 4= 1-3-4
110.
ILFILE item label file
Useful item identification greatly enhances the output tables.
You usually specify items labels between &amp;END and END LABELS in the control file. You may also use ILFILE=
for the initial or additional sets of item labels.
ILFILE= commands also add 'Edit ILFILE=' lines to the File pull-down menu.
Example: You wish to use abbreviated item labels for the initial analysis, but then switch to longer item labels to
produce the final reports.
In your control file specify the shorter labels, one per line,
(a) between &amp;END and END LABELS
or (b) between ILFILE=* and * in the control file
or (c) in a separate file identified by ILFILE=*
You can switch to the longer labels, in a file called &quot;Longer.txt&quot; by using the &quot;Specification&quot; menu item, and
entering
ILFILE=Longer.txt
If you have ILFILE= in your control file and your data is also in your control file , be sure that there is an &quot;END
LABELS&quot; before your data (or that you specify INUMB=YES).
Example: 4 item arithmetic test.
NI=4
ILFILE=*
Addition
; labels for the 4 items
Subtraction
Multiplication
Division
*
&amp;End
END LABELS
111.
IMAP item label on item maps Tables 1, 12
This specifies what part of the item label is to be used on the item map. The length of IMAP= overrides
NAMLMP=.
It's format is IMAP = $S..W.. or $S..E. etc. using the column selection rules.
Example: Item type is in columns 3 and 4 of the item label. Item content area is in columns 7 and 8.
IMAP= $S3W2+&quot;/&quot;+$S7W2
tfile=*
12 ; Item maps in Table 12 (or use Output Tables menu)
*
If the item label is &quot;KH323MXTR&quot;, the item label on the map will be &quot;32/XT&quot;
112.
INUMB label items by sequence numbers
Are item names provided, or are they the entry sequence numbers?
111
INUMB=Y
a name is given to each item based on its sequence number in your data records. The names are &quot;I0001&quot;,
&quot;I0002&quot;, ..., and so on for the NI= items. This is a poor choice as it produces noninformative output.
INUMB=N, the standard
Your item names are entered (by you) after the &quot;&amp;END&quot; at the end of the control variables. Entering detailed item
names makes your output much more meaningful to you.
The rules for supplying your own item names are:
1. Item names are entered, one per line, generally directly after &amp;END.
2. Item names begin in column 1.
3. Up to 300 characters (or ITLEN=) in each item name line will be used, but you may enter longer names in the
control file for your own record keeping.
4. The item names must be listed in exactly the same order as their responses appear in your data records.
5. There should be the same number of item names as there are items specified by NI=. If there are too many or
too few names, a message will warn you and sequence numbers will be used for the names of any unnamed
items. You can still proceed with the analysis.
6. Type END NAMES starting in column 1 of the line after the last item name.
Example:
An analysis of 4 items for which you supply identifying labels.
; these lines can start at any column
NI=4
four items
ITEM1=10
responses start in column 10
INUMB=N
item names supplied (the standard)
&amp;END
My first item name ; must start at column 1.
My second item label
My third item identifier
My fourth and last item name
END NAMES
; must start at column 1, in capital letters
Person A 1100 data records
|
Person Z 1001
113.
IPMATRIX response-level matrix
IPMATRIX= is only available from the Output Files pull-down menu. It constructs a rectangular matrix in which
there is one row for each person and one column for each item, or vice-versa. The entries in the matrix are
selected from the following screen:
The first rows and columns can be the entry numbers, measures, labels and/or fields from the labels.
The matrix must contain one of
3. Original response value (after keying/scoring) (I4) (OBS)
4. Observed response value (after recounting) (I4) (ORD)
5. Expected response value (F7.3) (EXPECT)
112
6. modeled variance of observed values around the expected value (F7.3) (VAR)
This is also the statistical information in the observation.
Square root(modeled variance) is the observation's raw score standard deviation.
7. Standardized residual: (Observed - Expected)/Square root Variance (F7.3) (ZSCORE)
8. Score residual: (Observed - Expected) (F7.3) (RESID)
11. Measure difference (Person measure - Item measure) (F7.3) (MEASURE)
12. Log-Probability of observed response (F7.3) (LOGe(PROB))
13. Predicted person measure from this response (F7.3) (PMEASURE)
14. Predicted item measure from this response (F7.3) (IMEASURE)
15. Response code in data file (A) (CODE)
Field numbers shown here are those for XFILE=.
Depending on CSV=, data values are separated by &quot;tab&quot; or comma characters. In fixed field format, all fields are 7
characters long, separated by a blank. Missing data codes are &quot;.&quot; as standard, but can be any character, or
nothing.
1234567-1234567-1234567-1234567-1234567-1234567-1234567 ; these indicate fields.
.
1
2
3
4
5
6
1
8.85
-.223
.
.
.
.
2
3.917
.
3.917
.
.
.
3
.
6.585
-.298
.
.
.
Example: I want a table of probabilities with items as the columns and possible scores as the rows, like the one
on page 166 of Doug Cizek's book, Setting Performance Standards, based on work by Mark Reckase.
From your main analysis, write out an IFILE=itemanc.txt
Create a data set with one record for each possible score (it doesn't matter what the actual pattern of 1's and 0's
is).
Enter the intended raw score as the person label.
In the control file for the new data set, put IAFILE=itemanc.txt
Analyze this second data set. Ignore any &quot;subset&quot; warning messages.
Check that the items are anchored in Table 14.
Check that the reported raw score for each person match that in the person label in Table 18.
Use the Output Files pull-down menu to write out an IPMATRIX=
Select &quot;Expected response value&quot;, &quot;Persons are rows, items are columns&quot;, &quot;Person label&quot;, &quot;Item entry number&quot;
Then &quot;Tab-separated&quot;, &quot;EXCEL&quot;
Then, for each observation in your data set, you have the probability for each score for each item.
Here it is from Exam1.txt (using the non-extreme items):
TITLE='KNOX CUBE TEST'
;
NAME1=1
;
ITEM1=11
;
NI=18
;
CODES=01
;
iafile = itemanc.txt
;
&amp;END
END NAMES
1
10000000000000
2
11000000000000
3
11100000000000
4
11110000000000
5
11111000000000
6
11111100000000
7
11111110000000
8
11111111000000
9
11111111100000
10
11111111110000
11
11111111111000
12
11111111111100
13
11111111111110
Report title
First column of person label in data file
First column of responses in data file
Number of items
Valid response codes in the data file
item calibrations from the original analysis
; score and a response string for it
Here's the array in EXCEL, after some editing. Add across the columns to confirm that the probabilities add up to
the scores.
113
Score Item 4 Item 5 Item 6 Item 7 Item 8 Item 9 Item 10 Item 11 Item 12 Item 13
1
0.29
0.19
0.13
0.19
0.05
0.13
0.02
0.00
0.00
0.00
2
0.51
0.37
0.28
0.37
0.12
0.28
0.06
0.01
0.00
0.00
3
0.68
0.55
0.44
0.55
0.22
0.44
0.11
0.01
0.00
0.00
4
0.81
0.71
0.60
0.71
0.35
0.60
0.20
0.02
0.01
0.01
5
0.90
0.83
0.76
0.83
0.53
0.76
0.34
0.05
0.01
0.02
6
0.95
0.92
0.88
0.92
0.73
0.88
0.55
0.10
0.03
0.04
7
0.98
0.97
0.96
0.97
0.89
0.96
0.79
0.26
0.08
0.10
8
1.00
0.99
0.99
0.99
0.96
0.99
0.93
0.54
0.21
0.27
9
1.00
1.00
1.00
1.00
0.99
1.00
0.97
0.76
0.43
0.50
10
1.00
1.00
1.00
1.00
1.00
1.00
0.99
0.89
0.65
0.72
11
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.95
0.82
0.86
12
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.98
0.92
0.94
13
1.00
1.00
1.00
1.00
1.00
1.00
1.00
0.99
0.97
0.98
114.
Item 14 Item 15 Item 16 Item 17
0.00
0.00
0.00
0.00
0.00
0.01
0.03
0.08
0.20
0.38
0.60
0.78
0.91
0.00
0.00
0.00
0.00
0.00
0.00
0.01
0.02
0.06
0.13
0.26
0.46
0.72
0.00
0.00
0.00
0.00
0.00
0.00
0.01
0.02
0.06
0.13
0.26
0.46
0.72
0.00
0.00
0.00
0.00
0.00
0.00
0.01
0.02
0.06
0.13
0.26
0.46
0.72
IREFER identifying items for recoding
Responses are revalued according to the matching codes in IVALUE=. If this implies that the items may have
different rating (or partial credit) scale structures, so ISGROUPS= may also be required.
IREFER= has three forms: IREFER=AABBCDAAD and IREFER=* list * and IREFER=*filename
Valid one-character IREFER= codes include: !#$%&amp;./123456789&lt;&gt;@ABCDEFGHIJKLMNOPQRSTUVWXYZ^_|~
A-Z are the same as a-z.
Characters with ASCII codes from 129-255 can also be used, but display peculiarly:
&Icirc;&Iuml;&ETH;&Ntilde;&Ograve;&Oacute;&Ocirc;&Otilde;&Ouml;&times;&Oslash;&Ugrave;&Uacute;&Ucirc;&Uuml;&Yacute;&THORN;&szlig;&agrave;&aacute;&acirc;&atilde; etc.
When XWIDE=2 or more, then
either (a) Use one character per XWIDE and blanks,
NI=8
XWIDE=2
IREFER=' A B C D D C B A'
or (b) Use one character per item with no blanks
NI=8
XWIDE=2
RESCORE='ABCDDCBA'
Item identifying codes can be letters or numbers.
&quot;A&quot; is the same as &quot;a&quot;, etc.
Example 1. There are 3 item types. Items are to rescored according to Type A and Type B. Other items to keep
original scoring.
CODES = 1234
IREFER = AAAAAAAABBBBBBBB******* 3 item types:
(&quot;a&quot;
is the same as &quot;A&quot; in these
codes)
IVALUEA = 1223
IVALUEB = 1123
IVALUE* = 1234
Recode Type A items
Recode Type B items
Recode Type * item. Can be omitted
or
IREFER=*
1-8 A
9-16 B
17-23 *
*
or
IREFER=*filename.txt
in filename.txt:
1-8 A
9-16 B
17-23 *
114
Example 2. There are 3 item types. Responses are 2 characters wide. Items are to rescored according to Type
A and Type B. Other items to keep original scoring.
XWIDE=2
CODES = '1 2 3 4 '
IREFER = AAAAAAAABBBBBBBB******* 3 item
IVALUEA = '1 2 2 3 '
Recode
IVALUEB = '1 1 2 3 '
Recode
IVALUE* = 1234
Recode Type * item. Can
types
Type A items
Type B items
be omitted
Example 3: All items are to be rescored the same way
NI = 100
IREFER=*
1-100 X
*
Codes = 12345
IVALUEX = 12223
100 ITEMS
FOR ALL 100 ITEMS, reference is X
rescore 12345
into 12223
Example 4: Items are to be rescored in 3 different ways, and then the items are to be divided into 4 rating scale
structures.
ISGROUPS=11112223312444
; 4 RATING SCALE GROUPINGS
IREFER =AAAABBBCCABBBB
; 3 RECODINGS
CODES
=01234 ; ORIGINAL CODES IN DATA
IVALUEA =01234 ; ORIGINAL CODING MAINTAINED - THIS LINE CAN BE OMITTED
IVALUEB =43210 ; CODING IS REVERSED
IVALUEC =*112* ; DICHOTOMIZED WITH EXTREME CATEGORIES MARKED MISSING
Example 5: Multiple-choice test with 4 options, ABCD
IREFER=ABCDDABADCDA
CODES =ABCD
IVALUEA=1000
IVALUEB=0100
IVALUEC=0010
IVALUED=0001
MISSCORVE=0
115.
;
;
;
;
;
;
;
SCORING KEY
VALID OPTIONS
A SCORED 1
B SCORED 1
C SCORED 1
D SCORED 1
EVERYTHING ELSE IN THE DATA SCORED 0
ISELECT item selection criterion
Items to be selected may be specified by using the ISELECT= instruction to match characters within the item
name. Items deleted by IDFILE= or similar are never selected by ISELECT=.
This can be done before analysis in the control file or with &quot;Extra specifications&quot;. It can also be done after the
analysis using the &quot;Specification&quot; pull-down menu.
Control characters to match item name:
? matches any character
{..} braces characters which can match a single character: {ABC} matches A or B or C.
{.. - ..} matches single characters in a range. {0-9} matches digits in the range 0 to 9.
{.. --..} matches a single &quot;-&quot; {AB--} matches A or B or &quot;-&quot;.
* matches any string of characters - must be last selection character.
Other alphanumeric characters match only those characters.
Each ISELECT= performed using the &quot;Specification&quot; pull-down menu selects from all those analyzed. For
incremental selections, i.e., selecting from those already selected, specify +ISELECT=
Example 1: Select for analysis only items with M in the 5th column of item name.
ISELECT=????M* M in column means Math items
0001M 2x4 selected
0002R the cat omitted
END NAMES
Example 2: Select for analysis all items with code &quot;A 4&quot; in columns 2-4 of their names.
115
ISELECT=&quot;?A 4*&quot; quotes because a blank is included. A is in col. 2 etc.
ZA 4PQRS selected
Example 3: Select all Math (M in column 2) items which are Addition or Division (A or D in column 4):
ISELECT=&quot;?M?{AD}*&quot;
1M3A456 2+2
selected
1M5D689 23/7
selected
1H2A123 George
omitted (History, American)
Example 3: Select codes A, 1,2,3,4,5,6 in column 3:
ISELECT=??{A1-6}*
Example 4: Select &quot;- &quot; in columns 2 and 3:
ISELECT=&quot;?- &quot;
Example 5: Select &quot;-&quot; or &quot;x&quot; in column 2 with &quot; &quot; in column 3:
ISELECT=&quot;?{--x} &quot;
Example 6: Analyze only math (column 4 or person-id). Then report only items in Topic C (column 1). Then only
report items in Strands 4 and 5 (column 2) in Topic C.
ISELECT=???M* in the Control file or at the Extra Specifications prompt.
ISELECT=C*
using the Specification pull-down menu, after the analysis
+ISELECT=?{45}* using the Specification pull-down menu.
116.
ISFILE item structure output file
Do not use this file for anchoring. ISFILE=filename produces an output file containing the category structure
measure information for each item. All measures are added to the corresponding item's calibration and rescaled
by USCALE= and UDECIMALS=. This file contains 4 heading lines (unless HLINES=N), followed by one line for
each item containing:
Columns:
Start
End Format Description
1
1 A1 Blank or &quot;;&quot; if no responses or deleted (status = -2, -3)
2
6 I5 1. The item sequence number (ENTRY)
7
11 I5 2. The item's status (STATUS)
1 = Estimated calibration
2 = Anchored (fixed) calibration
0 = Extreme minimum (estimated using EXTRSC=)
-1 = Extreme maximum (estimated using EXTRSC=)
-2 = No responses available for calibration
-3 = Deleted by user
12
5 I5 3. Number of active categories (MAXIMUM)
17
5 I5 4. Lowest active category number (CAT)
22
29 F8.2
5. Measure for an expected score of LOWADJ= (E.G., CAT+.25)
The following fields are repeated for the remaining active categories:
30
34 I5 6. Active category number (CAT)
35
39 I5 7. Ordered category number in structure (STRUCTURE) = &quot;Step counting from zero&quot;
40
47 F8.2
8. Structure measure (MEASURE) = Rasch-Andrich threshold + item measure = Dij, the
Rasch-Andrich threshold. The number of decimal places is set by UDECIMAL=. Do not use for
anchoring. Use the SFILE= and IFILE=.
48
55 F8.2
9. Rasch-Andrich threshold's standard error (ERROR) - reported if only one item in the
ISGROUPS=, otherwise 0.0.
56
63 F8.2
10. Measure for an expected score of category — 0.5 score points (CAT-0.5). This is the
Rasch-half-point threshold, the boundary between categories when conceptualized as average
performances. It is not a model parameter.
64
71 F8.2
11. Measure for an expected score of category score points (AT CAT). This is the measure
116
72
corresponding to a category when predicting for an individual or sample about which nothing else is
known. For the top category this value corresponds to the top category value less HIADJ= (e.g., CAT0.25), the measure for an expected score of HIADJ= score points less than the top category value.
79 F8.2
12. Measure at the 50% cumulative probability (50%PRB). This is the Rasch-Thurstone
threshold.
The &quot;AT CAT&quot; for the extreme categories are actually at -infinity and +infinity. So the value LOWADJ= above
minimum score on the item substitutes for -infinity. For the extreme high category, the value HIADJ= less than the
maximum score for that item substitutes for +inifinty. It is these values that are plotted on Table 2.2 for the
extreme categories.
Since the ISFILE= has the same number of category entries for every item, the repeated fields are filled out with
&quot;0&quot; for any further categories up to the maximum categories for any item.
The format descriptors are:
In = Integer field width n columns
Fn.m = Numeric field, n columns wide including n-m-1 integral places, a decimal point and m decimal places
An = Alphabetic field, n columns wide
When CSV=Y, commas separate the values with quotation marks around the &quot;Item name&quot;. When CSV=T, the
commas are replaced by tab characters.
When STKEEP=YES and there are intermediate null categories, i.e., with no observations, then the structure
calibration into the category is set 40 logits above the previous calibration. The structure calibration out of the
category, and into the next category, is set 40 logits above. Thus:
Category
0
1
2
3
TOTAL:
structure Calibration
Table 3.2
In SFILE
NULL
0.00
-1.00
-1.00
NULL
39.00
1.00
-38.00
0.00
0.00
Meanings of the columns
There are several ways of conceptualizing the category boundaries or thresholds of a rating (or partial credit)
scale item. Imagine a rating (or partial credit) scale with categories, 1, 2, 3:
From the &quot;expected score ogive&quot;, also called the &quot;model item characteristic curve&quot;
Average rating:
Measure (must be ordered)
1.25
Measure for an expected score of 0.25 (CAT+.25) when LOWADJ=0.25
1.5 Measure for an expected score of category — 0.5 score points (CAT-0.5)
2 Measure for an expected score of category score points (AT CAT)
2.5 Measure for an expected score of category — 0.5 score points (CAT-0.5)
2.75
Measure for an expected score of category score points (AT CAT)
since this is the top extreme category the reported values is for CAT-0.25 when HIADJ=0.25
From the &quot;category probability curves&quot; relative to the origin of the measurement framework (need not be ordered)
1-2 equal probability
Structure measure = Rasch-Andrich threshold + item measure = Dij (MEASURE)
standard error Rasch-Andrich threshold's standard error (ERROR)
2 maximum probability Measure for an expected score of category score points (AT CAT) - (yes, same as for the
ogive)
2-3 equal probability
Structure measure = Rasch-Andrich threshold + item measure = Dij (MEASURE)
standard error Rasch-Andrich threshold's standard error (ERROR)
From the &quot;cumulative probability curves&quot; (preferred by Thurstone) (must be ordered)
Category 1 at .5 probability Measure at the 50% cumulative probability (50%PRB)
117
Category 1+2 at .5 probability
Measure at the 50% cumulative probability (50%PRB)
Example 1: You wish to write a file on disk called &quot;ITEMST.FIL&quot; containing the item statistics reported in Table
2.2, for use in constructing your own tables:
ISFILE =
ISGROUPS
LOWADJ =
HIADJ =
;ENTRY STAT MAX CAT CAT+.25
ERROR CAT-0.5 CAT-.25 50%PRB
1
1
2
0
-2.47
.00
.79
1.68
.61
2
1
2
0
-2.78
.00
.48
1.37
.29
ITEMST.FIL
= 0
; each item has its own &quot;partial credit&quot; scale
0.25
; the standard for the low end of the rating scale
0.25
; the standard for the high end of the rating scale
CAT STRU MEASURE
ERROR CAT-0.5
AT CAT
50%PRB
CAT STRU MEASURE
1
1
-1.25
.00
-1.58
-.40
-1.40
2
2
.46
1
1
-1.57
.00
-1.89
-.71
-1.71
2
2
.15
&quot;ENTRY&quot; is the item entry number
&quot;STAT&quot; is the item status, see IFILE=
&quot;MAX&quot; is the highest category
&quot;CAT&quot; is the current category
&quot;CAT+0.25&quot; is the measure corresponding to an expected score of the lowest category+0.25 score points on the
item
&quot;STRU&quot; (structure calibration) or step measure is a Rasch model parameter estimate (Rasch-Andrich thresholds),
also the point at which adjacent categories are equally probable. See &quot;Category probability curves&quot; graph.
&quot;MEASURE&quot; is the item difficulty + structure calibration.
&quot;ERROR&quot; is an estimate of the standard error. It is reported as .00 if it is not known.
&quot;CAT-0.5&quot; is the location where the expected score on the item is the category half-point value, e.g., for a scale
for 0,1,2 the &quot;CAT-0.5&quot; values correspond to expected scores of 0.5 and 1.5. See the &quot;Expected score ICC&quot;
graph.
&quot;AT CAT&quot; is the location where the expected score on the item is the category point value, e.g., for a scale for
0,1,2 the &quot;At step&quot; values correspond to expected scores of 0.25, 1, and 1.75. Since the &quot;at step&quot; values 0 and 2
are infinite they are reported for 0.25 and 1.75. See the &quot;Expected score ICC&quot; graph.
&quot;50%PRB&quot; is the location of the Rasch-Thurstone threshold, the point at which the probability of all categories
below = the probability of all categories at or above. See the &quot;Cumulative probabilities&quot; graph.
Example 2: To produce a Table of expected measures per item-category similar to Pesudovs, K., E. Garamendi,
et al. (2004). &quot;The quality of life impact of refractive correction (QIRC) questionnaire: Development and
validation.&quot; Optometry and Vision Science 81(10): 769-777, write the ISFILE= to Excel. Then delete or hide
unwanted columns.
117.
ISORT column within item name for alphabetical sort in Table 15
Table 15 lists items alphabetically. Table 1 and Table 12 list them alphabetically within lines. As standard, the
118
whole item name is used. Select the sorting columns in the item labels with ISORT= using the column selection
rules, e.g., starting in column Snn and ending in column Enn or of width Wnn.
Example 1: The item name is entered in the specification file as sequence number, followed by identification in
column 6. Sort by identification for Table 15.
NI=4
TABLES=1111111111111111111111111
ISORT=5-30
; sort the items reported in Table 15 by item descriptions
&amp;END
0001 Addition Item
0002 Subtraction Item
0003 Multiplication item
0004 Division item
sort column
END NAMES
Example 2: The item name contains several important classifiers. Table 15 is required for each one:
TFILE=*
15 - - - 1
sort starts with column 1 of item name
15 - - - 6
sort starts with column 6
15 - - - 13 sort starts with column 13 of the item name and goes to the end of the item name
- entered as place-holders, see TFILE=
*
&amp;END
MCQU Geogrp 1995-0234
sort column
sort column
sort column
|
END NAMES
Example 3: A version of Table 15, sorted on item name column 13, is to be specified on the DOS command line
or on the Extra Specifications line. Commas are used as separators, and &quot;-&quot; as place-holders:
TFILE=* 15,-,-,-,13 *
118.
ISUBTOTAL columns within item label for subtotals in Table 27
This specifies what part of the data record is to be used to classify items for subtotal in Table 27.
Format 1: ISUBTOTAL = $S..W.. or $S..E.. using the column selection rules.
$S..W.. e.g., $S2W13 means that the label to be shown on the map starts in column 2 of the item label and is 13
columns wide.
$S..E.. e.g., $S3E6 means that the label to be shown on the map starts in column 3 of the item label and ends in
column 6.
These can be combined, and constants introduced, e.g,
ISUBTOTAL=$S3W2+&quot;/&quot;+$S7W2
If the item label is &quot;KH323MXTR&quot;, the subgrouping will be shown as &quot;32/XT&quot;
Format 2: ISUBTOTAL=*
This is followed by a list of subgroupings, each on a new line using the column selection rules:
ISUBTOTAL=*
$S1W1+$S7W2
; Subtotals reported for item classifications according to these columns
$S3E5
; Subtotals reported for item classifications according to these columns
*
119
Example:
Subtotal by first letter of item name:
ISUBTOTAL=$S1W1
TFILE=*
27 ; produce the subtotal report
*
Here is a subtotal report (Tables 27) for items beginning with &quot;R&quot;
&quot;R&quot; SUBTOTAL FOR 8 NON-EXTREME ITEMS
+-----------------------------------------------------------------------------+
|
RAW
MODEL
INFIT
OUTFIT
|
|
SCORE
COUNT
MEASURE
ERROR
MNSQ
ZSTD
MNSQ
ZSTD |
|-----------------------------------------------------------------------------|
| MEAN
28.1
25.0
4.04
3.48
.91
-.5
1.04
.0 |
| S.D.
5.5
.0
6.63
.14
.31
1.1
.54
1.4 |
| MAX.
38.0
25.0
16.30
3.82
1.61
2.0
2.37
3.4 |
| MIN.
19.0
25.0
-6.69
3.38
.64
-1.6
.60
-1.2 |
|-----------------------------------------------------------------------------|
| REAL RMSE
3.63 ADJ.SD
5.54 SEPARATION 1.52 PUPIL RELIABILITY .70 |
|MODEL RMSE
3.48 ADJ.SD
5.64 SEPARATION 1.62 PUPIL RELIABILITY .72 |
| S.E. OF PUPIL MEAN = 2.50
|
| WITH 2 EXTREME = TOTAL 10 PUPILS MEAN = 3.05, S.D. = 28.19
|
| REAL RMSE
8.88 ADJ.SD
26.75 SEPARATION 3.01 PUPIL RELIABILITY .90 |
|MODEL RMSE
8.83 ADJ.SD
26.77 SEPARATION 3.03 PUPIL RELIABILITY .90 |
| S.E. OF PUPIL MEAN = 9.40
|
+-----------------------------------------------------------------------------+
MAXIMUM EXTREME SCORE: 1 PUPILS
MINIMUM EXTREME SCORE: 1 PUPILS
LACKING RESPONSES: 1 PUPILS
DELETED: 1 PUPILS
119.
ITEM title for item labels
Up to 12 characters to use in table headings to describe the kind of items, e.g.
ITEM=MCQ.
Choose a word which makes its plural with an &quot;s&quot;, e.g. MCQS, since an S is added to whatever you specify. If you
say ITEM=mcq, then the plural will be &quot;mcqs&quot;.
120.
ITEM1 column number of first response
Specifies the column position where the response-string begins in your data file record, or the column where the
response-string begins in the new record formatted by FORMAT=.
If you have the choice, put the person-identifiers first in each record, and then the item responses with each
response taking one column.
Error messages regarding ITEM1= may be because your control file is not in &quot;Text with line breaks&quot; format.
It is easy to miscount the ITEM1= column. Scroll to the top of the Winsteps screen and check column
positions:
Input in process..
Input Data Record:
1
2
1234567890123456789012345678
Richard M 111111100000000000
^P
^I
^N
35 KID
Records Input.
^P marks the Name1=1 column position with ^.
^I marks the Item1=11 column position with ^.
120
^N marks the NI=18 column position with ^.
Example 1: The string of 56 items is contained in columns 20 to 75 of your data file.
ITEM1=20
NI=56
XWIDE=1
response to item 1 in column 20
for 56 items
one column wide (the standard)
Example 2: The string of 100 items is contained in columns 30 to 69 of the first record, 11 to 70 of the second
record, followed by 10 character person i.d.
XWIDE=1
one column wide (the standard)
FORMAT=(T30,40A,/,T11,60A,10A)
two records per person
ITEM1=1
item 1 in column 1 of reformatted record
NI=100
for 100 items
NAME1=101
person id starts in column 101
NAMLEN=10
person id starts in 10 columns wide
121.
ITLEN maximum length of item label
ITLEN= specifies the maximum number of columns in the control file that are to be used as item names. The
maximum possible is 300 characters.
Example 1: You only wish to use the first five columns of item information to identify the items on the output:
NI=4
ITLEN=5
&amp;END
AX123 This part is not shown on the output
BY246 Trial item
AZ476 This item may be biased
ZZ234 Hard item at end of test
END NAMES
Example 2: Your item names may be up to 50 characters long:
NI=4
ITLEN=50
&amp;END
This item
This item
This item
This item
END NAMES
122.
demonstrates ability for constructive reasoning
flags rigid thinking
detects moral bankruptcy
is a throw-away
IVALUEx recoding of data
Responses are revalued according to the matching codes in IREFER= (or ISGROUPS= if IREFER= is omitted).
Items in IREFER= not referenced by an IVALUEx= are not recoded.
IVALUEa= is the same as IVALUEA=
The recoded values in IVALUEx= line up vertically with the response codes in CODES=, if a data value does not
match any value in CODES= it will be treated as missing.
Valid one-character IVALUE= codes include: !#$%&amp;./123456789&lt;&gt;@ABCDEFGHIJKLMNOPQRSTUVWXYZ^_|~
A-Z are the same as a-z.
Characters with ASCII codes from 129-255 can also be used, but display peculiarly:
&Icirc;&Iuml;&ETH;&Ntilde;&Ograve;&Oacute;&Ocirc;&Otilde;&Ouml;&times;&Oslash;&Ugrave;&Uacute;&Ucirc;&Uuml;&Yacute;&THORN;&szlig;&agrave;&aacute;&acirc;&atilde; etc.
When XWIDE=2 or more, then
either (a) Use one character per XWIDE and blanks,
NI=8
121
XWIDE=2
IREFER=' A B C D D C B A'
or (b) Use one character per item with no blanks
NI=8
XWIDE=2
RESCORE='ABCDDCBA'
Layout is:
NI = 17
IREFER = AABCADBCDEEDEAABC ; the recoding type designators for the 17 items
; see the vertical line up here
CODES
= 0123456 ; valid codes across all items
IVALUEA = 012**** ; recodes for Grouping A
IVALUEB = *1224** ; recodes for Grouping B: &quot;2&quot; and &quot;3&quot; recoded to &quot;2&quot;
IVALUEC = *122226 ; 1-2-6 acts as 1-2-3 because STKEEP=NO
IVALUED = 012333*
IVALUEE = 00123**
STKEEP=NO
; missing intermediate codes are squeezed out
Example 1: Items identified by Y and Z in IREFER= are to be recoded.
Y-type items are 1-3, 7-8. Z-type items are 4-6, 9-10. All items have their own rating (or partial credit) scales,
NI = 10
IREFER = YYYZZZYYZZ ; items identified by type: item 1 is Y, item 4 is Z etc.
CODES = ABCD ; original codes in the data
IVALUEY= 1234 ; for Y-type items, this converts A to 1, B to 2, C to 3, D to 4
IVALUEZ= 4321 ; for Z-type items, this converts A to 4, B to 3, C to 2, D to 1
ISGROUPS=0
; allow each item to have its own rating (or partial credit) scale
structure
Example 2: Items identified by 1, 2, 3 in ISGROUPS= are to be recoded and given there own rating (or partial
credit) scales
Y-type items are 1-3, 7-8. Z-type items are 4-6, 9-10.
NI = 10
ISGROUPS = YYYZZZYYZZ
CODES= ABCD ; original codes in the data
IVALUEY= 1234
IVALUEZ= 4321
or
NI = 10
ISGROUPS = YYYZZZYYZZ
IREFER = YYYZZZYYZZ
CODES= ABCD ; original codes in the data
IVALUEY= 1234
IVALUEZ= 4321
Example 3: All items are to be recoded the same way.
NI = 100
100 ITEMS
IREFER=*
1-100 X
FOR ALL 100 ITEMS, reference is X
*
Codes = 12345
rescore 12345
IVALUEX = 12223
into 12223
123.
IWEIGHT item (variable) weighting
IWEIGHT= allows for differential weighting of items. The standard weights are 1 for all items. To change the
weighting of items, specify IWEIGHT=
Raw score, count, and standard error of measurement reflect the absolute size of weights as well as their relative
122
sizes. Measure, infit and outfit and correlations are sensitive only to relative weights.
Weighting is treated for estimation as that many independent observations. So, if you weight all items by two, you
will divide the S.E. by the square-root of 2, but will not change the measures or fit statistics.
If you want to do different weighting at different stages of an analysis, one approach is to use weighting to
estimate all the measures. Then anchor them all (IFILE= and IAFILE= etc.) and adjust the weighting to meet your
&quot;independent observation&quot; S.E. and reporting requirements.
If you want the standard error of the final weight-based measure to approximate the S.E. of the unweighted
measure, then ratio-adjust case weights so that the total of the weights is equal to the total number of
independent observations.
Formats are:
IWEIGHT=file name
item number
the weights are in a file of format:
weight
IWEIGHT=*
item number
....
*
weight
IWEIGHT=$S...$W... or $S...$E...
weights are in the item labels using the column selection rules, e.g.,starting in column S... with a width of W...
or starting in column S and ending in column E. This can be expanded, e.g, IWEIGHT = $S23W1+&quot;.&quot;+$S25W2
places the columns next to each other (not added to each other)
Example 1:
In a 20-item test, item 1 is to be given a weight of 2.5, all other items have a weight of 1.
IWEIGHT=*
1
2.5
2-20
1
*
A better weighting, which would make the reported person standard errors more realistic by maintaining
the original total sum of weights at 20 , is:
IWEIGHT=*
1
2.33
2-20
0.93
; 2.5 * 0.93
; the sum of all weights is 20.0
*
or adjust the weights to keep the sample-based &quot;test&quot; separation and reliability about the same - so that the reported
statistics are still reasonable:
e.g., original sample &quot;test&quot; reliability = .9, separation = 3, but separation with weighting = 4
Multiply all weights by (3/4)^2 to return separation to about 3.
Example 2:
The item labels contain the weights in columns 16-18.
IWEIGHT= $S16W3
&amp;END
Item 1 Hello
0.5
Item 2 Goodbye 0.7
......
END NAMES
; or $S16E18
Example 3:
Item 4 is a pilot or variant item, to be given weight 0, so that item statistics are computed, but this item does
not affect person measurement.
IWEIGHT=*
4 0
; Item 4 has weight 0, other items have standard weight of 1.
123
*
124.
KEYn scoring key
Usually only KEY1= is needed for an MCQ scoring key.
Up to 99 keys can be provided for scoring the response choices, with control variables KEY1= through KEY99=.
Usually KEY1= is a character string of &quot;correct&quot; response choices. The standard is one column per correct
response, or two columns if XWIDE=2.
As standard, responses matching the characters in KEY1= are scored 1. Other valid responses are scored 0.
KEY2= through KEY99= are character strings of successively &quot;more correct&quot; response choices to be used when
more than one level of correct response choice is possible for one or more items. The standard score value for
KEY2= is 2, and so on up to the standard score value for KEY99= which is 99. The values assigned to these
keys can be changed by means of KEYSCR=. If XWIDE=1, only the values assigned to KEY1= through KEY9=
can be changed, KEY10= through KEY99= retain their standard values of 10 through 99. If XWIDE=2, the all
KEYn= values can be changed.
Example 1: A key for a 20-item multiple choice exam, in which the choices are coded &quot;1&quot;, &quot;2&quot;, &quot;3&quot; and &quot;4&quot;, with
one correct choice per item.
CODES=1234
valid codes
KEY1 =31432432143142314324 correct answers
Example 2: A 20-item MCQ test with responses entered as &quot;a&quot;, &quot;b&quot;, &quot;c&quot;, &quot;d&quot;.
CODES=abcd
valid responses
KEY1 =cadcbdcbadcadbcadcbd correct answers
Example 3: A 20 item multiple choice exam with two somewhat correct response choices per item. One of the
correct choices is &quot;more&quot; correct than the other choice for each item, so the &quot;less correct&quot; choice will get a score
of &quot;1&quot; (using KEY1=) and the &quot;more correct&quot; choice will get a score of &quot;2&quot; (using KEY2=). All other response
choices will be scored &quot;0&quot;:
CODES=1234
valid responses
KEY1=23313141324134242113 assigns 1 to these responses
KEY2=31432432143142314324 assigns 2 to these responses
0 is assigned to other valid responses
Example 4: A 100 item multiple choice test key.
CODES= ABCD
KEY1 = BCDADDCDBBADCDACBCDADDCDBBADCDACBCDADDCA+
+DBBADCDACBCDADDCDBBADCDACBCDADDCDBBADCCD+
+ACBCDADDCDBBADCDACBC
continuation lines
Example 5: Multiple score key for items 1 to 10. Items 11 to 15 are on a rating scale of 1 to 5
CODES = abcd12345
KEY1
= bacdbaddcd*****
RESCORE= 111111111100000
; RESCORE= signals when to apply KEY1=
Example 6: A 10 item test. 5 MCQ items have responses entered as &quot;ABCD&quot;, with one of those being correct:
Item 1, correct response is B. Item 2 is C. 3 is D. 4 is A. 5 is C. Then 5 partial-credit performance items rated 0-5.
CODES =ABCD012345
ISGROUPS=1111100000
KEY1 = BCDAC11111 ; Key1= automatically has the value &quot;1&quot;, etc.
KEY2 = *****22222 ; * can be any character not in CODES=.
KEY3 = *****33333
KEY4 = *****44444
KEY5 = *****55555
124
125.
KEYFROM location of KEYn
Only use this if you have the scoring Key conveniently in data-record format.
Instructs where to find the KEYn= information.
KEYFROM=0
KEY1= through KEY99=, if used, are before &amp;END.
KEYFROM=1
KEY1 information follows after &amp;END, but before the item names. The key is formatted exactly like a data
record. It is helpful to place the name of the key, e.g. &quot;KEY1=&quot;, where the person name would usually go,
for reference.
KEYFROM=n
KEY1=, then KEY2=, and so on up to KEYn= (where n is a number up to 99) follow &amp;END, but placed
before the item names. Each key is formatted exactly like a data record. It is helpful to place the name of
the key, e.g. &quot;KEY2=&quot;, where the person name would usually go.
Example:
KEY1 and KEY2 information are to follow directly after &amp;END
NAME1=1
ITEM1=10
NI=7
CODES=abcd
KEYFROM=2
&amp;END
KEY1=****bacddba
KEY2=****cdbbaac
Item 1 name
|
Item 7 name
END NAMES
Mantovanibbacdba
|
126.
start of person-id (the standard)
start of response string
number of items
valid codes
two keys in data record format
keys formatted as data
item names
first data record
subsequent data records
KEYSCR reassign scoring keys
This is only needed for complicated rescoring.
Specifies the score values assigned to response choices which match KEY1= etc. To assign responses matching
key to the &quot;missing&quot; value of -1, make the corresponding KEYSCR= entry blank or some other non-numeric
character.
When XWIDE=1, each value can only take one position, so that only KEY1= through KEY9= can be reassigned.
KEY10= through KEY99= can also be used but keep their standard values of 10 through 99.
When XWIDE=2, each value takes two positions, and the values corresponding to all keys, KEY1= through
KEY99=, can be reassigned.
Example 1: Three keys are used, and XWIDE=1.
Response categories in KEY1= will be coded &quot;1&quot;
Response categories in KEY2= will be coded &quot;2&quot;
Response categories in KEY3= will be coded &quot;3&quot;
KEYSCR=123 (standard)
Example 2: Three keys are used, and XWIDE=1.
Response categories in KEY1= will be coded &quot;2&quot;
Response categories in KEY2= will be coded &quot;1&quot;
Response categories in KEY3= will be coded &quot;1&quot;
KEYSCR=211
125
Example 3: Three keys are used, and XWIDE=2
Response categories in KEY1= will be coded &quot;3&quot;
Response categories in KEY2= will be coded &quot;2&quot;
Response categories in KEY3= will be coded &quot;1&quot;
KEYSCR=030201
or
KEYSCR=&quot; 3 2 1&quot;
Example 4: Three keys are used, and XWIDE=1
Response categories in KEY3= will be coded &quot;1&quot;
Response categories in KEY6= will be coded &quot;missing&quot;
Response categories in KEY9= will be coded &quot;3&quot;
KEY3=BACDCACDBA
KEY6=ABDADCDCAB
KEY9=CCBCBBBBCC
KEYSCR=xx1xxXxx3
response keys
scores for keys
The &quot;x&quot;s correspond to unused keys, and so will be ignored.
The &quot;X&quot; corresponds to specified KEY6=, but is non-numeric and so will cause responses matching KEY6= to be
ignored, i.e. treated as missing.
Example 5: Some items in a test have two correct answers, so two keys are used. Since both answers are
equally good, KEY1= and KEY2= have the same value, specified by KEYSCR=. But some items have only one
correct answer so in one key &quot;*&quot;, a character not in CODES=, is used to prevent a match.
CODES=1234
KEY1 =23313141324134242113
KEY2 =31*324321*3142314*** * is not in CODES=
KEYSCR=11
both KEYs scored 1
Example 6: More than 9 KEYn= lines, together with KEYSCR=, are required for a complex scoring model for 20
items, but the original data are only one character wide.
Original data: Person name:
columns 1-10
20 Item responses: columns 21-40
Looks like: M. Stewart..........1321233212321232134
Solution: reformat from XWIDE=1 to XWIDE=2
TITLE=&quot;FORMAT= from XWIDE=1 to =2&quot;
FORMAT=(10A1,10X,20A1)
10 of Name, skip 10, 20 of responses
NI=20
NAME1=1
ITEM1=11
Responses in column 11 of reformatted record
XWIDE=2
CODES=&quot;1 2 3 4 &quot;
Original response now &quot;response blank&quot;
KEY1 =&quot;1 2 1 3 2 1 2 3 1 4 3 2 1 1 1 1 1 2 1 1 &quot;
Keying 20 items
KEY2 =&quot;2 1 2 1 1 2 1 1 2 1 1 1 2 3 3 2 2 * 2 1 &quot;
....
KEY10=&quot;3 3 3 2 3 3 4 2 3 * * * 4 4 4 4 4 4 4 4 &quot;
KEYSCR=&quot;1 2 3 2 2 2 3 4 1 4 &quot;
Renumbering 10 KEYn=
&amp;END
127.
LCONV logit change at convergence
Measures are only reported to 2 decimal places, so a change of less than .005 logits will probably make no visible
difference.
Specifies what value the largest change in any logit estimate for a person measure or item calibration or rating (or
partial credit) scale structure calibration must be less than, in the iteration just completed, for iteration to cease.
The current largest value is listed in Table 0 and displayed on your screen. See convergence considerations.
126
The standard setting is CONVERGE=&quot;E&quot;, so that iteration stops when either LCONV= or RCONV= is
satisfied. (Note: this depends on Winsteps version - and may explain differences in converged values.)
Example:
128.
To set the maximum change at convergence to be less or equal to .001 logits:
LCONV=.001
RCONV=0 ; set to zero, so does not affect convergence decision
CONVERGE=Logit
LINLEN length of printed lines in Tables 7, 10-16, 22
The misfitting responses, name maps, scalogram, and option frequency tables can be output in any convenient
width. Specify LINLEN=0 for the maximum page width (132 characters).
Example:
129.
You want to print the map of item names with up to 100 characters per line.
LINLEN=100 set line length to 100 characters
LOCAL locally restandardize fit statistics
LOCAL=N accords with large-sample statistical theory.
Standardized fit statistics test report on the hypothesis test: &quot;Do these data fit the model (perfectly)?&quot; With large
sample sizes and consequently high statistical power, the hypothesis can never be accepted, because all
empirical data exhibit some degree of misfit to the model. This can make t standardized statistics meaninglessly
large. t standardized statistics are reported as unit normal deviates. Thus ZSTD=2.0 is as unlikely to be observed
as a value of 2.0 or greater is for a random selection from a normal distribution of mean 0.0, standard deviation,
1.0. ZSTD (standardized as a z-score) is used of a t-test result when either the t-test value has effectively infinite
degrees of freedom (i.e., approximates a unit normal value) or the Student's t-distribution value has been adjusted
to a unit normal value.
LOCAL=N t standardized fit statistics are computed in their standard form. Even the slightest item misfit in tests
taken by many persons will be reported as very significant misfit of the data to the model. Columns reported with
this option are headed &quot;ZSTD&quot; for model-exact standardization. This is a &quot;significance test&quot; report on &quot;How
unexpected are these data if the data fit the model perfectly?&quot;
LOCAL=L Instead of t standardized statistics, the natural logarithm of the mean-square fit statistic is reported.
This is a linearized form of the ratio-scale mean-square. Columns reporting this option are headed &quot;LOG&quot;, for
mean-square logarithm.
LOCAL=Y t standardized fit statistics are transformed to reflect their level of unexpectedness in the context of
the amount of disturbance in the data being analyzed. The model-exact t standardized fit statistics are divided by
their local sample standard deviation. Thus their transformed sample standard deviation becomes 1.0. Columns
reported with this option are headed &quot;ZEMP&quot; for empirically restandardized. The effect of the local-rescaling is to
make the fit statistics more useful for interpretation. The meaning of ZEMP statistics is an &quot;acceptance test&quot;
report on &quot;How unlikely is this amount of misfit in the context of the overall pattern of misfit in these data?&quot;
Ronald A. Fisher (&quot;Statistical Methods and Scientific Inference&quot;New York: Hafner Press, 1973 p.81) differentiates
between &quot;tests of significance&quot; and &quot;tests of acceptance&quot;. &quot;Tests of significance&quot; answer hypothetical questions:
&quot;how unexpected are the data in the light of a theoretical model for its construction?&quot; &quot;Tests of acceptance&quot; are
concerned with whether what is observed meets empirical requirements. Instead of a theoretical distribution, local
experience provides the empirical distribution. The &quot;test&quot; question is not &quot;how unlikely are these data in the light of
a theory?&quot;, but &quot;how acceptable are they in the light of their location in the empirical distribution?&quot;
130.
LOGFILE accumulates control files
Specifying LOGFILE=file name causes the current control file to be appended to the log file, enabling an audit trail
of the Winsteps analysis. The contents of Table 0.3 are saved.
Example:
An audit trail of Winsteps analyses is to be maintained at c:\winsteps.log.txt
127
LOGFILE= c:\winsteps.log.txt
131.
LOWADJ correction for bottom rating scale categories
The Rasch model models the measure corresponding to a bottom rating (or partial credit) scale category as
infinite. This is difficult to think about and impossible to plot. Consequently, graphically in Table 2.2 and
numerically in Table 3.1 a measure is reported corresponding to a bottom category. This is the measure
corresponding to an imaginary rating LOWADJ= rating points above the bottom category. HIADJ= is the
corresponding instruction for top categories.
Example: The standard spread in Table 2.2 is based on LOWADJ=0.25. You wish the bottom category number
to be printed more to the right, close to the other categories.
LOWADJ=0.4
132.
MAKEKEY construct MCQ key
For multiple-choice and True-False questions, the analyst is usually provided with the answer key. When an
answer key is not available, MAKEKEY=YES constructs one out of the most frequent responses to each item.
The answer key is used in the analysis and reported at the end of Table 0.3 in the Report Output File. Inspect the
Item Tables, particularly the &quot;CATEGORY/OPTION/Distractor FREQUENCIES&quot;, to identify items for which this
scoring key is probably incorrect. The correct answer is expected to have the highest measure.
If you have no KEY1= at all, put in a dummy key, e.g., all A's or whatever, to get Winsteps to run.
Example:
The scoring key for Example5.con is lost.
MAKEKEY=YES
KEY1 = aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa ;
dummy key
Constructed key is:
KEY1 =
dabbbbadbdcacaadbabaabaaaacbdddabccbcacccbccccacbbcbbbacbdbacaccbcddb
Original key was:
KEY1 = dcbbbbadbdcacacddabadbaaaccbddddcaadccccdbdcccbbdbcccbdcddbacaccbcddb
The keys match for 48 of the 69 items. Item fit Tables suggest up to 29 items whose keys may not be correct.
The key is reported on the Iteration screenon the and after Table 0.3 in the Report Output file accessed by the
Edit File pull-down menu.
133.
MATRIX correlation output format
The correlation matrix ICORFILE= or PCORFILE= can be produced in list or matrix format.
MATRIX = NO is the list format
1
Item
Item
1
2
3 .05
Correlation
-.04
MATRIX = YES is the matrix format
1.0000
-.0451
.0447
......
-.0451
1.0000
-.0448
......
.0447
-.0448
1.0000
......
.0095
-.2024
-.0437
......
.....
.....
.....
.....
128
134.
MAXPAGE the maximum number of lines per page
For no page breaks inside Tables, leave MAXPAG=0
If you prefer a different number of lines per page of your output file, to better match the requirements of your
printer or word processor, give that value (see Using a Word Processor or Text Editor in Section 2). If you prefer
to have no page breaks, and all plots at their maximum size, leave MAXPAG=0.
On Table 1 and similar Tables, MAXPAG= controls the length of the Table.
Example: You plan to print your output file on standard paper with 60 lines per page (pages are 11 inches long,
less 1 inch for top and bottom margins, at 6 lpi):
MAXPAG=60 (set 60 lines per page)
FORMFD=^ (standard: Word Processor form feed)
135.
MFORMS reformat input data
MFORMS= supports the reformatting of input data records, and also equating multiple input files in different
formats, such as alternate forms of the same test. Data after END NAMES or END LABELS is processed first, as
is data specified by DATA= in the core control file.
Data reformatted by MFORMS= can be accessed, viewed, edited and &quot;saved as&quot; permanently using the &quot;Edit&quot;
pull-down menu. It has a file name of the form: ZMF.....txt
Here is the layout:
mforms=*
data=forma.txt
L=2
I1 = 20
I3-5 = 21
I16-20=11
P1=9
P3-8=1
C20-24=&quot;FORMA&quot;
C40-90 = 2:1
#
DATA=formb.txt
P3-7=1
.....
*
Details:
mforms=*
mforms=filename
; the name of an input data file
; there are 2 lines in input data file for each data record
; response to item 1 of the test is in column 20 of the input data file
; items 3, 4, 5 are in columns 21, 22, 23 of the input data file
; items 16, 17,18, 19, 20 are in columns 11, 12, 13, 14, 15
; the first character of person label is in column 9 of the input data file
; person label characters 3 through 8 start in column 1
; put in columns 20-24 the letters FORMA
; put in columns 40-90 the characters in the second line of the data record
; end of definition - start of next file reformat
; name of input data file
; information for columns 3-7 of person label starts in column 1 of data record
; end of mforms= command
instructions follow in control file, and end with another *.
instructions are in a file but in the same format as above.
data=filename
name of input file name to be reformatted.
The reformatted records are placed in a temporary work file. This may be accessed from the Edit pulldown menu, and saved into a permanent file.
This temporary file is processed after any Data Files specified with the master Data= instruction and in
the same way, e.g., any FORMAT= command will be applied also to the temporary work file.
L=nnn
Cnnn=....
nnn is the count of lines in each input data record
If L=1 this can be omitted
L=4 means that 4 input data lines are processed for each data record output.
nnn is the column number in the formatted data record.
129
XWIDE= does not apply. C10-12 means columns 10, 11, 12 of the formatted record.
C1= refers to column 1 of the formatted data record.
This can also be used to move item and person information.
Innn=....
nnn is the starting item number in the formatted data record
nnn-mmm are the starting to ending item numbers in the formatted data record
XWIDE= is applied, so that I3-5= with XWIDE=2 means 6 characters.
I1= points to column Item1= in the formatted data record.
Pnnn=....
nnn is the starting column number in the person label in the formatted person label.
XWIDE= is not applied. P6-8= always means 3 columns starting in column 6.
P1= points to column Name1= in the formatted data record.
.....=nnn
nnn is the starting column of the only, or the first, line in the input data record.
....=m:nnn
m is the line number in each data record
nnn is the starting column number of that line
....=&quot;xxxx&quot;
&quot;xxxx&quot; is a character constant to be placed in the formatted data record.
Note: for I18-20=&quot;abc&quot; with XWIDE=2, then response to Item 18 is &quot;ab&quot;, 19 is &quot;c &quot;, 20
is &quot; &quot;.
#
end of processing of one file, start of the next
*
end of Mforms= processing
Example 1:
See Exam10c.txt
Example 2:
Three data files with common items and one MCQ scoring key.
Datafile1.txt: (Items 1-6)
TOMY
ABCDAB
BILL
BCDADD
Datafile2.txt (Items 1-3 and 7-9)
TOTO
BBADAB
MOULA
BADADD
Datafile3.txt (Items 1-3 and 10-12)
IHSANI
ACCDAB
MALIK
CBDDCD
TITLE=&quot;Multiple MCQ forms with one
NI=12
; 12 ITEMS IN TOTAL
ITEM1=11
NAME1=1
CODES=&quot;ABCD&quot;
KEY1=BACCADACADDA
mforms=*
data=datafile1.txt ; name of data
L=1
; one line per
P1-10=1
; person label
I1-3=11
; items 1-3 in
I4-6=14
; items 4-6 in
#
data=datafile2.txt
L=1
P1-10=1
I1-3=11
I7-9=14
; items 8-9 in
#
data=datafile3.txt
L=1
scoring key&quot;
file
person
in columns 1-10
columns 11-13
columns 14-16
columns 14-16
130
P1-10=1
I1-3=11
I10-12=14
; items 10-12 in columns 14-16
*
&amp;END
; item identification here
END NAMES
Here is how the data appear to Winsteps for analysis:
TOMY
BILL
TOTO
MOULA
IHSANI
MALIK
ABCDAB
BCDADD
BBA
DAB
BAD
ADD
ACC
DAB
CBD
DCD
Example 3: Test 1 is a 4-item survey. Test 2 is a 4-item survey with two items in common with Test 1 which are to
be anchored to their Test 1 values.
Test 1 has 4 rating scale items. Each item has its own partial-credit structure:
title = &quot;Test 1&quot;
item1 = 1
; items start in column 1
ni = 4
; 4 items
name1 = 5
; person label starts in column 5
namlen = 14 ; length of person name
codes = 01234 ; rating scale
ISGROUPS = 0
; each item has its own rating scale structure
stkeep = YES
; this is probably what you want for these type of data
data = data1.txt
ifile = items1if.txt
; item calibrations from Test 1 for Test 2 (output)
sfile = items1sf.txt
; structure calibrations from Test 1 for Test 2
(output)
&amp;END
Test 1 item 1
Test 1 item 2
Test 1 item 3
Test 1 item 4
END NAMES
data 1.txt is:
1234Person 1-1
3212Person 1-2
......
Test 2 has 4 items. 1 and 4 are new - we will call these items 5 and 6 of the combined Test 1 and 2. Item 2 is Test
1 item 4, and item 3 is Test 1 item 2.
title = &quot;Test 2 (formatted to match Test 1)&quot;
item1 = 1
; items start in column 1
ni = 6
; 4 items in Test 1 + 2 more in Test 2
name1 = 7
; person label starts in column 7
namlen =14 ; length of person name
codes = 01234
; rating scale
stkeep = YES
; this is probably what you want for these type of data
ISGROUPS = 0
; each item has its own rating scale structure
iafile = items1if.txt
; item calibrations from Test 1 (input - unchanged)
safile = items1sf.txt
; structure calibrations from Test 1 (input unchanged)
MFORMS = *
; reformat the Test 2 data to align with Test 1
data = data2.txt ; the name of an input data file
131
L = 1
I2 = 3
; there is 1 line in input data file for each data record
; response to item 2 of Test 1 is in 3 of the data2.txt
I4 = 2
; response to item 4 of Test
file
1 is in 2 of the data2.txt
file
I5 = 1
; item 5 is in column 1 of data2.txt
I6 = 4
; item 6 is in column 4 of data2.txt
P1-14 = 5
; the first character of person label is in column 5 of
data2.txt for 14 columns.
*
; end of mforms= command
&amp;END
Test 1 item 1 (blank in Test 2)
Test 1 item 2 (Test 2 item 3)
Test 1 item 3 (blank in Test 2)
Test 1 item 4 (Test 2 item 2)
Item 5 (not in Test 1, Test 2 item 1)
Item 6 (not in Test 1, Test 2 item 4)
END NAMES
data2.txt is:
5426Person 2-1
1234Person 2-2
....
The formatted file (see Edit pull-down menu MFORMS==) is
2 456Person 2-1
3 214Person 2-2
....
136.
MHSLICE Mantel-Haenszel slice width
Differential item functioning (DIF) an be investigated using log-odds estimators, Mantel-Haenszel (1959) for
dichotomies or Mantel (1963) for polytomies. The sample is divided into difference classes (also called reference
groups and focal groups). These are produced for Table 30 specified with DIF=. In principle, when the data fit the
Rasch model, these estimators should concur with the DIF contrast measures. When DIF estimates disagree, it
indicates that the DIF in the data is non-uniform with ability level. The DIF contrast weights each observation
equally. Mantel-Haenszel weights each slice equally.
The simplest way to do a direct comparison of the MH and DIF Contrast methods is to anchor all persons at the
same ability:
PAFILE=*
1-NN 0
*
This will put everyone in one MH cross-tab.
MHSLICE= specifies the width of the slice (in logits) of the latent variable be included in each cross-tab. The lower
end of the first slice is always the lowest observed person measure.
MHSLICE = 0
bypasses Mantel-Haenszel or Mantel computation.
MHSLICE = .1 logits and smaller. The latent variable is stratified into thin slices.
MHSLICES = 1 logit and larger. The latent variable is stratified into thick slices.
For each slice, a cross-tabulation is constructed for each pair of person classifications against each scored
response level. An odds-ratio is computed from the cross-tab. Zero and infinite ratios are ignored. A homogeneity
chi-square is also computed when possible.
Thin slices are more sensitive to small changes in item difficulty across person classifications, but more persons
132
are ignored in inestimable cross-tabs. Thick slices are more robust because fewer persons are ignored. Use the
Specification pull-down menu to set different values of MHSLICE= and then produce the corresponding Table 30.
Person classications are A, B, ... They are compared pairwise. Starting from the lowest person measure, each
slice is MHSLICE= logits wide. There are K slices up through the highest person measure. For the target item, in
the kth slice and comparing classifications A and B:
ACk and BCk are the counts of persons in classifications A and B in slice k. ABCk = ACk + BCk.
ASk and BSk are the summed ratings or scores of persons in classifications A and B in slice k on the target item.
ABSk = ASk + BSk.
ABQk are the summed squared ratings or scores of persons in both classifications A and B in slice k on the target
item.
Then the Mantel or Mantel-Haenszel DIF chi-square for the target item is:
χ
2
1
 K AC kBSk − ASkBC
=  ∑
ABC k
 k =1
k



2
/
K
∑
AC kBC
k =1
k
(A B C
ABC
2
k
k
A B Q k − A B S k2
(A B C k
− 1)
)
For dichotomous items, the Mantel-Haenszel logit DIF size estimate for a dichotomous item is summed across
estimable slices:
 K
  K

ln(αˆ MH ) = log  ∑ AS k (BC k − BS k ) /  ∑ BS k (AC k − AS k ) 
  k =1

 k =1
For polytomous items using adjacent, transitional, sequential odds, the logit DIF size estimate becomes:
 m K
ln (αˆ M H ) = lo g   ∑ ∑ A S jk B S j − 1 .k
  j = 1 k = 1
  m K
 /  ∑ ∑ B S jk A S j − 1 .k
 
  j =1 k =1



 
where ASjk is the count of responses by Classification A in category j of slice k.
Mantel N. (1963) Chi-square tests with one degree of freedom: extensions of the Mantel Haenszel procedure. J
Amer Stat Assoc 58, 690-700.
Mantel, N. and Haenszel, W. (1959) Statistical aspects of the analysis of data from retrospective studies of
disease. J Natl Cancer Inst 22, 719-748.
Example:
+-----------------------------------------------------------------------------------------------------+
| PERSON
DIF
DIF
PERSON
DIF
DIF
DIF
JOINT
MantelHanzl ITEM
|
| CLASS MEASURE S.E. CLASS MEASURE S.E. CONTRAST S.E.
t d.f. Prob. Prob. Size Number Name
|
|-----------------------------------------------------------------------------------------------------|
| A
1.47
.28 P
2.75
.34
-1.28
.44 -2.94 104 .0041 .0040 -1.20
1 Response|
+-----------------------------------------------------------------------------------------------------+
Size of Mantel-Haenszel slice = .100 logits
title=&quot;MH computation&quot;
; d.f.=1 chi=8.3052 p=0.0040
; log-odds = 1.198
codes=01
clfile=*
1 Better
0 Same
*
item1=1
name1=1
NI=1
pweight=$s9w2 ; weighting substitutes for entering multiple records
PAFILE=$S6W1
; anchoring forces stratification
DIF = $4W1
; cross-tab by Gender, F or M
&amp;end
Response
;234567890
END LABELS
1 FA 1 16
0 FA 1 11
1 FP 1
5
0 FP 1 20
1 MA 2 12
0 MA 2 16
1 MP 2
7
133
0 MP 2
19
137.
MISSCORE scoring of missing data codes
This is NOT the missing-value code in your data. All codes NOT in CODES= are missing value codes. Use
this control specification when you want missing data to be treated as valid responses. Wnsteps and
Missing Data: No Problem!
Winsteps processes one observation at a time. For each observation, Xni by person n on item i, it computes an
expectation Eni, based on the current person measure estimate Bn and the current item measure Di and, if
relevant, the current rating (or partial credit) scale structure (calibrations) {Fk}. Pnik is the probability of observing
category k for person n on item i.
In this computation it skips over, omits, ignores &quot;missing&quot; data.
It then compares sum(Xni) with sum(Eni) for each person n, and adjusts Bn.
It then compares sum(Xni) with sum(Eni) for each item i, and adjusts Di
It then compares the count of (Xni=k) with the sum (Pnik) for each k, and adjusts Fk
These sums and counts are only over the observed data. There is no need to impute missing data.
There are no pairwise, listwise or casewise deletions associated with missing data.
MISSCORE= says what to do with characters that are not valid response codes, e.g. blanks and data entry errors.
Usually any characters not in CODES= are treated as missing data, and assigned a value of -1 which mean
&quot;ignore this response.&quot; This is usually what you want when such responses mean &quot;not administered&quot;. If they
mean &quot;I don't know the answer&quot;, you may wish to assign missing data a value of 0 meaning &quot;wrong&quot;, or, on a
typical attitude survey, 3, meaning &quot;neutral&quot; or &quot;don't know&quot;.
MISSING=0 is the same as MISSCORE=0 meaning that all codes in the data not listed in CODES= are to be
scored 0.
Non-numeric codes included in CODES= (without rescoring/recoding) or in NEWSCORE= or IVALUE= are always
assigned a value of &quot;not administered&quot;, -1.
Example 0a:
In my data file, missing data are entered as 9. I want to score them 0, wrong answers. Valid
codes are 0 and 1.
CODES = 01
do not specify a 9 as valid
MISSCORE = 0 specifies that all codes not listed in CODES=, e.g., 9's. are to be
scored 0.
Example 0b:
are 0 and 1.
In my data file, missing data are entered as 9. I want to ignore them in may analysis. Valid codes
CODES = 01
do not specify a 9 as valid
; the following line is the standard, it can be omitted.
MISSCORE = -1 specifies that all codes not listed in CODES=, e.g., 9's.
are to be treated as &quot;not administered&quot;
Example 1: Assign a code of &quot;0&quot; to any responses not in CODES=
MISSCORE=0
missing responses are scored 0.
Example 2: In an attitude rating scale with three categories (0, 1, 2), you want to assign a middle code of &quot;1&quot; to
missing values
MISSCORE=1
missing responses scored 1
Example 3: You want blanks to be treated as &quot;wrong&quot; answers, but other unwanted codes to be ignored items, on
a questionnaire with responses &quot;Y&quot; and &quot;N&quot;.
CODES=&quot;YN &quot; blank included as valid response
NEWSCORE=100 new response values
RESCORE=2
rescore all items
134
MISSCORE=-1
ignore missing responses (standard)
Example 4: Your optical scanner outputs an &quot;@&quot; if two bubbles are marked for the same response. You want to
ignore these for the analysis, but you also want to treat blanks as wrong answers:
CODES =&quot;1234 &quot;
blank is the fifth valid code
KEY1 =31432432143142314324
correct answers
MISSCORE=-1
applies to @ (standard)
Example 5: Unexpected codes are scored &quot;wrong&quot;, but 2's to mean &quot;not administered&quot;.
CODES = 012
NEWSCORE= 01X X is non-numeric, matching 2's ignored
MISSCORE= 0
all non-CODES= responses scored 0
Example 6: You have a long 4-option MCQ test with data codes ABCD. Most students do not have the time to
complete all the items. This requires a two-stage item analysis:
Stage 1.
Item calibration:
Deliberately skipped responses are coded &quot;S&quot; and scored incorrect. The student could not answer the
question.
Not-items are coded &quot;R&quot; and scored &quot;not administered&quot;. This prevents easy items at the end of the test
being calibrated as &quot;very difficult&quot;.
CODES=&quot;ABCDS&quot;
KEY1=&quot;CDBAD.....&quot;
MISSCORE=-1
IFILE=ITEMCAL.TXT
; write out the item calibrations
Stage 2.
Person measurement:
The convention with MCQ tests is that all missing responses are scored incorrect when measuring the
persons.
IAFILE=ITEMCAL.TXT ; anchor on the Stage 1 item calibrations
CODES=&quot;ABCDS&quot;
KEY1=&quot;CDBAD.....&quot;
MISSCORE=0
; all missing data are scored incorrect
138.
MJMLE maximum number of JMLE iterations
JMLE iterations may take a long time for big data sets, so initially set this to -1 for no JMLE iterations. Then set
MJMLE= to 10 or 15 until you know that more precise measures will be useful. The number of PROX iterations,
MPROX=, affects the number of JMLE iterations but does not affect the final estimates.
MJMLE= specifies the maximum number of JMLE iterations to be performed. Iteration will always cease when
both LCONV= and RCONV= criteria have been met, see CONVERGE=. To specify no maximum number
limitation, set MJMLE=0. Iteration always be stopped by Ctrl with F, see &quot;Stopping Winsteps&quot;.
Example 1: To allow up to 4 iterations in order to obtain rough estimates of a complex rating (or partial credit)
scale:
MJMLE=4 4 JMLE iterations maximum
Example 2: To allow up to as many iterations as needed to meet the other convergence criteria:
MJMLE=0 Unlimited JMLE iterations
Example 3: Perform no JMLE iterations, since the PROX estimates are good enough.
MJMLE=-1 No JMLE iteration
Example 4: Run as quick estimation as possible to check out control options.
MPROX=-1 ; minimal prox estimation iterations
MJMLE=-1 ; no JMLE iterations
139.
MNSQ show mean-square or standardized fit statistics
The mean-square or t standardized fit statistics are shown in Tables 7, 11 to quantify the unexpectedness in the
135
response strings, and in Tables 4, 5, 8, 9 for the fit plots.
MNSQ=N Show standardized (ZSTD) fit statistics. ZSTD (standardized as a z-score) is used of a t-test result
when either the t-test value has effectively infinite degrees of freedom (i.e., approximates a unit normal value) or
the Student's t-distribution value has been adjusted to a unit normal value.
MNSQ=Y
Show mean-square fit statistics. Use LOCAL=L for log scaling.
TABLE 7.1 TABLE OF POORLY FITTING PERSONS
( ITEMS IN ENTRY ORDER)
NUMBER - NAME -- POSITION ------ MEASURE - INFIT (MNSQ) OUTFIT
17 Rod
RESPONSE:
Z-RESIDUAL:
M
1:
0 0 2 4 1
-2
-1.41
4 3 1 3 3
2
2.4
A
1 4 3 2 3
-2
2.2
3 1 4 2 1
-2 2 -2
Mean-square:
TABLE 9.1
-5
-4
-3
-2
-1
0
1
2
3
++-------+-------+-------+-------+-------+-------+-------+-------++
2 +
|
+
I
|
|
|
T
|
|
|
E
|
|
B
A
|
M
|
| FDE
C
|
1 +-------------------------------|-----iIJg-Hj------G--------------+
O
|
| e
d f
|
U
|
c|
b
|
T
|
| a
h
|
F
|
|
|
I
0 +
|
+
T
++-------+-------+-------+-------+-------+-------+-------+-------++
-5
-4
-3
-2
-1
0
1
2
3
ITEM MEASURE
2
1
0
t standardized ZSTD:
I
T
E
M
O
U
T
F
I
T
Z
S
T
D
1
0
-1
-2
-3
-4
140.
-5
-4
-3
-2
-1
0
1
2
3
++-------+-------+-------+-------+-------+-------+-------+-------++
|
|
A
|
|
|
B
|
|
|
|
+
|
C
+
|
| DE
|
|
| F
|
|
|
H
|
+-------------------------------|------IJ---j------G--------------+
|
|
i
|
|
|
g
h
|
|
|
f
|
+
|
+
|
| e
d
|
|
c|
|
|
|
|
+-------------------------------|---------------------------------+
|
|
|
|
|
b
|
|
|
|
+
|
+
|
| a
|
|
|
|
|
|
|
+
|
+
++-------+-------+-------+-------+-------+-------+-------+-------++
-5
-4
-3
-2
-1
0
1
2
3
ITEM MEASURE
2
1
0
-1
-2
-3
-4
MODELS assigns model types to items
Winsteps estimates calibrations for four different ordered response category structures. Dichotomies are always
analyzed using the Rasch dichotomous model, regardless of what model is specified for polytomies.
136
MODELS=R (standard) is the default option, specifying standard Rasch analyses using the Rasch
dichotomous model, Andrich &quot;Rating Scale&quot; model and Masters' &quot;Partial Credit&quot; model, see ISGROUPS=
MODELS=S uses the Rasch dichotomous model and the Glas-Verhelst &quot;Success&quot; (growth) model, also called
the &quot;Steps&quot; Model (Verhelst, Glas, de Vries, 1997). If and only if the person succeeds on the first category,
another category is offered until the person fails, or the categories are exhausted, e.g. an arithmetic item, on
which a person is first rated on success on addition, then, if successful, on multiplication, then, if successful, on
division etc. &quot;Scaffolded&quot; items can function this way. This is a continuation ratio model parameterized as a
Rasch model with missing data on unreached categories. Verhelst N.D., Glas C.A.W. &amp; De Vries H.H. (1997) A
Steps model to analyze partial credit. In W.J. van der Linden &amp; R.K. Hambleton (Eds.), Handbook of modern item
response theory (pp. 123 - 138) New York: Springer.
MODELS=F uses the Rasch dichotomous model and the Linacre &quot;Failure&quot; (mastery) model. If a person
succeeds on the first category, top rating is given and no further categories are offered. On failure, the next lower
category is administered until success is achieved, or categories are exhausted. This is a continuation ratio
model parameterized as a Rasch model with missing data on unreached categories. The Success and Failure
model computations were revised at Winsteps version 3.36, August 2002.
MODELS= has three forms: MODELS=RRSSFR and MODELS=* list * and MODELS=*filename. When only one
letter is specified with MODELS=, e.g., MODELS=R, all items are analyzed using that model. Otherwise
MODELS=some combination of R's, F's, S's, and G's, e.g., MODELS=RRSF
Items are assigned to the model for which the serial location in the MODELS string matches the item sequence
number. The item grouping default becomes each item with its own rating scale, ISGROUPS=0.
When XWIDE=2 or more, then
either (a) Use one character per XWIDE and blanks,
NI=8
XWIDE=2
MODELS=' R S R F R S R R' ; this also forces ISGROUPS=0 to be the default
or (b) Use one character per item with no blanks
NI=8
XWIDE=2
RESCORE='RSRFRSRR' ; this also forces ISGROUPS=0 to be the default
Example 1: All items are to be modeled with the &quot;Success&quot; model.
MODELS=S
the Success model
Example 2: A competency test consists of 3 success items followed by 2 failure items and then 10 dichotomies.
The dichotomies are to be reported as one grouping.
NI=15
fifteen items
MODELS=SSSFFRRRRRRRRRR ; matching models: ; forces ISGROUPS=0 to be the default
ISGROUPS=000001111111111 ; dichotomies grouped: overriding the default ISGROUPS=0
or
MODELS=*
1-3 S
4F
5F
6-15 R
*
141.
MODFROM location of MODELS
This command has not proved productive. It is maintained for backwards compatibility.
Only use this if you have too many items to put conveniently on one line of the MODELS= control variable. It is
easier to us &quot;+&quot; continuation lines
Instructs where to find the MODELS= information.
137
MODFROM=N MODELS= is a control variable before &amp;END (the standard).
MODFROM=Y
MODELS= information follows just after &amp;END but before the item names. It is formatted exactly like a
data record. It is helpful to enter &quot;MODELS=&quot; where the person name would go.
Example: A test consists of 10 three-category items. The highest level answer is scored with KEY2=. The next
level with KEY1=. Some items have the &quot;Success&quot; structure, where the higher level is administered only after
success has been achieved on the lower level. Some items have the &quot;Failure&quot; structure, where the lower level is
administered only after failure at the higher level. The MODELS=, KEY1=, KEY2= are formatted exactly like data
records. The data records are in a separate file.
NAME1 = 5
start of person-id
ITEM1 = 20
start of responses
NI = 10
ten items
CODES = ABCDE valid codes
MODFRM = Y
MODELS= in data format
KEYFRM = 2
two keys in data format
DATA = DATAFILE
location of data
;
1
2 columns
;2345678901234567890
&amp;END
MODELS=
SSSFFFSSSS data format
KEY1=
BCDABCDABC starts in column ITEM1 = 20
KEY2=
ABCDDBCBAA
Item name 1
first item name
|
Item name 10
END NAMES
142.
MPROX maximum number of PROX iterations
Specifies the maximum number of PROX iterations to be performed. PROX iterations will always be performed
so long as inestimable parameters have been detected in the previous iteration, because inestimable parameters
are always dropped before the next iteration. At least 2 PROX iterations will be performed. PROX iteration
ceases when the spread of the persons and items no longer increases noticeably (0.5 logits). The spread is the
logit distance between the top 5 and the bottom 5 persons or items.
If you wish to continue PROX iterations until you intervene with Ctrl and S, set MPROX=0. JMLE iterations will
then commence.
Example: To set the maximum number of PROX iterations to 20, in order to speed up the final JMLE estimation
of a symmetrically-distributed set of parameters,
MPROX=20
143.
MRANGE half-range of measures on plots
Specifies the measure (X-axis on most plots) half-range, (i.e., range away from the origin or UMEAN=), of the
maps, plots and graphs. This is in logits, unless USCALE= is specified, in which case it must be specified in the
new units defined by USCALE=. To customize particular tables, use the Specification pull-down menu, or see
TFILE=.
Example 1: You want to see the category probability curves in the range -3 to +3 logits:
MRANGE=3
Example 2: With UMEAN=500 and USCALE=100, you want the category probability curves to range from 250 to
750:
UMEAN=500 new item mean calibration
USCALE=100 value of 1 logit
MRANGE=250 to be plotted each way from UMEAN=
138
144.
NAME1 first column of person label
NAME1= gives the column position where the person label information starts in your data file or in the new record
formatted by FORMAT=.
It is easy to miscount the NAME1= column. Scroll to the top of the Winsteps screen and check column
positions:
Input in process..
Input Data Record:
1
2
1234567890123456789012345678
Richard M 111111100000000000
^P
^I
^N
35 KID
Records Input.
^P marks the Name1=1 column position with ^.
^I marks the Item1=11 column position with ^.
^N marks the NI=18 column position with ^.
Example 1: The person-id starts in column 10, data responses are 1 column wide, in columns 1-8:
NAME1=10
XWIDE=1
NI=8
starting column of person-id
width of response
number of responses
Example 2: The person-id in column 10, there are 4 data responses are 2 columns wide, in columns 1-8:
NAME1=10
XWIDE=2
NI=4
starting column of person-id
width of response
number of responses
Example 3: The person id starts in column 23 of the second record.
FORMAT=(80A,/,80A)
NAME1=103
concatenate two 80 character records
starts in column 103 of combined record
Example 4: The person id starts in column 27 of a record with XWIDE=2 and FORMAT=.
This becomes complicated, see FORMAT=
145.
NAMLEN length of person label
Use this if too little or too much person-id information is printed in your output tables.
NAMLEN= allows you define the length of the person-id name with a value in the range of 1 to 30 characters.
This value overrides the value obtained according to the rules which are used to calculate the length of the
person-id. These rules are:
1) Maximum person-id length is 300 characters
2) Person-id starts at column NAME1=
3) Person-id ends at ITEM1= or end of data record.
4) If NAME1= equals ITEM1= then length is 30 characters.
Example 1: The 9 characters including and following NAME1= are the person's Social Security number, and are
to be used as the person-id.
NAMLEN=9
Example 2: We want to show the responses in Example0.txt as the person label to help diagnose the fit statistics:
ITEM1 = 1
NI = 25
NAME1 = 1
NAMLEN = 25
+-------------------------------------------------------------------------------------------+
|ENTRY
RAW
MODEL|
INFIT | OUTFIT |PTMEA|
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.| KID
|
139
|------------------------------------+----------+----------+-----+--------------------------|
|
12
17
25
-.85
.36|1.90
2.7|2.86
3.9|A .24| 0100220101210011021000101|
|
6
24
25
-.02
.34|1.48
1.7|1.83
2.3|B .30| 1011211011121010122101210|
|
15
27
25
.31
.33| .86
-.5|1.72
1.9|C .10| 1111111111111111112211111|
|
7
44
25
2.62
.47|1.71
1.7|1.00
.3|D .41| 2220022202222222222222222|
|
14
23
25
-.14
.34|1.62
2.1|1.53
1.6|E .54| 2110020212022100022000120|
146.
NAMLMP name length on map for Tables 1, 12, 16
The id fields are truncated for Tables 12 and 16. The name-length for maps variable, NAMLMP=, overrides the
calculated truncation.
This is ignored when IMAP= or PMAP= is specified.
Example: The 9 characters including and following NAME1= are the person&quot;s Social Security number, and are
to be used as the person-id on the maps.
NAMLMP=9
147.
NEWSCORE recoding values
NEWSCORE= says which values must replace the original codes when RESCORE= is used. If XWIDE=1 (the
standard), use one column per code. If XWIDE=2, use two columns per code. The length of the NEWSCORE=
string must match the length of the CODES= string. For examples, see RESCORE=. NEWSCORE= is ignored
when KEYn= is specified.
The responses in your data file may not be coded as you desire. The responses to some or all of the items can
be rescored or keyed using RESCORE=. RESCORE= and NEWSCORE= are ignored when KEYn= is specified,
except as below.
RESCORE=&quot; &quot; or 2 or is omitted
All items are recoded using NEWSCORE=. RESCORE=2 is the standard when NEWSCORE= is specified.
RESCORE= some combination of 1's and 0's
Only items corresponding to 1's are recoded with NEWSCORE= or scored with KEYn=. When KEYn= is
specified, NEWSCORE= is ignored.
If some, but not all, items are to be recoded or keyed, assign a character string to RESCORE= in which &quot;1&quot;
means &quot;recode (key) the item&quot;, and &quot;0&quot; (or blank) means &quot;do not recode (key) the item&quot;. The position of the &quot;0&quot; or
&quot;1&quot; in the RESCORE= string must match the position of the item-response in the item-string.
Example 1: The original codes are &quot;0&quot; and &quot;1&quot;. You want to reverse these codes, i.e., 1 0 and 0 1, for all items.
XWIDE=1
CODES =01
NEWSCORE=10
RESCORE=2
one character wide responses (the standard)
valid response codes are 0 and 1 (the standard)
desired response scoring
rescore all items - this line can be omitted
NI = 100
IREFER=*
1-100 X
*
Codes = 01
IVALUEX = 10
100 ITEMS
or
FOR ALL 100 ITEMS, reference is X
recode 01
into 10
Example 2: Your data is coded &quot;0&quot; and &quot;1&quot;. This is correct for all 10 items except for items 1 and 7 which have
the reverse meaning, i.e. 1 0 and 0 1.
NI=10
CODES =01
ten items
the standard, shown here for clarity
(a) old method - which still works:
NEWSCORE=10
RESCORE=1000001000
revised scoring
only for items 1 and 7
140
(b) new method - recommended:
IVALUE1 =10
revised scoring
IVALUE0 =01
scoring unchanged, so this line can be omitted.
IREFER
=1000001000
only for items 1 and 7
If XWIDE=2, use one or two columns per RESCORE= code, e.g., &quot; 1&quot; or &quot;1 &quot; mean recode (key). &quot; 0&quot; or &quot;0 &quot;
mean do not recode (key).
Example 3: The original codes are &quot; 0&quot; and &quot; 1&quot;. You want to reverse these codes, i.e., 1 0 and 0 1, for items 1
and 7 of a ten item test.
NI
=10
XWIDE =2
CODES = &quot; 0 1&quot;
NEWSCORE=&quot; 1 0&quot;
RESCORE =&quot; 1 0 0 0 0 0 1 0
ten items
two characters wide
original codes
new values
0 0&quot; rescore items 1 &amp; 7
Example 4: The original codes are &quot;0&quot;, &quot;1&quot;, and &quot;2&quot;. You want 0 0, 1 1, and 2 1 for all items
XWIDE=1
CODES =012
NEWSCORE=011
one character wide (standard)
valid codes
desired scoring
Example 5: The original codes are &quot;0&quot;, &quot;1&quot;, and &quot;2&quot;. You want to make 0 2, 1 1, and 2 0, for even-numbered
items in a twenty item test.
NI=20
twenty items
CODES =012
three valid codes
NEWSCORE=210
desired scoring
RESCORE=01010101010101010101
rescore &quot;even&quot; items
Example 6: The original codes are &quot;0&quot;, &quot;1&quot;, &quot;2&quot;, &quot;3&quot; and some others. You want to make all non-specified codes
into &quot;0&quot;, but to treat codes of &quot;2&quot; as missing.
CODES = 0123
NEWSCORE= 01X3
MISSCORE=0
four valid codes
response code 2 will be ignored
treat all invalid codes as 0
Example 7: The original codes are &quot;0&quot;, &quot;1&quot;, &quot;2&quot;, &quot;3&quot;. You want to rescore some items selectively using KEY1= and
KEY2= and to leave the others unchanged - their data codes will be their rating values. For items 5 and 6, 0 0, 1
0, 2 1, 3 2; for item 7, 0 0, 1 0, 2 0, 3 1. Responses to other items are already entered correctly as 0, 1, 2, or 3.
CODES =0123 valid codes
RESCORE=0000111000 rescore items 5,6,7
KEY1
=****223*** keyed for selected items
KEY2
=****33X*** the X will be ignored
^ read these columns vertically
148.
NI number of items
The total number of items to be read in (including those to be deleted by IDFILE= etc.). NI= is limited to about
30000 for one column responses or about 15000 for two column responses in the standard program. NI= is
usually the length of your test (or the total number of items in all test forms to be combined into one analysis).
Example:
If there are 230 items in your test, enter
NI=230
; 230 items
It is easy to miscount the NI= column. Scroll to the top of the Winsteps screen and check column
positions:
Input in process..
Input Data Record:
1
2
1234567890123456789012345678
Richard M 111111100000000000
^P
^I
^N
35 KID
Records Input.
141
^P marks the Name1=1 column position with ^.
^I marks the Item1=11 column position with ^.
^N marks the NI=18 column position with ^.
149.
NORMAL normal distribution for standardizing fit
The standard generally matches the statistics used in BTD and RSA.
Specifies whether distribution of squared residuals is hypothesized to accord with the chi-square or the normal
distribution. Values of t standardized fit statistics are obtained from squared residuals by means of one of these
distributions.
NORMAL=N
t standardized fit statistics are obtained from the squared residuals by means of the chi-square
distribution and the Wilson-Hilferty transformation (the standard).
NORMAL=Y
t standardized fit statistics are obtained from the squared residuals by means of an asymptotic
normal distribution (F.A.G. Windmeijer, The asymptotic distribution of the sum of weighted squared residuals in
binary choice models, Statistica Neerlandica, 1990, 44:2, 69-78).
150.
OSORT option/distractor sort
Specifies the order in which categories, options and distractors are listed within items in the Item
Category/Option/Distractor Tables, such as Table 13.3
OSORT = D
Options within items are listed in their Data order in CODES= (the standard).
OSORT = S or V or &quot; &quot; Score or Value: Options are listed in their Score Value order.
OSORT = A or M Average or Measure: Options are listed in their Average Measure order.
Example: List distractors in original CODES= order:
OSORT=D
CODES=000102
XWIDE=2
ACTS CATEGORY/OPTION/Distractor FREQUENCIES: MISFIT ORDER
+-----------------------------------------------------------------------------+
|ENTRY
DATA SCORE |
DATA
| AVERAGE S.E. OUTF|
|
|NUMBER CODE VALUE | COUNT
% | MEASURE MEAN MNSQ| ACT
|
|--------------------+------------+--------------------+----------------------|
|
23 A 00
0 |
44 59 |
9.57 1.37 1.7 |WATCH A RAT
|
|
01
1 |
20 27 |
10.35 3.79 2.4 |
|
|
02
2 |
11 15 |
8.92 6.48 7.6 |
|
|
MISSING *** |
1
1 |
|
|
|
|
|
|
|
|
5 B 00
0 |
47 63 |
9.57 1.48 1.8 |FIND BOTTLES AND CANS |
|
01
1 |
19 25 |
8.34 3.45 2.9 |
|
|
02
2 |
9 12 |
13.14 8.10 6.1 |
|
151.
00 dislike
01 neutral
02 like
00 dislike
01 neutral
02 like
OUTFIT sort misfits on infit or outfit
Other Rasch programs may use infit, outfit or some other fit statistic. There is no one &quot;correct&quot; statistic. Use the
one you find most useful.
Specifies whether mean-square infit or mean-square outfit is used as your output sorting and selection criterion
for the diagnosis of misfits.
OUTFIT=Y For each person, the greater of the outfit mean-square and infit mean-square is used as the fit
statistic for sorting and selection (the standard).
OUTFIT=N Infit mean-square only is used as the fit statistic for sorting and selection.
142
152.
PAFILE person anchor file
The PFILE= from one run can be used unedited as the person anchor file, PAFILE=, of another.
The person parameter values (thetas) can be anchored (fixed) using PAFILE=. Person anchoring can also
facilitate test form equating. The persons common to two test forms can be anchored at the values for one form.
Then the measures constructed from the second form will be equated to the measures of the first form. Other
measures are estimated in the frame of reference defined by the anchor values.
In order to anchor persons, an anchor file must be created of the following form:
1. Use one line per person-to-be-anchored.
2. Type the sequence number of the person, a blank, and the measure value (in logits if USCALE=1, otherwise
your user-rescaled units) at which to anchor the person.
Anything after &quot;;&quot; is treated as a comment.
PAFILE = filename
Person anchor information is in a file containing lines of format
person entry number
anchor value
person entry number
anchor value
PAFILE=*
Person anchor information is in the control file in the format
PAFILE=*
person entry number anchor value
person entry number anchor value
*
PAFILE=$SnnEnn or PAFILE=$SnnWnn
Person anchor information is in the person data recordsusing the column selection rules, e.g., starting in
column Snn and ending in column Enn or of width Wnn. Blanks of non-numeric values indicate no anchor
value. PAFILE=$S10E12 or PAFILE=$S10W2 means anchor information starts in column 10 and ends
in column 12 of the person's data record (not person label). This can be expanded, e.g, PAFILE =
$S23W1+&quot;.&quot;+$S25W2 places the columns next to each other (not added to each other)
Example: The third person in the test is to be anchored at 1.5 logits, and the eighth at -2.7.
1. Create a file named, say, &quot;PERSON.ANC&quot;
2. Enter the line &quot;3 1.5&quot; into this file, meaning &quot;person 3 is fixed at 1.5 logits&quot;.
3. Enter the line &quot;8 -2.7&quot;, meaning &quot;person 8 is fixed at -2.7 logits&quot;.
4. Specify, in the control file,
PAFILE=PERSON.ANC
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
or, enter directly into the control file
PAFILE=*
3 1.5
8 -2.7
*
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
or, include the anchor information in the data record
PAFILE=$S1E4
NAME1=5
ITEM1=11
NI=12
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
143
&amp;END
.....
END LABELS
Fred
Mary
1.5 Jose
Jo
etc.
111010111001
011010111100
101001111010
111110011101
; this is the first data line
; this data line has an PAFILE= anchor value
To check: &quot;A&quot; after the measure means &quot;anchored&quot;
+----------------------------------------------------------------------------------------+
|ENTRY
RAW
|
INFIT | OUTFIT |PTMEA|
|
|
|NUMBER SCORE COUNT MEASURE ERROR|MNSQ ZSTD|MNSQ ZSTD|CORR.|DISPLACE| PERSONS
|
|------------------------------------+----------+----------+-----+--------+--------------|
|
3
32
35
1.5A
.05| .80
-.3| .32
.6| .53|
.40| Jose
|
153.
PAIRED correction for paired comparison data
Paired comparison data is entered as only two observations in each row (or each column). The raw score of
every row (or column) is identical. In the simplest case, the &quot;winner&quot; receives a '1', the &quot;loser&quot; a '0', and all other
column (or rows) are left blank, indicating missing data.
Example: Data for a chess tournament is entered. Each row is a player. Each column a match. The winner is
scored '2', the loser '0' for each match. For draws, each player receives a '1'.
PAIRED=YES ; paired comparisons
CODES=012
; valid outcomes
NI=56
; number of matches
154.
PANCHQU anchor persons interactively
If your system is interactive, persons to be anchored can be entered interactively by setting PANCHQ=Y before
the &amp;END line. If you specify this, you will be asked if you want to anchor any persons. If you respond &quot;yes&quot;, it will
ask if you want to read these anchored persons from a file; if you answer &quot;yes&quot; it will ask for the file name and
process that file in the same manner as if PAFILE= had been specified. If you answer &quot;no&quot;, you will be asked to
enter the sequence number of each person to be anchored, one at a time, along with the logit (or user-rescaled
by UANCHOR=, USCALE=, UMEAN=) calibration. When you are finished, enter a zero.
Example: You are doing a number of analyses, anchoring a few, but different, persons each analysis. This
time, you want to anchor person 4.
Enter on the DOS control line, or in the control file:
PANCHQ=Y
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
You want to anchor person 4:
WINSTEPS asks you:
DO YOU WANT TO ANCHOR ANY PERSONS?
respond YES(Enter)
DO YOU WISH TO READ THE ANCHORED PERSONS FROM A FILE?
respond NO(Enter)
INPUT PERSON TO ANCHOR (0 TO END):
respond 4(Enter) (the first person to be anchored)
INPUT VALUE AT WHICH TO ANCHOR PERSON:
respond 1.45(Enter) (the first anchor value)
INPUT PERSON TO ANCHOR (0 TO END): 0(Enter) (to end anchoring)
155.
PCORFIL person residual correlation file
This writes out the Table of inter-person correlations which is the basis of the principal components analysis of
144
residuals.
Missing data: for these Winsteps substitutes their expectations when possible. For residuals and standardized
residuals, these are 0. Items with extreme scores (minimum possible or maximum possible): Winsteps drops
these from the correlation computation. The reason for these choices is to make the principal components
analysis of residuals as meaningful as possible.
PCORFILE = file name
Example:
SF.txt,
Write out the Table of inter-person residual correlations. PCORFIL=file.txt - Then file.txt contains, for
Person
1
1
1
.....
3
3
.....
Person
3
4
5
Correlation
.23
-.31
.13
4
5
-.10
-.02
Example 2: When PCORFILE= is selected on the Output Files menu or MATRIX=YES, the Data Format: Matrix
option can be selected:
This produces:
1.0000
.2265
-.3147
......
156.
.2265
1.0000
-.1048
......
-.3147
-.1048
1.0000
......
.1306
-.0222
.0403
......
....
....
....
....
PDELETE person one-line item deletion
A one-line list of persons to be deleted or reinstated can be conveniently specified with PDELETE=. This is
designed to be used in the post-analysis Specification pull-down menu box.
The formats are:
PDELETE= 3 ; an entry number: delete person 3
PDELETE= 6 1 ; delete persons 6 and 1
PDELETE= 2-5 ; delete persons 2, 3, 4, 5
PDELETE= +3-10 ; delete all persons, then reinstate persons 3 to 10.
PDELETE= 4-20 +8 ; delete persons 4-20 then reinstate person 8
PDELETE=3,7,4,10 ; delete persons 3, 7, 4, 10. Commas, blanks and tabs are valid separators. Commas are
useful at the &quot;Extra information prompt.
Example 1: After an analysis is completed, delete criterion-performance synthetic cases from the reporting.
In the Specification pull-down box:
PDELETE=16 25 87
Example 2: Delete all except persons 5-10 and report
Specification menu box: PDELETE=+5-10
Output Tables Menu
Now reinstate item 11 and report items 5-11.
Specification menu box: PDELETE= 1 +11 ; item 1 is already deleted, but prevents deletion of all except
+11.
Output Tables Menu
157.
PDELQU delete persons interactively
Persons to be deleted or selected can be entered interactively by setting PDELQU=Y. If you specify this, you will
145
be asked if you want to delete any persons. If you respond &quot;yes&quot;, it will ask if you want to read these deleted
persons from a file; if you answer &quot;yes&quot; it will ask for the file name and process that file in the same manner as if
PDFILE= had been specified. If you answer &quot;no&quot;, you will be asked to enter the sequence number or numbers of
persons to be deleted or selected one line at a time, following the rules specified for IDFILE=. When you are
finished, enter a zero.
Example: You are doing a number of analyses, deleting a few, but different, persons each analysis. You don't
want to create a lot of small delete files, but rather just enter the numbers at the terminal, so specify:
PDELQU=Y
You want to delete persons 23 and 50.
WINSTEPS asks you:
DO YOU WANT TO DELETE ANY PERSONS?
respond YES(Enter)
DO YOU WISH TO READ THE DELETED PERSONS FROM A FILE?
respond NO(Enter)
INPUT PERSON TO DELETE (0 TO END):
respond 23(Enter) (the first person to be deleted)
INPUT PERSON TO DELETE (0 TO END): 50(Enter)
INPUT PERSON TO DELETE (0 TO END): 0(Enter) (to end deletion)
158.
PDFILE person deletion file
Deletion or selection of persons from a test to be analyzed, but without removing their responses from your data
file, is easily accomplished by creating a file in which each line contains the sequence number of a person or
persons to be deleted or selected (according to the same rules given under IDFILE=), and then specifying this file
by means of the control variable, PDFILE=, or enter the deletion list in the control file using PDFILE=*.
Example 1: You wish to delete the fifth and tenth persons from this analysis.
1. Create a file named, say, &quot;PERSON.DEL&quot;
2. Enter into the file, the lines:
5
10
3. Specify, in the control file,
PDFILE=PERSON.DEL
or, enter directly into the control file,
PDFILE=*
5
10
*
Example 2: The analyst wants to delete the most misfitting persons reported in Table 6.
1. Set up a standard control file.
2. Specify
PDFILE=*
*
3. Copy the target portion of Table 6.
4. Paste it between the &quot;*&quot;
5. Delete characters before the entry numbers.
6. Type ; after the entry numbers to make further numbers into comments.
TITLE = 'Example of person deletion list from Table 6'
IDFILE = *
Delete the border character before the entry number
; ENTRY RAW
INFIT OUTFIT
; NUM SCORE COUNT MEASURE ERROR MNSQ ZSTD MNSQ ZSTD PTBIS PUP
73 ; 21 22
.14 .37 .95 -.3 1.03 .2 B-.19 SAN
75 ; 16 22 -.56 .39 .95 -.3 1.03 .2 C-.19 PAU
Enter the ; to make other numbers into comments
146
*
closure of IDFILE=*
159.
PDROPEXTREME drop persons with extreme scores
Unanchored persons with extreme (zero or perfect, minimum possible or maximum possible) scores provide no
information for estimating item measures, but they are reported and included in summary statistics. To remove
them:
PDROPEXTREME = No
; do not drop extreme persons (standard)
PDROPEXTREME = Yes or All ; drop zero and perfect scores
PDROPEXTREME = Zero or Low or Bottom or Minimum ; drop zero or minimum-possible scores
PDROPEXTREME = Perfect or High or Top or Maximum ; drop perfect or maximum-possible scores
Example: The data file contains many data records of persons who did not attempt this test and so were scored
0. They are skewing the test statistics:
PDROPEXTREME = Zero
160.
PERSON title for person labels
Up to 12 characters to use in table headings to describe the persons, e.g.
PERSON=KID
Choose a word which makes its plural with an &quot;s&quot;, e.g. KIDS.
If you specify, PERSON=kid, then the plural will be &quot;kids&quot;
161.
PFILE person output file
PFILE=filename produces an output file containing the information for each person. This file contains 4 heading
lines (unless HLINES=N), followed by one line for each person containing:
Columns:
Start
End Format Description
1 1 A1 Blank or &quot;;&quot; if HLINES=Y and there are no responses or deleted or extreme (status =0,-1, -2, -3)
2 6 I5 1. The person sequence number (ENTRY)
7 14 F8.2
2. Person's measure (user-rescaled by UMEAN=, USCALE=, UDECIM=) (MEASURE)
If entry number exceeds 5 digits then:
0
1
Columns
123456789101234567
; 1234
-3.27 -1
entry 1234 measure -3.27: extreme minimum
1234
-3.27 1
entry 1234 measure -3.27: not extreme
;12345
-3.27 -1
entry 12345 measure -3.27: extreme minimum
12345
-3.27 1
entry 12345 measure -3.27: not extreme
;123456
-3.27 -1
entry 123456 measure -3.27: extreme minimum: when HLINES=Yes
123456
-3.27 -1
entry 123456 measure -3.27: extreme minimum: when HLINES=No
123456
-3.27 1
entry 123456 measure -3.27: not extreme
;1234567 -3.27 -1
entry 1234567 measure -3.27: extreme minimum: when HLINES=Yes
1234567
-3.27 -1
entry 1234567 measure -3.27: extreme minimum: when HLINES=No
1234567
-3.27 1 entry 1234567 measure -3.27: not extreme
15 17 I3 3. The person's status (STATUS)
2 = Anchored (fixed) measure
1 = Estimated measure
0 = Extreme maximum (estimated using EXTRSC=)
-1 = Extreme minimum (estimated using EXTRSC=)
-2 = No responses available for measure
-3 = Deleted by user
18 25 F7.1
4. The number of responses used in measuring (COUNT) or the observed count (TOTAL=Y)
26 34 F8.1
5. The raw score used in measuring (SCORE) or the observed score (TOTAL=Y)
147
35 41 F7.2
6. Person measure's standard error (user-rescaled by USCALE, UDECIM) (ERROR)
42 48 F7.2
7. Person mean square infit (IN.MSQ)
49 55 F7.2
8. Person infit: t standardized (ZSTD), locally t standardized (ZEMP) or log-scaled (LOG)
56 62 F7.2
9. Person mean square outfit (OUT.MSQ)
63 69 F7.2
10. Person outfit: t standardized (ZSTD), locally t standardized (ZEMP) or log-scaled (LOG)
70 76 F7.2
11. Person displacement (user-rescaled by USCALE=, UDECIM=) (DISPLACE)
77 83 F7.2
12. Person by test-score correlation: point-biserial (PTBS) or point-measure (PTME)
84 90 F7.2
13. Person weight (WEIGHT)
91 96 F6.1
14. Observed percent of observations matching prediction (OBSMA)
97 102 F6.1
15. Expected percent of observations matching prediction (EXPMA)
103 103 1X 15. Blank
104 133 A30+ 16. Person name (NAME)
The format descriptors are:
In = Integer field width n columns
Fn.m = Numeric field, n columns wide including n-m-1 integral places, a decimal point and m decimal places
An = Alphabetic field, n columns wide
Nx = n blank columns.
When CSV=Y, commas separate the values with quotation marks around the &quot;Person name&quot;. When CSV=T, the
commas are replaced by tab characters.
Example: You wish to write a file on disk called &quot;STUDENT.MES&quot; containing the person statistics for import
later into a student information database:
PFILE=STUDENT.MES
When W300=Yes, then this is produced in Winsteps 3.00, 1/1/2000, format:
Columns:
Start
End Format Description
1 1 A1 Blank or &quot;;&quot; if HLINES=Y and there are no responses or deleted (status = -2, -3)
2 6 I5 1. The Person sequence number (ENTRY)
7 14 F8.2
2. Person's calibration (user-rescaled by UMEAN=, USCALE=, UDECIM) (MEASURE)
15 17 I3 3. The Person's status (STATUS)
2 = Anchored (fixed) calibration
1 = Estimated calibration
0 = Extreme minimum (estimated using EXTRSC=)
-1 = Extreme maximum (estimated using EXTRSC=)
-2 = No responses available for calibration
-3 = Deleted by user
18 23 I6 4. The number of responses used in calibrating (COUNT) or the observed count (TOTAL=Y)
24 30 I6 5. The raw score used in calibrating (SCORE) or the observed score (TOTAL=Y)
31 37 F7.2
6. Person calibration's standard error (user-rescaled by USCALE=, UDECIM=) (ERROR)
38 44 F7.2
7. Person mean square infit (IN.MSQ)
45 51 F7.2
8. Person infit: t standardized (ZSTD), locally t standardized (ZEMP) or log-scaled (LOG)
52 58 F7.2
9. Person mean square outfit (OUT.MS)
59 65 F7.2
10. Person outfit: t standardized (ZSTD), locally t standardized (ZEMP) or log-scaled (LOG)
66 72 F7.2
11. Person displacement (user-rescaled by USCALE=, UDECIM=) (DISPLACE)
73 79 F7.2
12. Person by test-score correlation: point-biserial (PTBS) or point-measure (PTME)
80 80 1X 15. Blank
81 132+ A30+ 16. Person name (NAME)
Example of standard PFILE=
; PERSON Knox Cube Test (Best Test Design p.31) Nov 10 15:40 2005
;ENTRY MEASURE STTS COUNT
SCORE ERROR IN.MSQ IN.ZSTD OUT.MS OUT.ZSTD DISPL PTME WEIGHT OBSMA EXPMA
NAME
1
-3.08 1
14.0
4.0
.83
.61 -1.31
.29
-.14
.00
.80
1.00 92.9 84.8
Richard M
;
35
-6.78 -1
14.0
.0
1.88
1.00
.00
1.00
.00
.00
.61
1.00
.0
.0
148
Helen
F
162.
PMAP person label on person map: Tables 1, 16
This specifies what part of the data record is to be used on the person map.
It's format is PMAP = $S..W.. or $S..E.. using the column selection rules.
$S..W.. e.g., $S2W13 means that the person label to be shown on the map starts in column 2 of the person label
and is 13 columns wide.
$S..E.. e.g., $S3E6 means that the person label to be shown on the map starts in column 3 of the person label
and ends in column 6.
These can be combined, and constants introduced, e.g,
PMAP= $S3W2+&quot;/&quot;+$S7W2
If the person label is &quot;KH323MXTR&quot;, the person label on the map will be &quot;32/XT&quot;
The length of PMAP= overrides NAMLMP=
163.
PRCOMP residual type for principal components analyses in Tables 23, 24
Principal components analysis of item-response or person-response residuals can help identify structure in the
misfit patterns across items or persons. The measures have been extracted from the residuals, so only
uncorrelated noise would remain, if the data fit the Rasch model.
PRCOMP=S or Y Analyze the standardized residuals, (observed - expected)/(model standard error).
Simulation studies indicate that PRCOMP=S gives the most accurate reflection of secondary dimensions in the
items.
PRCOMP=R
Analyze the raw score residuals, (observed - expected) for each observation.
PRCOMP=L
Analyze the logit residuals, (observed - expected)/(model variance).
PRCOMP=O
Analyze the observations themselves.
Example 1: Perform a Rasch analysis, and then see if there is any meaningful other dimensions in the residuals:
PRCOMP=S
Standardized residuals
Example 2: Analysis of the observations themselves is more familiar to statisticians.
PRCOMP=O
Observations
164.
PSELECT person selection criterion
Persons to be selected may be specified by using the PSELECT= instruction to match characters within the
person name. Persons deleted by PDFILE= etc. are never selected by PSELECT=.
This can be done before analysis in the control file or with &quot;Extra specifications&quot;. It can also be done after the
analysis using the &quot;Specification&quot; pull-down menu.
Control characters to match person name:
? matches any character
{..} braces characters which can match a single character: {ABC} matches A or B or C.
{.. - ..} matches single characters in a range. {0-9} matches digits in the range 0 to 9.
{.. --..} matches a single &quot;-&quot; {AB--} matches A or B or &quot;-&quot;.
* matches any string of characters - must be last selection character.
Other alphanumeric characters match only those characters.
Each PSELECT= performed using the &quot;Specification&quot; pull-down menu selects from all those analyzed. For
149
incremental selections, i.e., selecting from those already selected, specify +PSELECT=
PSELECT= works best for single column indicators. It is usually possible to convert 2 columns (such as age) to
one column using the rectangular copy feature in Word or in EXCEL =code(age+65) converts &quot;age&quot; to letters,
starting with A as 0.
Example 1: Select for analysis only persons with M in the 5th column of person name. Person name starts in
column 6 of the data record:
NAME1=6
Person name field starts in col. 6
NAMLEN=8
Person name field is 8 characters long
PSELECT=????M* Column 5 of person name is sex
|
END NAMES
xxxxxBPL M J 01101000101001
xxxxxMEL F S 01001000111100
-----1234selection column
selected
omitted
Example 2: Select for analysis all persons with code &quot;A 4&quot; in columns 2-4 of their names. Person name starts in
column 23, so target characters starts in column 24:
NAME1=23
person name starts in column 23
PSELECT=&quot;?A 4*&quot; quotes because a blank is included. A is in col. 2 etc.
ZA 4PQRS selected
Example 3: Select all Male (M in column 2) persons who are Asian or Hispanic (A or H in column 4):
PSELECT=&quot;?M?{AH}*&quot;
1M3A456 MPQRS
selected
1M5H689 ABCDE
selected
1X2A123 QWERT
omitted
Example 4: Select Males (M in column 8) in School 23 (023 in column 14-16):
PSELECT=???????M?????023*
Selects: 1234567MABCDE023XYZ
Example 5: Select codes 1,2,3,4,5,6,7, in column 2:
PSELECT=?{1-7}*
Example 6: Analyze only males (column 4 or person-id). Then report only School C (column 1). Then only report
Grades 4 and 5 (column 2) in School C.
PSELECT=???M* in the Control file or at the Extra Specifications prompt.
PSELECT=C*
using the Specification pull-down menu, after the analysis
+PSELECT=?{45}* using the Specification pull-down menu.
165.
PSORT column within person label for alphabetical sort in Table 19
Table 19 lists persons alphabetically. Table 1 and Table 16 list them alphabetically within lines. Ordinarily, the
whole person name is used. Select the sorting columns in the person labels using the column selection rules,
e.g., starting in column Snn and ending in column Enn or of width Wnn.
Example 1: The person name is entered in the data file starting in column 20. It is a 6-digit student number
followed by a blank and then gender identification in column 8 of the person name. Sort by gender identification
for Table 19, then by student number.
NAME1=20
NAMLEN=8
; student number + gender
PSORT=8+1-6
; same as $S8W1+$S8W6 alphabetical sort on gender
TABLES=1111111111111111111111111
&amp;END
|
END NAMES
150
xxxxxxxxxxxxxxxxxxx123456 M 0010101101001002010102110011
xxxxxxxxxxxxxxxxxxx229591 F 1102010020100100201002010021
sort column
Example 2: The person name contains several important classifiers. Table 19 is needed for each one:
NAME1=14 Person name starts in column 14
ITEM1=24 Response start in column 24
TFILE=*
19 - - - 1
sort starts with column 1 of person name
19 - - - 8
sort starts with column 8 of person name
19 - - - 6
sort starts with column 6 of person name upto the end of the person name
- entered as place-holders, see TFILE=
*
&amp;END
|
END NAMES
xxxxxxxxxxxxx1234 M 12 0010101101001002010102110011
xxxxxxxxxxxxx2295 F 09 1102010020100100201002010021
Example 3: A version of Table 19, sorted on person name column 6, is to be specified on the DOS command line
or on the Extra Specifications line. Commas are used as separators, and &quot;-&quot; as place-holders:
TFILE=* 19,-,-,-,6 *
166.
PSUBTOTAL columns within person label for subtotals in Table 28
This specifies what part of the data record is to be used to classify persons for subtotal in Table 28.
With tab-separated data and the subtotal indicator in a separate field from the Person Iabel, specify the subtotal
field as the person label field using NAME1=, then PSUBTOTAL=$S1W1
Format 1: PSUBTOTAL = $S..W.. or $S..E.. using the column selection rules.
$S..W.. e.g., $S2W13 means that the label to be shown on the map starts in column 2 of the person label and is
13 columns wide.
$S..E.. e.g., $S3E6 means that the label to be shown on the map starts in column 3 of the person label and ends
in column 6.
These can be combined, and constants introduced, e.g,
PSUBTOTAL=$S3W2+&quot;/&quot;+$S7W2
If the person label is &quot;KH323MXTR&quot;, the subgrouping will be shown as &quot;32/XT&quot;
Format 2: PSUBTOTAL=*
This is followed by a list of subgroupings, each on a new line:
PSUBTOTAL=*
$S1W1+$S7W2
; Subtotals reported for person classifications according to these columns
$S3E5
; Subtotals reported for person classifications according to these columns
*
Example:
Subtotal by first letter of person name:
PSUBTOTAL=$S1W1
TFILE=*
27 ; produce the subtotal report
*
Here is a subtotal report (Table 28) for person beginning with &quot;R&quot;
151
&quot;R&quot; SUBTOTAL FOR 8 NON-EXTREME PUPILS
+-----------------------------------------------------------------------------+
|
RAW
MODEL
INFIT
OUTFIT
|
|
SCORE
COUNT
MEASURE
ERROR
MNSQ
ZSTD
MNSQ
ZSTD |
|-----------------------------------------------------------------------------|
| MEAN
28.1
25.0
4.04
3.48
.91
-.5
1.04
.0 |
| S.D.
5.5
.0
6.63
.14
.31
1.1
.54
1.4 |
| MAX.
38.0
25.0
16.30
3.82
1.61
2.0
2.37
3.4 |
| MIN.
19.0
25.0
-6.69
3.38
.64
-1.6
.60
-1.2 |
|-----------------------------------------------------------------------------|
| REAL RMSE
3.63 ADJ.SD
5.54 SEPARATION 1.52 PUPIL RELIABILITY .70 |
|MODEL RMSE
3.48 ADJ.SD
5.64 SEPARATION 1.62 PUPIL RELIABILITY .72 |
| S.E. OF PUPIL MEAN = 2.50
|
| WITH 2 EXTREME = TOTAL 10 PUPILS MEAN = 3.05, S.D. = 28.19
|
| REAL RMSE
8.88 ADJ.SD
26.75 SEPARATION 3.01 PUPIL RELIABILITY .90 |
|MODEL RMSE
8.83 ADJ.SD
26.77 SEPARATION 3.03 PUPIL RELIABILITY .90 |
| S.E. OF PUPIL MEAN = 9.40
|
+-----------------------------------------------------------------------------+
MAXIMUM EXTREME SCORE: 1 PUPILS
MINIMUM EXTREME SCORE: 1 PUPILS
LACKING RESPONSES: 1 PUPILS
DELETED: 1 PUPILS
167.
PTBIS compute point-biserial correlation coefficients
PTBIS=Y Compute and report conventional point bi-serial correlation coefficients, rpbis. These are reported not
only for items but also for persons. Extreme measures are included in the computation, but missing observations
are omitted. In Rasch analysis, rpbis is a useful diagnostic indicator of data miscoding or item miskeying:
negative or zero values indicate items or persons with response strings that contradict the variable. The Biserial
correlation can be computed from the Point-biserial.
PTBIS=N (or PTBIS=RPM). Compute and report point-measure correlation coefficients, rpm or RPM, shown as
PTMEA. These are reported in Tables 14 and 18 for items and persons. They correlate an item's (or person's)
responses with the measures of the encountered persons (or items). rpm maintains its meaning better than rpbis
in the presence of missing observations. Extreme measures are included in the computation. Negative or zero
values indicate response strings that contradict the variable.
The formula for this product-moment correlation coefficient is:
rpbis = {sum {(x-x bar)(y-y bar``)}} over {sort {{sum {(x-x bar``)} sup 2}`` {sum {(y-y bar``)} sup 2}} }
where x = observation for this item (or person), y = total score for this person omitting this item (or for this item
omitting this person).
Conventional computation of rpbis includes persons with extreme scores. These correlation can be obtained by
forcing observations for extreme measures into the analysis. The procedure is:
1) Perform a standard analysis but specify PFILE=, IFILE= and, if a polytomous analysis, SFILE=.
2) Perform a second analysis setting PAFILE= to the name of the PFILE= of the first analysis, IAFILE= to
the name of the IFILE=, and SAFILE= to the name of the SFILE=. The PTBIS of this second analysis is the
conventional rpbis, which includes extreme scores.
RPM (PTMEA) is reported instead of PTBIS when PTBIS=N or PTBIS=RPM is specified. RPM is the pointmeasure correlation, rpm. It is computed in the same way as the point bi-serial, except that Rasch measures
replace total scores.
Example:
For rank-order data, point-biserials are all -1. So specify Point-measure correlations.
PTBIS=NO
152
168.
PVALUE proportion correct or average rating
An important statistic in classical item analysis is the item p-value, the proportion of the sample succeeding on a
dichotomous item. In Winsteps, this is interpreted as the average of the responses to the item whether
dichotomous or polytomous.
PVALUE = NO Do not report the item p-value.
PVALUE=YES Report item p-values in the IFILE= and Tables 6, 10, etc.
Example: To parallel a classical analysis of an MCQ data set, it is desired to report the raw scores (including
extreme persons and items), the point-biserials and the p-values.
TOTALSCORE= YES ; report original total raw scores
PTBIS = YES
; report point-biserial, not point-measure correlations
PVALUE = YES
; report p-values
+----------------------------------------------------------------------------------------+
|ENTRY
TOTAL
|
INFIT | OUTFIT |PTBIS| P- |
|
|NUMBER SCORE COUNT MEASURE ERROR|MNSQ ZSTD|MNSQ ZSTD|CORR.|VALUE| ITEM
|
|------------------------------------+----------+----------+-----+-----+-----------------|
|
1
35
35
-6.59
1.85| MINIMUM ESTIMATED MEASURE | 1.00| 1= 1-4
|
|
4
32
35
-4.40
.81| .90
.0| .35
.8| .48| .91| 4= 1-3-4
|
|
12
6
35
2.24
.55|1.16
.6|1.06
.5| .26| .17| 12=1-3-2-4-3
|
|
18
0
35
6.13
1.84| MAXIMUM ESTIMATED MEASURE | .00| 18=4-1-3-4-2-1-4|
169.
PWEIGHT person (case) weighting
PWEIGHT= allows for differential weighting of persons. The standard weights are 1 for all persons. To change
the weighting of items, specify PWEIGHT=
Raw score, count, and standard error of measurement reflect the absolute size of weights as well as their relative
sizes.
Measure, infit and outfit and correlations are sensitive only to relative weights.
If you want the standard error of the final weight-based measure to approximate the S.E. of the unweighted
measure, then ratio-adjust case weights so that the total of the weights is equal to the total number of
independent observations.
Formats are:
PWEIGHT=file name
the weights are in a file of format:
person entry number
weight
PWEIGHT=*
person entry number
....
*
weight
PWEIGHT=$S...$W... or $S...$E...
weights are in the data records using the column selection rules, e.g., starting in column S... with a width of
W... or starting in column S and ending in column E. This can be expanded, e.g, PWEIGHT =
$S23W1+&quot;.&quot;+$S25W2 places the columns next to each other (not added to each other)
Example 1:
In a sample of 20 respondents, person 1 is to be given a weight of 2.5, all other persons have a weight of 1.
PWEIGHT=*
1
2.5
2-20
1
*
153
A better weighting, which would make the reported item standard errors more realistic by maintaining the
original total sum of weights at 20 , is:
PWEIGHT=*
1
2.33
2-20
0.93
; 2.5 * 0.93
; the sum of all weights is 20.0
*
+---------------------------------------------------------------------------------+
|ENTRY
RAW
MODEL|
INFIT | OUTFIT |PTMEA|
|
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.|WEIGH| KID
|
|------------------------------------+----------+----------+-----+-----+----------|
|
1
4
10
-1.20
.85| .65
-.9| .41
.3| .82| 2.33| Richard M|
|
2
7
10
1.70
1.17| .18 -1.2| .10
-.4| .94| .93| Tracie F|
|
3
7
10
1.70
1.17|2.04
1.2|1.32
.7| .84| .93| Walter M|
Example 2:
The data records contain the weights in columns 16-18 of the record. The person label starts in column 15, so
this is also column 2 of the person label
PWEIGHT= $C16W3
; or $C16E18 ; column in data record
NAME1 = 15
NAMELEN = 20
PWEIGHT= $S2W3
; start of person label
; length of person label
; location in person label
or
&amp;END
...
END NAMES
10110110011001 0.5 Person A
01001100011011 0.7 Person B
......
Example 3:
Person 4 is a dummy case, to be given weight 0.
PWEIGHT=*
4 0
; Person 4 has weight 0, other persons have standard weight of 1.
*
170.
QUOTED quote-marks around labels
Non-numeric values in the output files can be placed within quote-marks. This is required by some software in
order to decode internal blanks within labels correctly. These apply to comma-separated and tab-separated
output files.
QUOTED=Y
&quot;non-numeric values within quotation marks&quot;
QUOTED=N
non-numeric values without quotation marks.
Example: Produce an SFILE=
CSV=Y
; produce a comma-separated output file
QUOTED=Y
; with labels in quotation marks
&quot;;STRUCTURE MEASURE ANCHOR FILE&quot;
&quot;;CATEGORY&quot;,&quot;Rasch-Andrich Threshold&quot;
0,.00
1,-.86
2,.86
QUOTED=N
; labels without quotation marks
;STRUCTURE MEASURE ANCHOR FILE
;CATEGORY,Rasch-Andrich Threshold
0,.00
1,-.86
2,.86
154
171.
RCONV score residual at convergence
Scores increment in integers so that 0.1 is about as precise a recovery of observed data as can be hoped for.
Specifies what value the largest score residual, corresponding to any person measure or item calibration, must be
less than in the iteration just completed for iteration to cease. The current largest value is listed in Table 0, and
displayed on your screen. In large data sets, the smallest meaningful logit change in estimates may correspond
to score residuals of several score points. See convergence considerations.
The standard setting is CONVERGE=&quot;E&quot;, so that iteration stops when either LCONV= or RCONV= is
satisfied. (Note: this depends on Winsteps version - and may explain differences in converged values.)
Example: To set the maximum score residual, when convergence will be accepted, at 5 score points and
maximum logit change in estimates of .01 logits. Your data consists of the responses of 5,000 students to a test
of 250 items.
NI=250
; 250 items
RCONV=5 ; score residual convergence at 5 score points
LCONV=.01
; this is the standard.
CONVERGE=Both ;convergence when both RCONV= and LCONV= are met
172.
REALSE inflate S.E. for misfit
The modeled, REALSE=N, standard errors of measure estimates (abilities and difficulties) are the smallest
possible errors. These always overstate the measurement precision.
Controls the reporting of standard errors of measures in all tables.
REALSE=N
Report modeled, asymptotic, standard errors (the standard).
REALSE=Y
Report the modeled standard errors inflated by the square root of the infit mean square, when it is greater than
1.0. This inflates the standard error to include uncertainty due to overall lack of fit of data to model.
See Standard Errors: Model and Real for more details.
173.
RESCORE response recoding
The responses in your data file may not be coded as you desire. The responses to some or all of the items can
be rescored or keyed using RESCORE=. RESCORE= and NEWSCORE= are ignored when KEYn= is specified,
except as below. If rescoring implies that the items have different rating (or partial credit) scale structures,
ISGROUPS= may also be required.
RESCORE= has three forms: RESCORE=1101110 and RESCORE=* list * and RESCORE=*filename
RESCORE=&quot; &quot; or 2 or is omitted
All items are recoded using NEWSCORE=. RESCORE=2 is the standard when NEWSCORE= is specified.
RESCORE= some combination of 1's and 0's
Only items corresponding to 1's are recoded with NEWSCORE= or scored with KEYn=. When KEYn is specified,
NEWSCORE= is ignored.
If some, but not all, items are to be recoded or keyed, assign a character string to RESCORE= in which &quot;1&quot;
means &quot;recode (key) the item&quot;, and &quot;0&quot; (or blank) means &quot;do not recode (key) the item&quot;. The position of the &quot;0&quot; or
&quot;1&quot; in the RESCORE= string must match the position of the item-response in the item-string.
When XWIDE=2 or more, then
either (a) Use one character per XWIDE and blanks,
155
NI=8
XWIDE=2
RESCORE=' 1 0 1 0 1 0 1 1'
or (b) Use one character per item with no blanks
NI=8
XWIDE=2
RESCORE='10101011'
Example 1: The original codes are &quot;0&quot; and &quot;1&quot;. You want to reverse these codes, i.e., 1 0 and 0 1, for all items.
XWIDE=1
one character wide responses (the standard)
CODES =01
valid response codes are 0 and 1 (the standard)
NEWSCORE=10
desired response scoring
RESCORE=2
rescore all items - this line can be omitted
Example 2: Your data is coded &quot;0&quot; and &quot;1&quot;. This is correct for all 10 items except for items 1 and 7 which have
the reverse meaning, i.e. 1 0 and 0 1.
NI=10
ten items
CODES =01
standard, shown here for clarity
(a) old method - which still works:
NEWSCORE=10
revised scoring
RESCORE=1000001000
only for items 1 and 7
or
NEWSCORE=20
RESCORE=*
1 1
item 1 is to be rescored
7 1
item 7 is to be rescored
*
(b) new method - recommended:
IVALUE1 =10
revised scoring
IVALUE0 =01
scoring unchanged, so this line can be omitted.
IREFER =1000001000 only for items 1 and 7
If XWIDE=2, use one or two columns per RESCORE= code, e.g., &quot; 1&quot; or &quot;1 &quot; mean recode (key). &quot; 0&quot; or &quot;0 &quot;
mean do not recode (key).
Example 3: The original codes are &quot; 0&quot; and &quot; 1&quot;. You want to reverse these codes, i.e., 1 0 and 0 1, for items 1
and 7 of a ten item test.
NI
=10
XWIDE =2
CODES
=&quot; 0 1&quot;
NEWSCORE=&quot; 1 0&quot;
RESCORE =&quot; 1 0 0 0 0 0 1 0
;
;
;
;
0
ten items
two characters wide
original codes
new values
0&quot; ; rescore items 1 &amp; 7
Example 4: The original codes are &quot;0&quot;, &quot;1&quot;, and &quot;2&quot;. You want 0 0, 1 1, and 2 1 for all items
XWIDE=1
one character wide (standard)
CODES =012
valid codes
NEWSCORE=011 desired scoring
Example 5: The original codes are &quot;0&quot;, &quot;1&quot;, and &quot;2&quot;. You want to make 0 2, 1 1, and 2 0, for even-numbered
items in a twenty item test.
NI=20
twenty items
CODES =012
three valid codes
NEWSCORE=210
desired scoring
RESCORE=01010101010101010101 rescore &quot;even&quot; items
Example 6: The original codes are &quot;0&quot;, &quot;1&quot;, &quot;2&quot;, &quot;3&quot; and some others. You want to make all non-specified codes
into &quot;0&quot;, but to treat codes of &quot;2&quot; as missing.
156
CODES = 0123
four valid codes
NEWSCORE= 01X3
response code 2 will be ignored
MISSCORE=0
treat all invalid codes as 0
Example 7: The original codes are &quot;0&quot;, &quot;1&quot;, &quot;2&quot;, &quot;3&quot;. You want to rescore some items selectively using KEY1= and
KEY2= and to leave the others unchanged - their data codes will be their rating values. For items 5 and 6, 0 0, 1
0, 2 1, 3 2; for item 7, 0 0, 1 0, 2 0, 3 1. Responses to other items are already entered correctly as 0, 1, 2, or 3.
CODES =0123 valid codes
RESCORE=0000111000
rescore items 5,6,7
KEY1 =****223*** keyed for selected items
KEY2 =****33X*** the X will be ignored
^ read these columns vertically
Example 8: Multiple score key for items 1 to 10. Items 11 to 15 are on a rating scale of 1 to 5
CODES = abcd12345
KEY1
= bacdbaddcd*****
RESCORE= 111111111100000
; RESCORE= signals when to apply KEY1=
174.
RESFROM location of RESCORE
Only use this if you have too many items to put conveniently on one line of the RESCORE= control variable.
Instructs where to find the RESCORE= information.
RESFRM=N
RESCORE= is a control variable between before &amp;END (the standard).
RESFRM=Y
RESCORE= information follows after &amp;END but before the item names, if any, and is formatted exactly like a
data record. It is helpful, for reference, to enter the label &quot;RESCORE=&quot; where the person name would go in your
data record.
Example: KEY1= and KEY2= information follows the RESCORE= information, all are formatted like your data.
No item names are provided,
NAME1 = 1 start of person-id
ITEM1 = 10 start of responses
NI = 10 ten items
INUMB = Y use item sequence numbers as names
CODES = ABCDE valid codes
RESFRM = Y rescore information in data record format
KEYFRM = 2 two keys in data record format
&amp;END
RESCORE= 0000110000
RESCORE= looks like data
KEY1=
****AB****
KEY1= looks like data
KEY2=
****CA****
KEY2= looks lie data record
George
ABCDABCDAB
first data record
|
subsequent data records
175.
RFILE scored response file
Useful for reformatting data from a family of test forms, linked by a network of common items, into a single
common structure suitable for one-step item banking.
If this parameter is specified in the control file with RFILE=filename, a file is output which contains a scored/keyed
copy of the input data. This file can be used as input for later analyses. Items and persons deleted by PDFILE=
or the like are replaced by blank rows or columns in the scored response file. The file format is:
1. Person-idA30 or maximum person-id length
2. Responses one per item:
A1 if largest scored response is less than or equal to 9
157
A2 if largest scored response is more than 9.
The width of the responses is not determined by XWIDE=.
176.
SAFILE item structure anchor file
The SFILE= of one analysis may be used unedited as the SAFILE= of another.
The rating-scale structure parameter values (taus, Rasch-Andrich thresholds, steps) can be anchored (fixed)
using SAFILE=. The anchoring option facilitates test form equating. The structure in the rating (or partial credit)
scales of two test forms, or in the item bank and in the current form, can be anchored at their other form or bank
values. Then the common rating (or partial credit) scale calibrations are maintained. Other measures are
estimated in the frame of reference defined by the anchor values.
In order to anchor category structures, an anchor file must be created of the following form:
1. Use one line per category Rasch-Andrich threshold to be anchored.
2. If all items use the same rating scale (i.e. ISGROUPS=&quot; &quot;, the standard, or you assign all items to the same
grouping, e.g ISGROUPS=222222..), then type the category number, a blank, and the &quot;structure measure&quot; value
(in logits or your user-rescaled units) at which to anchor the Rasch-Andrich threshold measure corresponding to
that category (see Table 3). If you wish to force category 0 to stay in an analysis, anchors its calibration at 0.
Specify SAITEM=Yes to use the multiple ISGROUP= format
or
If items use different rating (or partial credit) scales (i.e. ISGROUPS=0, or items are assigned to different
groupings, e.g ISGROUPS=122113..), then type the sequence number of any item belonging to the grouping, a
blank, the category number, a blank, and the &quot;structure measure&quot; value (in logits if USCALE=1, otherwise your
user-rescaled units) at which to anchor the Rasch-Andrich threshold up to that category for that grouping. If you
wish to force category 0 to stay in an analysis, anchor its calibration at 0.
This information may be entered directly in the control file using SAFILE=*
Anything after &quot;;&quot; is treated as a comment.
Example 1: A rating scale, common to all items, of three categories numbered 2, 4, and 6, is to be anchored at
pre-set calibrations. The calibration of the Rasch-Andrich threshold from category 2 to category 4 is -1.5, and of
the Rasch-Andrich threshold to category 6 is +1.5.
1. Create a file named, say, &quot;STANC.FIL&quot;
2. Enter the lines
20
place holder for bottom category of this rating scale
4 -1.5
Rasch-Andrich threshold from category 2 to category 4, anchor at -1.5 logits
6 1.5
Rasch-Andrich threshold from category 4 to category 6, anchor at +1.5 logits
Note: categories are calibrated pair-wise, so the Rasch-Andrich threshold values do not have to advance.
3. Specify, in the control file,
ISGROUPS=&quot; &quot;
SAFILE=STANC.FIL
(the standard)
structure anchor file
or, enter directly in the control file,
SAFILE=*
4 -1.5
6 1.5
*
If you wish to use the multiple grouping format, i.e., specify an example item, e.g., 13
SAITEM=YES
SAFILE=*
13 4 -1.5
13 6 1.5
*
To check this: &quot;A&quot; after the structure measure
158
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ|| MEASURE | MEASURE|
|-------------------+------------+------------++---------+--------+
| 4
4
620 34|
.14
.36|
.87
.72||
-1.50A|
.00 |
Example 2: A partial credit analysis (ISGROUPS=0) has a different rating scale for each item. Item 15 has four
categories, 0,1,2,3 and this particular response structure is to be anchored at pre-set calibrations.
1. Create a file named, say, &quot;PC.15&quot;
2. Enter the lines
15 0 0 Bottom categories are always at logit 0
15 1 -2.0
item 15, Rasch-Andrich threshold to category 1, anchor at -2 logits
15 2 0.5
15 3 1.5
3. Specify, in the control file,
ISGROUPS=0
SAFILE=PC.15
Example 3: A grouped rating scale analysis (ISGROUPS=21134..) has a different rating scale for each grouping
of items. Item 26 belongs to grouping 5 for which the response structure is three categories, 1,2,3 and this
structure is to be anchored at pre-set calibrations.
1. Create a file named, say, &quot;GROUPING.ANC&quot;
2. Enter the lines
26 2 -3.3
for item 26, representing grouping 5, Rasch-Andrich threshold to category 2, anchored at
-3.3
26 3 3.3
3. Specify, in the control file,
ISGROUPS =21134..
SAFILE=GROUPING.ANC
Example 4: A partial-credit scale has an unobserved category last time, but we want to use those anchor values
where possible.
We have two choices.
a) Treat the unobserved category as a structural zero, i.e., unobservable. If so...
Rescore the item using IVALUE=, removing the unobserved category from the category hierarchy, and use a
matching SAFILE=.
In the run generating the anchor values, which had STKEEP=NO,
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ|| MEASURE | MEASURE|
|-------------------+------------+------------++---------+--------+
| 1
1
33
0| -.23 -.15|
.91
.93|| NONE
|( -.85)| 1
| 2
2
23
0|
.15
.05|
.88
.78||
-1.12 |
1.44 | 2
| 4
3
2
0|
.29
.17|
.95
.89||
1.12 |( 3.73)| 4
|-------------------+------------+------------++---------+--------+
In the anchored run:
IREFER=A......
CODES=1234
IVALUEA=12*3
SAFILE=*
1 1
.00
1 2
-1.12
1 3
1.12
*
; item 1 is an &quot;A&quot; type item
; valid categories
; rescore &quot;A&quot; items from 1,2,4 to 1,2,3
If the structural zeroes in the original and anchored runs are the same then, the same measures would result
from:
STKEEP=NO
SAFILE=*
159
1
1
1
*
1
2
4
.00
-1.12
1.12
b) Treat the unobserved category as an incidental zero, i.e., very unlikely to be observed.
Here is Table 3.2 from the original run which producted the anchor values. The NULL indicates an incidental or
sampling zero.
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ|| MEASURE | MEASURE|
|-------------------+------------+------------++---------+--------+
| 1
1
33
0| -.27 -.20|
.91
.95|| NONE
|( -.88)|
| 2
2
23
0|
.08 -.02|
.84
.68||
-.69 |
.72 |
| 3
3
0
0|
|
.00
.00|| NULL
|
1.52 |
| 4
4
2
0|
.22
.16|
.98
.87||
.69 |( 2.36)|
|-------------------+------------+------------++---------+--------+
1
2
3
4
Here is the matching SAFILE=
SAFILE=*
1 1
.00
1 2
-.69
1 3
46.71
1 4 -46.02
*
; flag category 3 with a large positive value, i.e., unlikely to be observed.
; maintain sum of structure measures (step calibrations) at zero.
Example 5: Score-to-measure Table 20 is to be produced from known item and rating scale structure difficulties.
Specify:
IAFILE=
; the item anchor file
SAFILE= ; the structure/step anchor file (if not dichotomies)
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
STBIAS=NO
; anchor values do not need estimation bias correction.
The data file comprises two dummy data records, so that every item has a non extreme score, e.g.,
For dichotomies:
Record 1: 10101010101
Record 2: 01010101010
For a rating scale from 1 to 5:
Record 1: 15151515151
Record 2: 51515151515
Pivot Anchoring
Pivots are the locations in the dichotomy, rating (or partial credit) scale at which the categories would be
dichotomized, i.e., the place that indicates the transition from &quot;bad&quot; to &quot;good&quot;, &quot;unhealthy&quot; to &quot;healthy&quot;. Ordinarily
the pivot is placed at the point where the highest and lowest categories of the response structure are equally
probable. Pivot anchoring redefines the item measures.
Dichotomies (MCQ, etc.):
Example 1: To set mastery levels at 75% on dichotomous items (so that maps line up at 75%, rather than 50%)
set
SAFILE=*
1 -1.1 ; set the Rasch-Andrich threshold point 1.1 logits down, so that the item ability appears at 75%
success.
; If you are using USCALE=, then the value is -1.1 * USCALE=
*
160
Polytomies (rating scales, partial credit, etc.:
When a variety of rating (or partial credit) scales are used in an instrument, their different formats perturb the item
hierarchy. This can be remedied by choosing a point along each rating (or partial credit) scale that dichotomizes
its meaning (not its scoring) in an equivalent manner. This is the pivot point. The effect of pivoting is to move the
structure calibrations such that the item measure is defined at the pivot point on the rating (or partial credit) scale,
rather than the standard point (at which the highest and lowest categories are equally probable).
Example 2: Pivoting with ISGROUPS=. Positive (P) items pivot at an expected score of 2.5. Negative (N) items at
an expected score of 2.0
ISGROUPS=PPPPPNNNNN
SAFILE=*
1 2 0.7 ; put in the values necessary to move the center to the desired spot
5 2 0.5 ; e.g., the &quot;structure calibration&quot; - &quot;score-to-measure of pivot point&quot;
*
Example 3: To set a rating (or partial credit) scale turning point: In the Liking for Science, with 0=Dislike,
1=Neutral, 2=Like, anything less than an expected score of 1.5 indicates some degree of lack of liking:
SAFILE=*
1 -2.22 ; put in the step calibration necessary to move expected rating of 1.5 to the desired spot
*
RATING SCALE PIVOTED AT 1.50
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ||CALIBRATN| MEASURE|
|-------------------+------------+------------++---------+--------+
| 0
0
197 22| -2.29 -2.42| 1.05
.99|| NONE
|( -3.42)| dislike
| 1
1
322 36| -1.17 -.99|
.90
.79||
-2.22 | -1.25 | neutral
| 2
2
368 41|
.89
.80|
.98 1.29||
-.28 |(
.92)| like
|-------------------+------------+------------++---------+--------+
|MISSING
1
0|
.04
|
||
|
|
+-----------------------------------------------------------------AVERAGE MEASURE is mean of measures in category.
+-------------------------------------------------------------------+
|CATEGORY
STRUCTURE
| SCORE-TO-MEASURE
|CUMULATIV| COHERENCE|
| LABEL CALIBRATN S.E. | AT CAT. ----ZONE----|PROBABLTY| M-&gt;C C-&gt;M|
|------------------------+---------------------+---------+----------|
|
0
NONE
|( -3.42) -INF
-2.50|
| 63% 44%| dislike
|
1
-2.22
.10 | -1.25 -2.50
.00|
-2.34 | 55% 72%| neutral
|
2
-.28
.09 |(
.92)
.00 +INF |
-.16 | 84% 76%| like
+-------------------------------------------------------------------+
Values of .00 for scores of 1.5 show effect of pivot anchoring on the rating (or partial credit) scale. The structure
calibrations are offset.
TABLE 21.2 LIKING FOR SCIENCE (Wright &amp; Masters p.18) sf.out Aug 1 21:31 2000
EXPECTED SCORE OGIVE: MEANS
++------+------+------+------+------+------+------+------++
2 +
2222222222+
|
22222222
|
|
2222
|
|
222
|
E
|
22
|
X 1.5 +
12
+
P
|
11|
|
E
|
11 |
|
C
|
11
|
|
T
|
1
|
|
E
1 +
11
|
+
D
|
11*
|
|
|
1 *
|
|
S
|
11
*
|
|
C
|
11
*
|
|
O
.5 +
01
*
|
+
R
|
00|
*
|
|
E
|
000 |
*
|
|
|00000
|
*
|
|
|
|
*
|
|
0 +
|
*
|
+
++------+------+------+------+------+------+------+------++
-4
-3
-2
-1
0
1
2
3
4
161
PUPIL [MINUS] ACT MEASURE
Example 4: A questionnaire includes several rating (or partial credit) scales, each with a pivotal transitionstructure between two categories. The item measures are to be centered on those pivots.
1. Use ISGROUPS= to identify the item response-structure groupings.
2. Look at the response structures and identify the pivot point:
e.g., here are categories for &quot;grouping A&quot; items, after rescoring, etc.
Strongly Disagree 1
Disagree
2
Neutral
3
Agree
4
Strongly Agree 5
If agreement is wanted, pivot between 3 and 4, identified as transition 4.
If no disagreement is wanted, pivot between 2 and 3, identified as transition 3.
3. Anchor the transition corresponding to the pivot point at 0, e.g., for agreement:
e.g., for
ISGROUPS=AAAAAAABBBBAACCC
SAFILE=*
640
6 is an item in grouping A, pivoted at agreement (Rasch-Andrich threshold from category 3 into
category 4)
820
8 is an item in grouping B, pivoted at Rasch-Andrich threshold from category 2 into category 3
; no pivoting for grouping C, as these are dichotomous items
*
Example 5: Anchor files for dichomotous and partial credit items. Use the IAFILE= for anchoring the item
difficulties, and SAFILE= to anchor partial credit structures. Winsteps decomposes the Dij of partial credit items
into Di + Fij.
The Di for the partial credit and dichotomous items are in the IAFILE=
The Fij for the partial credit files are in the SAFILE=
Suppose the data are A,B,C,D, and there are two partial credit items, scored 0,1,2, and two merely right-wrong.
0,1 then: :
CODES=ABCD
KEY1=BCBC
KEY2=DA**
ISGROUPS=0
; SCORE OF 1 ON THE 4 ITEMS
; SCORE OF 2 ON THE PARTIAL CREDIT ITEMS
If the right-wrong MCQ items are to be scored 0,2, then
CODES=ABCD
KEY1=BC**
KEY2=DABC
ISGROUPS=0
; SCORE OF 1 ON THE 4 ITEMS
; SCORE OF 2 ON THE PARTIAL CREDIT ITEMS
but better psychometrically is:
CODES=ABCD
KEY1=BCBC
KEY2=DA**
IWEIGHT=*
3-4 2
*
ISGROUPS=0
; SCORE OF 1 ON THE 4 ITEMS
; SCORE OF 2 ON THE PARTIAL CREDIT ITEMS
; items 3 and 4 have a weight of 2.
Then write out the item and partial credit structures
IFILE= items.txt
SFILE=pc.txt
In the anchored run:
CODES= ... etc.
IAFILE=items.txt
SAFILE=pc.txt
CONVERGE=L
; only logit change is used for convergence
162
LCONV=0.005
; logit change too small to appear on any report.
Anchored values are marked by &quot;A&quot; in the Item Tables, and also Table 3.2
177.
SAITEM item numbers in SAFILE with one grouping
Step Files SFILE= and Step Anchor Files SAFILE= (Rating (or partial credit) scale structure anchoring files) have
the format:
For only one active grouping in ISGROUPS=, or no ISGROUPS=
step number step calibration
For multiple groupings in ISGROUPS=, or no ISGROUPS=
example item number in grouping step number step calibration
To specify that the multiple grouping format be used when there is only one active grouping, specify
SAITEM=YES
Example: &quot;Liking for Science&quot; step anchor file:
Standard format for the Rating Scale model:
ISGROUPS= ; no groupings specified
SAFILE=*
1 -1.8 ; Rasch-Andrich threshold from category 0 to category 1
2 1.8 ; Rasch-Andrich threshold from category 1 to category 2
*
Alternative format allowing an example item number to identify the grouping, say item 10
SAITEM=Yes
SAFILE=*
10 1 -1.8
10 2 1.8
*
178.
SANCHQ anchor category structure interactively
If your system is interactive, steps to be anchored can be entered interactively by setting SANCHQ=Y before the
&amp;END line. If you specify this, you will be asked if you want to anchor any steps. If you respond &quot;yes&quot;, it will ask if
you want to read these anchored items from a file; if you answer &quot;yes&quot; it will ask for the file name and process that
file in the same manner as if SAFILE= had been specified. If you answer &quot;no&quot;, you will be asked to enter the step
measures (found in Table 3).
If there is only one rating (or partial credit) scale, enter the category numbers for which the Rasch-Andrich
thresholds are to be anchored, one at a time, along with their logit (or user-rescaled by USCALE=) structure
measure calibrations. Bypass categories without measures. Enter 0 where there is a measure of &quot;NONE&quot;. When
you are finished, enter -1 in place of the category number.
If there are several rating (or partial credit) scales, enter one of the item numbers for each rating (or partial credit)
scale, then the structure measures corresponding to its categories. Repeat this for each category of an item for
each rating (or partial credit) scale. Enter 0 where there is a structure measure for a category of &quot;NONE&quot;.
Entering 0 as the item number completes anchoring.
Example 1: You are doing a number of analyses, anchoring the common rating (or partial credit) scale to different
values each time. You want to enter the numbers at your PC:
SANCHQ=Y
You want to anchor items 4 and 8.
WINSTEPS asks you:
DO YOU WANT TO ANCHOR ANY STRUCTURES? respond YES(Enter)
DO YOU WISH TO READ THE ANCHORED STRUCTURES FROM A FILE?
163
respond NO(Enter)
INPUT STRUCTURE TO ANCHOR (-1 TO END):
respond 2(Enter) (the first category Rasch-Andrich threshold to be anchored)
INPUT VALUE AT WHICH TO ANCHOR Rasch-Andrich threshold:
respond 0(Enter) (the first anchor value)
INPUT Rasch-Andrich threshold TO ANCHOR (-1 TO END): 4(Enter)
INPUT VALUE AT WHICH TO ANCHOR Rasch-Andrich threshold:-1.5(Enter)
INPUT Rasch-Andrich threshold TO ANCHOR (-1 TO END): 6(Enter)
INPUT VALUE AT WHICH TO ANCHOR Rasch-Andrich threshold:1.5(Enter)
INPUT Rasch-Andrich threshold TO ANCHOR (-1 TO END): -1(Enter) (to end anchoring)
Example 2: You wish to enter the structure measures for several rating scales, each comprising a grouping of
items: SANCHQ=Y
WINSTEPS asks you:
DO YOU WANT TO ANCHOR ANY Rasch-Andrich thresholdS? YES(Enter)
DO YOU WANT TO READ THE ANCHORED Rasch-Andrich thresholdS FROM A FILE? NO
Item 1 represents the first grouping of items, sharing a common rating scale:
INPUT AN ITEM, REPRESENTING A GROUPING (0 TO END): 1
INPUT Rasch-Andrich threshold TO ANCHOR (-1 TO END): 0
bottom category
INPUT VALUE AT WHICH TO ANCHOR Rasch-Andrich threshold: 0 &quot;NONE&quot;
INPUT AN ITEM, REPRESENTING A GROUPING (0 TO END): 1
INPUT Rasch-Andrich threshold TO ANCHOR (-1 TO END): 1
INPUT VALUE AT WHICH TO ANCHOR Rasch-Andrich threshold: -0.5
INPUT AN ITEM, REPRESENTING A GROUPING (0 TO END): 1
INPUT Rasch-Andrich threshold TO ANCHOR (-1 TO END): 2
INPUT VALUE AT WHICH TO ANCHOR Rasch-Andrich threshold: 0.5
Item 8 represents the second grouping of items, sharing a common rating scale:
INPUT AN ITEM, REPRESENTING A GROUPING (0 TO END): 8
INPUT Rasch-Andrich threshold TO ANCHOR (-1 TO END): 0
bottom category
When all are anchored, enter 0 to end:
INPUT AN ITEM, REPRESENTING A GROUPING (0 TO END): 0
179.
SCOREFILE person score file
If SCOREFILE=filename is specified, a file is output which contains the measure and model standard error
corresponding to every possible score on a test consisting of all the items. This is also shown in Table 20. It has
3 heading lines (unless HLINES=N), and has the format:
; PERSON SCORE FILE FOR
; EXAMPLE ANALYSES
;Apr 28 22:37 2005 UMEAN=.000 USCALE=1.000
; TABLE OF SAMPLE NORMS (500/100) AND FREQUENCIES CORRESPONDING TO COMPLETE TEST
;SCORE MEASURE
S.E.
INFO NORMED S.E. FREQUENCY %
CUM.FREQ. % PERCENTILE
0 -5.2891 1.4230
.49
-76 125
0
.0
0
.0
0
...
58
1.8144
.3351
8.90
546
29
15
2.9
363 71.2
70
59
1.9303
.3461
8.35
556
30
20
3.9
383 75.1
73
60
2.0544
.3588
7.77
567
31
21
4.1
404 79.2
77
...
70
5.3232 1.4245
.49
853 125
3
.6
510 100.0
99
1. SCORE: Score on test of all items
The score file shows integer raw scores, unless there are decimal weights for IWEIGHT=. In which case, scores
to 1 decimal place are shown. To obtain other decimal raw scores for short tests, go to the Graphs pull-down
menu. Select &quot;Test Characteristic Curve&quot;. This displays the score-to-measure ogive. Click on &quot;Copy data to
clipboard&quot;. Open Excel. Paste. There will be to three columns. The second column is the measure, the third
column is the raw score.
2. MEASURE: Measure (user-scaled bu USCALE=)
3. S.E.: Standard error (user scaled by USCALE=) - model, because empirical future misfit is unknown.
4. INFO: Statistical information in measure (=1/Logit S.E.&sup2;)
Measures locally-rescaled, so that sample mean=500, sample S.D.=100
5. NORMED: Measure (rescaled)
164
6. S.E.: Standard error (rescaled)
Sample distribution:
7. FREQUENCY: Count of sample at this measure
8. %: Percent of sample at this measure
9. CUM.FREQ.: Count of sample at or below this measure
10. %: Percent of sample at or below this measure
11. PERCENTILE: Percentile of this sample lower than the current measure (range is 0-99).
If CSV=Y, these values are separated by commas. When CSV=T, the commas are replaced by tab characters.
Example 1:
You wish to write a file on disk called &quot;MYDATA.SCF.txt&quot; containing a score-to-measure table for
the complete test.
SCOREFILE=MYDATA.SCF.txt
Example 2: You want a score-to-measure file for items with known difficulties.
ITEM1=1
; start of response string
NI=10
; number of items
CODES=01
; valid item codes
IAFILE=*
; known item difficulties here
1 0.53
*
SAFILE=*
; known structure &quot;step&quot; calibrations here, if rating scale or partial credit items
*
SCOREFILE=sm.txt
; the score-to-measure file - also see Table 20
&amp;END
END LABELS
0101010101
; two dummy data records
1010101010
; give every item a non-extreme score
180.
SDELQU delete category structure interactively
This is better performed with IREFER= and IVALUE=. SDELQU= is not supported in this version of Winsteps.
If your system is interactive, categories to be deleted can be entered interactively by setting SDELQU=Y before
the &amp;END line. If you specify this, you will be asked if you want to delete any category structure.
If you respond &quot;yes&quot;, it will ask if you want to read these deleted categories from a file; if you answer &quot;yes&quot; it will
ask for the file name and process that file in the same manner as if SDFILE= had been specified.
If you answer &quot;no&quot;, you will be asked to enter
a) the sequence number of each item (representing a grouping, as described under SDFILE=). This question is
omitted if all items are in one grouping.
b) the score value of one category to be deleted from that item and its grouping.
Enter these deletions one at a time. When you are finished, enter a zero.
181.
SDFILE category structure deletion file
This is better performed with IREFER= and IVALUE=. SDFILE= is not supported in this version of Winsteps.
Deletion of categories from a test analysis (i.e. conversion of responses in these categories to &quot;missing data&quot;), but
without removing these responses from your data file, is easily accomplished by creating a file in which each line
contains the number of the category to be deleted from that item and its grouping. If there is more than one
grouping, the sequence number of an item representing a grouping followed by a blank.
165
Specify this file by means of the control variable, SDFILE=, or this information may be specified in the control file
using SDFILE=*.
Since rating (or partial credit) scales may be shared by groupings of items, deletion of categories is performed by
grouping:
a) If no ISGROUPS= control variable is specified, no item need be entered, since specifying deletion of a
category deletes that category for all items.
b) If a ISGROUPS= control variables is specified, then specifying deletion of a category for one item deletes that
category for all items in the same grouping.
c) If ISGROUPS=0 is specified, only the specified category for the specified item is deleted.
Example:
You wish to delete particular categories for the fifth and tenth &quot;partial credit&quot; items for this analysis.
1. Create a file named, say, &quot;CAT.DEL&quot;
2. Enter into the file, the lines:
5 3 (item 5 category 3)
10 2 (item 10 category 2)
3. Specify, in the control file,
SDFILE=CAT.DEL
ISGROUPS=0
or, enter in the control file,
SDFILE=*
53
10 2
*
ISGROUPS=0
182.
SFILE category structure output file
If SFILE=filename is specified, a file is output which contains the item and category information needed for
anchoring structures. It has 4 heading lines (unless HLINES=N), and has the format:
1. The item sequence number (I6)
Only if there are multiple groupings in ISGROUPS= or SAITEM=Yes, otherwise this is omitted.
2. The category value (I3) (STRU) - the Rasch-Andrich threshold.
3. structure calibration (F7.2) (user-rescaled by USCALE=) (number of decimals by UDECIM=) (MEASURE)
If the category is an intermediate category with no observations, kept in the analysis by STKEEP=YES, then
its structure calibration is shown as very large. The next higher calibration is shown as very low.
Use this for anchoring, together with IFILE=
If CSV=Y, these values are separated by commas. When CSV=T, the commas are replaced by tab characters.
Anchoring with Unobserved Categories
When STKEEP=YES and there are intermediate null categories, i.e., with no observations, then the RaschAndrich threshold into the category is set 40 logits above the highest calibration. The Rasch-Andrich threshold
out of the category, and into the next category, is set the same amount down. Thus:
Category
0
structure Calibration
Table 3.2
In SFILE
NULL
0.00
166
1
2
3
TOTAL:
-1.00
NULL
1.00
0.00
-1.00
41.00
-40.00
0.00
This is an attempt to model unobserved categories for the next time, when they may be observed.
If categories will not be observed next time, then please specify STKEEP=NO, which automatically drops the
category from the rating (or partial credit) scale.
If categories may be observed next time, then it is better to include a dummy data record in your data file
which includes an observation of the missing category, and reasonable values for all the other item responses
that accord with that missing category. This one data record will have minimal impact on the rest of the analysis.
183.
SIFILE simulated data file
This uses the estimated (or anchored) person, item and structure measures to simulate a data file equivalent to
the raw data. This can be used to investigate the stability of measures, distribution of fit statistics and amount of
statistical bias. Each time SIFILE= is run, or selected from the Output Files pull-down menu, a different simulated
data file is produced. Do simulated analyses with several simulated datasets to verify their overall pattern. When
SIFILE= is specified, missing data values are filled in, where possible.
Missing data: From the Output Files menu, missing data can be imputed or left missing in the simulated data file.
When there are missing data in the original data set, the simulated data can have the same missing data pattern
or be complete.
Example 1: It is desired to investigate the stability of the &quot;Liking for Science&quot; measures.
(1) Estimate measures from SF.txt
(2) Choose SIFILE= from the Output Files menu. SIFILE=SFSIMUL.TXT
(3) Rerun Winsteps with DATA=SFSIMUL.TXT on the &quot;Extra Specifications&quot; line.
(4) Compare person, item and structure measures.
The file format matches the input data file if both are in fixed-field format. When SIFILE= is written with CSV=Y,
comma-separated or CSV=T, tab-separated, the item responses precede the person label.
Example: KCT.txt simulated with fixed field format:
Richard M 111101101000000000
Tracie F 111111111000000000
Walter M 111111101100000000
KCT.txt simulated with comma-separated, CSV=Y, format:
1,1,1,0,0,1,1,0,1,0,0,0,0,0,0,0,0,0,Richard M
1,1,1,1,1,1,1,1,1,0,1,0,0,0,0,0,0,0,Tracie F
1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,Walter M
Example 2. To estimate the measure standard errors in a linked equating design.
Do a concurrent calibration with Winsteps and then simulate data files SIFILE= from the Output Files menu.
Specify &quot;missing data&quot; as &quot;missing&quot; to maintain the same data pattern. Save 10 simulated sets as S1.txt S2.txt .....
Then rerun your Winsteps analysis 10 times specifying in Extra Specifications &quot;DATA=S1.txt PFILE=P1.txt
CSV=TAB&quot; etc. This will produce 10 PFILE=s, P1.txt P2.txt ....., in tab-separated format. These are easy to
import into Excel. So you will have 10 measures for every person. Compute the standard deviation of the
measures for each person based on the 10 person measures - this is their model standard error for the equating
design for each measure. Inflate these values by 20%, say, to allow for systematic equating errors, misfit, etc.
167
184.
SPFILE supplementary control file
There is often a set of control instructions that you wish to make a permanent or temporary part of every control
file. Such files can be specified with SPFILE=. Multiple SPFILE= specifications can be included in one control
file. Supplemental files called with SPFILE= can also include SPFILE= specifications.
Example 1: The analyst has a standard set of convergence criteria and other control instructions to include in
every control file.
a) Enter these into a standard DOS TEXT/ASCII file, e.g, SUPPL.TXT
The analyst's SUPPL.TXT contains:
LCONV=.01
ITEM=TASK
PERSON=CLIENT
TABLES=101110011
b) Specify this supplemental file in the main control file, say, MAIN.TXT
TITLE='NEW YORK CLIENTS'
SPFILE=SUPPL.TXT
ITEM1=37
NI=100
Example 2: The analyst has a set of control instructions that are used only for the final run. These are coded in a
separate DOS TEXT file called FINAL.SPC
C:&gt;WINSTEPS CONTROL.FIL OUTPUT.FIL SPFILE=FINAL.SPC
Example 3: Keyn= is a particularly useful application of SPFILE=.
Put the KEY1= instruction for each test form in its own DOS TEXT file, then reference that file rather than
including the key directly in the control file.
Here is FORMA.KEY:
NI=23
CODES=ABCD
KEY1=ABCDDADBCDADDABBCAADBBA
Here is the control file:
TITLE='FORM A READING RASCH ANALYSIS'
ITEM1=20
SPFILE=FORMA.KEY
TABLES=111011011
|
185.
STBIAS correct for estimation bias
STBIAS=Y causes an approximate correction for estimation bias in JMLE estimates to be applied to measures
and calibrations. This is only relevant if an exact probabilistic interpretation of logit differences is required for short
tests or small samples. Set STBIAS=NO when using IWEIGHT=, PWEIGHT=, anchoring or artificially
lengthened tests or augmented samples, e.g., by replicating item or person response strings.
Fit statistics are computed without this estimation-bias correction. Eastimation-bias correction makes the
measures more central, generally giving a slight overfit condition to Outfit and Infit. Correct &quot;unbiased&quot;
computation of INFIT and OUTFIT needs not only unbiased measures, but also probabilities adjusted for the
possibility of extreme score vectors (which is the cause of the estimation bias). The XMLE algorithm
(implemented experimentally) attempts to do this - but it has other drawbacks that often make it impractical.
Example 1: You have a well-behaved test of only a few items, for which you judge the statistical bias correction to
be useful because you are planning to make exact probabilistic inferences based on differences between
logit measures:
STBIAS=Y
168
186.
STKEEP keep non-observed intermediate categories in structure
Unobserved categories can be dropped from rating (or partial credit) scales and the remaining category recounted
during estimation. For intermediate categories only, recounting can be prevented and unobserved categories
retained in the analysis. This is useful when the unobserved categories are important to the rating (or partial
credit) scale logic or are usually observed, even though they happen to have been unused this time. Category
Rasch-Andrich thresholds for which anchor calibrations are supplied are always maintained wherever
computationally possible, even when there are no observations of a category in the current data set.
Use STKEEP=YES when there may be intermediate categories in your rating (or partial credit) scale that
aren't observed in this data set, i.e., incidental zeroes.
Use STKEEP=NO when your category numbering deliberately skips over intermediate categories, i.e.,
structural zeroes.
STKEEP=N Eliminate unused categories and close up the observed categories.
STKEEP=Y Retain unused non-extreme categories in the ordinal categorization.
When STKEEP=Y, missing categories are retained in the rating (or partial credit) scale, so maintaining the raw
score ordering. But missing categories require locally indefinite structure calibrations. If these are to be used for
anchoring later runs, compare these calibrations with the calibrations obtained by an unanchored analysis of the
new data. This will assist you in determining what adjustments need to be made to the original calibrations in
order to establish a set of anchor calibrations that maintain the same rating (or partial credit) scale structure.
To remind yourself, STKEEP=YES can be written as STRUCTUREKEEP=YES, STRKEEP=YES or
STEPKEEP=YES and other similar abbreviations starting STK, STR and STEPK.
Example 1: Incidental unobserved categories. Keep the developmentally important rating (or partial credit)
scale categories, observed or not. Your small Piaget rating scale goes from 1 to 6. But some levels may not
have been observed in this data set.
STKEEP=Y
Example 2: Structural unobserved categories. Responses have been coded as &quot;10&quot;, &quot;20&quot;, &quot;30&quot;, &quot;40&quot;, but they
really mean 1,2,3,4
CODES = &quot;10203040&quot;
XWIDE = 2
STKEEP=NO
; if STKEEP=YES, then data are analyzed as though categories 11, 12, 13, 14, etc. could exist, which
would distort the measures.
; for reporting purposes, multiply Winsteps SCOREs by 10 to return to the original 10, 20, 30
categorization.
Example 3: Some unobserved categories are structural and some incidental. Rescore the data and use
STKEEP=YES. Possible categories are 2,4,6,8 but only 2,6,8 are observed this time.
(a) Rescore 2,4,6,8 to 1,2,3,4 using IVALUE= or NEWSCORE=
(b) Set STKEEP=YES, so that the observed 1,3,4 and unobserved 2 are treated as 1,2,3,4
(c) For reporting purposes, multiply the Winsteps SCORE by 2 using Excel or similar software.
CODES=2468
NEWSCORE=1234
STKEEP=YES
Incidental and Structural Zeroes: Extreme and Intermediate
For missing intermediate categories, there are two options.
If the categories are missing because they cannot be observed, then they are &quot;structural zeroes&quot;. Specify
&quot;STKEEP=NO&quot;.
This effectively recounts the observed categories starting from the bottom category, so that 1,3,5,7 becomes
169
1,2,3,4.
If they are missing because they just do not happen to have been observed this time, then they are &quot;incidental or
sampling zeros&quot;. Specify &quot;STKEEP=YES&quot;. Then 1,3,5,7 is treated as 1,2,3,4,5,6,7.
Categories outside the observed range are always treated as structural zeroes.
When STKEEP=Y, unobserved intermediate categories are imputed using a mathematical device noticed by Mark
Wilson. This device can be extended to runs of unobserved categories.
187.
STEPT3 include category structure summary in Table 3 or 21
The structure summary statistics usually appear in Table 3. For grouped analysis this part of Table 3 can become
long, in which case it can be moved to Table 21.
Example:
188.
Don't output partial credit structure summaries in Table 3. Move them to Table 21:
ISGROUPS=0 each item has own rating scale
STEPT3=N report scale statistics in Table 21
T1I# number of items summarized by &quot;#&quot; symbol in Table 1
For ease in comparing the outputs from multiple runs, force consistent x-axis scaling by using MRANGE=, T1I#=
and T1P#=. Choose T1I#= to be the largest number of items summarized by one &quot;#&quot; from any of the separate
runs.
Example:
In one run, the bottom of Table 1 states that
EACH &quot;#&quot; IN THE ITEM COLUMN IS 20 ITEMS
In another run: EACH &quot;#&quot; IN THE ITEM COLUMN IS 15 ITEMS
To make the runs visually comparable, specify the bigger value:
T1I#=20
189.
T1P# number of persons summarized by &quot;#&quot; symbol in Table 1
For ease in comparing the outputs from multiple runs, force consistent x-axis scaling by using MRANGE=, T1I#=
and T1P#=. Choose T1P#= to be the largest number of persons summarized by one &quot;#&quot; from any of the separate
runs.
Example:
In one run, the bottom of Table 1 states that
EACH &quot;#&quot; IN THE PERSON COLUMN IS 250 PERSONS
In another run: EACH &quot;#&quot; IN THE PERSON COLUMN IS 300 PERSON
To make the runs visually comparable, specify the bigger value:
T1P#=300
190.
TABLES output tables
For more flexibility, use the Output Tables pull-down menu.
Tables= causes the specified Tables to be written into the report output file. It is has no effect on the pull-down
menus.
A character string that tells WINSTEPS which output tables to prepare for printing. The sequence number of the
&quot;1&quot; or &quot;0&quot; in the TABLES= string matches the table number. For more elaborate table selection, use TFILE=.
&quot;1&quot; means prepare the corresponding table.
&quot;0&quot; or anything else means do not prepare the corresponding table.
Example 1: You want only Tables 2,4,6,8,10 and 20 to be prepared
TABLES=010101010100000000001000
170
This is the same as specifying:
TFILE=*
2
4
6
8
10
20
*
Example 2: You want only Tables 1-4.
TABLES=1111
191.
TARGET estimate using information-weighting
TARGET=Y lessens the effect of guessing on the measure estimates, but increases reported misfit. A big
discrepancy between the measures produced by TARGET=N and TARGET=Y indicates much anomalous
behavior disturbing the measurement process.
Unwanted behavior (e.g. guessing, carelessness) can cause unexpected responses to off-target items. The
effect of responses on off-target items is lessened by specifying TARGET=Y. This weights each response by its
statistical information during estimation. Fit statistics are calculated as though the estimates were made in the
usual manner. Reported displacements show how much difference targeting has made in the estimates.
Example: Some low achievers have guessed wildly on a MCQ test. You want to reduce the effect of their lucky
guesses on their measures and on item calibrations.
TARGET=Y
How Targeting works:
a) for each observation:
calculate probability of each category (0,1 for dichotomies)
calculate expected score (= probability of 1 for dichotomy)
calculate variance = information
= probability of 1 * probability of 0 for dichotomies,
so maximum value is 0.25 when person ability measure = item difficulty measure
b) for targeting:
weighted observation = variance * observation
weighted expected score = variance * expected score
c) sum these across persons and items (and structures)
d) required &quot;targeted&quot; estimates are obtained when, for each person, item, structure sum (weighted
observations) = sum (weighted expected scores)
e) for calculation of fit statistics and displacement, weights of 1 are used but with the targeted parameter
estimates. Displacement size and excessive misfit indicate how much &quot;off-target&quot; aberrant behavior exists in the
data.
For targeting, there are many patterns of responses that can cause infinite measures, e.g. all items correct except
for the easiest one. The convergence criteria limit how extreme the reported measures will be.
192.
TFILE input file listing tables to be output
Omit TFILE= and use the pull-down menu.
TFILE= causes the specified Tables to be written into the report output file. It is has no effect on the pull-down
menus.
171
TABLES= selects the tables in a fixed sequence, and prints only one copy. TFILE= allows the analyst to print
multiple copies of tables to different specifications. TFILE= specifies the name of an input ASCII file. Each line of
the file contains a table number or table number.sub-table and other control parameters, separated by blanks or
commas. Unused control values are specified with &quot;-&quot;. The list may be entered directly into the control file with
TFILE=* (see Example 2).
TFILE= Parameters: (enter unused parameters with &quot;-&quot;)
1: Table number. subtable
2
3
4
5
6
7
8
Distribution map 1.0, 1.1
Lowest measure Highest measure
Rows per unit Persons per `#'
Items per `#'
Response plot 2, 7.2 Lowest measure Highest measure
Columns per marked division
Reference
category for sorting High rating adjustment Low rating adjustment Unexpected only
Person fit plots 4, 5 Lowest measure Highest measure
Columns per marked division
Person/Item list: 6, 10, 13, 14, 15, 17, 18, 19
Low fit bar
High fit bar
Item fit plots 8, 9, 23
Lowest measure Highest measure
Columns per marked division
Item map 12, 1.2
Lowest measure Highest measure
Rows per unit Sort column within item name
Items per `#'
Item list alphabetical 15
Sort column within item name
Person map 16, 1.3
Lowest measure Highest measure
Rows per unit Sort column within person
name
Persons per `#'
Person list alphabetical 19
Sort column within person name
Score table 20 Columns per marked division
Category curves 21
Lowest measure Highest measure
Columns per marked division
Example 1: The analyst wishes to select and print several tables:
TFILE=TABLES.TF
TABLES.TF is a DOS (ASCII) file with the following lines:
; Table Low
High
Columns
; number Range Range per Unit
2
; print Tables 2.1, 2.2, 2.3
10.2
0.5 1.5
; print Table 10.2 with fit bars at 0.5 and 1.5
8
-5 5
; print Table 8 with range -5 to +5 logits
9
-2 7 10 ; range -2 to +7 logits, 10 columns per logit
9
-5 5 10 ; print Table 9 again, different range
15
- - - 4 ; print Table 15, sorted on column 4 of item name
or enter directly into the control file,
TFILE=*
2
10.2
8 -5 5
9 -2 7 10
9 -5 5 10
15 - - - 4
*
Example 2: Analyst wishes to specify on the DOS control line, Table 15 sorted on item name column 4. Values
are separated by commas, because blanks act as end-of-line separators.
C:&gt;WINSTEPS SF.TXT SFO.TXT TFILE=* 15,-,-,-,4 *
193.
TITLE title for output listing
Use this option to label output distinctly and uniquely.
Up to 60 characters of title. This title will be printed at the top of each page of output.
Example:
You want the title to be: Analysis of Math Test
TITLE=&quot;Analysis of Math Test&quot;
Quote marks &quot; &quot; or ' ' are required if the title contains any blanks.
172
194.
TOTALSCORE show total scores with extreme observations
TOTALSCORE=N, the standard
Winsteps uses an adjusted raw score for estimation, from which observations that form part of extreme
scores have been dropped. This is displayed in Table 13, PFILE=, IFILE=, etc.
TOTALSCORE=Y
The total raw score from the data file, after any recoding and weighting, is shown. This usually matches
the numbers used in raw-score analysis.
This can be changed from the Specification pull-down menu.
Example: KCT.txt with TOTALSCORE=N, the standard. This shows scores on the 15 measurable items.
+----------------------------------------------------------------------+----------+
|ENTRY
RAW
|
INFIT | OUTFIT |SCORE|
|
|
|NUMBER SCORE COUNT MEASURE ERROR|MNSQ ZSTD|MNSQ ZSTD|CORR.|WEIGH| KID
|
|------------------------------------+----------+----------+-----+-----+----------|
|
24
12
15
3.50
.91|1.81
1.4| .79
.0| .52| 1.00| Rick
M|
|
7
11
15
2.68
.90|1.64
1.1|1.49
.1| .59| 1.00| Susan
F|
|
15
11
15
2.68
.90| .35 -1.7| .15
-.4| .76| 1.00| Frank
M|
With TOTALSCORE=Y. This shows scores on all 18 items.
+----------------------------------------------------------------------+----------+
|ENTRY
TOTAL
|
INFIT | OUTFIT |SCORE|
|
|
|NUMBER SCORE COUNT MEASURE ERROR|MNSQ ZSTD|MNSQ ZSTD|CORR.|WEIGH| KID
|
|------------------------------------+----------+----------+-----+-----+----------|
|
24
15
18
3.50
.91|1.81
1.4| .79
.0| .52| 1.00| Rick
M|
|
7
14
18
2.68
.90|1.64
1.1|1.49
.1| .59| 1.00| Susan
F|
|
15
14
18
2.68
.90| .35 -1.7| .15
-.4| .76| 1.00| Frank
M|
195.
UANCHOR anchor values supplied in user-scaled units
This simplifies conversion from previously computed logit measures to user-scaled measures.
UANCHOR=A or N or L specifies that the anchor values are in UASCALE= units per logit. Reported measures,
however, will be user-rescaled by UMEAN= (or UIMEAN= or UPMEAN=) and USCALE=.
UANCHOR=Y specifies that anchor values are in USCALE= units per logit.
If UASCALE&lt;&gt;1.0 then UANCHOR=A is forced.
Example 1: Your item bank calibrations are user-scaled with 10 units per logits, but you want to report person
measures in CHIPS (BTD p.201):
UASCALE=10
; user-scaling of anchor values
UANCHOR=A
UMEAN=50
; user-scaling of reported values
USCALE=4.55
Example 2: Your previous test was in logits, but now you want the sample mean to be 500, with user-scaling 100
per logit.
UASCALE=1
; user-scaling of anchor values: logits
UANCHOR=Logits
UPMEAN=500
USCALE=100
; user-scaling of reported values
173
196.
UASCALE the anchor user-scale value of 1 logit
Specifies the number of units per logit of the anchor values
If UASCALE=1, then specify UANCHOR=A to confirm that this is the anchor user-scaling.
Example 1: The anchor values are user-scaled such that 1 logit is 45.5 units, so that differences of -100, -50, 0,
+50, +100 correspond to success rates of 10%, 25%, 50%, 75%, 90%:
UASCALE = 45.5
Example 2: The anchor values are on one scaling, but you want the reported values to be on another scaling.
UASCALE= 5
; there are 5 units per logit in the scaled values
USCALE=10
; you want 10 units per logit on the reported values
UPMEAN=50
; you want the reported person mean to be 50
197.
UCOUNT number of unexpected responses: Tables 6, 10
This sets the maximum number of &quot;most unexpected responses&quot; to report in Tables 6.6, 10.6. Also the maximum
number of persons and items to report in the anti-Guttman matrices in Tables 6.4, 6.5, 10.4, 10.5
MOST UNEXPECTED RESPONSES
+-------------------------------------------------------------------------------------------+
| DATA |OBSERVED|EXPECTED|RESIDUAL|ST. RES.| ACT
| KID
| ACT
| KID
|
|------+--------+--------+--------+--------+-------+--------+-------------+-----------------|
|
0 |
0 |
1.93 | -1.93 | -7.66 |
18 |
73 | GO ON PICNIC| SANDBERG, RYNE |
|
2 |
2 |
.07 |
1.93 |
7.57 |
23 |
72 | WATCH A RAT | JACKSON, SOLOMON|
|
2 |
2 |
.07 |
1.93 |
7.57 |
23 |
29 | WATCH A RAT | LANDMAN, ALAN
|
|
0 |
0 |
1.93 | -1.93 | -7.40 |
19 |
71 | GO TO ZOO
| STOLLER, DAVE
|
Example: Show 100 &quot;Most Unexpected Responses&quot;:
UCOUNT=100
198.
UDECIMALS number of decimal places reported
This is useful for presenting your output measures and calibrations in a clear manner by removing meaningless
decimal places from the output. Range is 0 (12345.) to 4 (1.2345).
How small is meaningless? Look at the Standard Error columns. Any value clearly less than a standard error has
little statistical meaning.
Use the &quot;Specification&quot; pull-down menu to alter the value of UDECIMALS= for individual reports.
Example 1: You want to report measures and calibrations to the nearest integer:
UDECIMALS = 0
Example 2: You want to report measures and calibrations to 4 decimal places because of a highly precise, though
arbitrary, pass-fail criterion level:
UDECIMALS = 4
199.
UIMEAN the mean or center of the item difficulties
Assigns your chosen numerical value to the average measure of the non-extreme items, i.e., a criterionreferenced measure. Previous UPMEAN= values are ignored.
UMEAN= and UIMEAN= are the same specification. Anchor values are treated according to UANCHOR=
Table 20 gives the UMEAN= and USCALE= values for a conversion that gives the measures a range of 0-100.
Example 1: You want to recenter the item measures at 10 logits, and so add 10 logits to all reported measures, to
avoid reporting negative measures for low achievers:
UIMEAN = 10
174
Example 2: You want to recenter and user-rescale the item measures, so that the range of observable measures
goes from 0 to 100.
Look at Table 20.1. Beneath the Table are shown the requisite values, e.g.,
UMEAN = 48.3 ; this is the same as UIMEAN=48.3
USCALE = 9.7
For more examples, and how to compute this by hand, see User-friendly rescaling
200.
UPMEAN the mean or center of the person abilities
Assigns your chosen numerical value to the average of the non-extreme abilities for persons, i.e., this provides a
norm-referenced user-scaling. Previous UIMEAN= values are ignored. Anchor values are treated according to
UANCHOR=
Example 1: You want to used conventional IRT norm-referenced user-scaling with person mean of 0 and person
S.D. of 1.
UPMEAN = 0
USCALE = 1 / (person S.D. in logits) ; find this from Table 3.1
If there are extreme person scores, see User-friendly rescaling
Example 2: I want to compare the mean performance of random samples of examinees from my database. Will
UPMEAN= help in this?
UPMEAN=0 sets the local mean of the persons to zero (excluding extreme scores) regardless of the
sampling.
If you wish to investigate the behavior of the person mean for different person samples, then
(1) analyze all persons and items: set UPMEAN=0, for convenience, and write IFILE=. For better
comparison, set STBIAS=NO.
(2) anchor the items using IAFILE=
(3) analyze samples of persons with the anchored items.
The person means reported in Table 3.1 now show person means (with or without extreme scores) in the
one frame of reference across all analyses defined by the anchored items.
For more examples, and how to compute this by hand, see User-friendly rescaling
201.
USCALE the user-scaled value of 1 logit
Specifies the number of reported user-scaled units per logit. When USCALE=1 (or USCALE= is omitted) then all
measures are in logits.
Table 20 gives the UMEAN= and USCALE= values for a conversion that gives the measures a range of 0-100.
Example 1: You want to user-rescale 1 logit into 45.5 units, so that differences of -100, -50, 0, +50, +100
correspond to success rates of 10%, 25%, 50%, 75%, 90%:
USCALE = 45.5
Example 2: You want to reverse the measurement directions, since the data matrix is transposed so that the
`items' are examinees and the `persons' are test questions:
USCALE = -1
KEYn=, RESCORE=, ISGROUPS= will still apply to the columns, not the rows, of the data matrix.
Centering will still be on the column measures.
Example 3: You want to approximate the &quot;probit&quot; measures used in many statistical procedures.
UPMEAN = 0
; set the person sample mean to zero
USCALE = 0.59
; probits * 1.7 = logits
Example 4: You want to report measures in &quot;log-odds units to the base 10&quot; instead of the standard &quot;logits to the
base e&quot;.
USCALE=0.434294482 ; the conversion between &quot;natural&quot; and base-10 logarithms.
175
For more examples, and how to compute this by hand, see User-friendly rescaling
202.
W300 Output files in Winsteps 3.00 format
New or revised statistics are added as extra or changed columns in IFILE= and PFILE= when they are introduced
into Winsteps.
To revert to an earlier format of IFILE= and PFILE=, specify W300=Yes. This produces the PFILE= and IFILE= in
the format of Winsteps 3.00 1/1/2000.
Example: IFILE= in current format:
;ACT
FILE FOR
;ENTRY MEASURE STTS COUNT
NAME
1
-.89 1
75.0
WATCH BIRDS
2
-.61 1
75.0
READ BOOKS ON ANIMALS
SCORE
ERROR IN.MSQ IN.ZSTD OUT.MS OUT.ZSTD DISPL PTME WEIGHT
DISCR G M
109.0
.23
.74
-1.97
.67
-1.89
.00
.64
1.00
1.06 0 R
116.0
.20
.76
-1.54
.56
-1.55
.00
.58
1.00
1.07 0 R
IFILE= in W300=Yes format:
;ACT
FILE FOR
;ENTRY MEASURE ST COUNT
1
-.89 1
75
2
-.61 1
75
ANIMALS
SCORE
109
116
ERROR IN.MSQ IN.ZSTD OUT.MS OUT.ZSTD DISPL CORR G M NAME
.23
.74 -1.97
.67 -1.89
.00
.64 0 R WATCH BIRDS
.20
.76 -1.54
.56 -1.55
.00
.58 0 R READ BOOKS ON
PFILE= in current format:
;PUPIL
FILE FOR
;ENTRY MEASURE STTS COUNT
1
.49 1
25.0
ROSSNER, MARC DANIEL
;
2
5.99 0
25.0
ROSSNER, LAWRENCE F.
SCORE
30.0
50.0
ERROR IN.MSQ IN.ZSTD OUT.MS OUT.ZSTD DISPL PTME WEIGHT
.35
.96
-.15
.84
-.43
.00
.69
1.00
1.84
1.00
.00
1.00
.00
.00
.00
DISCR NAME
.00
1.00
.00
PFILE= in W300=Yes format:
;PUPIL FILE FOR
;ENTRY MEASURE ST COUNT
1
.49 1
25
2
5.99 0
25
SCORE
30
50
ERROR IN.MSQ IN.ZSTD OUT.MS OUT.ZSTD DISPL CORR NAME
.35
.96
-.15
.84
-.43
.00
.69 ROSSNER, MARC DANIEL
1.84
1.00
.00
1.00
.00
.00
.00 ROSSNER, LAWRENCE F.
Notes:
TOTAL=YES is active for both current and old formats.
&quot;;&quot; shown for extreme scores, such as person 2, in current format, but not in old format.
COUNT and SCORE are shown rounded to nearest integer in old format.
203.
WHEXACT Wilson-Hilferty exact normalization
Some versions of Winsteps have the standard WHEXACT=NO.
ZSTD INFIT is the &quot;t standardized Weighted Mean Square&quot; shown at the bottom of RSA p. 100. ZSTD
(standardized as a z-score) is used of a t-test result when either the t-test value has effectively infinite degrees of
freedom (i.e., approximates a unit normal value) or the Student's t-distribution value has been adjusted to a unit
normal value.
ZSTD OUTFIT is the &quot;t standardized Unweighted Mean Square&quot; based on the terms on RSA p. 100.
The Wilson-Hilferty transformation converts mean-square values to their equivalent &quot;t standardized&quot; normal
deviates. See RSA p. 101
176
Under certain circumstances, it can correctly report the paradoxical finding that the mean-square apparently
reports an overfit condition, but the normal deviate an underfit.
To allow this possibility, specify WHEXACT=Y
To suppress it, specify WHEXACT=N
The final q/3 term is omitted from the transformation.
Example:
A person takes a test of 20 dichotomous items and obtains an unweighted chi-square value of 19.5.
WHEXACT=Y
The OUTFIT mean-square is 0.975, i.e., apparently slightly overfitting. The exact normal deviate is .03,
i.e., very slightly underfitting.
WHEXACT=N
The OUTFIT mean-square is 0.975, i.e., apparently slightly overfitting. The reported normal deviate is .08, i.e., slightly overfitting.
204.
XFILE analyzed response file
If XFILE=filename is specified in the control file, a file is output which enables a detailed analysis of individual
response anomalies. This file contains 4 heading lines (unless HLINES=N) followed by one line for each personby-item response used in the estimation. Each line contains:
1. Person number (I7) (PERSON)
2. Item number (I7) (ITEM)
3. Original response value (after keying/scoring) (I4) (OBS)
4. Observed response value (after recounting) (I4) (ORD)
5. Expected response value (F7.3) (EXPECT)
6. modeled variance of observed values around the expected value (F7.3) (VARIAN)
This is also the statistical information in the observation.
Square root(modeled variance) is the observation's raw score standard deviation.
7. Standardized residual: (Observed - Expected)/Square root (Variance) (F7.3) (ZSCORE)
8. Score residual: (Observed - Expected) (F7.3) (RESIDL)
9. Person measure in USCALE= units (F7.2*) (PERMEA)
10. Item measure in USCALE= units (F7.2*) (ITMMEA)
11. Measure difference (Person measure - Item measure) in USCALE= units (F7.2*) (MEASDF)
12. Log-Probability of observed response (F7.3) (L-PROB)
13. Predicted person measure from this response alone in USCALE= units (F7.2*) (PPMEAS)
14. Predicted item measure from this response alone in USCALE= units (F7.2*) (PIMEAS)
15. Response code in data file (A) (CODE)
16. Person label (A) (PLABEL)
17. Item label (A) (ILABEL)
2* means decimal places set by UDECIM=.
Fields can be selected interactively, see below.
If CSV=Y, the values are separated by commas. When CSV=T, the commas are replaced by tab characters. For
&quot;non-numeric values in quotation marks&quot;, specify QUOTED=Y.
This file enables a detailed analysis of individual response anomalies. The response residual can be analyzed in
three forms:
1) in response-level score units, from [(observed value - expected value)].
2) in logits, from [(observed value - expected value)/variance].
3) in standard units, [(observed value - expected value)/(square root of variance)].
177
The log-probabilities can be summed to construct log-likelihood and chi-square tests. Asymptotically, &quot;chi-square
= -2*log-likelihood&quot;.
Predicted person measure: Imagine that this observation was the only observation made for the person ... this
value is the measure we would predict for that person given the item measure.
Predicted item measure: Imagine that this observation is the only observation made for this item ... this value is
the measure we would predict for that item given the person measure.
The formulas are the same as for a response string of more than 1 observation. For dichotomies, see
www.rasch.org/rmt/rmt102t.htm and for polytomies www.rasch.org/rmt/rmt122q.htm
Example: You wish to write a file on disk called &quot;MYDATA.XF&quot; containing response-level information for use in
examining particularly response patterns:
XFILE=MYDATA.XF
Example: You wish to compute differential item functioning, DIF, for a specific subset of people:
If Table 30 is not suitable, here is a simple approximation:
Since one item does not have enough information to measure a person, for item bias we have to do it on the
basis of a subset of people.
From the XFILE,
add the &quot;score residuals&quot; (not standardized) for everyone in subset &quot;A&quot; on a particular item.
Add the &quot;modelled variance&quot; for everyone in the subset.
Divide the residual sum by the variance sum. This gives an estimate of the DIF for subset &quot;A&quot; relative to
the grand mean measure.
Do the same for subset &quot;B&quot; on the same item.
To contrast subset &quot;A&quot; with subset &quot;B&quot; then
DIF size &quot;AB&quot; =DIF estimate for &quot;A&quot; - DIF estimate for &quot;B&quot;
A significance t-test is t =DIF size &quot;AB&quot; / square root ( 1/variance sum for subset A + 1/variance sum for
subset B))
When called from the &quot;Output Files&quot; menu, you can select what fields to include in the file. And you can also
specify particular persons and/or items. If this is too big for your screen see Display too big.
Only for Person (Item) nnn or range nnn-mmm: this allows selection of particular items and persons.
Example: &quot;4 11-13 6 20-23&quot; will list those persons in that order,
205.
XMLE consistent, almost unbiased, estimation
Experimental! This implements Linacre's (1989) XCON algorithm as XMLE &quot;Exclusory Maximum Likelihood
Estimation&quot;.
The reason that JMLE is statistically inconsistent under some conditions, and noticeably estimation-biased for
178
short tests or small samples, is that it includes the possibility of extreme scores in the estimation space, but
cannot actually estimate them. The XMLE algorithm essentially removes the possibility of extreme response
vectors from the estimation space. This makes XMLE consistent, and much less biased than JMLE. In fact it is
even less biased than CMLE for small samples, this is because CMLE only eliminates the possibility of extreme
person response vectors, not the possibility of extreme item response vectors.
Considerations with XMLE=YES include:
(1) Anchoring values changes the XMLE probabilities. Consequently, measures from, say, a Table 20 score table
do not match measures from the estimation run. Consequently, it may be necessary to estimate item calibrations
with XMLE=YES. Then anchor the items and perform XMLE=NO.
(2) Items and persons with extreme (zero and perfect) scores are deleted from the analysis.
(3) For particular data structures, measures for finite scores may not be calculable.
Selecting XMLE=YES, automatically makes STBIAS=NO and PAIRED=NO, because XMLE is a more powerful
bias correction technique..
Example: Produce XMLE estimates, to compare with JMLE estimates, and so investigate the size of the JMLE
estimation bias.
XMLE=YES
206.
XWIDE columns per response
The number of columns taken up by each response in your data file (1 or 2). If possible enter your data one
column per response. If there are two columns per response, make XWIDE=2. If your data includes responses
entered in both 1 and 2 character-width formats, use FORMAT= to convert all to XWIDE=2 format. When
XWIDE=2, these control variables require two columns per item or per response code: CODES=, KEYn=,
KEYSCR=, NEWSCORE=, IVALUE=. Either 1 or XWIDE= columns can be used for RESCORE=, ISGROUPS=,
RESCORE= and IREFER=
Example 1: The responses are scanned into adjacent columns in the data records,
XWIDE=1
Observations 1 column wide
Example 2: Each response is a rating on a rating scale from 1 to 10, and so requires two columns in the date
record,
XWIDE=2
2 columns per datum
Example 3: Some responses take one column, and some two columns in the data record. Five items of 1character width, code &quot;a&quot;, &quot;b&quot;, &quot;c&quot;, or &quot;d&quot;, then ten items of 2-character width, coded &quot;AA&quot;, &quot;BB&quot;, &quot;CC&quot;, &quot;DD&quot;.
These are preceded by person-id of 30 characters.
XWIDE=2
Format to two columns per response
FORMAT=(30A1,5A1,10A2)
Name 30 characters, 5 1-chars, 10 2-chars
CODES
=&quot;a b c d AABBCCDD&quot;
&quot;a&quot; becomes &quot;a &quot;
NEWSCORE=&quot;1 2 3 4 1 2 3 4 &quot;
response values
RESCORE=2
rescore all items
NAME1=1
person id starts in column 1
ITEM1=31
item responses start in column 31
NI=15
15 items all now XWIDE=2
207.
@Field name for location in label
@Fieldname= allows for user-defined names for locations with the person or item labels to be specified with the
column selection rules.
@Fieldname = value
Field name:
Value:
a user-specified name which can include letters and numbers, but not = signs. Field names are
converted to capital letters, and must be referenced in full.
a user-specified values which must accord with the column selection rules.
179
Example 1: The gender of persons is in column 14 of the person label. A DIF report on gender is wanted.
@GENDER = 14 ; gender indicator in column 14 of the person label
DIF = @GENDER ; DIF classification is by Gender column
or
DIF = @gender
; lower case letters in field names are allowed
but not
DIF = @GEN
; abbreviations of field names are not allowed
TFILE=*
30
; produce the DIF Table
*
This can also be done by the pull-down menus
Specification menu box: @GENDER = 14
Output Tables menu: 30. Items: DIF
Right-click on DIF selection box: @GENDER
Click on OK box
208.
&amp;END end of control variables
The first section in a control file contains the control variables, one per line. Its end is indicated by &amp;END.
TITLE = &quot;Example control file&quot;
ITEM1 = 1
NI = 10
NAME1 = 12
&amp;END
...... ; Item labels here
END LABELS
209.
&amp;INST start of control instructions
&amp;INST is ignored by current versions of Winsteps. It is maintained for backward compatibility with earlier
versions, where it was required to be the first control instruction. It is still present in some example files, again for
backwards compatibility.
&amp;INST
; this is allowed for compatibility
TITLE = &quot;Old control file&quot;
....
210.
The Iteration Screen
While WINSTEPS is running, information about the analysis is displayed on the screen. The iterative estimation
process is by logistic curve-fitting. Here is an example based on the &quot;Liking for Science&quot; data. The analysis was
initiated with:
C:&gt; WINSTEPS SF.TXT SFO.TXT(Enter)
The &quot;====&quot; is a horizontal bar-chart which moves from left to right to show progress through the work file during
each phase of the analysis.
The screen display includes:
WINSTEPS Version: 2.58
Reading Control Variables ..
Reading keys, groupings etc..
Input in process ..
Program running - shows version number
Processing your control variables
Processing special scoring instructions
Reading in your data: each . is 1,000 persons
1
2
3
4
5
6
7
1234567890123456789012345678901234567890123456789012345678901234567890
180
column in data record
---------------------------------------------------------------------01020101010002000102020202000201010202000201000200ROSSNER, MARC DANIEL
^I
^N^P
^I where item response string starts, ITEM1=
^N last item response, NI=
^P where person label starts, NAME1=
76 person records input
Writing response file
Total person records found
if RFILE= specified
To stop iterations: Press Ctrl with S
press Ctrl with F to cancel program
CONVERGENCE TABLE for more details see Table 0.2
Control: sf.TXT
Output: sfO.TXT
PROX
ACTIVE COUNT
EXTREME 5 RANGE
MAX LOGIT CHANGE
ITERATION
PERSONS ITEMS
CATS
PERSONS ITEMS
MEASURES STRUCTURES
1
76
25
3
2
74
25
3
DROPPING OUT OF RANGE OBSERVATIONS
This is reported for CUTLO= and CUTHI=
3
74
25
3
4
74
25
3
3.78
4.53
3.20
3.67
3.8918
.7628
.0740
-.6167
4.73
4.82
3.85
3.90
.2143
.0846
-.0991
-.0326
WARNING: DATA MAY BE AMBIGUOUSLY CONNECTED INTO 6 SUBSETS,
see Connection Ambiguities
Control: sf.TXT
Output: sfO.TXT
JMLE
MAX SCORE
MAX LOGIT
LEAST CONVERGED
ITERATION
RESIDUAL*
CHANGE
PERSON ITEM
CAT
1
3.01
-.4155
60
24*
2
2
.50
-.0258
53
24*
1
3
-.37
.0292
53
5*
1
4
.26
.0206
53
21*
1
5
.20
.0154
53
21*
0
6
.15
.0113
53
21*
0
7
.11
.0083
53
21*
0
Check that values diminish, and are near 0.
Calculating Fit Statistics
CATEGORY STRUCTURE
RESIDUAL
CHANGE
27.64
-.0184
6.88
.0198
3.10
.0091
2.74
.0079
-1.90
.0056
-1.42
.0042
-1.05
.0030
an extra pass to calculate fit statistics
If Tables= is specified:
Processing Misfitting PERSONS for Table 7
Calculating Correlations for Table 10
Calculating Principal Components for Table 10
...............*............*.* one . per iteration, one * per contrast.
Processing Misfitting ITEMS for Table 11
Sorting ITEMS
for Table 12
Sorting ITEMS
for Table 15
Sorting PERSONS for Table 16
Sorting PERSONS for Table 19
Calculating Scores for Table 20
Writing Sorted Responses in Table 22
Guttman Scalogram
Analysis completed of SF.TXT
name of control file
LIKING FOR SCIENCE-(Wright-&amp;-Masters-p.18)--------------+-----------------------------------------------------------------------------+
|
76 KIDS
IN
74 KIDS
MEASURED
INFIT
OUTFIT
|
|
SCORE
COUNT
MEASURE
ERROR
IMNSQ
ZSTD OMNSQ
ZSTD |
| MEAN
26.4
16.8
56.99
5.74
1.01
-.2
.82
-.3 |
| S.D.
11.9
5.7
23.67
1.33
.65
1.4
.78
1.2 |
| REAL RMSE
5.90 ADJ.SD
22.93 SEPARATION 3.89 KID
RELIABILITY .94 |
|-----------------------------------------------------------------------------|
|
25 TAPS
IN
25 TAPS
MEASURED
INFIT
OUTFIT
|
| MEAN
78.2
49.7
50.00
3.48
1.06
.0
.89
-.1 |
| S.D.
43.0
22.5
27.56
1.04
.36
1.3
.43
.7 |
| REAL RMSE
3.63 ADJ.SD
27.32 SEPARATION 7.53 TAP
RELIABILITY .98 |
+-----------------------------------------------------------------------------+
181
|
|
|
|
|
|
|
|
|
|
NUM SCORE COUNT MEASURE ERROR|IMNSQ ZSTD|OMNSQ ZSTD|CORR.| KID
73
15
14
33.18
4.75|3.46
4.5|4.52
5.4|A-.16| SANDBERG, RYN
71
24
19
47.29
4.32|3.25
4.8|5.19
5.1|B .14| STOLLER, DAVE
14
15
11
37.24
5.48|2.09
2.1|1.77
1.3|C .45| HWA, NANCY MA
32
21
13
50.17
5.77|1.14
.3|2.04
1.2|D .13| ROSSNER, JACK
NUM SCORE COUNT MEASURE ERROR|IMNSQ ZSTD|OMNSQ ZSTD|CORR.| TAPS
23
8
10
103.87
5.68|2.06
2.1|2.18
2.3|A .19| WATCH A RAT
9
49
37
62.83
3.12|1.67
2.5|1.81
1.4|B .47| LEARN WEED NA
16
53
40
60.93
3.01|1.73
2.8|1.54
1.0|C .38| MAKE A MAP
7
40
29
66.62
3.51|1.20
.7|1.31
.6|D .47| WATCH ANIMAL
Summary statistics from Table 3 and the largest misfitting persons and items
Output written to SFO.TXT
name of output file
For details of the numbers, see the description of output Tables 0, 3, 10 and 14.
211.
Table heading
At the top of each output Table is basic information about the analysis:
TABLE 1.0 LIKING FOR SCIENCE (Wright &amp; Masters p. ZOU214ws.txt Feb 1 14:38 2005
INPUT: 75 KIDS, 25 ACTS MEASURED: 75 KIDS, 25 ACTS, 3 CATS
WINSTEPS 3.55.2
-------------------------------------------------------------------------------TABLE 1.0 is identifies the current Table and sub-table.
LIKING FOR SCIENCE (Wright &amp; Masters is set by TITLE=
ZOU214ws.txt is the name of the disk file containing this Table.
Feb 1 14:38 2005 is the date and time of this analysis.
INPUT: 75 KIDS, 25 ACTS
75 KIDS gives the number of cases in the data file(s) and the case (row) identification PERSON=
25 ACTS gives the number of items specified by NI= and the column (item) identification ITEM=
MEASURED: 75 KIDS, 25 ACTS, 3 CATS
shows how many rows, columns and categories are reported as measurable, including extreme scores.
The number of categories is determined by ISGROUPS= and the data structure. For details, see Table
3.1 and Table 3.2
Previously, ANALYZED was used to mean &quot;used for item analysis&quot;, omitting extreme scores.
WINSTEPS 3.55.2 is the program name and version number performing this analysis.
212.
Table 1.0, 1.2, 1.3, 1.10, 1.12 Distribution maps
(controlled by MRANGE=, MAXPAG=, NAMLMP=, ISORT=, PSORT=)
Table 1.2 is printed if the item map can be squeezed into one page. Table 1.3 is printed if the person map can be
squeezed into one page. If person and item maps can be squeezed into one page, Table 1.0 is printed. You can
use NAMLMP= to limit the number of characters of each name reported.
&quot;TSMST&quot; summarize the left-hard and right-hand distributions. An &quot;M&quot; marker represents the location of the
mean measure. &quot;S&quot; markers are placed one sample standard deviation away from the mean. &quot;T&quot; markers are
placed two sample standard deviations away.
MAXPAG= controls the length of the Table. MRANGE= controls the displayed range of measures. ISORT= and
PSORT= control the sort order within rows.
In subtables 10 and above, the items are arranged by easiness. The item hierarchy is reversed.
Items arranged by measure: Look for the hierarchy of item names to spell out a meaningful construct from
182
easiest at the bottom to hardest at the top.
3
2
1
0
-1
-2
-3
PUPILS MAP OF ACTS
&lt;more&gt;|&lt;rare&gt;
X +
|
X |T
XX |
XX | FIND BOTTLES AND
XX S| WATCH A RAT
XXX +
XX | WATCH BUGS
XX | LOOK IN SIDEWALK
XX |
XX |S
XXXXXX |
XXX + WATCH ANIMAL MOVE
XXX M|
XXXXX | LEARN WEED NAMES
XXX | TALK W/FRIENDS AB
XXXXXXXXXXX | LOOK AT PICTURES
XXXX | WATCH WHAT ANIMAL
XXXXXXX +M FIND OUT WHAT ANI
X |
S| FIND OUT WHAT FLO
X | READ ANIMAL STORI
XX | READ BOOKS ON ANI
X | WATCH BIRD MAKE N
X + FIND WHERE ANIMAL
|
XX |S GROW GARDEN
T| LISTEN TO BIRD SI
X |
|
+ GO TO MUSEUM
|
|
| GO TO ZOO
|T
|
+ GO ON PICNIC
&lt;less&gt;|&lt;frequ&gt;
WATCH GRASS CHANG
MAKE A MAP
LOOK UP STRANGE A
READ BOOKS ON PLA
WATCH BIRDS
Subtables .10 and above:
Items arranged by easiness: Look for the hierarchy of item names to spell out a meaningful construct from
easiest at the top to hardest at the bottom.
The double line || indicates the two sides have opposite orientations. This is useful if the items and persons are
being compared to the response structures.
3
2
1
0
PUPILS MAP OF ACTS
&lt;more&gt;||&lt;frequ&gt;
X ++ GO ON PICNIC
||
X ||T
XX || GO TO ZOO
XX ||
XX S||
XXX ++ GO TO MUSEUM
XX ||
XX ||
XX || LISTEN TO BIRD SI
XX ||S GROW GARDEN
XXXXXX ||
XXX ++ FIND WHERE ANIMAL
XXX M|| WATCH BIRD MAKE N
XXXX || READ BOOKS ON ANI
XXXX || READ ANIMAL STORI
XXXXXXXXXXX || FIND OUT WHAT FLO
XXXX ||
XXXXXXX ++M FIND OUT WHAT ANI
X || WATCH WHAT ANIMAL
S|| LOOK AT PICTURES
X || MAKE A MAP
WATCH BIRDS
LOOK UP STRANGE A
TALK W/FRIENDS AB
183
READ BOOKS ON PLA
-1
-2
-3
213.
XX
X
X
|| LEARN WEED NAMES
||
++ WATCH ANIMAL MOVE
||
XX ||S
T||
X || LOOK IN SIDEWALK
|| WATCH BUGS
++
|| WATCH A RAT
|| FIND BOTTLES AND
||
||T
||
++
&lt;less&gt;||&lt;rare&gt;
WATCH GRASS CHANG
Table 1.1, 1.4 Distribution maps
(controlled by MRANGE=, ITEM=, PERSON=, MAXPAG= )
These tables show the distribution of the persons and items. The variable is laid out vertically with the most able
persons, and most difficult items at the top.
TABLE 1.1 LIKING FOR SCIENCE (Wright &amp; Masters p.18)
SFO.TXT Dec 1 18:33 1996
76 PUPILS
25 ACTS
ANALYZED:
74 PUPILS
25 ACTS
3 CATEGS v2.67
76 PUPILS 25 ACTS Number of persons in data file and items in NI= specification.
ANALYZED: 74 PUPILS 25 ACTS 3 CATEGS Number of persons, items and categories that with non-extreme
scores.
v2.67
Winsteps version number.
MAP OF KIDS
AND TAPS
MEASURE
|
MEASURE
&lt;more&gt; --------------------- KIDS
-+- TAPS
--------------------- &lt;rare&gt;
5.0
+ X
5.0
|
|
| XXX (items too hard for persons)
|
|
4.0
+
4.0
|
XX |
|
|
| X
3.0
+
3.0
X |
|
|
|
| X
2.0
XXXX +
2.0
| X
|
|
|
|
1.0
XXXXX +
1.0
| X
|
|
|
|
.0
+
.0
|
XXXXXXXXXXXX | (gap!)
|
|
|
-1.0
+
-1.0
184
-2.0
-3.0
&lt;less&gt; ---------------------
|
XXX |
| X
|
|
+
XX | X
|
|
X | X
| XX
XX + X
KIDS
-+- TAPS
-2.0
-3.0
------------------&lt;frequent&gt;
In Table 1, each person or item is indicated by an &quot;X&quot;, or, when there are too many &quot;X&quot;s to display on one line,
several persons or items are represented by a &quot;#&quot;. Less than that number by &quot;.&quot;. So that if &quot;#&quot; represents 4, then
&quot;.&quot; represents 1 to 3.
The left-hand column locates the person ability measures along the variable. For dichotomous items, the righthand column locates the item difficulty measures along the variable. Look for an even spread of items along the
variable (the Y-axis) with no gaps, indicating poorly defined or tested regions of the variable. The persons often
show a normal distribution. Good tests usually have the items targeted (lined up with) the persons.
For rating (or partial credit) scales, each item is shown three times in Table 1.4. In the center item column, each
item is placed at its mean calibration, i. e., this is the location of the center of the rating (or partial credit) scale the location at which being ratings in the top and bottom category are equally probable. In the left-hand item
column, the item is shown at the measure level corresponding to a probability of .5 of exceeding (or being rated
in) the bottom rating (or partial credit) scale category. In the right-hand item column, the item is shown at the
measure level corresponding to a probability of .5 of being rated in (or falling below) the top rating (or partial
credit) scale category. These locations are also shown in Table 2.3. Dichotomous items, &quot;D&quot;, have only one
location.
MAP OF PUPILS AND ACTS
MEASURE
|
P=50%|
|
&lt;more&gt; ------PUPILS -+-ACTS
BOTTOM+-ACTS
CENTER+-ACTS
4.0
X +
+
+
|
|
|
|
|
|
X |
|
|
|
|
| X
X |
|
| X
3.0
+
+
+
X |
|
| X
X |
|
| XX
XX |
|
|
XX |
| X
|
XX |
| X
|
2.0
XXX +
+
+ X
XX |
| X
|
XX |
| XX
| X
XX |
|
| XX
XX | X
|
| XXX
XXX | X
|
| X
1.0
XXXXXX +
+ X
+ X
XXX | X
|
|
XXXXX | XX
| X
| XX
XXX |
| XX
| X
XXXXXXXXXXX |
| XXX
| X
XXXX |
| X
| X
.0
XXXXXXX + X
+ X
+ X
X |
|
|
| X
| XX
| X
X | XX
| X
| X
XX | XXX
| X
|
X | X
| X
|
-1.0
X + X
+ X
+ X
|
|
|
XX | XX
| X
|
| X
| X
| X
X | X
|
|
185
P=50% MEASURE
TOP -- &lt;rare&gt;
4.0
3.0
2.0
1.0
.0
-1.0
| X
+ X
|
| X
| X
|
|
-3.0
+ X
|
|
| X
|
|
-4.0
+ X
&lt;less&gt; ------PUPILS -+-ACTS
-2.0
|
+ X
|
|
| X
|
|
+ X
|
|
|
|
|
+
BOTTOM+-ACTS
|
+ X
|
|
|
|
|
+
|
|
|
|
|
+
CENTER+-ACTS
-2.0
-3.0
-4.0
TOP -- &lt;frequ&gt;
Observe that the top pupil (left column) is well above the top category of the most difficult act item (right-most
column), but that all pupils are above the top category of the easiest item (bottom X in right-most column).
&quot;Above&quot; means with greater than 50% chance of exceeding.
214.
Table 2 Multiple-choice distractor plot
Here are the Tables produced by selecting &quot;2.0 Measure Forms (All)&quot; on the Output Tables pull-down menu. They
reflect different conceptualizations of the category structure.
The codes for the response options (distractors) are located according to the measures corresponding to them.
Each subtable is presented two ways: with the response code itself (or one of them if several would be in the
same place), e.g., Table 2.1, and with the score corresponding to the option, e.g. Table 2.11 (numbered 10
subtables higher).
Table 2.1: shows the most probable response on the latent variable. In this example, for item &quot;al07&quot;, &quot;a&quot; (or any
other incorrect option) is most probable up to 3.2 logits, when &quot;d&quot;, the correct response, becomes most probable
according to the Rasch model.
TABLE 2.1: MOST PROBABLE RESPONSE: MODE (BETWEEN &quot;0&quot; AND &quot;1&quot; IS &quot;0&quot;, ETC.) (ILLUSTRATED BY
AN OBSERVED CATEGORY)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
a
d
d
55 al07 newspaper
a
c
c
64 sa01 magazine
......
b
a
a
12 nm07 sign on wall
a
d
d
10 nm05 public place
|------+------+------+------+------+------+------+------| NUM
TOPIC
-4
-3
-2
-1
0
1
2
3
4
1
T
11 1111 111 212 3 2 12
S
M
12 1
S
1
2
T
STUDENTS
M = Mean, the average of the person measures, S = one Standard Deviation from the mean, T = Two S.D.s. from
the mean
Table 2.11 is the same as Table 1, but the options are shown by their scored values, not by their codes in the
data.
TABLE 2.11: MOST PROBABLE RESPONSE: MODE (BETWEEN &quot;0&quot; AND &quot;1&quot; IS &quot;0&quot;, ETC.) (BY CATEGORY
SCORE)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
0
1
1
55 al07 newspaper
0
1
1
64 sa01 magazine
Table 2.2: shows the predicted average response on the latent variable. In this example, for item &quot;al07&quot;, &quot;a&quot; (or
any other incorrect option) is the predicted average response up to 3.2 logits, then &quot;d&quot;, the correct response,
becomes the average predictions. The &quot;:&quot; is at the transition from an average expected wrong response to an
186
average expected &quot;right&quot; response, i.e., where the predicted average score on the item is 0.5, the Rasch-halfpoint thresholds. The &quot;a&quot; below &quot;2&quot; is positions where the expected average score on the item is 0.25. Similarly
&quot;d&quot; would be repeated where the expected average score on the item is 0.75, according to the Rasch model.
TABLE 2.2 EXPECTED SCORE: MEAN (&quot;:&quot; INDICATES HALF-POINT THRESHOLD) (ILLUSTRATED BY AN
OBSERVED CATEGORY)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
a
a
:
d
55 al07 newspaper
a
a
:
c
64 sa01 magazine
Table 2.12 is the same as Table 2, but the options are shown by their scored values, not by their codes in the
data.
TABLE 2.12 EXPECTED SCORE: MEAN (&quot;:&quot; INDICATES HALF-POINT THRESHOLD) (BY CATEGORY SCORE)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
0
0
:
1
55 al07 newspaper
0
0
:
1
64 sa01 magazine
Table 2.3 shows the 50% cumulative probability points, the Rasch-Thurstone thresholds. The lower category (&quot;a&quot;
and other wrong answers) has a greater than than 50% probability of being observed up to 3.2 logits, when &quot;d&quot;,
the correct answer, has a higher than 50% probability.
TABLE 2.3 50% CUMULATIVE PROBABILITY: MEDIAN (ILLUSTRATED BY AN OBSERVED CATEGORY)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
a
d
d
55 al07 newspaper
a
c
c
64 sa01 magazine
Table 2.13 is the same as Table 3, but the options are shown by their scored values, not by their codes in the
data.
TABLE 2.13 50% CUMULATIVE PROBABILITY: MEDIAN (BY CATEGORY SCORE)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
0
1
1
55 al07 newspaper
0
1
1
64 sa01 magazine
Table 2.4 shows the item difficulties (or more generally the Rasch-Andrich thresholds) coded by the option of the
higher category. For item &quot;al07&quot; this is &quot;d&quot;, the correct option.
TABLE 2.4 STRUCTURE MEASURES (Rasch model parameters: equal-adjacent-probability thresholds)
(ILLUSTRATED BY AN OBSERVED CATEGORY)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
|
d
|
55 al07 newspaper
|
c
|
64 sa01 magazine
Table 2.14 is the same as Table 4, the Rasch-Andrich thresholds, but the options are shown by their scored
values, not by their codes in the data.
TABLE 2.14 STRUCTURE MEASURES (Rasch model parameters: equal-adjacent-probability thresholds)
(BY CATEGORY SCORE)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
|
1
|
55 al07 newspaper
|
1
|
64 sa01 magazine
Table 2.5 shows the average measures of persons choosing wrong distractors (illustrated by one of the wrong
distractors, &quot;a&quot;) and the average measures or persons choosing a correct distractor (illustrated by one of the
correct distractors, &quot;d&quot;).
TABLE 2.5 OBSERVED AVERAGE MEASURES FOR STUDENTS (scored) (ILLUSTRATED BY AN OBSERVED
187
CATEGORY)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------|
|
a
d
|
a
c |
NUM
55
64
TOPIC
al07 newspaper
sa01 magazine
Table 2.15 is the same as Table 5, but the options are shown by their scored values, not by their codes in the
data.
TABLE 2.15 OBSERVED AVERAGE MEASURES FOR STUDENTS (scored) (BY CATEGORY SCORE)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
|
0
1
55 al07 newspaper
|
0
1 |
64 sa01 magazine
Table 2.6, shown first from the Diagnosis menu, shows the average measures of the persons choosing each
distractor. &quot;m&quot; usually indicates the average measure of persons with missing data.
TABLE 2.6 OBSERVED AVERAGE MEASURES FOR STUDENTS (unscored) (BY OBSERVED CATEGORY)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
|
m
ab
c
d
55 al07 newspaper
|
m
d
c |
64 sa01 magazine
Code for unidentified missing data: m
Table 2.16 is the same as Table 6, but the options are shown by their scored values, not by their codes in the
data.
TABLE 2.16 OBSERVED AVERAGE MEASURES FOR STUDENTS (unscored) (BY CATEGORY SCORE)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
|
m
00
0
1
55 al07 newspaper
|
m
0
1 |
64 sa01 magazine
Table 2.7 shows the measures that would be predicted to be observed for incorrect and correct responses if the
persons responded exactly as the Rasch model predicts. &quot;a&quot; (an incorrect distractor) shows the average measure
for persons in the sample who would be predicted to fail the item, and &quot;d&quot; (a correct distractor) shows the verage
measure for persons in the sample who would be predicted to succeed on the item.
TABLE 2.7 EXPECTED AVERAGE MEASURES FOR STUDENTS (scored) (ILLUSTRATED BY AN OBSERVED
CATEGORY)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
|
a
d|
55 al07 newspaper
|
a
c
64 sa01 magazine
Table 2.17 is the same as Table 7, but the options are shown by their scored values, not by their codes in the
data.
TABLE 2.17 EXPECTED AVERAGE MEASURES FOR STUDENTS (scored) (BY CATEGORY SCORE)
-4
-3
-2
-1
0
1
2
3
4
|------+------+------+------+------+------+------+------| NUM
TOPIC
|
0
1|
55 al07 newspaper
|
0
1
64 sa01 magazine
215.
Table 2 Most probable, expected, cumulative, structure, average measures
(controlled by MRANGE=, CATREF=, CURVES=)
Each plot answers a different question:
What category is most likely? The maximum probability (mode) plot.
What is the average or expected category value? The expected score (mean) plot.
What part of the variable corresponds to the category? The cumulative probability (median) plot.
188
The numeric information for these plots is in ISFILE=
Which Table should be used for a standard setting procedure?
Most standard setting is based on &quot;average&quot; or &quot;frequency&quot; considerations. For instance,
&quot;If we observed 1000 candidates whose measures are known to be exactly at the pass-fail point, ...
..., we would expect their average score to be the pass-fail score.&quot; If this is how you think, then the Table
you want is Table 2.2 (matches Table 12.5)
..., we would expect 50% to pass and 50% to fail the pass-fail score.&quot; If this is how you think, then the Table
you want is Table 2.3 (matches 12.6)
..., we would expect more to be in the criterion pass-fail category of each item than any other category.&quot; If
this is how you think, then the Table you want is Table 2.1 (no matching 12.)
&quot;Our current sample is definitive, ...
..., we would expect the next sample to behave in exactly the same way this sample did.&quot; If this is how you
think, then the Table you want is Table 2.5 (or Table 2.6, if the responses have been rescored.)
..., we would expect the next sample to behave the way this sample should have behaved, if this sample
had conformed to the Rasch model.&quot; If this is how you think, then the Table you want is Table 2.7.
The left-side of this table lists the items in descending order of measure. Anchored items are indicated by an *
between the sequence number and name. A particular category can be used as the reference for sorting the
items by specifying the CATREF= variable.
Across the bottom is the logit (or user-rescaled) variable with the distribution of the person measures shown
beneath it. An &quot;M&quot; marker represents the location of the mean person measure. &quot;S&quot; markers are placed one
sample standard deviation away from the mean. &quot;T&quot; markers are placed two sample standard deviations away.
An &quot;M&quot; inside a plot indicates the measure corresponding to missing data.
To produce all subtables of Table 2, request Table 2.0
Tables 2.1 &amp; 2.11: The &quot;Most Probable Response&quot; Table, selected with CURVES=001, answers the question
&quot;which category is a person of a particular measure most likely to choose?&quot; This is the most likely category with
which the persons of logit (or user-rescaled) measure shown below would respond to the item shown on the left.
The area to the extreme left is all &quot;0&quot;; the area to the extreme right is at the top category. Each category number
is shown to the left of its modal area. If a category is not shown, it is never a most likely response. An item with
an extreme, perfect or zero, score is not strictly estimable, and is omitted here. Blank lines are used to indicate
large gaps between items along the variable.
This table presents in one picture the results of this analysis in a form suitable for inference. We can predict for
people of any particular measure measure what responses they would probably make. &quot;M&quot; depicts an &quot;average&quot;
person. The left &quot;T&quot; a low performer. The right &quot;T&quot; a high performer. Look straight up from those letters to read
off the expected response profiles.
Table 2.1 to 2.7 reports with observed categories, i.e., those in the CODES= statement.
Table 2.11 to 2.17 report with scored categories, i.e., after IVALUE=, RESCORE=, KEY1=, etc., but only if
different from Table 2.1 to 2.7.
MOST PROBABLE RESPONSE: MODE (BETWEEN &quot;0&quot; AND &quot;1&quot; IS &quot;0&quot;, ETC.)
NUM ITEM
-5
-4
-3
-2
-1
0
1
2
3
4
5
------------------------------------------------------------5 FIND BOTTLES 0
1
2
2
20 WATCH BUGS
0
1
2
2
8 LOOK IN SIDE 0
1
2
2
7 WATCH ANIMAL 0
1
2
2
17 WATCH WHAT A 0
1
2
2
189
deliberate space
21 WATCH BIRD M
10 LISTEN TO BI
12 GO TO MUSEUM
18 GO ON PICNIC
|
|
0
1
2
2
0
1
2
2
0
1
2
2
0
1
2
2
-------------------------------------------------------------5
-4
-3
-2
-1
0
1
2
3
4
5
1
PERSON
1 2 1121 174135336222232221 11 1
T
S
M
S
T
1
11
Tables 2.2 &amp; 2.12: In the &quot;Expected Score&quot; Table, the standard output (or selected with CURVES=010)
answers the question &quot;what is the average rating that we expect to observer for persons of a particular measure?&quot;
This rating information is expressed in terms of expected scores (with &quot;:&quot; at the half-point thresholds). Extreme
scores are located at expected scores .25 score points away from the extremes.
EXPECTED SCORE:MEAN (&quot;:&quot; INDICATES HALF-POINT THRESHOLDS)
-5
-4
-3
-2
-1
0
1
2
3
4
5
------------------------------------------------------------5 FIND BOTTLES 0
:
1
:
2
23 WATCH A RAT 0
:
1
:
2
9 LEARN WEED N 0
:
1
:
2
21 WATCH BIRD M 0
:
1
:
2
11 FIND WHERE A 0
:
1
:
2
19 GO TO ZOO
0
:
1
:
2
18 GO ON PICNIC 0
:
1
:
2
-------------------------------------------------------------5
-4
-3
-2
-1
0
1
2
3
4
5
1
PERSON
1 2 1121 174135336222232221 11 1 1
11
T
S
M
S
T
Tables 2.3 &amp; 2.13: The &quot;Cumulative Probability&quot; Table: Rasch-Thurstone thresholds, selected with
CURVES=001, answers the question &quot;whereabouts in the category ordering is a person of a particular measure
located?&quot; This information is expressed in terms of median cumulative probabilities (the point at which the
probability of scoring in or above the category is .5).
5
23
9
21
11
19
18
CUMULATIVE PROBABILITY: MEDIAN
-6
-4
-2
0
2
4
6
------------------------------------------------------------FIND BOTTLES 0
1
2
2
WATCH A RAT 0
1
2
2
LEARN WEED N 0
1
2
2
WATCH BIRD M 0
1
2
2
FIND WHERE A 0
1
2
2
GO TO ZOO
0
1
2
2
GO ON PICNIC 0
1
2
2
-------------------------------------------------------------6
-4
-2
0
2
4
6
Tables 2.4 &amp; 2.14 show Rasch structure calibrations: Rasch-Andrich thresholds (step parameters, step
measures, step difficulties, rating (or partial credit) scale calibrations). These are the relationships between
adjacent categories, and correspond to the points where adjacent category probability curves cross, i.e., are
equally probable of being observed according to a Rasch model.
STRUCTURE MEASURES (Rasch Parameters)
-40 -30
-20
-10
0
10
20
30
40
50
|-----+-----+-----+-----+-----+-----+-----+-----+-----|
|
1
2
|
|
1
2
|
|
2
1
|
|
|
|
2
|
|
|
|
1
2
|
|
|
|
1
2
|
|
|
|
|
|
1 2
|
190
NUM
5
1
13
ACT
FIND BOTTLES AND CANS
WATCH BIRDS
GROW GARDEN
12
GO TO MUSEUM
11
FIND WHERE ANIMAL LIVES
10
LISTEN TO BIRD SING
19
GO TO ZOO
|
2
1
|
|-----+-----+-----+-----+-----+-----+-----+-----+-----|
-40 -30
-20
-10
0
10
20
30
40
50
18
NUM
GO ON PICNIC
ACT
Tables 2.5 &amp; 2.15 plot the observed average person measures for each scored category. It reflects how this
sample used these categories. The plotted values cannot fall outside the range of the sample.
OBSERVED AVERAGE MEASURES BY SCORED CATEGORY FOR PUPILS
-2
-1
0
1
2
3
4
5
|-------+-------+-------+-------+-------+-------+-------|
|
000102
|
|
|
|
000102
|
|
|
|
00 01
02
|
|
00
01
02
|
|
0001
02
|
|-------+-------+-------+-------+-------+-------+-------|
-2
-1
0
1
2
3
4
5
1 2
T
11 2 114346533233332222 322 2 1 1 1 1
S
M
S
T
1
11
NUM
5
23
20
4
8
NUM
ACT
FIND BOTTLES AND CANS
WATCH A RAT
WATCH BUGS
WATCH GRASS CHANGE
LOOK IN SIDEWALK CRACKS
ACT
PUPILS
Tables 2.6 &amp; Table 2.16 plot the observed average person measures for each observed category. It
reflects how this sample used these categories. The plotted values cannot fall outside the range of the sample.
&quot;m&quot; in the plot indicates the average measure of those for whom their observation is missing on this item. This
Table is shown first from the Diagnosis pull-down menu.
OBSERVED AVERAGE MEASURES BY OBSERVED CATEGORY FOR PUPILS
-2
-1
0
1
2
3
4
5
|-------+-------+-------+-------+-------+-------+-------|
|
000102
|
|
|
|
000102
m
|
|
|
|
00 01
02
|
|
00
01
02
|
|
0001
02
|
|-------+-------+-------+-------+-------+-------+-------|
-2
-1
0
1
2
3
4
5
Code for unidentified missing data: m
1 2
T
11 2 114346533233332222 322 2 1 1 1 1
S
M
S
T
1
11
NUM
5
23
20
4
8
NUM
ACT
FIND BOTTLES AND CANS
WATCH A RAT
WATCH BUGS
WATCH GRASS CHANGE
LOOK IN SIDEWALK CRACKS
ACT
PUPILS
Tables 2.7 &amp; 2.17 plot the expected average person measures for each category score. It reflects how this
sample were expected to use these categories. The plotted values cannot fall outside the range of the sample.
This Table applies the empirical person distribution to Table 2.2.
EXPECTED AVERAGE MEASURES BY CATEGORY FOR PUPILS
-2
-1
0
1
2
3
4
5
|-------+-------+-------+-------+-------+-------+-------|
|
00
01
02
|
|
|
|
00
01
02
|
|
|
|
00
01
02
|
|
00
01
02
|
|
00
01
02
|
|-------+-------+-------+-------+-------+-------+-------|
-2
-1
0
1
2
3
4
5
1 2
T
216.
11 2 114346533233332222 322 2 1 1 1 1
S
M
S
T
1
11
NUM
5
23
20
4
8
NUM
ACT
FIND BOTTLES AND CANS
WATCH A RAT
WATCH BUGS
WATCH GRASS CHANGE
LOOK IN SIDEWALK CRACKS
ACT
PUPILS
Table 3.1, 27.3, 28.3 Summaries of persons and items
(controlled by REALSE=, UMEAN=, USCALE=, ISUBTOTAL=, PSUBTOTAL=)
This table summarizes the person, item and structure information. Extreme scores (zero and perfect scores)
have no exact measure under Rasch model conditions, so they are dropped from the main summary statistics.
191
Using a Bayesian technique, however, reasonable measures are reported for each extreme score, see
EXTRSC=. Totals including extreme scores are also reported, but are necessarily less inferentially secure than
those totals only for non-extreme scores.
Table 3: Gives summaries for all persons and items.
Table 27: Gives subtotal summaries for items, controlled by ISUBTOT=
Table 28: Gives subtotal summaries for persons, controlled by PSUBTOT=
SUMMARY OF
34 MEASURED (NON-EXTREME) KIDS
+-----------------------------------------------------------------------------+
|
RAW
MODEL
INFIT
OUTFIT
|
|
SCORE
COUNT
MEASURE
ERROR
MNSQ
ZSTD
MNSQ
ZSTD |
|-----------------------------------------------------------------------------|
| MEAN
6.9
14.0
-.20
1.03
1.03
-.3
.73
-.3 |
| S.D.
2.1
.0
2.07
.11
1.01
1.2
1.45
.5 |
| MAX.
11.0
14.0
3.89
1.15
4.43
2.3
6.86
1.3 |
| MIN.
2.0
14.0
-4.48
.82
.17
-1.6
.08
-.8 |
|-----------------------------------------------------------------------------|
| REAL RMSE
1.23 ADJ.SD
1.66 SEPARATION 1.35 KID
RELIABILITY .65 |
|MODEL RMSE
1.03 ADJ.SD
1.79 SEPARATION 1.73 KID
RELIABILITY .75 |
| S.E. OF KID MEAN = .36
|
+-----------------------------------------------------------------------------+
MINIMUM EXTREME SCORE: 1 KIDS
MINIMUM EXTREME SCORE: 46 PUPILS
LACKING RESPONSES: 8 PUPILS
SUMMARY OF
35 MEASURED(EXTREME AND NON-EXTREME)KIDS
+-----------------------------------------------------------------------------+
|
RAW
MODEL
INFIT
OUTFIT
|
|
SCORE
COUNT
MEASURE
ERROR
MNSQ
ZSTD
MNSQ
ZSTD |
|-----------------------------------------------------------------------------|
| MEAN
6.7
14.0
-.38
1.05
|
| S.D.
2.4
.0
2.31
.18
|
| MAX.
11.0
14.0
3.89
1.88
|
| MIN.
.0
14.0
-6.79
.82
|
|-----------------------------------------------------------------------------|
| REAL RMSE
1.25 ADJ.SD
1.95 SEPARATION 1.56 KID
RELIABILITY .71 |
|MODEL RMSE
1.07 ADJ.SD
2.05 SEPARATION 1.92 KID
RELIABILITY .79 |
| S.E. OF KID MEAN = .40
|
+-----------------------------------------------------------------------------+
KID RAW SCORE-TO-MEASURE CORRELATION = 1.00
CRONBACH ALPHA (KR-20) KID RAW SCORE RELIABILITY = .73
UMEAN=.000 USCALE=1.000
476 DATA POINTS. APPROXIMATE LOG-LIKELIHOOD CHI-SQUARE: 221.61
For valid observations used in the estimation,
SCORE is the raw score (number of correct responses).
COUNT is the number of responses made.
MEASURE is the estimated measure (for persons) or calibration (for items).
ERROR is the standard error of the estimate.
RAW SCORE-TO-MEASURE CORRELATION is the Pearson correlation between raw scores and measures,
including extreme scores. When data are complete, this correlation is expected to be near 1.0 for persons and
near -1.0 for items.
INFIT is an information-weighted fit statistic, which is more sensitive to unexpected behavior affecting responses
to items near the person's measure level.
MNSQ is the mean-square infit statistic with expectation 1. Values substantially below 1 indicate dependency in
your data; values substantially above 1 indicate noise.
ZSTD is the infit mean-square fit statistic t standardized to approximate a theoretical mean 0 and variance 1
distribution. ZSTD (standardized as a z-score) is used of a t-test result when either the t-test value has effectively
infinite degrees of freedom (i.e., approximates a unit normal value) or the Student's t-distribution value has been
adjusted to a unit normal value. When LOCAL=Y, then EMP is shown, indicating a local {0,1} standardization.
When LOCAL=L, then LOG is shown, and the natural logarithms of the mean-squares are reported.
192
OUTFIT is an outlier-sensitive fit statistic, more sensitive to unexpected behavior by persons on items far from the
person's measure level.
MNSQ is the mean-square outfit statistic with expectation 1. Values substantially less than 1 indicate
dependency in your data; values substantially greater than 1 indicate the presence of unexpected outliers.
ZSTD is the outfit mean-square fit statistic t standardized to approximate a theoretical mean 0 and variance 1
distribution. ZSTD (standardized as a z-score) is used of a t-test result when either the t-test value has effectively
infinite degrees of freedom (i.e., approximates a unit normal value) or the Student's t-distribution value has been
adjusted to a unit normal value. When LOCAL=Y, then EMP is shown, indicating a local {0,1} standardization.
When LOCAL=L, then LOG is shown, and the natural logarithms of the mean-squares are reported.
MEAN is the average value of the statistic.
S.D. is its sample standard deviation.
MAX. is its maximum value.
MIN. is its minimum value.
RMSE is the square-root of the average error variance. It is the Root Mean Square standard Error computed over
the persons or over the items.
MODEL RMSE is computed on the basis that the data fit the model, and that all misfit in the data is merely a
reflection of the stochastic nature of the model. This is a &quot;best case&quot; reliability, which reports an upper limit to the
reliability of measures based on this set of items for this sample.
REAL RMSE is computed on the basis that misfit in the data is due to departures in the data from model
specifications. This is a &quot;worst case&quot; reliability, which reports a lower limit to the reliability of measures based on
this set of items for this sample.
ADJ. S.D. is the &quot;adjusted&quot; standard deviation, i.e., the &quot;true&quot; standard deviation. This is the sample standard
deviation of the estimates after subtracting the error variance (attributable to their standard errors of
measurement) from their observed variance.
2
2
2
(ADJ.S.D.) = (S.D. of MEASURE) - (RMSE)
The ADJ.S.D. is an estimate of the &quot;true&quot; sample standard deviation from which the bias caused by measurement
error has been removed.
SEPARATION is the ratio of the PERSON (or ITEM) ADJ.S.D., the &quot;true&quot; standard deviation, to RMSE, the error
standard deviation. It provides a ratio measure of separation in RMSE units, which is easier to interpret than the
2
reliability correlation. SEPARATION is the signal-to-noise ratio, the ratio of &quot;true&quot; variance to error variance.
RELIABILITY is a separation reliability. The PERSON (or ITEM) reliability is equivalent to KR-20, Cronbach
Alpha, and the Generalizability Coefficient. See much more at Reliability.
S.E. OF MEAN is the standard error of the mean of the person (or item) measures for this sample.
WITH 1 EXTREME KIDS = 75 KIDS
MEAN is the mean of the measures including measures corresponding to extreme scores
S.D. is the sample standard deviation of those measures.
MEDIAN is the median measure of the sample including extreme scores (in Tables 27, 28).
The separation and reliability computations are repeated, but including any elements with extreme measures.
Since the measures for extreme scores are imprecise, these statistics are often lower than their non-extreme
equivalents. Conventional computation of a reliability coefficient (KR-20, Cronbach Alpha) includes persons with
extreme scores. The PERSON SEP REL. of this second analysis is the conventional reliability, and is usually
between the MODEL and REAL values, closer to the MODEL.
KID RAW SCORE-TO-MEASURE CORRELATION is the correlation between the marginal scores (person raw
scores and item scores) and the corresponding measures. The item correlation is expected to be negative
because higher measure implies lower probability of success and so lower item scores.
CRONBACH ALPHA (KR-20) KID RAW SCORE RELIABILITY is the conventional &quot;test&quot; reliability index. It reports
an approximate test reliability based on the raw scores of this sample. It is only reported for complete data. See
more at Reliability.
193
UMEAN=.000 USCALE=1.000 are the current settings of UMEAN= and USCALE=.
476 DATA POINTS is the number of observations that are used for standard estimation, and so are not missing
and not in extreme scores.
APPROXIMATE LOG-LIKELIHOOD CHI-SQUARE: 221.61 is the approximate value of the global fit statistic. The
accuracy of the approximation depends on how close the reported estimated measures are to their &quot;true&quot;
maximum likelihood estimates.The degrees of freedom, d.f., of the chi-square are the number of data points less
the number of free parameters, where number of free parameters for complete data = the minimum of ((number of
different non-extreme person raw scores) or (number of different non-extreme item raw scores)) - 1 (for
identifiability of local origin) + the sum, across rating scales, of the number of categories in each scale - 2 for each
scale. Increase this by 1 for each for response string with a different missing data pattern.
217.
Table 3.2 Summary of rating scale category structure
(controlled by STEPT3=, STKEEP=, MRANGE=)
The average measures and category fit statistics are how the response structure worked &quot;for this sample&quot;
(which might have high or low performers etc.). For each observation in category k, there is a person of measure
Bn and an item of measure Di. Then:
average measure = sum( Bn - Di ) / count of observations in category. These are not estimates of parameters.
The probability curves are how the response structure is predicted to work for any future sample, provided it
worked satisfactorily for this sample.
Our logic is that if the average measures and fit statistics don't look reasonable for this sample, why should they in
any future sample? If they look OK for this sample, then the probability curves tell us about future samples. If
they don't look right now, then we can anticipate problems in the future.
a) For dichotomies,
SUMMARY OF MEASURED STRUCTURE
FOR GROUPING &quot;0&quot;, MODEL &quot;R&quot;, ACT NUMBER:
12
GO TO MUSEUM
ACT
MEASURE OF -1.07 ADDED TO MEASURES
+--------------------------------------------------------+-----+
|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT| COHERENCE|ESTIM|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ| M-&gt;C C-&gt;M|DISCR|
|-------------------+------------+------------+----------|-----+
| 1
1
13 18| -.38
.01|
.83
.52| 75% 23%|
|01 neutral
| 2
2
61 82| 1.12 1.03|
.78
.85| 85% 98%| 1.23|02 like
+--------------------------------------------------------+-----+
AVERAGE MEASURE is mean of measures in category.
M-&gt;C = Does Measure imply Category?
C-&gt;M = Does Category imply Measure?
CATEGORY LABEL is the number of the category in your data set after scoring/keying.
CATEGORY SCORE is the ordinal value of the category used in computing raw scores - and in Table 20.
OBSERVED COUNT and % is the count of occurrences of this category used in the estimation (i.e., for nonextreme persons and items). Counts of all occurrences of categories are given in the distractor Tables,
e.g., Table 14.3.
OBSVD AVERGE is the average of the measures that are modelled to produce the responses observed in the
category. The average measure is expected to increase with category value. Disordering is marked by &quot;*&quot;. This
is a description of the sample, not a Rasch parameter. For each observation in category k, there is a person
of measure Bn and an item of measure Di. Then:
average measure = sum( Bn - Di ) / count of observations in category.
SAMPLE EXPECT is the expected value of the average measure for this sample. These values always advance
194
with category. This is a description of the sample, not a Rasch parameter.
INFIT MNSQ is the average of the INFIT mean-squares associated with the responses in each category. The
expected values for all categories are 1.0.
OUTFIT MNSQ is the average of the OUTFIT mean-squares associated with the responses in each category.
The expected values for all categories are 1.0. This statistic is sensitive to grossly unexpected
responses.
Note: Winsteps always reports the MNSQ values in Table 3.2. An approximation to their standardized values can
be obtained by using the number of observations in the category as the degrees of freedom, and then
looking at the plot below.
COHERENCE
M-&gt;C shows what percentage of the measures that were expected to produce observations in this category
actually did. Do the measures imply the category?
Guttman's Coefficient of Reproducibility is the count-weighted average of the M-&gt;C, i.e.,
Reproducibility = sum (COUNT * M-&gt;C) / sum(COUNT * 100)
C-&gt;M shows what percentage of the observations in this category were produced by measures corresponding to
the category. Does the category imply the measures?
ESTIM DISCR is an estimate of the local discrimination when the model is parameterized in the form: log-odds =
aj (Bn - Di - Fj)
RESIDUAL (when shown) is the residual difference between the observed and expected counts of observations in
the category. Shown as % of expected, unless observed count is zero. Then residual count is shown.
Only shown if residual count is &gt;= 1.0. Indicates lack of convergence, structure anchoring, or large data
set.
CATEGORY CODES and LABELS are shown to the right based on CODES=, CFILE= and CLFILE=.
Measures corresponding to the dichotomous categories are not shown, but can be computed using the Table at
&quot;What is a Logit?&quot; and LOWADJ= and HIADJ=.
b) For rating (or partial credit) scales, the structure calibration table lists:
SUMMARY OF CATEGORY STRUCTURE. Model=&quot;R&quot;
FOR GROUPING &quot;0&quot; ITEM NUMBER: 1 A. EATING
ITEM ITEM DIFFICULTY MEASURE OF -.61 ADDED TO MEASURES
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ||CALIBRATN| MEASURE|
|-------------------+------------+------------++---------+--------+
| 5
5
5 14| -.51 -.42|
.89
.68|| NONE
|( -2.22)| 5 Supervision
| 6
6
9 26|
.39
.04| 1.45 1.63||
-.18 |
-.61 | 6 Device
| 7
7
21 60|
.73
.86| 1.34 1.32||
.18 |( 1.00)| 7 Independent
+-----------------------------------------------------------------AVERAGE MEASURE is mean of measures in category.
+---------------------------------------------------------------------------------------------+
|CATEGORY
STRUCTURE
| SCORE-TO-MEASURE
| 50% CUM.| COHERENCE|ESTIM| OBSERVED-EXPECTED |
| LABEL
MEASURE S.E. | AT CAT. ----ZONE----|PROBABLTY| M-&gt;C C-&gt;M|DISCR|RESIDUAL DIFFERENCE|
|------------------------+---------------------+---------+----------+-----+-------------------|
|
5
NONE
|( -2.22) -INF
-1.50|
|
0%
0%|
|
-1.4%
-.1 | 5
Supervision
|
6
-.79
.52 |
-.61 -1.50
.28|
-1.18 | 31% 66%| 1.22|
.2%
.0 | 6 Device
|
7
-.43
.39 |( 1.00)
.28 +INF |
-.04 | 81% 61%| .59|
.2%
.0 | 7
Independent
+---------------------------------------------------------------------------------------------+
M-&gt;C = Does Measure imply Category?
C-&gt;M = Does Category imply Measure?
195
ITEM MEASURE OF -.64 ADDED TO MEASURES
When there is only one item in a grouping (the Partial Credit model), the item measure is added to the
reported measures.
CATEGORY LABEL, the number of the category in your data set after scoring/keying.
CATEGORY SCORE is the value of the category in computing raw scores - and in Table 20.
OBSERVED COUNT and %, the count of occurrences of this category used in the estimation.
OBSVD AVERGE is the average of the measures that are modelled to produce the responses observed in the
category. The average measure is expected to increase with category value. Disordering is marked by &quot;*&quot;. This
is a description of the sample, not the estimate of a parameter. For each observation in category k, there is a
person of measure Bn and an item of measure Di. Then: average measure = sum( Bn - Di ) / count of
observations in category.
SAMPLE EXPECT is the expected value of the average measure for this sample. These values always advance
with category. This is a description of the sample, not a Rasch parameter.
INFIT MNSQ is the average of the INFIT mean-squares associated with the responses in each category. The
expected values for all categories are 1.0.
OUTFIT MNSQ is the average of the OUTFIT mean-squares associated with the responses in each category.
The expected values for all categories are 1.0. This statistic is sensitive to grossly unexpected
responses.
Note: Winsteps always reports the MNSQ values in Table 3.2. An approximation to their standardized values can
be obtained by using the number of observations in the category as the degrees of freedom, and then
looking at the plot below.
STRUCTURE CALIBRATN, the calibrated measure of the transition from the category below to this category.
This is an estimate of the Rasch model parameter, Fj. Use this for anchoring in Winsteps. (This
corresponds to Fj in the Di+Fj parameterization of the &quot;Rating Scale&quot; model, and is similarly applied as
the Fij of the Dij=Di+Fij of the &quot;Partial Credit&quot; model.) The bottom category no prior transition, and so that
the measure is shown as NONE. The Rasch-Andrich threshold is expected to increase with category
value. This parameter, sometimes called the Step Difficulty, Step Calibration, Rasch-Andrich
threshold, Tau or Delta, indicates how difficult it is to observe a category, not how difficult it is to
perform it. Disordering of these estimates (so that they do not ascend in value up the rating scale),
sometimes called &quot;disordered deltas&quot;, indicates that the category is relatively rarely observed, i.e.,
occupies a narrow interval on the latent variable, and so may indicate substantive problems with the
rating (or partial credit) scale category definitions. These Rasch-Andrich thresholds are relative pair-wise
measures of the transitions between categories. They are the points at which adjacent category
probability curves intersect. They are not the measures of the categories. See plot below.
CATEGORY MEASURE, the sample-free measure corresponding to this category. ( ) is printed where the
matching calibration is infinite. The value shown corresponds to the measure .25 score points (or
LOWADJ= and HIADJ=) away from the extreme. This is the best basis for the inference: &quot;ratings
averaging x imply measures of y&quot; or &quot;measures of y imply ratings averaging x&quot;. This is implied by
the Rasch model parameters.
STRUCTURE MEASURE, item measure add to the calibrated measure of this transition from the category below
to this category. For structures with only a single item, this is an estimate of the Rasch model
parameter, Dij = Di + Fij. (This corresponds to the Dij parameterization of the &quot;Partial Credit&quot; model.)
The bottom category has no prior transition, and so that the measure is shown as NONE. The RaschAndrich threshold is expected to increase with category value, but these can be disordered. &quot;Dgi + Fgj&quot;
locations are plotted in Table 2.4, where &quot;g&quot; refers to the ISGROUPS= assignment.
STRUCTURE S.E. is an approximate standard error of the Rasch-Andrich threshold measure.
SCORE-TO-MEASURE
These values are plotted in Table 21, &quot;Expected Score&quot; ogives. They are useful for quantifying category
196
measures. This is implied by the Rasch model parameters.
AT CAT is the measure (on an item of 0 logit measure) corresponding to an expected score equal to the category
label, which, for the rating (or partial credit) scale model, is where this category has the highest
probability. See plot below.
( ) is printed where the matching calibration is infinite. The value shown corresponds to the measure .25 score
points (or LOWADJ= and HIADJ=) away from the extreme.
--ZONE-- is the range of measures from an expected score from 1/2 score-point below to the category to 1/2
score-point above it, the Rasch-half-point thresholds. Measures in this range (on an item of 0
measure) are expected to be observed, on average, with the category value. See plot below.
50% CUMULATIVE PROBABILITY gives the location of median probabilities, i.e. these are Rasch-Thurstone
thresholds, similar to those estimated in the &quot;Graded Response&quot; or &quot;Proportional odds&quot; models. At
these calibrations, the probability of observing the categories below equals the probability of observing
the categories equal or above. The .5 or 50% cumulative probability is the point on the variable at which
the category interval begins. This is implied by the Rasch model parameters.
COHERENCE
M-&gt;C shows what percentage of the measures that were expected to produce observations in this category
actually did. Do the measures imply the category?
Guttman's Coefficient of Reproducibility is the count-weighted average of the M-&gt;C, i.e., Reproducibility = sum
(COUNT * M-&gt;C) / sum(COUNT * 100)
C-&gt;M shows what percentage of the observations in this category were produced by measures corresponding to
the category. Does the category imply the measures?
ESTIM DISCR (when DISCRIM=Y) is an estimate of the local discrimination when the model is parameterized in
the form: log-odds = aj (Bn - Di - Fj)
OBSERVED - EXPECTED RESIDUAL DIFFERENCE (when shown) is the residual difference between the
observed and expected counts of observations in the category.
residual difference % = (observed count - expected count) * 100 / (expected count)
residual difference value = observed count - expected count
These are shown if at least one residual percent &gt;=1%. This indicates that the Rasch estimates have
not converged to their maximum-likelihood values, due to lack of convergence, anchoring, or a large data
set. For example,
(a) iteration was stopped early using Ctrl+F or the pull-down menu option.
(b) iteration was stopped when the maximum number of iterations was reached MJMLE=
(c) the convergence criteria LCONV= and RCONV= are not small enough for this data set.
(d) anchor values (PAFILE=, IAFILE= and/or SAFILE=) are in force which do not allow maximum
likelihood estimates to be obtained.
ITEM MEASURE ADDED TO MEASURES, is shown when the rating (or partial credit) scale applies to only one
item, e.g., when ISGROUPS=0. Then all measures in these tables are adjusted by the estimated item
measure.
P
R
O
B
A
B
I
L
I
T
CATEGORY PROBABILITIES: MODES - Structure measures at intersections
++---------+---------+---------+---------+---------+---------++
1.0 +
+
|
|
|00
22|
| 0000
2222 |
.8 +
000
222
+
|
000
222
|
|
00
22
|
|
00
22
|
.6 +
00
22
+
197
Y
|
00
1111111
22
|
.5 +
0 1111
1111 2
+
|
1**
**1
|
.4 +
11
00
22
11
+
|
111
00
22
111
|
|
11
00
22
11
|
|
111
0*2
111
|
.2 +
111
22 00
111
+
|
1111
222
000
1111
|
|111
2222
0000
111|
|
2222222
0000000
|
.0 +22222222222222
00000000000000+
++---------+---------+---------+---------+---------+---------++
-3
-2
-1
0
1
2
3
PUPIL [MINUS] ACT
MEASURE
O
F
R
E
S
P
O
N
S
E
Curves showing how probable is the observation of each category for measures relative to the item measure.
Ordinarily, 0 logits on the plot corresponds to the item measure, and is the point at which the highest and lowest
categories are equally likely to be observed. The plot should look like a range of hills. Categories which never
emerge as peaks correspond to disordered Rasch-Andrich thresholds. These contradict the usual interpretation
of categories as a being sequence of most likely outcomes.
Null, Zero, Unobserved Categories
STKEEP=YES and Category 2 has no observations:
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ||CALIBRATN| MEASURE|
|-------------------+------------+------------++---------+--------+
| 0
0
378 20| -.67 -.73|
.96 1.16|| NONE
|( -2.01)|
| 1
1
620 34| -.11 -.06|
.81
.57||
-.89 |
-.23 |
| 2
2
0
0|
|
.00
.00|| NULL
|
.63 |
| 3
3
852 46| 1.34 1.33| 1.00 1.64||
.89 |( 1.49)|
+-----------------------------------------------------------------Category 2 is an incidental (sampling)zero. The category is maintained in the response structure.
STKEEP=NO and Category 2 has no observations:
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ||CALIBRATN| MEASURE|
|-------------------+------------+------------++---------+--------+
| 0
0
378 20| -.87 -1.03| 1.08 1.20|| NONE
|( -2.07)|
| 1
1
620 34|
.13
.33|
.85
.69||
-.86 |
.00 |
| 3
2
852 46| 2.24 2.16| 1.00 1.47||
.86 |( 2.07)|
+-----------------------------------------------------------------Category 2 is a structural (unobservable) zero. The category is eliminated from the response structure.
218.
Table 4.1, 5.1, 8.1, 9.1 Fit plots
(controlled by FRANGE=, LOCAL=, MNSQ=, OUTFIT=)
P
E
R
S
O
N
5
4
3
2
I
N
F
I
T
1
0
-4
-3
-2
-1
0
1
2
3
4
5
|-----------------------------------------------------------------|
|
|
B
|
|
| A
|
|
|
|
|
|
|
|
C
|
|
|
D
I|
|
|--------------------------F-G|--Q-M-------------U----------------|
|
E
H|
|
|
J|T
|
|
R
|1S
11 1
1
1
|
|------------------K-----1----|1-Z1-21-1--1--12-----------------W-|
|
O
2 L1
1 x
y 1
1
|
198
5
4
3
2
1
0
S
T
D
-1 |
1N t22 lnp r u
1
| -1
|
|vVh
j i
o
|
-2 |-----------------------------|-w--eg--f----k---------------------| -2
|
| b
d
|
-3 |
| a
c
| -3
|-----------------------------------------------------------------|
-4
-3
-2
-1
0
1
2
3
4
5
PERSON MEASURE
ITEMS
1
1
T
1
11 111 21 21221
S
M
1
21 1 1
S
T
These tables are plots of the t standardized fit statistics, INFIT or OUTFIT, against the parameter estimates.
INFIT is a t standardized information-weighted mean square statistic, which is more sensitive to unexpected
behavior affecting responses to items near the person's measure level. OUTFIT is a t standardized outliersensitive mean square fit statistic, more sensitive to unexpected behavior by persons on items far from the
person's measure level. The standardization is approximate. Its success depends on the distribution of persons
and items. Consequently, the vertical axis is only a guide and should not be interpreted too rigidly. The
NORMAL= variable controls the standardization method used.
Letters on the plot indicate the misfitting person or items. Numbers indicate non-extreme fit or multiple
references. The letters appear on Tables 6, 17, 18, 19 for persons, and Tables 10, 13, 14, 15 for items.
Use MNSQ= to change the y-axis to mean-squares.
219.
Table 5.2, 9.2 Fit plots
If both infit and outfit plots are requested, then a plot of outfit against infit is also produced to assist with the
identification of the different patterns of observations they diagnose. For interpretation of fit statistics, see
dichotomous and polytomous fit statistics.
6
P
E
R
S
O
N
5
4
3
I
N
F
I
T
2
S
T
D
-1
1
0
-2
-3
220.
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
--------------------------------------------------------------------|
|
.
6
|
|
.B
|
|
.
A
5
|
|
.
|
|
.
4
|
|
.
|
|
.
C
3
|
|I .
D
----------+--------- -U-Q---M-+-------2--------------------------2
|
. | H
E
|
T .
J
1
|
24
2
|
453W
K
0
|
31.3
PO L |
|
2721
N
|
-1
| .3ov
V
|
----------+-2gk-w--- ---------+-----------------------------------2
b d
|
a
c|
|
-3
-------------------------------------------------------------------4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
PERSON OUTFIT ZSTD
Table 6.1, 10.1 Person and item statistics
(controlled by FITI=, FITP=, USCALE=, UIMEAN=, UPMEAN=, UDECIM=, LOCAL=, ISORT=, OUTFIT=,
PSORT=, TOTAL=)
PERSON STATISTICS: MISFIT ORDER
+---------------------------------------------------------------------------------------------------+
|ENTRY
RAW
MODEL|
INFIT | OUTFIT |PTMEA|EXACT MATCH|
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.| OBS% EXP%| KID
|
|------------------------------------+----------+----------+-----+-----------+----------------------|
|
72
14
25
-1.32
.37|2.02
2.9|5.16
5.7|A .04| 60.0 65.8| JACKSON, SOLOMON
|
|
71
33
25
.97
.35|3.15
5.4|4.95
5.8|B-.03| 36.0 61.8| STOLLER, DAVE
|
|
BETTER FITTING OMITTED
+----------+----------+-----|
|
|
199
|
22
35
25
1.23
.36| .34 -3.2| .33 -2.1|b .80| 92.0 62.2| HOGAN, KATHLEEN
|
|
21
28
25
.38
.34| .29 -3.9| .31 -3.0|a .86| 88.0 60.2| EISEN, NORM L.
|
|------------------------------------+----------+----------+-----+-----------+----------------------|
| MEAN
31.7
25.0
.97
.40| .99
-.2|1.08
.1|
| 68.8 65.2|
|
| S.D.
8.6
.0
1.35
.19| .50
1.6|1.04
1.9|
| 13.5
8.4|
|
+---------------------------------------------------------------------------------------------------+
ITEMS STATISTICS: MEASURE ORDER
+-------------------------------------------------------------------------------------------------------------+
|ENTRY
RAW
MODEL|
INFIT | OUTFIT |PTMEA|EXACT MATCH|
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.| OBS% EXP%| ACT
|
|------------------------------------+----------+----------+-----+-----------+--------------------------------|
|
23
40
74
2.18
.21|2.41
6.3|4.11
9.0|A .00| 40.5 65.0| WATCH A RAT
|
|
5
35
74
2.42
.22|2.30
5.6|3.62
7.3|B .05| 52.7 68.1| FIND BOTTLES AND CANS
|
|
BETTER FITTING OMITTED
+----------+----------+-----|
|
|
|
3
86
74
.42
.19| .57 -3.5| .54 -3.0|b .72| 73.0 57.7| READ BOOKS ON PLANTS
|
|
1
107
74
-.40
.21| .55 -3.5| .49 -2.5|a .64| 77.0 61.7| WATCH BIRDS
|
|------------------------------------+----------+----------+-----+-----------+--------------------------------|
| MEAN
93.0
74.0
.00
.23|1.02
-.2|1.08
.0|
| 68.8 65.2|
|
| S.D.
30.9
.0
1.41
.06| .45
2.3| .87
2.8|
| 13.1 10.2|
|
+-------------------------------------------------------------------------------------------------------------+
MISFIT ORDER for Tables 6 and 10 is controlled by OUTFIT=. The reported MEAN and S.D. include extreme
scores, which are not shown in these Tables.
ENTRY NUMBER is the sequence number of the person, or item, in your data, and is the reference number used
for deletion or anchoring.
&quot;PERSONS&quot; or &quot;ITEMS&quot;, etc. is the item name or person-identifying label.
RAW SCORE is the raw score corresponding to the parameter, i.e., the raw score by a person on the test, or the
sum of the scored responses to an item by the persons.
COUNT is the number of data points used to construct measures.
MEASURE is the estimate (or calibration) for the parameter, i.e., person ability (theta, B, beta, etc.), or the item
difficulty (b, D, delta, etc.). Values are reported in logits with two decimal places, unless rescaled by
USCALE=, UIMEAN=, UPMEAN=, UDECIM=. If the score is extreme, a value is estimated, but as MAXIMUM
(perfect score) or MINIMUM (zero score). No measure is reported if the element is DROPPED (no valid
observations remaining) or DELETED (you deleted the person or item). The difficulty of an item is defined to
be the point on the latent variable at which its high and low categories are equally probable. SAFILE= can be
used to alter this definition.
If unexpected results are reported, check whether TARGET= or CUTLO= or CUTHI= are specified.
INESTIMABLE is reported if all observations are eliminated as forming part of extreme response strings.
ERROR is the standard error of the estimate. For anchored values, an &quot;A&quot; is shown on the listing and the error
reported is that which would have been obtained if the value had been estimated. Values are reported in logits
with two decimal places, unless rescaled by USCALE=, UDECIM=
INFIT is a t standardized information-weighted mean square statistic, which is more sensitive to unexpected
behavior affecting responses to items near the person's measure level.
MNSQ is the mean-square infit statistic with expectation 1. Values substantially less than 1 indicate dependency
in your data; values substantially greater than 1 indicate noise. See dichotomous and polytomous fit statistics.
Value
Meaning
&gt;2.0
Off-variable noise is greater than useful information. Degrades measurement.
&gt;1.5
Noticeable off-variable noise. Neither constructs nor degrades measurement
0.5 - 1.5
Productive of measurement
&lt;0.5
Overly predictable. Misleads us into thinking we are measuring better than we really are.
(Attenuation paradox.)
Always remedy the large misfits first. Misfits &lt;1.0 are only of concern when shortenning a test.
ZSTD is the infit mean-square fit statistic t standardized to approximate a theoretical &quot;unit normal&quot;, mean 0 and
variance 1, distribution. ZSTD (standardized as a z-score) is used of a t-test result when either the t-test value
has effectively infinite degrees of freedom (i.e., approximates a unit normal value) or the Student's tdistribution value has been adjusted to a unit normal value. The standardization is shown on RSA, p.100-101.
When LOCAL=Y, then EMP is shown, indicating a local {0,1} standardization. When LOCAL=LOG, then LOG
200
is shown, and the natural logarithms of the mean-squares are reported. More exact values are shown in the
Output Files.
Ben Wright advises: &quot;ZSTD is only useful to salvage non-significant MNSQ&gt;1.5, when sample size is
small or test length is short.&quot;
OUTFIT is a t standardized outlier-sensitive mean square fit statistic, more sensitive to unexpected behavior by
persons on items far from the person's measure level.
Value
&gt;2.0
&gt;1.5
0.5 - 1.5
&lt;0.5
Meaning
Off-variable noise is greater than useful information. Degrades measurement.
Noticeable off-variable noise. Neither constructs nor degrades measurement
Productive of measurement
Overly predictable. Misleads us into thinking we are measuring better than we really are.
(Attenuation paradox.)
Always remedy the large misfits first. Misfits &lt;1.0 are usually only of concern when shortenning a test.
MNSQ is the mean-square outfit statistic with expectation 1. Values substantially less than 1 indicate
dependency in your data; values substantially greater than 1 indicate the presence of unexpected outliers.
See dichotomous and polytomous fit statistics.
ZSTD is the outfit mean-square fit statistic t standardized similarly to the INFIT ZSTD. ZSTD (standardized as a zscore) is used of a t-test result when either the t-test value has effectively infinite degrees of freedom (i.e.,
approximates a unit normal value) or the Student's t-distribution value has been adjusted to a unit normal
value.
Ben Wright advises: &quot;ZSTD is only useful to salvage non-signficant MNSQ&gt;1.5, when sample size is
small or test length is short.&quot;
PTBIS CORR (reported when PTBIS=Yes) is the point-biserial correlation, rpbis, between the individual item (or
person) response &quot;scores&quot; and the total person (or item) test score (less the individual response &quot;scores&quot;).
Negative values for items often indicate mis-scoring, or rating (or partial credit) scale items with reversed
direction. Letters indicating the identity of persons or items appearing on the fit plots appear under PTBIS.
For adaptive tests, an rpbis near zero is expected.
The formula for this product-moment correlation coefficient is:
rpbis = {sum {(x-x bar)(y-y bar)}} over {sqrt {{sum {(x-x bar)}^2} {sum {(y-y bar)}^2}} }
where x = observation for this item (or person), y = total score (including extreme scores) for person omitting
this item (or for item omitting this person).
PTMEA CORR. (reported when PTBIS=N) is the point-measure correlation, rpm or RPM, between the
observations on an item (as fractions with the range 0,1) and the corresponding person measures, or viceversa. Since the point-biserial loses its meaning in the presence of missing data, specify PTBIS=N when there
are missing data or when CUTLO= or CUTHI= are specified. The point-measure correlation has a range of -1
to +1.
EXACT MATCH
OBS% Observed% is the percent of data points which are within 0.5 score points of their expected values, i.e.,
that match predictions.
EXP% Expected% is the percent of data points that are predicted to be within 0.5 score points of their expected
values.
ESTIM DISCRIM is an estimate of the item discrimination, see DISCRIM=
ASYMPTOTE LOWER and UPPER are estimates of the upper and lower asymptotes for dichotomous items, see
ASYMPTOTE=
201
WEIGH is the weight assigned by IWEIGHT= or PWEIGHT=
DISPLACE is the displacement of the reported MEASURE from its data-derived value. This should only be shown
with anchored measures.
G is the grouping code assigned with ISGROUPS=
M is the model code assigned with MODELS=
&quot;BETTER FITTING OMITTED&quot; appears in fit-ordered Tables, where items better than FITI=, or persons better
than FITP=, are excluded.
Above the Table are shown the &quot;real&quot; separation and reliability coefficients from Table 3.
Selection of persons and items for misfit tables
Report measure in Tables 6 and 10, if any of:
Statistic
Less than Greater than
t standardized INFIT
-(FITP or FITI)
FITP or FITI
t standardized OUTFIT -(FITP or FITI)
FITP or FITI
mean-square INFIT 1 - (FITP or FITI)/10 1 + (FITP or FITI)/10
mean-square OUTFIT 1 - (FITP or FITI)/10 1 + (FITP or FITI)/10
point-biserial correlation negative
To include every person, specify FITP=0. For every item, FITI=0.
Usually OUTFIT=N, and Tables 6 and 10 are sorted by by maximum (Infit mean-square, Outfit mean-square)
When OUTFIT=Y, Tables 6 and 10 are sorted by Infit Mean-square.
221.
Table 6.2, 10.2, 13.2, 14.2, 15.2, 17.2, 18.2, 19.2, 25.2 Person and item statistics
Controlled by USCALE=, UMEAN=, UDECIM=, LOCAL=, FITLOW=, FITHIGH=, MNSQ=
Specify CHART=YES to produce Tables like this.
With MNSQ=Yes:
PUPIL FIT GRAPH: OUTFIT ORDER
+----------------------------------------------------------------------------------+
|ENTRY| MEASURE | INFIT MEAN-SQUARE | OUTFIT MEAN-SQUARE |
|
|NUMBR| + |0
0.7 1 1.3
2|0
0.7 1 1.3
2| PUPIL
|
|-----+---------+---------------------+---------------------+----------------------|
|
72|*
|
: . :
* |A
: . :
*| JACKSON, SOLOMON
|
|
47| *
|
: . *:
|J
: . :
*
| VAN DAM, ANDY
|
|
53|*
|
: *. :
|K
: . :
* | SABOL, ANDREW
|
|
32| *
|
*: . :
|w
* . :
| ROSSNER, JACK
|
|
21| *
| *
: . :
|a *
: . :
| EISEN, NORM L.
|
+----------------------------------------------------------------------------------+
The fit information is shown in graphical format to aid the eye in identifying patterns and outliers. The fit bars are
positioned by FITLOW= and FITHIGH=. They may also be repositioned using TFILE=.
With MNSQ=No:
+----------------------------------------------------------------------------+
|ENTRY | MEASURE |INFIT t standardized |OUTFIT t standardized|
|
|NUMBER| + |-3 -2 -1 0 1 2 3 |-3 -2 -1 0 1 2 3 | TAPS
|
|------+---------+---------------------+---------------------+---------------|
|
18|
E|
:
.
:
|
:
.
:
| 4-1-3-4-2-1-4 |
|
15|
* |
:
*.
:
|
:
*
:
| 1-3-2-4-1-3
|
|
14|
* |
:
. * :
|
:
*
:
| 1-4-2-3-4-1
|
|
12|
* |
:
.*
:
|
:
*
:
| 1-3-2-4-3
|
202
|
13|
*
|
: * .
:
|
:
* .
:
| 1-4-3-2-4
|
|
11|
*
|
:
*
:
|
:
*.
:
| 1-3-1-2-4
|
|
10|
*
|
:
*
:
|
:
*
:
| 2-4-3-1
|
|
7| *
|
:
. *
:
|
:
.*
:
| 1-4-3-2
|
|
4| *
|
:
*
:
|
:
*
:
| 1-3-4
|
|
1|E
|
:
.
:
|
:
.
:
| Cube 1, then 4|
+----------------------------------------------------------------------------+
222.
Table 6.5, 10.5 Most unexpected observations
These tables display the unexpected responses in Guttman scalogram format. The Guttman Scalogram of
unexpected responses shows the persons and items with the most unexpected data points (those with the largest
standardized residuals) arranged by measure, such that the high value observations are expected in the top-left
of the data matrix, near the &quot;high&quot;, and the low values are expected in the bottom of the matrix, near the &quot;low&quot;.
The category values of unexpected observations are shown. Expected values (with standardized residuals less
than |2|) are shown by &quot;.&quot;. Missing values, if any, are left blank.
UCOUNT= sets the maximum number of persons and items to report in the anti-Guttman matrices in Tables 6.4,
6.5, 10.4, 10.5.
MOST UNEXPECTED RESPONSES
PUPIL
MEASURE
41
17
71
53
223.
FATALE, NATASHA
SCHATTNER, GAIL
STOLLER, DAVE
SABOL, ANDREW
|ACT
|1111112 122 1
22
|89203112542669784035
high-------------------4.77 |..................1.
3.55 |...................0
.96 B|.0.10....0.......222
-1.59 L|..............1....1
|-----------------low
|11111122122619784225
|8920311 542 6
03
Table 6.4, 10.4 Most misfitting response strings
These tables display the unexpected responses in the most misfitting response strings in a Guttman scalogram
format. UCOUNT= sets the maximum number of persons and items to report in the anti-Guttman matrices in
Tables 6.4, 6.5, 10.4, 10.5.
MOST MISFITTING RESPONSE STRINGS
ACT
OUTMNSQ
23 WATCH A RAT
20 WATCH BUGS
9 LEARN WEED NAMES
|PUPIL
|4314 452371 17667422 315231 4 4 31725
|1475789534189175348742556649769502293
high------------------------------------5.68 A|1110.0.......2.2...2....1112.2211122.
1.99 B|......0......2..2.....2.......11..21.
1.29 D|....0....0..........................1
|----------------------------------low
|4314745237181766742243152319464531725
|1475 895341 91753487 255664 7 9 02293
The items (or persons) are ordered by descending mean-square misfit. Each column corresponds to a person.
The entry numbers are printed vertically. The responses are ordered so that the highest expected responses are
to the left (high), the lowest to the right (low). The category values of unexpected observations are shown.
Expected values (with standardized residuals less than |2|) are shown by &quot;.&quot;. Missing values, if any, are left blank.
224.
Table 6.6, 10.6 Most unexpected response list
This shows the most unexpected responses sorted by unexpectedness (standardized residual). Large
standardized residuals contribute to large outfit mean-square fit statistics. UCOUNT= sets the maximum number
of &quot;most unexpected responses&quot; to report in Tables 6.6, 10.6.
TABLE 6.6 LIKING FOR SCIENCE (Wright &amp; Masters
INPUT: 75 KIDS, 25 ACTS MEASURED: 75 KIDS, 25 ACTS, 3 CATS
203
WINSTEPS 3.58.0
-------------------------------------------------------------------------------MOST UNEXPECTED RESPONSES
+------------------------------------------------------------------------------------------------------+
| DATA |OBSERVED|EXPECTED|RESIDUAL|ST. RES.|MEASDIFF| ACT
| KID
| ACT
| KID
|
|------+--------+--------+--------+--------+--------+-------+--------+--------------+------------------|
|
0 |
0 |
1.93 | -1.93 | -7.66 |
3.53 |
18 |
73 | GO ON PICNIC | SANDBERG, RYNE
|
|
2 |
2 |
.07 |
1.93 |
7.57 | -3.50 |
23 |
72 | WATCH A RAT | JACKSON, SOLOMON |
|
2 |
2 |
.07 |
1.93 |
7.57 | -3.50 |
23 |
29 | WATCH A RAT | LANDMAN, ALAN
|
|
0 |
0 |
1.93 | -1.93 | -7.41 |
3.46 |
19 |
71 | GO TO ZOO
| STOLLER, DAVE
|
DATA is the response code in the data file
OBSERVED is the code's value after rescoring
EXPECTED is the predicted observation based on the person and item estimated measures
RESIDUAL is (OBSERVED - EXPECTED), the difference between the observed and expected values
ST. RES. is the standardized residual, the unexpectedness of the residual expressed as a unit normal deviate
MEASDIFF is the difference between the ability and difficulty estimates. This produces the EXPECTED value.
ACT is the item entry number
KID is the person entry number
ACT is the item label
KID is the person label
225.
Table 7.1, 11.1 Misfitting responses
(controlled by FITI=, FITP=, MNSQ=, OUTFIT=)
These tables show the persons or items for which the t standardized outfit (or infit, if OUTFIT=N) statistic is
greater than the misfit criterion (FITP= or FITI=). Persons or items are listed in descending order of misfit. The
response codes are listed in their sequence order in your data file. The residuals are standardized response
score residuals, which have a modelled expectation of 0, and a variance of 1. Negative residuals indicate that the
observed response was less correct (or, for rating (or partial credit) scales, lower down the rating scale) than
expected. The printed standardized residual is truncated, not rounded, so that its actual value is at least as
extreme as that shown. Standardized residuals between -1 and 1 are not printed. For exact details, see XFILE=.
&quot;X&quot; indicates that the item (or person) obtained an extreme score. &quot;M&quot; indicates a missing response.
For Table 7, the diagnosis of misfitting persons, persons with a t standardized fit greater than FITP= are reported.
Selection is based on the OUTFIT statistic, unless you set OUTFIT=N in which case the INFIT statistic is used.
For Table 11, the diagnosis of misfitting items, items with a t standardized fit greater than FITI= are reported.
Selection is based on the OUTFIT statistic, unless you set OUTFIT=N in which case the INFIT statistic is used.
TABLE OF POORLY FITTING ITEMS
NUMBER
NAME
POSITION
23 WATCH A RAT
RESPONSE:
Z-RESIDUAL:
RESPONSE:
Z-RESIDUAL:
1:
0 2 1 1 1
X
2
26:
1 2 0 2 1
3
6 2
5 FIND BOTTLES AND CANS
RESPONSE:
1: 1 2 0 0 1
Z-RESIDUAL:
X
2
226.
(PERSONS IN ENTRY ORDER)
MEASURE
INFIT (ZSTD) OUTFIT
MISFIT OVER 2.0
2.00
5.8
A
8.1
2 2 0 2 0
0 1 0 1 1 0 1 0 0 0
3
3
2
M 0 0 1 1
1 0 1 0 0
1 0 0 0 0
-2
-2
0 1 1 0 0
0 0 0 2 1
4
/ This letter on fit plots
2.21
5.2
B
6.5
2 0 0 1 0
0 2 0 0 1 1 0 0 0 0
0 1 1 0 1
4
6
-2
Table 7.2, 17.3, 18.3, 19.3 Diagnostic KeyForms for Persons
This table displays a version of Table 2.2 with the responses filled in for each person. These can be useful for
diagnosis and individual reporting. The responses are the original values in the data file, as far as possible.
TABLE 7.2 LIKING FOR SCIENCE (Wright &amp; Masters p.18)
sf.out Sep 26 8:59 2000
INPUT: 76 PUPILS, 25 ACTS ANALYZED: 74 PUPILS, 25 ACTS, 36 CATS WINSTEPS v3.08
-------------------------------------------------------------------------------KEY: .1.=OBSERVED, 1=EXPECTED, (1)=OBSERVED, BUT VERY UNEXPECTED.
204
NUMBER - NAME ------------------ MEASURE - INFIT (MNSQ) OUTFIT - S.E.
72 JACKSON, SOLOMON
-1.81
1.9
A
5.4
.36
-3
-2
-1
0
1
2
3
4
5
|------+------+------+------+------+------+------+------|
0
(1)
.0.
0
(2)
.0.
.0.
0
(1)
.0.
.0.
.0.
.0.
.1.
0
(2)
0
.1.
0.
1
.0.
PLANT
0.
1
.0.
.0.
0.
1
.1.
0.
1
.1.
.1.
.2.
.1.
|------+------+------+------+------+------+------+------|
-3
-2
-1
0
1
2
3
4
5
NUM
4
5
23
9
16
8
3
14
17
7
1
20
25
2
6
ACT
WATCH GRASS CHANGE
FIND BOTTLES AND CANS
WATCH A RAT
LEARN WEED NAMES
MAKE A MAP
LOOK IN SIDEWALK CRACKS
READ BOOKS ON PLANTS
LOOK AT PICTURES OF PLANTS
WATCH WHAT ANIMALS EAT
WATCH ANIMAL MOVE
WATCH BIRDS
WATCH BUGS
TALK W/FRIENDS ABOUT PLANTS
READ BOOKS ON ANIMALS
LOOK UP STRANGE ANIMAL OR
11
22
24
10
15
21
13
12
18
19
NUM
FIND WHERE ANIMAL LIVES
FIND OUT WHAT ANIMALS EAT
FIND OUT WHAT FLOWERS LIVE ON
LISTEN TO BIRD SING
READ ANIMAL STORIES
WATCH BIRD MAKE NEST
GROW GARDEN
GO TO MUSEUM
GO ON PICNIC
GO TO ZOO
ACT
The vertical line of numbers corresponds to the person measure, and indicates the expected (average)
responses. Responses marked .0. or .1. or .2. are observed. Responses (1) and (2) are observed and statistically
significantly unexpected, |t|≥2, p&lt;5%. Responses shown merely as 0 or 1 are expected, but not observed.
227.
Table 10.3, 13.3, 14.3, 15.3, 25.3 Item option/distracter frequencies
(controlled by DistractorS=Y, OSORT=, CFILE=)
ITEM OPTION FREQUENCIES are output if DistractorS=Y. These show occurrences of each of the valid data
codes in CODES=, and also of MISSCORE= in the input data file. Counts of responses forming part of extreme
scores are included. Only items included in the corresponding main table are listed.
OSORT= controls the ordering of options within items. The standard is the order of data codes in CODES=.
ACTS CATEGORY/OPTION/Distractor FREQUENCIES: ENTRY ORDER
+-------------------------------------------------------------------------+
|ENTRY
DATA SCORE |
DATA
| AVERAGE S.E. OUTF PTMEA|
|
|NUMBER CODE VALUE | COUNT
% | MEASURE MEAN MNSQ CORR.| ACT
|
|--------------------+------------+--------------------------+------------|
|
1
0
0 |
3
4 |
-.87
.44
.5 -.28 |WATCH BIRDS | 0 Dislike
|
1
1 |
35 47 |
.23
.11
.4 -.51 |
| 1 Neutral
|
2
2 |
36 49 |
1.83
.22
.7
.62 |
| 2 Like
|
MISSING *** |
1
1*|
1.04
.01 |
|
ENTRY NUMBER is the item sequence number.
The letter next to the sequence number is used on the fit plots.
DATA CODE is the response code in the data file.
MISSING means that the data code is not listed in the CODES= specification.
Codes with no observations are not listed.
205
SCORE VALUE is the value assigned to the data code by means of NEWSCORE=, KEY1=, IVALUEA=, etc.
*** means the data code is ignored, i.e., regarded as not administered. MISSCORE=1 scores missing data as
&quot;1&quot;.
DATA COUNT is the frequency of the data code in the data file - this includes observations for both non-extreme
and extreme persons and items. For counts only for non-extreme persons and items, see the DISFILE=
DATA % is the percent of scored data codes. For dichtomies, the % are the p-values for the options.
For data with score value &quot;***&quot;, the percent is of all data codes, indicated by &quot;*&quot;.
AVERAGE MEASURE is the observed, sample-dependent, average measure of persons in this analysis who
responded in this category (adjusted by PWEIGHT=). This is a quality-control statistic for this analysis. (It is not
the sample-independent value of the category, which is obtained by adding the item measure to the &quot;score at
category&quot;, in Table 3.2 or higher, for the rating (or partial credit) scale corresponding to this item.) For each
observation in category k, there is a person of measure Bn and an item of measure Di. Then: average measure =
sum( Bn - Di ) / count of observations in category.
An &quot;*&quot; indicates that the average measure for a higher score value is lower than for a lower score value. This
contradicts the hypothesis that &quot;higher score value implies higher measure, and vice-versa&quot;.
S.E. MEAN is the standard error of the mean (average) measure of the sample of persons who responded in this
category (adjusted by PWEIGHT=).
OUTFIT MEAN-SQUARE is the ratio of observed variance to expected variance for observed responses in this
category. Values greater than 1.0 indicate unmodeled noise. Values less than 1.0 indicate loss of information.
PTMEA CORR is the correlation between the occurence, scored 1, or non-occurrence, scored 0, of this category
or distractor and the person measures. With PTBIS=Yes, the correlation is between the occurrence and the
person raw score, indicated by PTBIS CORR. When this correlation is high positive for a correct MCQ option,
then the item exhibits convergent validity. When this correlation is low or negative for incorrect MCQ options, then
the item exhibits discriminant validity. Krus, D. J. &amp; Ney, R. G. (1978) Convergent and discriminant validity in item
analysis. Educational and Psychological Measurement, 38, 135-137.
ITEM (here, ACT) is the name or label of the item.
Data codes and Category labels are shown to the right of the box, if CLFILE= or CFILE= is specified.
228.
Table 12.2, 12.12 Item distribution maps
(controlled by MRANGE=, MAXPAG=, NAMLMP=, ISORT=)
In Table 12.2, the full item names are shown located at their calibrations, along with the person distribution. You
can use NAMLMP= to control the number of characters of each name reported. &quot;QSMSQ&quot; summarize the
distributions. An &quot;M&quot; marker represents the location of the mean measure. &quot;S&quot; markers are placed one sample
standard deviation away from the mean. &quot;T&quot; markers are placed two sample standard deviations away.
MAXPAG= controls the length of the Table. MRANGE= controls the displayed range of measures. ISORT=
controls the sort order within rows. You can adjust these values from the &quot;Specification&quot; pull-down menu and then
select Table 12 from the Output Tables menu. Thus you can experiment without needing to rerun the analysis.
Where there are more items than can be shown on one line, the extra items are printed on subsequent lines, but
the latent variable &quot;|&quot; does not advance and is left blank.
In subtable 12.12, the items are arranged by easiness. The item hierarchy is reversed.
Items arranged by measure: Look for the hierarchy of item names to spell out a meaningful construct from
easiest (highest p-value or highest average rating) at the bottom to hardest (lowest p-value or lowest average
rating) at the top.
KIDS MAP OF ACTS
&lt;more&gt;|&lt;rare&gt;
3
+
X |
206
X
2
XX
XX
XX
XXX
XX
|T
|
|
|
S+
|
|
FIND BOTTLES AND CANS
WATCH A RAT
WATCH BUGS
LOOK IN SIDEWALK CRACKS
WATCH GRASS CHANGE
(multiple items with same measure)
XXXX |
XX |S
XXX |
1
XXXXXX + WATCH ANIMAL MOVE
XXX M|
XXXX | LEARN WEED NAMES
MAKE A MAP
XXXX | TALK W/FRIENDS ABOUT PLANTS
XXXXXXXXXXX | LOOK AT PICTURES OF PLANTS
LOOK UP STRANGE ANIMAL OR PLANT
READ BOOKS ON PLANTS
XXXX | FIND OUT WHAT ANIMALS EAT
WATCH WHAT ANIMALS EAT
0
XXXXXXX +M
X |
X S| FIND OUT WHAT FLOWERS LIVE ON
WATCH BIRDS
| READ ANIMAL STORIES
XX | READ BOOKS ON ANIMALS
X | FIND WHERE ANIMAL LIVES
WATCH BIRD MAKE NEST
-1
X +
&lt;less&gt;|&lt;frequ&gt;
Subtable 12.12:
Items arranged by easiness: Look for the hierarchy of item names to spell out a meaningful construct from
easiest at the top to hardest at the bottom.
The double line || indicates the two sides have opposite orientations.
3
2
1
0
-1
229.
KIDS MAP OF ACTS
&lt;more&gt;||&lt;frequ&gt;
X T|| GO ON PICNIC
++
X ||
X ||T
|| GO TO ZOO
XX ||
XX ||
XX S++ GO TO MUSEUM
XXX ||
XX ||
XXXX || LISTEN TO BIRD SING
XX ||S
XXX || GROW GARDEN
XXXXXX ++
XXX M|| FIND WHERE ANIMAL LIVES
WATCH BIRD MAKE NEST
XXXX || READ BOOKS ON ANIMALS
XXXX || READ ANIMAL STORIES
XXXXXXXXXXX || FIND OUT WHAT FLOWERS LIVE ON
WATCH BIRDS
XXXX ||
XXXXXXX ++M
X || FIND OUT WHAT ANIMALS EAT
WATCH WHAT ANIMALS EAT
X S|| LOOK AT PICTURES OF PLANTS
LOOK UP STRANGE ANIMAL OR PLANT
READ BOOKS ON PLANTS
|| TALK W/FRIENDS ABOUT PLANTS
XX || LEARN WEED NAMES
MAKE A MAP
X ||
X ++ WATCH ANIMAL MOVE
&lt;less&gt;||&lt;rare&gt;
Table 12.5 Item map with expected score zones
This Table shows the items positioned at the lower edge of each expected score zone. The expected score zone
above label.2 extends from 1.5 to 2. Above label.1, the zone extends from 0.5 to 1.5. Lower than label.1 is the
207
zone from 0 to 0.5. If you put the item number at the start of the item labels after &amp;END, you can show only the
item numbers on this plot by using NAMLMP= or IMAP=. Column headings are the category labels that match the
(rescored) category numbers in CFILE=.
Where there are more items than can be shown on one line, the extra items are printed on subsequent lines, but
the latent variable &quot;|&quot; does not advance and is left blank.
5
4
3
2
1
KIDS MAP OF ACTS - Expected score zones
&lt;more&gt;| Neutral
X +
X |
|
|
|
|
X +
|
X |
|
T|
X |
X +
|T
X |
XX |
XX |
XX S|
XXX +
XX |
XX
|
XX
XX
|
|S
XXXXXX |
XXX +
XXX M|
XXXXX
XXX
XXXXXXXXXXX
XXXX
0
XXX
XXXXX
-1
-2
-3
-4
-5
230.
Like
|
|
FIND BOTTLES AND CANS
.2
WATCH A RAT
.2
WATCH BUGS
WATCH GRASS CHANGE
LOOK IN SIDEWALK CRACKS
.2
WATCH ANIMAL MOVE
.2
.2
LEARN WEED NAMES
.2
MAKE A MAP
LOOK AT PICTURES OF PLANTS
.2
READ BOOKS ON PLANTS
TALK W/FRIENDS ABOUT PLANTS
LOOK UP STRANGE ANIMAL OR PLANT.2
FIND OUT WHAT ANIMALS EAT
.2
WATCH WHAT ANIMALS EAT
FIND BOTTLES AND CANS
WATCH A RAT
.1
.1
WATCH BUGS
LOOK IN SIDEWALK CRACKS
WATCH GRASS CHANGE
.1
.1
|
|
+M
| WATCH ANIMAL MOVE
.1
S|
X | LEARN WEED NAMES
.1
XX | MAKE A MAP
.1
TALK W/FRIENDS ABOUT PLANTS
| LOOK AT PICTURES OF PLANTS
.1
LOOK UP STRANGE ANIMAL OR PLANT
READ BOOKS ON PLANTS
XX + WATCH WHAT ANIMALS EAT
.1
| FIND OUT WHAT ANIMALS EAT
.1
XX |S
T| FIND OUT WHAT FLOWERS LIVE ON .1
WATCH BIRDS
X | READ ANIMAL STORIES
.1
| READ BOOKS ON ANIMALS
.1
+ WATCH BIRD MAKE NEST
.1
| FIND WHERE ANIMAL LIVES
.1
|
| GROW GARDEN
.1
| LISTEN TO BIRD SING
.1
|T
+
| GO TO MUSEUM
.1
|
|
| GO TO ZOO
.1
|
+
|
| GO ON PICNIC
.1
|
|
|
+
&lt;less&gt;|&lt;frequ&gt;
FIND OUT WHAT FLOWERS LIVE ON
WATCH BIRDS
READ ANIMAL STORIES
READ BOOKS ON ANIMALS
.2
WATCH BIRD MAKE NEST
FIND WHERE ANIMAL LIVES
.2
.2
GROW GARDEN
LISTEN TO BIRD SING
.2
.2
GO TO MUSEUM
.2
GO TO ZOO
.2
GO ON PICNIC
.2
.2
.2
Table 12.6 Item map with 50% cumulative probabilities
This Table shows the items positioned at median 50% cumulative probability (the Rasch-Thurstone thresholds) at
the lower edge of each rating probability zone. Above label.2, the most probable category is 2. Below label.1, the
most probable category is 0. Between label.1 and label.2 is the zone which can be thought of as corresponding
to a rating of 1. If you put the item number at the start of the item labels after &amp;END, you can show only the item
208
numbers on this plot by using NAMLMP= or IMAP=. Columns are headed by the (rescored) categories in CFILE=.
Where there are more items than can be shown on one line, the extra items are printed on subsequent lines, but
the latent variable &quot;|&quot; does not advance and is left blank.
5
4
3
2
KIDS MAP OF ACTS - 50% Cumulative probabilities
&lt;more&gt;| Neutral
X +
X |
|
|
|
|
X +
|
X |
|
T|
X |
X +
|T
X |
XX |
XX |
XX S|
XXX +
XX |
XX |
XX
XX
XXXXXX
1
-2
-3
-4
-5
231.
|S FIND BOTTLES AND CANS
| WATCH A RAT
XXX +
XXX M|
XXXXX |
XXX
XXXXXXXXXXX
XXXX
0
XXX
XXXXX
-1
|
WATCH BUGS
LOOK IN SIDEWALK CRACKS
WATCH GRASS CHANGE
.1
.1
.1
.1
|
|
| WATCH ANIMAL MOVE
.1
+M
|
S| LEARN WEED NAMES
.1
MAKE A MAP
X | TALK W/FRIENDS ABOUT PLANTS
.1
XX | LOOK AT PICTURES OF PLANTS
.1
LOOK UP STRANGE ANIMAL OR PLANT
READ BOOKS ON PLANTS
| WATCH WHAT ANIMALS EAT
.1
XX + FIND OUT WHAT ANIMALS EAT
.1
|
XX |S FIND OUT WHAT FLOWERS LIVE ON .1
WATCH BIRDS
T| READ ANIMAL STORIES
.1
X | READ BOOKS ON ANIMALS
.1
| WATCH BIRD MAKE NEST
.1
+ FIND WHERE ANIMAL LIVES
.1
|
| GROW GARDEN
.1
| LISTEN TO BIRD SING
.1
|
|T
+ GO TO MUSEUM
.1
|
|
| GO TO ZOO
.1
|
|
+
| GO ON PICNIC
.1
|
|
|
|
+
&lt;less&gt;|&lt;frequ&gt;
Like
FIND BOTTLES AND CANS
.2
WATCH A RAT
.2
WATCH BUGS
WATCH GRASS CHANGE
LOOK IN SIDEWALK CRACKS
.2
WATCH ANIMAL MOVE
.2
.2
LEARN WEED NAMES
.2
MAKE A MAP
LOOK AT PICTURES OF PLANTS
.2
READ BOOKS ON PLANTS
TALK W/FRIENDS ABOUT PLANTS
LOOK UP STRANGE ANIMAL OR PLANT.2
FIND OUT WHAT ANIMALS EAT
.2
WATCH WHAT ANIMALS EAT
FIND OUT WHAT FLOWERS LIVE ON
WATCH BIRDS
READ ANIMAL STORIES
READ BOOKS ON ANIMALS
WATCH BIRD MAKE NEST
FIND WHERE ANIMAL LIVES
.2
GROW GARDEN
.2
LISTEN TO BIRD SING
.2
GO TO MUSEUM
.2
GO TO ZOO
.2
GO ON PICNIC
.2
.2
.2
.2
.2
Table 13.1, 14.1, 15.1, 25.1, 26.1 Item statistics
(controlled by USCALE=, UMEAN=, UDECIM=, LOCAL=, ISORT=, TOTAL=, DISCRIMINATION=,
ASYMPTOTE=, PVALUE=)
TAP STATISTICS:
MEASURE ORDER
+-------------------------------------------------------------------------------------------------------------------+
|ENTRY
RAW
MODEL|
INFIT | OUTFIT |PTMEA|EXACT MATCH|ESTIM| ASYMPTOTE | P- |
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.| OBS% EXP%|DISCR|LOWER UPPER|VALUE| TAP
|
|------------------------------------+----------+----------+-----+-----------+-----+-----------+-----+--------------|
|
18
0
34
6.14
1.84| MAXIMUM ESTIMATED MEASURE |
|
|
| .00| 4-1-3-4-2-1-4|
|
15
1
34
4.82
1.07| .75
-.1| .11
1.4| .25| 97.1 97.0| 1.18| .00
.00| .02| 1-3-2-4-1-3 |
209
|
16
1
34
4.82
1.07| .75
-.1| .11
1.4| .25| 97.1 97.0| 1.18| .00
.00| .02| 1-4-2-3-1-4 |
|
17
1
34
4.82
1.07| .75
-.1| .11
1.4| .25| 97.1 97.0| 1.18| .00
.00| .02| 1-4-3-1-2-4 |
|
14
3
34
3.38
.70|1.57
1.2|1.51
1.0| .26| 85.3 92.1| .61| .01 1.00| .06| 1-4-2-3-4-1 |
|
12
6
34
2.24
.55|1.17
.6|1.06
.5| .42| 85.3 86.8| .82| .00 1.00| .12| 1-3-2-4-3
|
|
13
7
34
1.95
.52| .70 -1.0| .38
-.3| .54| 88.2 84.6| 1.35| .00 1.00| .14| 1-4-3-2-4
|
|
11
12
34
.79
.45|1.08
.4| .79
-.1| .59| 76.5 79.2| .96| .00 1.00| .24| 1-3-1-2-4
|
|
10
24
34
-1.58
.48|1.07
.3| .83
.0| .79| 79.4 83.0| .96| .01 1.00| .48| 2-4-3-1
|
|
8
27
34
-2.36
.54| .59 -1.3| .43
-.2| .87| 94.1 86.5| 1.35| .00 1.00| .54| 1-4-2-3
|
|
6
29
33
-3.37
.65|1.18
.6| .98
.7| .84| 90.9 89.8| .85| .00 1.00| .59| 3-4-1
|
|
9
30
34
-3.39
.64| .62 -1.0| .21
.0| .89| 91.2 90.1| 1.34| .00 1.00| .60| 1-3-2-4
|
|
5
31
34
-3.85
.71|1.04
.2| .52
.6| .87| 88.2 91.8| 1.01| .00 1.00| .62| 2-1-4
|
|
7
31
34
-3.85
.71|1.34
.9|2.24
1.2| .82| 94.1 91.8| .54| .00
.98| .62| 1-4-3-2
|
|
4
32
34
-4.42
.81| .90
.0| .35
.8| .88| 94.1 94.0| 1.09| .00 1.00| .64| 1-3-4
|
|
1
34
34
-6.60
1.85| MINIMUM ESTIMATED MEASURE |
|
|
| 1.00| 1-4
|
|
2
34
34
-6.60
1.85| MINIMUM ESTIMATED MEASURE |
|
|
| 1.00| 2-3
|
|
3
34
34
-6.60
1.85| MINIMUM ESTIMATED MEASURE |
|
|
| 1.00| 1-2-4
|
|------------------------------------+----------+----------+-----+-----------+-----+-----------+-----+--------------|
| MEAN
18.7
33.9
-.76
.97| .96
.0| .69
.6|
| 89.9 90.0|
|
|
|
|
| S.D.
13.7
.2
4.27
.51| .28
.7| .59
.6|
| 6.3
5.3|
|
|
|
|
+-------------------------------------------------------------------------------------------------------------------+
Above the Table are shown the &quot;real&quot; separation and reliability coefficients from Table 3.
ENTRY NUMBER is the sequence number of the item, in your data, and is the reference number used for
deletion or anchoring.
RAW SCORE is the raw score corresponding to the parameter, i.e., the sum of the scored responses to an item
by the persons. If TOTALSCORE=Yes, then RAW SCORE includes extreme (zero, perfect) scores. If
TOTALSCORE=No, then RAW SCORE excludes extreme (zero, perfect) scores.
COUNT is the number of data points summed to make the RAW SCORE.
MEASURE is the estimate (or calibration) for the parameter, i.e., the item difficulty (b, D, delta, etc.). If the score
is extreme, a value is estimated, but as MAXIMUM (zero score) or MINIMUM (perfect score). No measure is
reported if the element is DROPPED (no valid observations remaining) or DELETED (you deleted the item).
The difficulty of an item is defined to be the point on the latent variable at which its high and low categories are
equally probable. SAFILE= can be used to alter this definition.
If unexpected results are reported, check whether TARGET= or CUTLO= or CUTHI= are specified.
INESTIMABLE is reported if all observations are eliminated as forming part of extreme response strings. To
make such measures estimable, further data (real or artificial) is required including both extreme and nonextreme observations.
MINIMUM ESTIMATE MEASURE - the sample obtained the extreme maximum (perfect) score on this item, so
it has been estimated with an extreme minimum measure, see Extreme scores. Fit statistics are not
computable, but they would correspond to perfect fit to the Rasch model.
MAXIMUM ESTIMATE MEASURE - the sample obtained the extreme minimum (zero) score on this item, so it
has been estimated with an extreme maximum measure, see Extreme scores. Fit statistics are not
computable, but they would correspond to perfect fit to the Rasch model.
ERROR is the standard error of the estimate. For anchored values, an &quot;A&quot; is shown on the listing and the error
reported is that which would have been obtained if the value had been estimated.
INFIT is a t standardized information-weighted mean square statistic, which is more sensitive to unexpected
behavior affecting responses to items near the person's measure level.
MNSQ is the mean-square infit statistic with expectation 1. Values substantially less than 1 indicate dependency
in your data; values substantially greater than 1 indicate noise. See dichotomous and polytomous fit statistics.
Value Meaning
&gt;2.0
Off-variable noise is greater than useful information. Degrades measurement.
&gt;1.5
Noticeable off-variable noise. Neither constructs nor degrades measurement
0.5 - 1.5
Productive of measurement
&lt;0.5
Overly predictable. Misleads us into thinking we are measuring better than we really are. (Attenuation
paradox.)
Always remedy the large misfits first. Misfits &lt;1.0 are only of concern when shortenning a test.
ZSTD is the infit mean-square fit statistic t standardized to approximate a theoretical &quot;unit normal&quot;, mean 0 and
210
variance 1, distribution. ZSTD (standardized as a z-score) is used of a t-test result when either the t-test value
has effectively infinite degrees of freedom (i.e., approximates a unit normal value) or the Student's tdistribution value has been adjusted to a unit normal value. The standardization is shown on RSA, p.100-101.
When LOCAL=Y, then EMP is shown, indicating a local {0,1} standardization. When LOCAL=LOG, then LOG
is shown, and the natural logarithms of the mean-squares are reported. More exact values are shown in the
Output Files.
Ben Wright advises: &quot;ZSTD is only useful to salvage non-significant MNSQ&gt;1.5, when sample size is
small or test length is short.&quot;
OUTFIT is a t standardized outlier-sensitive mean square fit statistic, more sensitive to unexpected behavior by
persons on items far from the person's measure level.
Value
&gt;2.0
&gt;1.5
0.5 - 1.5
&lt;0.5
Meaning
Off-variable noise is greater than useful information. Degrades measurement.
Noticeable off-variable noise. Neither constructs nor degrades measurement
Productive of measurement
Overly predictable. Misleads us into thinking we are measuring better than we really are.
(Attenuation paradox.)
Always remedy the large misfits first. Misfits &lt;1.0 are usually only of concern when shortenning a test.
MNSQ is the mean-square outfit statistic with expectation 1. Values substantially less than 1 indicate
dependency in your data; values substantially greater than 1 indicate the presence of unexpected outliers.
See dichotomous and polytomous fit statistics.
ZSTD is the outfit mean-square fit statistic t standardized similarly to the INFIT ZSTD. ZSTD (standardized as a zscore) is used of a t-test result when either the t-test value has effectively infinite degrees of freedom (i.e.,
approximates a unit normal value) or the Student's t-distribution value has been adjusted to a unit normal
value.
Ben Wright advises: &quot;ZSTD is only useful to salvage non-signficant MNSQ&gt;1.5, when sample size is
small or test length is short.&quot;
PTMEA CORR. (reported when PTBIS=N) is the point-measure correlation, rpm or RPM, between the
observations on an item (as fractions with the range 0,1) and the correspondng person measures, or viceversa. Since the point-biserial loses its meaning in the presence of missing data, specify PTBIS=N when data
are missing or CUTLO= or CUTHI= are specified.
PTBIS CORR (reported when PTBIS=Yes) is the point-biserial correlation, rpbis, between the individual item
response &quot;scores&quot; and the total person test score (less the individual item response &quot;scores&quot;). Negative values
for items often indicate mis-scoring, or rating (or partial credit) scale items with reversed direction. Letters
indicating the identity of items appearing on the fit plots appear under PTBIS. For adaptive tests, an rpbis near
zero is expected.
The formula for this product-moment correlation coefficient is:
rpbis = {sum {(x-x bar)(y-y bar)}} over {sqrt {{sum {(x-x bar)}&sup2;} {sum {(y-y bar)}&sup2;}} }
where x = observation for this item (or person), y = total score (including extreme scores) for person omitting
this item
EXACT MATCH
OBS% Observed% is the percent of data points which are within 0.5 score points of their expected values, i.e.,
that match predictions.
EXP% Expected% is the percent of data points that are predicted to be within 0.5 score points of their expected
values.
ESTIM DISCRIM is an estimate of the item discrimination, see DISCRIM=
ASYMPTOTE LOWER and UPPER are estimates of the upper and lower asymptotes for dichotomous items, see
ASYMPTOTE=
211
P-VALUE is the observed proportion correct or average rating, see PVALUE=
WEIGH is the weight assigned by IWEIGHT= or PWEIGHT=
DISPLACE is the displacement of the reported MEASURE from its data-derived value. This should only be shown
with anchored measures.
G is the grouping code assigned with ISGROUPS=
M is the model code assigned with MODELS=
232.
Table 16.3 Person distribution map
(controlled by MRANGE=, MAXPAG=, NAMLMP=, PSORT=)
In Table 16.3, the full person names are shown with an item distribution. You can use NAMLMP= to control the
number of characters of each name reported. &quot;QSMSQ&quot; summarize the distributions. An &quot;M&quot; marker represents
the location of the mean measure. &quot;S&quot; markers are placed one sample standard deviation away from the mean.
&quot;T&quot; markers are placed two sample standard deviations away. MAXPAG= controls the length of the Table.
MRANGE= controls the displayed range of measures. PSORT= controls the sort order within rows.
Where there are more persons than can be shown on one line, the persons are printed on subsequent lines, but
the latent variable &quot;|&quot; does not advance and is left blank.
Persons arranged by measure: Look for the hierarchy of person names to spell out a meaningful distribution
from highest scoring at the top to lowest scoring at the bottom.
4
3
2
1
0
ACTS MAP OF KIDS
&lt;rare&gt;|&lt;more&gt;
+ PASTER, RUTH
|
| SCHATTNER, GAIL
|
T|
| DOEPPNER, TOM
+ MCLOUGHLIN, BILLY
|T
| WRIGHT, BENJAMIN
X | BUFF, MARGE BABY
| CLAPP, LB
X S| ERNST, RICHARD MAX
+ CLAPP, CHARLIE
VROOM, JEFF
X | BADENOV, BORIS
XX | MOOSE, BULLWINKLE
| MAN, SPIDER
|S CORLEONE, VITO
X | BECKER, SELWIN
HOGAN, KATHLEEN
REISS, STEVE
+ BLOFELD, VILLAIN
STOLLER, DAVE
M| FONTANILLA, HAMES
TOZER, AMY ELIZABETH
XX | BABBOO, BABOO
PINHEAD, ZIPPY
X | ALLEN, PETER
ROSSNER, MARC DANIEL
XXX | AIREHEAD, JOHN
CIANCI, BUDDY
DYSON, STEPHIE NINA
HSIEH, DANIEL SEB
ROSSNER, MICHAEL T.
STULTZ, NEWELL
X | ANGUIANO, ROB
PAULING, LINUS
X +M AMIRAULT, ZIPPY
NEIMAN, RAYMOND
CHAZELLE, BERNIE
SEILER, KAREN
SQURREL, ROCKY J.
EASTWOOD, CLINT
KENT, CLARK
ROSSNER, BESS
PATRIARCA, RAY
SQUILLY, MICHAEL
CLAPP, DOCENT
HSIEH, PAUL FRED
ROSSNER, TOBY G.
LAMBERT, MD., ROSS W.
LEADER, FEARLESS
CORLEONE, MICHAEL
SABILE, JACK
MALAPROP, MRS.
SQUILLY, BAY OF
BEISER, ED
DENNY, DON
EISEN, NORM L.
RINZLER, JAMES
SANDBERG, RYNE
MULLER, JEFF
ROSSNER, JACK
DRISKEL, MICHAEL
212
|
-1
-2
233.
HWA, NANCY MARIE
ROSSNER, TR CAT
VAN DAM, ANDY
XX S|
X | BAUDET, GERARD
X | BOND, JAMES
X |
X + LIEBERMAN, DANIEL
|
X |S JACKSON, SOLOMON
X T|
| SABOL, ANDREW
|
X +
&lt;frequ&gt;|&lt;less&gt;
LIEBERMAN, BENJAMIN
SCHULZ, MATTHEW
ROSSNER, REBECCA A.
NORDGREN, JAN SWEDE
LANDMAN, ALAN
Table 17.1, 18.1, 19.1 Person statistics
(controlled by USCALE=, UMEAN=, UDECIM=, LOCAL=, PSORT=, TOTAL=)
KID STATISTICS: ENTRY ORDER
+---------------------------------------------------------------------------+
|ENTRY
TOTAL
MODEL|
INFIT | OUTFIT |PTMEA|
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.| KID
|
|------------------------------------+----------+----------+-----+----------|
|
1
7
18
-2.94
.82| .61 -1.2| .29
2.0| .80| Richard M|
......
|
34
12
18
1.94
.98|1.74
1.4| .71
.9| .77| Elsie
F|
|
35
3
18
-6.62
1.85| MINIMUM ESTIMATED MEASURE | Helen
F|
|------------------------------------+----------+----------+-----+----------|
| MEAN
6.7
14.0
-.37
1.03| .99
-.2| .68
.9|
|
|
| S.D.
2.4
.0
2.22
.17| .94
1.2|1.29
1.1|
|
|
+---------------------------------------------------------------------------+
Above the Table are shown the &quot;real&quot; separation and reliability coefficients from Table 3.
ENTRY NUMBER is the sequence number of the person in your data, and is the reference number used for
deletion or anchoring.
RAW SCORE is the raw score corresponding to the parameter, i.e., the raw score by a person on the test. If
TOTALSCORE=Yes, then RAW SCORE includes extreme (zero, perfect) scores. If TOTALSCORE=No, then
RAW SCORE excludes extreme (zero, perfect) scores.
COUNT is the number of data points summed to make the RAW SCORE.
MEASURE is the estimate for the parameter, i.e., person ability (theta, B, beta, etc.). If the score is extreme, a
value is estimated, but as MAXIMUM (perfect score) or MINIMUM (zero score). No measure is reported if the
element is DROPPED (no valid observations remaining) or DELETED (you deleted the person or item). The
difficulty of an item is defined to be the point on the latent variable at which its high and low categories are
equally probable. SAFILE= can be used to alter this definition.
If unexpected results are reported, check whether TARGET= or CUTLO= or CUTHI= are specified.
INESTIMABLE is reported if all observations are eliminated as forming part of extreme response strings. To
make such measures estimable, further data (real or artificial) is required including both extreme and nonextreme observations.
MINIMUM ESTIMATE MEASURE - the person obtained the extreme minimum (zero) score on the items, so
has been estimated with an extreme minimum measure, see Extreme scores. Fit statistics are not computable,
but they would correspond to perfect fit to the Rasch model.
MAXIMUM ESTIMATE MEASURE - the person obtained the extreme maximum (perfect) score on the items,
so has been estimated with an extreme maximum measure, see Extreme scores. Fit statistics are not
computable, but they would correspond to perfect fit to the Rasch model.
ERROR is the standard error of the estimate. For anchored values, an &quot;A&quot; is shown on the listing and the error
reported is that which would have been obtained if the value had been estimated.
INFIT is a t standardized information-weighted mean square statistic, which is more sensitive to unexpected
behavior affecting responses to items near the person's measure level.
213
MNSQ is the mean-square infit statistic with expectation 1. Values substantially less than 1 indicate dependency
in your data; values substantially greater than 1 indicate noise. See dichotomous and polytomous fit statistics.
Value Meaning
&gt;2.0
Off-variable noise is greater than useful information. Degrades measurement.
&gt;1.5
Noticeable off-variable noise. Neither constructs nor degrades measurement
0.5 - 1.5
Productive of measurement
&lt;0.5
Overly predictable. Misleads us into thinking we are measuring better than we really are. (Attenuation
paradox.)
Always remedy the large misfits first. Misfits &lt;1.0 are only of concern when shortenning a test.
ZSTD is the infit mean-square fit statistic t standardized to approximate a theoretical &quot;unit normal&quot;, mean 0 and
variance 1, distribution. ZSTD (standardized as a z-score) is used of a t-test result when either the t-test value
has effectively infinite degrees of freedom (i.e., approximates a unit normal value) or the Student's tdistribution value has been adjusted to a unit normal value. The standardization is shown on RSA, p.100-101.
When LOCAL=Y, then EMP is shown, indicating a local {0,1} standardization. When LOCAL=LOG, then LOG
is shown, and the natural logarithms of the mean-squares are reported. More exact values are shown in the
Output Files.
Ben Wright advises: &quot;ZSTD is only useful to salvage non-significant MNSQ&gt;1.5, when sample size is
small or test length is short.&quot;
OUTFIT is a t standardized outlier-sensitive mean square fit statistic, more sensitive to unexpected behavior by
persons on items far from the person's measure level.
Value
&gt;2.0
&gt;1.5
0.5 - 1.5
&lt;0.5
Meaning
Off-variable noise is greater than useful information. Degrades measurement.
Noticeable off-variable noise. Neither constructs nor degrades measurement
Productive of measurement
Overly predictable. Misleads us into thinking we are measuring better than we really are.
(Attenuation paradox.)
Always remedy the large misfits first. Misfits &lt;1.0 are usually only of concern when shortenning a test.
MNSQ is the mean-square outfit statistic with expectation 1. Values substantially less than 1 indicate
dependency in your data; values substantially greater than 1 indicate the presence of unexpected outliers.
See dichotomous and polytomous fit statistics.
ZSTD is the outfit mean-square fit statistic t standardized similarly to the INFIT ZSTD. ZSTD (standardized as a zscore) is used of a t-test result when either the t-test value has effectively infinite degrees of freedom (i.e.,
approximates a unit normal value) or the Student's t-distribution value has been adjusted to a unit normal
value.
Ben Wright advises: &quot;ZSTD is only useful to salvage non-signficant MNSQ&gt;1.5, when sample size is
small or test length is short.&quot;
PTMEA CORR. (reported when PTBIS=N) is the point-measure correlation, rpm or RPM, between the
observations on an item (as fractions with the range 0,1) and the correspondng item measures. Since the
point-biserial loses its meaning in the presence of missing data, specify PTBIS=N when data are missing or
CUTLO= or CUTHI= are specified.
PTBIS CORR (reported when PTBIS=Yes) is the point-biserial correlation, rpbis, between the individual person
response &quot;scores&quot; and the total item test score (less the individual person response &quot;scores&quot;). Letters
indicating the identity of persons appearing on the fit plots appear under PTBIS. For adaptive tests, an rpbis
near zero is expected.
The formula for this product-moment correlation coefficient is:
rpbis = {sum {(x-x bar)(y-y bar)}} over {sqrt {{sum {(x-x bar)}&sup2;} {sum {(y-y bar)}&sup2;}} }
where x = observation for this person, y = total score (including extreme scores) for item omitting this
214
person.
WEIGH is the weight assigned by IWEIGHT= or PWEIGHT=
DISPLACE is the displacement of the reported MEASURE from its data-derived value. This should only be shown
with anchored measures.
234.
Table 20 Complete score-to-measure table on test of all items
A measure and standard error is estimated for every possible score on a test composed of all non-extreme items
included in the analysis. This can also be written with SCOREFILE=. The measures corresponding to extreme
scores (all items right, or all items wrong) are marked by &quot;E&quot; and estimated using the EXTRSC= criterion. A
graph of the score to measure conversion is also reported. `*' indicates the conversion. Since the 'S' and 'F'
models specify that not all item levels are encountered, measures complete tests are only approximated here. In
the Table of Measures on Complete Test:
SCORE
raw score on a complete test containing all calibrated items.
MEASURE
measure corresponding to score. The convergence criterion used are LCONV=*.01 and
RCONV=*.01 - these are considerably tighter than for the main analysis. So Table 20 is a more
precise estimate of the person measures based on the final set of item difficulties. If
STBIAS=YES or XMLE=YES are used, then score table measures are dependent on the data
array, even if items are anchored.
S.E.
standard error of the measure (model).
Table 20.1 gives the score-to-measure conversion for a complete test, when going from the y-axis (score) to the
x-axis (measure). When going from the x-axis (measure) to the y-axis(score), it predicts what score on the
complete test is expected to be observed for any particular measure on the x-axis. For CAT tests and the like, no
one takes the complete test, so going from the y-axis to the x-axis does not apply. But going from the x-axis to
the y-axis predicts what the raw score on the complete bank would have been, i.e., the expected total score, if the
whole bank had been administered.
Score-to-measure tables for subtests can be produced by using ISELECT= or IDELETE= before requesting Table
20.
TEST SLOPE and INTERCEPT
These are estimated from the relationship:
log ( SCORE / (MAXIMUM - SCORE) ) = TEST SLOPE * (MEASURE - INTERCEPT) / USCALE
with extreme scores (zero and maximum) omitted.
TEST SLOPE is the slope of the Test ICC relative to a standard logistic ogive.
INTERCEPT usually approximates the mean item difficulty.
TABLE OF MEASURES ON COMPLETE TEST
+--------------------------------------------------------------------------+
| SCORE MEASURE
S.E. | SCORE MEASURE
S.E. | SCORE MEASURE
S.E. |
|------------------------+------------------------+------------------------|
|
0
-6.17E
1.83 |
5
-1.97
.83 |
10
2.55
.89 |
|
1
-4.86
1.08 |
6
-1.19
.92 |
11
3.37
.89 |
|
2
-3.94
.85 |
7
-.23
1.00 |
12
4.21
.93 |
|
3
-3.27
.79 |
8
.80
.97 |
13
5.23
1.12 |
|
4
-2.64
.78 |
9
1.72
.92 |
14
6.60E
1.84 |
+--------------------------------------------------------------------------+
CURRENT VALUES, UMEAN=0.00 USCALE=1.00
TO SET MEASURE RANGE AS 0-100, UMEAN=48.307 USCALE=7.827
TO SET MEASURE RANGE TO MATCH RAW SCORE RANGE, UMEAN=6.763 USCALE=1.096
TEST SLOPE=.40 INTERCEPT=-.02
E
X
RAW SCORE-MEASURE OGIVE FOR COMPLETE TEST
++-------+-------+-------+-------+-------+-------+-------++
14 +
E +
13 +
*
+
12 +
*
+
215
P
E
C
T
E
D
S
C
O
R
E
11
10
9
8
7
6
5
4
3
2
1
0
+
*
+
+
*
+
+
*
+
+
*
+
+
*
+
+
*
+
+
*
+
+
*
+
+
*
+
+
*
+
+
*
+
+
E
+
++-------+-------+-------+-------+-------+-------+-------++
-7
-5
-3
-1
1
3
5
7
MEASURE
KIDS
1 2
T
TAPS
T
1 22
S
2 2
3
S
1
1
2
M
5
4
1
S
1
1
M
11
2
T
1
S
3
T
The statistical information is (USCALE/S.E.)&sup2; For the test information function, plot (USCALE/S.E.)&sup2; against
the &quot;MEASURE&quot; or the &quot;SCORE&quot;. The plot can be drawn with EXCEL. It is easier to obtain the information
directly from the SCOREFILE=.
TABLE 20.2
TABLE OF SAMPLE NORMS (500/100) AND FREQUENCIES CORRESPONDING TO COMPLETE TEST
+---------------------------------------------------------------------------+
| SCORE
MEASURE
S.E.|NORMED S.E. FREQUENCY %
CUM.FREQ. % PERCENTILE|
|------------------------+--------------------------------------------------|
|
0
-6.17E
1.83| 147 107
0
.0
0
.0
0
|
|
1
-4.86
1.08| 225
63
0
.0
0
.0
0
|
|
2
-3.94
.85| 278
50
1
2.9
1
2.9
1
|
|
3
-3.27
.79| 318
46
2
5.9
3
8.8
6
|
|
4
-2.64
.78| 355
46
2
5.9
5 14.7
12
|
|
5
-1.97
.83| 394
49
2
5.9
7 20.6
18
|
|
6
-1.19
.92| 440
54
3
8.8
10 29.4
25
|
|
7
-.23
1.00| 496
59
12 35.3
22 64.7
47
|
|
8
.80
.97| 557
57
5 14.7
27 79.4
72
|
|
9
1.72
.92| 610
54
4 11.8
31 91.2
85
|
|
10
2.55
.89| 660
52
1
2.9
32 94.1
93
|
|
11
3.37
.89| 707
52
2
5.9
34 100.0
97
|
|
12
4.21
.93| 756
54
0
.0
34 100.0
100
|
|
13
5.23
1.12| 817
66
0
.0
34 100.0
100
|
|
14
6.60E
1.84| 897 108
0
.0
34 100.0
100
|
+---------------------------------------------------------------------------+
216
The columns in the Table of Sample Norms and Frequencies are:
Measures on the Complete Test:
SCORE
raw score on a complete test containing all calibrated items.
MEASURE measure corresponding to score.
If a person did not take all items or items are weighted, then that person is stratified with the measure on
the complete test nearest the person's estimated measure (as reported in Table 18), regardless of that
person's observed score.
S.E.
standard error of the measure (model).
The statistical information is (USCALE/S.E.)
2
Statistics for this sample:
NORMED measures linearly locally-rescaled so that the mean person measure for this sample is 500 and the
sample measure sample standard deviation is 100. Equivalent to UPMEAN=500, USCALE=100/(Person S.D.)
S.E.
standard error of the normed measure.
FREQUENCY count of sample with measures at or near (for missing data) the complete test measure
%
percentage of sample included in FREQUENCY.
CUM.FREQ.
count of sample with measures near or below the test measure, the cumulative frequency.
%
percentage of sample include in CUM. FREQ.
PERCENTILE mid-range percentage of sample below the test measure, constrained to the range 1-99.
Logit measures support direct probabilistic inferences about relative performances between persons and absolute
performances relative to items. Normed measures support descriptions about the location of subjects within a
sample (and maybe a population). Report the measures which are most relevant to your audience.
This Table is easy to paste into Excel. use Excel's &quot;data&quot;, &quot;text to columns&quot; feature to put the scores and
measures into columns.
Table 20 shows integer raw scores, unless there are decimal weights for IWEIGHT=. In which case, scores to 1
decimal place are shown.
To obtain other decimal raw scores for short tests, go to the Graphs pull-down menu. Select &quot;Test Characteristic
Curve&quot;. This displays the score-to-measure ogive. Click on &quot;Copy data to clipboard&quot;. Open Excel. Paste. There
will be to three columns. The second column is the measure, the third column is the raw score.
Score-to-measure Table 20 can be produced from known item measures (IFILE=) and, for polytomies, known
rating scale structure difficulties (SFILE=).
IAFILE= (IFILE=) ; the item anchor file
SAFILE= (SFILE=) ; the structure/step anchor file (if not dichotomies)
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
STBIAS=NO
; no estimation bias correction with anchor values
The data file comprises two dummy data records, so that every item has a non extreme score, e.g.,
For dichotomies:
Record 1: 10101010101
Record 2: 01010101010
For a rating scale from 1 to 5:
Record 1: 15151515151
Record 2: 51515151515
235.
Table 20.3 Complete score-to-calibration table for tests based on whole sample
This Table, which must be selected explicitly with TFILE=, or as a subtable with &quot;Request Subtable&quot; on the
&quot;Output Tables&quot; menu, shows an estimated item calibration for all possible rates of success by persons, i.e., the
item measure corresponding to every observable p-value for the entire sample. To select this Table, use &quot;Request
subtables&quot; in the Output Tables menu,
217
Output Tables
Request Subtables
20.3
OK
or enter into your control file:
TFILE=*
20.3
*
TABLE OF ITEM MEASURES ON COMPLETE SAMPLE
FOR GROUPING &quot;0&quot;, MODEL &quot;R&quot;, ACT
NUMBER:
1 WATCH BIRDS
+--------------------------------------------------------------------------+
| SCORE MEASURE
S.E. | SCORE MEASURE
S.E. | SCORE MEASURE
S.E. |
|------------------------+------------------------+------------------------|
|
0
8.71E
1.81 |
50
1.91
.22 |
100
-.51
.22 |
|
1
7.47
1.02 |
51
1.86
.22 |
101
-.56
.22 |
|
2
6.72
.74 |
52
1.82
.22 |
102
-.61
.22 |
|
3
6.26
.62 |
53
1.77
.22 |
103
-.66
.23 |
+--------------------------------------------------------------------------+
SCORE raw score on this item of a complete sample containing all calibrated persons.
MEASURE measure corresponding to score.
S.E.
standard error of the measure.
The statistical information is (USCALE/S.E.)&sup2;
If an item is not taken by all persons or persons are weighted, then that item is stratified with the measure on the
complete sample nearest the estimated measure, regardless of the observed score.
236.
Table 21 Probability curves
(controlled by MRANGE=, CURVES=)
To produce these using other software, see GRFILE=
The probability of each response is shown across the measurement continuum. The measure to be used for
determining the probability of any particular response is the difference between the measure of the person and
the calibration of the item. For dichotomies, only one curve is shown plotting the probability of scoring a &quot;1&quot;
(correct), and also of scoring a &quot;0&quot; (incorrect) for any measure relative to item measure. For 'S' and 'F' models
these curves are approximations.
P
R
O
B
A
B
I
L
I
T
Y
1.0
.8
.6
.4
.2
.0
DICHOTOMOUS CURVES
------------------------------------------------------------0000000000000000
1111111111111111
000000
111111
000
111
00
11
00
11
***
11
00
11
00
111
000
111111
000000
1111111111111111
0000000000000000
-------------------------------------------------------------6
-4
-2
0
2
4
6
PERSON [MINUS] ITEM MEASURE
When there are more than two categories, the probability of each category is shown.
P
R
O
B
1.0
.8
CATEGORY PROBABILITIES: MODES - Structure measures at intersections
------------------------------------------------------------00000000
22222222
0000000
2222222
000
222
218
A
B
I
L
I
T
Y
000
222
00
22
00*111111111*22
111 00
22 111
111
00 22
111
.2
1111
22*00
1111
1111111
22222
00000
1111111
.0 ********222222222222222
000000000000000********
-------------------------------------------------------------5
-4
-3
-2
-1
0
1
2
3
4
5
PERSON [MINUS] ITEM MEASURE
.6
.5
.4
For response structures with three or more categories, two further graphs can be drawn. The second graph
depicts the expected score ogive. The vertical &quot;*&quot; characters correspond to integer expected scores, and the &quot;|&quot;
characters correspond to half-point expected scores, the Rasch-half-point thresholds. The intervals between the
Rasch-half-point thresholds can be thought of as the intervals corresponding to the observed categories. For the
purposes of inference, measures in the zone on the x-axis between `|' and `|' correspond, on average, to the
rating given on the `y' axis, `1'. Similarly ratings on the y-axis can be thought of as corresponding to measures in
the matching zone on the x-axis. The degree to which the data support this is given by the COHERENCE
statistics in Table 3. Empirical item characteristic curves are shown in Table 29 and from the Graphs menu.
E
X
P
E
C
T
E
D
2
1.5
1
.5
S
C
O
R
E
0
EXPECTED SCORE OGIVE: MEANS
------------------------------------------------------------222222222222
2222222
2222
111
111 |
111
|
111 *
|
111
*
|
0000|
*
|
0000000
|
*
|
000000000000
|
*
|
-------------------------------------------------------------5
-4
-3
-2
-1
0
1
2
3
4
5
PERSON [MINUS] ITEM MEASURE
The third graph is of the zone curves which indicate the probability of an item score at or below the stated
category for any particular difference between person measure and item calibration. The area to the left of the &quot;0&quot;
ogive corresponds to &quot;0&quot;. The right-most area corresponds to the highest category. The P=0.5 intercepts are the
median cumulative probabilities. &quot;|&quot; indicate the Rasch-Thurstone thresholds.
P
R
O
B
A
B
I
L
I
T
Y
1.0
.8
.6
.5
.4
.2
.0
237.
MEDIANS - Cumulative probabilities
------------------------------------------------------------********222222222222222
0
1111111
22222
0
111
222
0
111
22
0
11
22
0----------------------111---------222----------------------0
| 11
| 22
0
|
11
|
222
0
|
111
|
222
0
|
11111
2222222
0
|
| 111111111111111********
-------------------------------------------------------------5
-4
-3
-2
-1
0
1
2
3
4
5
PERSON [MINUS] ITEM MEASURE
Table 22.1 Sorted observed data matrix (scalogram)
The observations are printed in order of person and item measures, with most able persons listed first, the easiest
items printed on the left. This scalogram shows the extent to which a Guttman pattern is approximated. The
zoned responses are in Table 22.2 and the original responses in Table 22.3.
GUTTMAN SCALOGRAM OF RESPONSES:
PERSON ITEM
1111112 1 221 1 21
22
GUTTMAN SCALOGRAM OF ZONED RESPONSES:
PUPIL |ACT
|111111 2 1221 121
2 2
219
8920311251427643569784035
41
17
45
40
65
1
2222222222222222222222212
2222222222222222222222210
2222222222222222222221200
2222222222222122212101100
2222222211011101020101122
2222222221211001011201001
1111112211221613219784225
8920311 5 427 4 56
03
238.
41
17
45
40
65
1
|8901231125427634569780435
|------------------------+22222222222222222222222B2
+22222222222222222222222BA
+22222222222222222222CC1AA
+2222222222222B222121A11AA
+222222122BA1111AACA1A11CC
+222222122CC11A1AA11CA010B
|------------------------|1111111221221631219782425
|890123 1 5427 456
0 3
Table 22.2 Guttman scalogram of zoned responses.
The scalogram is that of Table 22.1, but with each observation marked as to whether it conforms with its
expectation or not. Observations within 0.5 rating points of their expectation are deemed to be in their expected
categories, and are reported with their category values, e.g., `1', `2', etc. These ratings support the overall
inferential relationship between observations and measures. Observations more than 0.5 rating points away from
their expectations, i.e., in a &quot;wrong&quot; category, are marked with a letter equivalent: `A' = `0',`B' = `1',`C' = `2', etc.
These contradict observation-to-measure inferences. The proportion of in- and out-of-category observations are
reported by the COHERENCE statistics in Table 3.
GUTTMAN SCALOGRAM OF RESPONSES:
PERSON ITEM
1111112 1 221 1 21
22
8920311251427643569784035
41
17
45
40
65
1
2222222222222222222222212
2222222222222222222222210
2222222222222222222221200
2222222222222122212101100
2222222211011101020101122
2222222221211001011201001
1111112211221613219784225
8920311 5 427 4 56
03
239.
GUTTMAN SCALOGRAM OF ZONED RESPONSES:
PUPIL |ACT
|111111 2 1221 121
2 2
|8901231125427634569780435
|------------------------41 +22222222222222222222222B2
17 +22222222222222222222222BA
45 +22222222222222222222CC1AA
40 +2222222222222B222121A11AA
65 +222222122BA1111AACA1A11CC
1 +222222122CC11A1AA11CA010B
|------------------------|1111111221221631219782425
|890123 1 5427 456
0 3
Table 22.3 Guttman scalogram of original codes
The observations are printed in order of person and item measures, with most able persons listed first, the easiest
items printed on the left. This scalogram shows the original codes in the data file. Here is the Scalogram for
Example 5, a computer-adaptive, multiple-choice test.
GUTTMAN SCALOGRAM OF ORIGINAL RESPONSES:
STUDENT |TOPIC
| 11 3 11 1 3212232212132425654421434145625 36555366356465464633654
|640215038189677748390992315641640517264579268221076889430372139553458
|--------------------------------------------------------------------3 +
d
b
c
db
d
c b cd
6 +
c
b
c
d
d c
cd
27 +
c c c
c
cbb a dd c b dd
4 +
c
b
a d cb c dd
d b d dc
9 +
d
c
a a cbc
c ba b dca c db
dc bbdbb
c
25 +
c b
d c
bc
a
a d a c dba d
cbccbd c
24 +
a
d
ac
b
c d
c b a
a
12 +
c b
c
c
b b
b
b a
cb cc c
28 +
c
c a bcc b c acc d
b bc
b c bdcb
d
21 + b
c b a
c b
a bc bb ab a db bbd
cc
b
b
11 +
a d c b cc cbad d ca c
d
ca ba b ac
bac
bc d c cc
17 +
d
d c
c
acac cac b c a
a b
c c bc b ac ac
7 +
d c d ccaaac d bcd ca
b cb c b a a a a ab
b
bb ac
1 +
c
c
ab
ca
dc b
a a
c
a
c
20 +
bd
b
c b cad cd a
bac
b
a b
c
b
22 + d
a d c b
d cdabcc bbc cbdb a
a dbac
c dc
d a
30 +
d
a a cab cdd
d d
da
c
d c
c
db
14 + b
d c db caa caa
da c
dcd b c
a cd
d
c c
d b
23 +
d cab dcaa cadbbbba a
c a db
c b
a
c c
ab c
26 +
d
a
cc b ac cdd b
a d db a d
b d
d
cb c
15 + d
d
a a caba dcbddc c
d b
b a
c
bb
13 + d
d
d c bac dc bd
d dcd
d a
d d c b
b
29 +b dbd bdcabd baaacbd c d c
ab
a
dd
d
c b c
8 +bbdadbd abaa acbcbccacbcadab
c
d cdab a d
a a c c
cb
16 +b da b bdccbbbcccbadaabba ccd
c ac bd a a
b
18 +b a a d bca bc a
cccc a
db
c
a a
d
d
10 +b badbdbcccadacbababcda cba a
a acd a
d
a
b
19 +bbdad
dacbdb cabdbc c ad
d ab b
c a
b
2 +bbda a d aaaa c c a
cbd
ab
a
d
c
5 +
aad a
ab
a b
a
|---------------------------------------------------------------------
220
S
S
S
S
S
S
AP
A
IH
AP
S
IH
IH
IM
IL
IM
IL
IL
IL
IL
NH
NH
NH
IL
NM
NH
NM
NM
NM
NL
STA
CHU
ERI
CHE
MEL
DAN
SAR
ALL
GRE
MIL
CHR
FRA
JAN
CAT
JAM
BAR
HOL
TAK
MEI
MIT
MAR
KEL
RIC
ROB
HEL
JOE
JON
THE
KAT
SMI
| 11 3 11 1 3212232212132425654421434145625 36555366356465464633654
|640215038189677748390992315641640517264579268221076889430372139553458
240.
Table 23.0, 24.0 Variance components scree plot for items or persons
This Table shows the variance decomposition in the observations.
STANDARDIZED RESIDUAL VARIANCE SCREE PLOT
Table of STANDARDIZED RESIDUAL variance (in Eigenvalue units)
Empirical
Modeled
Total variance in observations
=
142.7 100.0%
100.0%
Variance explained by measures
=
117.7 82.5%
84.5%
Unexplained variance (total)
=
25.0 17.5% 100.0%
15.5%
Unexpl var explained by 1st contrast =
4.5
3.1% 17.8%
Unexpl var explained by 2nd contrast =
3.0
2.1% 11.9%
Unexpl var explained by 3rd contrast =
2.4
1.7%
9.6%
Unexpl var explained by 4th contrast =
1.7
1.2%
6.9%
Unexpl var explained by 5th contrast =
1.6
1.1%
6.4%
Table of STANDARDIZED RESIDUAL variance: the standardized residuals form the basis of this computation,
set by PRCOMP=
(in Eigenvalue units): variance components are rescaled so that the total unexplained variance has its expected
summed eigenvalue.
Empirical: variance components for the observed data
Model: variance components expected for the data if exactly fit the Rasch model
Total variance in observations: total variance in the observations around their Rasch expected values in
standardized residual units
Variance explained by measures: variance explained by the item difficulties, person abilities and rating scale
structures.
Unexplained variance (total): variance not explained by the Rasch measures
Unexpl var explained by 1st, 2nd, ... contrast: size of the first, second, ... contrast (component) in the principal
component decomposition of residuals
Variance Explained
In this example, 82.5% is explained by the measures based on the standardized residuals from the observations.
If the data fit the model perfectly, 84.5% would be explained. The unexplained variance in the data is 17.5%. This
includes the Rasch-predicted randomness and any departures in the data from Rasch criteria, e.g.,
multidimensionality.
The variance terms are computed in the following way. The average of the responses made by all persons to all
items is computed. Extreme scores and missing data are omitted. Adjustment is made of number of categories in
polytomies. This average &quot;standard&quot; response is what every response would be if all items were equally difficult
and all persons were equally able. The variance to be explained by the Rasch measures is caused by the spread
of the item difficulties, person abilities, and disparate polytomous structures.
i) The empirical total observed response variance is the sum of the (observed - standard responses)**2.
ii) The modeled variance explained by the measures is the sum of (Rasch expectations - standard responses)**2.
iii) The modeled unexplained variance is sum of (Rasch-model response variances).
iv) The modeled total observed response variance is the sum of (ii) and (iii).
(v) The empirical unexplained variance is the larger of:
(a) The sum of the (observed - Rasch expectation)**2.
(b) The absolute value of (i) - (ii).
(vi) The empirical explained variance is (i) - (v).
All values are locally-rescaled so that the total unexplained variance matches the sum of the eigenvalues, i.e., the
variance to be explained, by the PCA of residuals. In general, the empirical variances are more conservative,
misfit- or dimensionality-attenuated, values. The modeled values correspond to unidimensional data fitting the
221
Rasch model. This is analgous to &quot;real&quot; and &quot;model&quot; standard errors.
Effect of misfit or gross dimensionality
Here is part of Table 23 for an analysis which contains 3 items negatively-correlated with the others - symptomatic
of miskeying or gross multi-dimensionality. Tthe empirical explained variance is conspicuously less than the
model explained variance.
Table of STANDARDIZED RESIDUAL variance (in Eigenvalue units)
Empirical
Modeled
Total variance in observations
=
25.7 100.0%
100.0%
Variance explained by measures
=
5.7
22.1%
47.8%
Unexplained variance (total)
=
20.0
77.9%
52.2%
Unexpl var explained by 1st contrast =
5.3
20.6%
VARIANCE COMPONENT SCREE PLOT
+--+--+--+--+--+--+--+--+--+
100%+ T
+
|
M
|
V 63%+
+
A
|
|
R 40%+
+
I
|
|
A 25%+
+
N
|
|
C 16%+
U
+
E
|
|
10%+
+
L
|
|
O 6%+
+
G
|
|
| 4%+
+
S
|
|
C 3%+
1
+
A
|
2
|
L 2%+
3
+
E
|
|
D 1%+
4 5 +
|
|
0.5%+
+
+--+--+--+--+--+--+--+--+--+
TV MV UV U1 U2 U3 U4 U5
VARIANCE COMPONENTS
Scree plot of the variance component percentage sizes, logarithmically scaled.
T, TV: total variance in the observations, always 100%
M, MV: variance in the observations explained by the Rasch measures
U, UV: unexplained variance
1, U1: first contrast (component) in the residuals
2, U2: second contrast (component) in the residuals, etc.
241.
Table 23.2, 24.2 Principal components plots of item or person loadings
Please do not interpret this as a usual factor analysis. These plots show contrasts between opposing
factors, not loadings on one factor. For more discussion, see dimensionality and contrasts.
Quick summary:
(a) the X-axis is the measurement axis. So we are not concerned about quadrants, we are concerned about
vertical differences. The Table 23 plots show contrasts between types of items: those at the top vs. those at the
bottom. The Table 24 plots show contrasts between types of persons: those at the top vs. those at the bottom.
(b) &quot;How much&quot; is important. See the Variance Table explained in Table 23.0. Important differences have
eigenvalues greater than 2.0.
(c) If the difference is important, it suggests that we divide the test into two pieces: the items in the top half of the
plot and the items in the bottom half. Perform two separate analyses and cross-plot and correlate the person
measures. We will then see for whom the differences are important. Usually, for a carefully designed instrument,
222
it is such a small segment that we decide it is not worth thinking of the test as measuring two dimensions. Tables
23.4 and 24.4 also help us think about this.
These plots show the contrasts by plotting the loading on each component against the item calibration (or person
measure). The contrast shows items (or persons) with different residual patterns. A random pattern with few high
loadings is expected.
The horizontal axis is the Rasch dimension. This has been extracted from the data prior to the analysis of
residuals.
Letters &quot;A&quot; and &quot;a&quot; identify items (persons) with the most opposed loadings on the first contrast in the residuals.
On subsequent contrasts, the items retain their first contrast identifying letters.
In the residuals, each item (person) is modeled to contribute one unit of randomness. Thus, there are as many
residual variance units as there are items (or persons). For comparison, the amount of person (item) variance
explained by the item (person) measures is approximated as units of that same size.
In this example, based on the FIM™ data, Example 10 using exam10hi.txt data, the first contrast in the
standardized residuals separates the items into 3 clusters. To identify the items, see Tables 23.3, 24.3. You will
see that items A and B have a psycho-social aspect.
In this example, the dimension is noticeable, with strength of around 3 out of 13 items. This is in the residual
variance, i.e., in the part of the obsesrvations unexplained by the measurement model. But, hopefully, most of the
variance in the observations has been explained by the model. The part of that explained variance attributable to
the Persons is shown in variance units locally-rescaled to accord with the residual variances. In this example,
the variance explained by the measures is equivalent to 16 items. Consequently, though the secondary
dimension in the items is noticeable, we do not expect it to have much practical impact on person measurement.
For items:
STANDARDIZED RESIDUAL CONTRAST 1 PLOT
Table of STANDARDIZED RESIDUAL variance (in Eigenvalue units)
Empirical
Modeled
Total variance in observations
=
29.5 100.0%
100.0%
Variance explained by measures
=
16.5 56.0%
56.3%
Unexplained variance (total)
=
13.0 44.0% 100.0%
43.7%
Unexplned variance in 1st contrast =
3.3 11.1% 25.2%
.9
.8
.7
C
O
N
T
R
A
S
T
1
.6
.5
.4
.3
.2
L .1
O
A .0
D
I -.1
N
G -.2
-.3
-.4
-.5
-.6
-2
-1
0
1
2
++---------------+---------------+---------------+---------------++
|
|
|
+
A
|
+
|
B
|
|
+
|
+
|
|
|
+
|
+
|
|
|
+
|
+
|
|
|
+
|
+
|
|
|
+
C
|
+
|
|
|
+
|
+
|
|
|
+
|
+
|
|
D
|
+
|
+
|
|
|
+--------------------------------E--------------------------------+
|
|
|
+
|
G
F
+
|
|
|
+
f
|
+
|
|
|
+
|
+
|
|
|
+
| e
+
|
|
|
+
d
|
+
|
|
|
+
b
|
c
+
|
|
a
|
223
++---------------+---------------+---------------+---------------++
-2
-1
0
1
2
ITEM MEASURE
For persons:
STANDARDIZED RESIDUAL CONTRAST 1 PLOT
.7
.6
C
O
N
T
R
A
S
T
.5
.4
.3
.2
.1
1
.0
L
O -.1
A
D -.2
I
N -.3
G
-.4
-.5
-.6
-.7
-6
-4
-2
0
2
4
6
8
++--------+--------+--------+--------+--------+--------+--------++
|
|
A
|
+
B
|
C
+
|
HFD
|
G
E
|
+
I
J
+
|
|
N LM
K
O
|
+
|
RT YP SQ
+
|
|
W 1U
XVZ
|
+
1 |1
1
1
+
|
|
1 2
1
|
+
11
1
11
1
+
|
1
|
1 1
11
|
+
111 |
1 111
111
+
|
|
21 1 1
1
|
+
1
1|
2 1 1
1
+
|
11 1 2
11 1 111
1
|
+-----------------------112-|--1--1-11-----1------1-----1--------+
|
1
|
1 1
2
1 11
1
1
|
+
11
|
11121
+
| 1
1 1
1 11 | 1 1 2 1 1
1
1
|
+
1
1 1 | 11
2 111
+
|
1 |
1 11
1
1
|
+
|
11
1 1
1
+
|
1
| 11 1
3 1 2
|
+
| v
y x z w
+
|
r
| s
t u
q
p
|
+
|
o
+
|
l|
nj
m
k
|
+
|
h i
+
|
f
c|
g
e d
|
+
a |
b
+
++--------+--------+--------+--------+--------+--------+--------++
-6
-4
-2
0
2
4
6
8
PERSON MEASURE
The plot shows a contrast in the residuals for PERSONS. Each letter is a person up to a maximum of 52 persons,
A-Z and a-z. For persons 53-181, &quot;1&quot; means that there is one person at that location on the plot. &quot;2&quot; means that
there are two persons, etc.
242.
Table 23.3, 24.3 Principal components analysis/contrast of residuals
Please do not interpret this as a usual factor analysis. These plots show contrasts between opposing
factors, not loadings on one factor. For more discussion, see dimensionality and contrasts.
This Table decomposes the matrix of item (Table 23) or person (Table 24) correlations based on residuals to
identify possible other contrasts (dimensions) that may be affecting response patterns. Specify PRCOMP=S or
=R or =L to obtain this Table.
Prior to this first contrast, the Rasch dimension has been extracted from the data. Residuals are those parts of
the observations not explained by the Rasch dimension. According to Rasch specifications, these should be
random and show no structure. The contrasts show conflicting local patterns in inter-item (or inter-person)
correlations based on residuals or their transformations. Letters &quot;E&quot;, &quot;b&quot;, etc. relate items (or persons) to their
loadings on the first contrast. In this Table, Bladder and Bowel contrast with Dressing. Since Bladder and Bowel
misfit, they load on a second dimension in the data.
+-----------------------------------------------------------------+
|
|
|
INFIT OUTFIT| ENTRY
|
|contrast|LOADING|MEASURE MNSQ MNSQ |NUMBER ITEM
|
|------+-------+-------------------+------------------------------|
| 1
|
.90 |
-.32 2.07 1.96 |A
7 G. BLADDER
|
| 1
|
.84 |
-.22 1.66 1.64 |B
8 H. BOWEL
|
| 1
|
.40 |
-.42 .89 .87 |C
9 I. BED TRANSFER
|
| 1
|
.15 |
1.26 .87 .86 |D
13 M. STAIRS
|
| 1
|
.01 |
-.02 .58 .58 |E
12 L. WALK/WHEELCHAIR
|
|
|-------+-------------------+------------------------------|
224
| 1
| -.64 |
.57 .61 .60 |a
5 E. LOWER BODY DRESSING |
| 1
| -.59 |
-.47 .86 .88 |b
4 D. UPPER BODY DRESSING |
| 1
| -.58 |
.66 .79 .76 |c
3 C. BATHING
|
| 1
| -.52 |
-.58 2.02 2.14 |d
2 B. GROOMING
|
| 1
| -.40 |
.16 .46 .47 |e
10 J. TOILET TRANSFER
|
| 1
| -.18 |
-1.71 1.23 1.14 |f
1 A. EATING
|
| 1
| -.10 |
.26 .77 .76 |G
6 F. TOILETING
|
| 1
| -.08 |
.83 .52 .54 |F
11 K. TUB, SHOWER
|
+-----------------------------------------------------------------+
To copy numbers out of this Table, use WORD to copy a rectangle of text.
243.
Table 23.4, 24.4 Person or item contrast
Please do not interpret this as a usual factor analysis. These plots show contrasts between opposing
factors, not loadings on one factor. For more discussion, see dimensionality and contrasts.
The effect of the contrast between the oppositely loading items (Table 23) or persons (Table 24) at the top and
bottom of the contrast plot is shown here.
Responses by persons (or on items) to the items (or by the persons) with extreme positive loadings and the items
(or persons) with extreme negative loadings are identified. These responses are seen to be higher, as expected,
or lower than the model predicts. Counts of these are obtained for each person (or item). The persons (or items)
showing the biggest impact of this contrast are listed first. Items and persons showing the most contrast are
chosen for this Table based on the &quot;Liking for Science&quot; data.
Table 23.4, showing the impact of the first contrast in the item residuals on persons:
ACT contrast 1 CONTRASTING RESPONSES BY PUPILS
+---------------------------------+
|
FAVORS TOP
|
| TOP 7 ACTS
| BOTTOM 7 ACTS |
| HIGH EXP. LOW | HIGH EXP. LOW |
+---------------------------------+
|
5
2
0 |
0
3
4 |
|
5
2
0 |
0
3
4 |
+---------------------------------+
+---------------------------------+
|
FAVORS BOTTOM
|
| TOP 7 ACTS
| BOTTOM 7 ACTS |
| HIGH EXP. LOW | HIGH EXP. LOW |
+---------------------------------+
|
0
3
4 |
0
7
0 |
|
0
3
4 |
0
7
0 |
|
0
5
2 |
3
3
1 |
|
0
3
4 |
0
7
0 |
+---------------------------------+
Persons scoring higher than expected on the top 7 items
15
26
DYSON, STEPHIE NINA
NEIMAN, RAYMOND
Persons scoring higher than expected on the bottom 7 items
8
11
32
51
LAMBERT, MD., ROSS W.
HSIEH, PAUL FRED
ROSSNER, JACK
REISS, STEVE
Table 24.4, showing the impact of the first contrast in the person residuals on items:
PUPIL contrast 1 CONTRASTING RESPONSES BY ACTS
+---------------------------------+
|
FAVORS TOP
|
| TOP 9 PUPILS
| BOTTOM 9 PUPILS|
| HIGH EXP. LOW | HIGH EXP. LOW |
+---------------------------------+
|
9
0
0 |
0
0
9 |
|
8
1
0 |
0
1
8 |
|
7
2
0 |
1
7
1 |
+---------------------------------+
+---------------------------------+
|
FAVORS BOTTOM
|
| TOP 9 PUPILS
| BOTTOM 9 PUPILS|
| HIGH EXP. LOW | HIGH EXP. LOW |
+---------------------------------+
|
0
4
5 |
0
9
0 |
|
0
5
4 |
0
9
0 |
|
0
5
4 |
0
9
0 |
Items biased towards top 9 persons
23
5
20
WATCH A RAT
FIND BOTTLES AND CANS
WATCH BUGS
Items biased towards bottom 9 persons
13
2
15
GROW GARDEN
READ BOOKS ON ANIMALS
READ ANIMAL STORIES
225
|
0
5
4 |
0
9
0 | 24
+---------------------------------+
244.
FIND OUT WHAT FLOWERS LIVE ON
Table 23.99, 24.99 Largest residual correlations for items or persons
These Tables show items (Table 23.99, formerly Table 23.1) and persons (Table 24.99, formerly 24.1) that may
be locally dependent. Specify PRCOMP=R (for score residuals) or PRCOMP=S or Y (for standardized residuals)
or PRCOMP=L (for logit residuals) to obtain this Table. Residuals are those parts of the data not explained by the
Rasch model. High correlation of residuals for two items (or persons) indicates that they may not be locally
independent, either because they duplicate some feature of each other or because they both incorporate some
other shared dimension.
Missing data are deleted pairwise if both of a pair are missing or PRCOMP=O (for observations), otherwise
missing data are replaced by their Rasch expected residuals of 0.
LARGEST STANDARDIZED RESIDUAL CORRELATIONS
USED TO IDENTIFY DEPENDENT ITEMS
+---------------------------------------------------------------------+
|RESIDUL| ENTRY
| ENTRY
|
|CORRELN|NUMBER ITEM
|NUMBER ITEM
|
|-------+------------------------------+------------------------------|
|
.71 |
17 Q. PROBLEM SOLVING
|
18 R. MEMORY
|
|
.69 |
14 N. COMPREHENSION
|
15 O. EXPRESSION
|
|
.61 |
9 I. BED TRANSFER
|
10 J. TOILET TRANSFER
|
|
.61 |
7 G. BLADDER
|
8 H. BOWEL
|
|
.54 |
4 D. UPPER BODY DRESSING |
5 E. LOWER BODY DRESSING |
|
.46 |
16 P. SOCIAL INTERACTION |
17 Q. PROBLEM SOLVING
|
|
.45 |
16 P. SOCIAL INTERACTION |
18 R. MEMORY
|
|
.40 |
10 J. TOILET TRANSFER
|
11 K. TUB, SHOWER
|
|-------+------------------------------+------------------------------|
| -.37 |
10 J. TOILET TRANSFER
|
18 R. MEMORY
|
| -.36 |
10 J. TOILET TRANSFER
|
17 Q. PROBLEM SOLVING
|
+---------------------------------------------------------------------+
Note: Redundant correlations of 1.0 are not printed. If A has a correlation of 1.0 with B, and also with C, assume
that B and C also have a correlation of 1.0.
This Table is used to detect dependency between pairs of items or persons. When raw score resiudal correlations
are computed, it corresponds to Wendy Yen's Q3 statistic. is used to detect dependency between pairs of items
or persons. Yen suggests a small positive adjustment to the correlation of size 1/(L-1) where L is the test length.
Yen, W. M. (1984). Effects of local item dependence on the fit and equating performance of the three-parameter
logistic model. Applied Psychological Measurement, 8, 125-145. Yen, W. M. (1993). Scaling performance
assessments: Strategies for managing local item dependence. Journal of Educational Measurement, 30, 187-213.
Yen suggest a small adjustment for bias
245.
Table 27.1, 28.1 Subtotal summaries on one line
These summarize the measures from the main analysis for all items or persons selected by ISUBTOT= (Table 27)
or PSUBTOT= (Table 28), including extreme scores. Histograms are shown in Tables 27.2 and 28.2.
PSUBTOTAL= is useful for quantifying the impact of a test on different types of test-takers.
Subtotal specification is: PSUBTOTAL=$S9W1
+---------------------------------------------------------+
| KID
MEAN
S.E. OBSERVED MEDIAN
REAL
|
| COUNT MEASURE
MEAN
S.D.
SEPARATION CODE |
|---------------------------------------------------------|
|
35
-.36
.37
2.15
-.25
1.55
*
|
|
18
-.36
.52
2.14
-.25
1.47
F
|
|
17
-.37
.54
2.17
-.25
1.63
M
|
+---------------------------------------------------------+
UMEAN=0 USCALE=1
The first line, &quot;*&quot;, is the total for all persons (or items)
226
The remaining codes are those in the person (or item) columns specified by $S9W1or whatever, using the column
selection rules. In this example, &quot;F&quot; is the code for &quot;Female&quot; in the data file. &quot;M&quot; for &quot;Male&quot;. It is seen that the two
distributions are almost identical.
The statistical significance of the difference between the two subtotal means is:
t = ( mean measure of &quot;F&quot; - mean measure of &quot;M&quot;) / sqrt ( (S.E. Mean &quot;F&quot;)&sup2; + (S.E. Mean &quot;M&quot;)&sup2; )
REAL SEPARATION is the separation coefficent defined as &quot;true&quot; sample standard deviation / measurement
error based on misfit-inflated error variance.
246.
Table 27.2, 28.2 Subtotal measure histograms
These show the distributions of the measures from the main analysis for each sub-sample. Summaries are shown
in Tables 27.1 (Items, ISUBTOT=) and 28.1 (Persons, PSUBTOT=).
Here is the measure distribution of the total sample:
1
2
5
4
1
2
T
S
M
S
T
|-----+-----+-----+-----+-----+-----+-----+-----+-----+-----|
-5
-4
-3
-2
-1
0
1
2
3
4
5
1
1
2
2
2
3
* TOTAL
GLOBAL SAMPLE COUNT
Read the counts vertically so that the center count is 12 observed near -0.4.
M = Mean, S = one sample standard deviation from mean, T = two sample standard deviations from mean.
Here is the measure distribution of the total sample standardized to a sample size of 1000:
3
1
1
4
4
1
2
5
3
3
5
9
8
T
S
M
S
T
|-----+-----+-----+-----+-----+-----+-----+-----+-----+-----|
-5
-4
-3
-2
-1
0
1
2
3
4
5
2
9
2
9
5
8
5
8
5
8
8
6
* TOTAL
PER THOUSAND IN GLOBAL SAMPLE
Here is the measure distribution of one sub-sample, specifed as $S9W1=&quot;F&quot;, using the column selection rules.
1
1
1
2
7
3
2
1
T
S
M
S
T
|-----+-----+-----+-----+-----+-----+-----+-----+-----+-----|
-5
-4
-3
-2
-1
0
1
2
3
4
5
F
LOCAL SAMPLE COUNT
Here is the measure distribution of the sub-sample standardized to a sample size of 1000:
3
1
1
8
6
1
5
9
7
2
6
T
S
M
S
T
|-----+-----+-----+-----+-----+-----+-----+-----+-----+-----|
-5
-4
-3
-2
-1
0
1
2
3
4
5
5
6
5
6
5
6
1
1
2
F
PER THOUSAND IN LOCAL SAMPLE
Here is the measure distribution of the sub-sample standardized so that the total at any measure is 1,000, but in
proportion to the observed counts in each sub-sample:
1
0
5
5
6
5
6
5
5
0
0
0
6
8
0
0
0
0
0
0
7
4
0
0
0
T
S
M
S
T
|-----+-----+-----+-----+-----+-----+-----+-----+-----+-----|
-5
-4
-3
-2
-1
0
1
2
3
4
5
SAMPLES
F
PER THOUSAND AT MEASURE FOR COUNTED LOCAL
Here is the measure distribution of the sub-sample standardized so that the total any measure is 1,000, but based
on the proportion observed when the sizes of all sub-samples are adjusted to be equal:
1
227
0
4
4
6
5
5
4
4
0
8
8
5
7
8
8
8
0
6
6
4
0
7
6
6
T
S
M
S
T
|-----+-----+-----+-----+-----+-----+-----+-----+-----+-----|
-5
-4
-3
-2
-1
0
1
2
3
4
5
SAMPLES
247.
F
PER THOUSAND AT MEASURE FOR EQUAL LOCAL
Table 29 Empirical ICCs and option frequencies
This Table display both the model expected score ogive, the predicted Item Characteristic Curve (ICC), and also
the empirical observed average score per measure, the empirical ICC. See also Graphs Menu.
It also shows the relative frequency of each item response code for each measure level.
O
B
S
E
R
V
E
D
1.0
.8
.6
A
V
E
R
A
G
E
.5
.4
.2
S
C
O
R
E
.0
EMPIRICAL &amp; MODEL ICCs : 1. What is the capital of Burundi?
++---------+---------+---------+---------+---------+---------++
+
xxxxxxx *...*...*......+
|
......
|
|
xx ..
|
|
x .*
|
+
x x..
+
|
.
|
|
x..
|
|
. x
|
+
*
+
|
x*
|
+
.
+
|
.
|
+
x .
+
|
x x . x
|
|
x.xx
|
|
.
|
+
..
+
|
..x
|
|
...
|
|
......
|
+.....*...
x x
+
++---------+---------+---------+---------+---------+---------++
-6
-4
-2
0
2
4
6
PERSON MEASURE
The model ICC is shown by '.'. this is the expected average score on the item for persons at each measure
relative to the item.
Observed average scores on the items are shown by 'x'.
When '.' and 'x' coincide, '*' is shown.
You can use this plot to compute the empirical item discrimination.
For a multiple-choice item, with C as an incorrect distractor, this could like the plot below. The dots are the Rasch
ogive for an incorrect answer.
F
R
E
Q
U
E
N
C
Y
O
F
R
100
96
92
88
84
80
76
72
68
64
60
56
52
48
44
40
EMPIRICAL CODE FREQUENCIES: &quot;C&quot; : 2. 2
++------+------+------+------+------+------+------+------++
+
+
+.......
+
+
....
+
+
..
+
+
..
+
+
..
+
+
..
+
+
.
+
+
.
+
+
..
+
+
.
+
+
.
+
+
.
+
+
.
+
+
.
+
+
.
+
228
E
S
P
O
N
S
E
%
36
32
28
24
20
16
12
8
4
0
+
..
+
+
C
.
+
+
C C
.
+
+
C
C
..
+
+
C
C
..
+
+
C
C CC C
..
+
+
C
..
+
+
CC
....
+
+
........
+
+
....+
++------+------+------+------+------+------+------+------++
-4
-3
-2
-1
0
1
2
3
4
PERSON MEASURE
For a polytomous item, the percentage frequencies of responses, and the model ordinal probability, of each
category are shown at each measure level. The dots are the model curve. &quot;4&quot; is the empirical frequency for
category 4. &quot;*&quot; means a &quot;4&quot; and a &quot;.&quot;/
F
R
E
Q
U
E
N
C
Y
O
F
R
E
S
P
O
N
S
E
%
100
96
92
88
84
80
76
72
68
64
60
56
52
48
44
40
36
32
28
24
20
16
12
8
4
0
248.
EMPIRICAL CODE FREQUENCIES: &quot;4&quot; : 1. A. FEEDING
++-------+-------+-------+-------+-------+-------+-------++
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
44
+
+
.** 4
+
+
..4 .. 4
+
+
4 *
..4
+
+
.. 44
.
+
+
4
. 4
*.
+
+
.* 4
4*.
+
+
....
4**.
+
+....
4*******...................+
++-------+-------+-------+-------+-------+-------+-------++
-3
-2
-1
0
1
2
3
4
PERSON MEASURE
Table 30 Differential item functioning DIF bias analysis
Table 30 supports the investigation of item bias, Differential Item Functioning (DIF), i.e., interactions between
individual items and types of persons. Specify DIF= for person classifying indicators in person labels. Item bias
and DIF are the same thing. The widespread use of &quot;item bias&quot; dates to the 1960's, &quot;DIF&quot; to the 1980's. The
reported DIF is corrected to test impact, i.e., differential average performance on the whole test. Use ability
stratification to look for non-uniform DIF using the selection rules. Tables 30.1 and 30.2 present the same
information from different prespectives.
Table 31 supports person bias, Differential Person Functioning (DPF), i.e., interactions between individual
persons and classifications of items.
Table 33 reports bias or interactions between classifications of items and classifications of persons.
In these analyses, persons and items with extreme scores are excluded, because they do not exhibit
differential ability across items. For background discussion, see DIF and DPF considerations.
Example output:
229
You want to examine item bias (DIF) between Females and Males. You need a column in your Winsteps person
label that has two (or more) demographic codes, say &quot;F&quot; for female and &quot;M&quot; for male (or &quot;0&quot; and &quot;1&quot; if you like
dummy variables) in column 9.
Table 30.1 is best for pairwise comparisons, e.g., Females vs. Males.
DIF specification is: DIF=$S9W1
+-------------------------------------------------------------------------------------------------------+
| KID
DIF
DIF
KID
DIF
DIF
DIF
JOINT
MantelHanzl TAP
|
| CLASS MEASURE S.E. CLASS MEASURE S.E. CONTRAST S.E.
t d.f. Prob. Prob. Size Number Name
|
|-------------------------------------------------------------------------------------------------------|
| F
2.85
.89 M
1.24
.70
1.61 1.13 1.42 32 .1639 .2049 1.95
13 1-4-3-2-4
|
| F
-5.25&gt; 1.90 M
-3.89
.90
-1.37 2.10 -.65 32 .5188 .2528 -.
4 1-3-4
|
| F
-2.78
.89 M
-3.89
.90
1.10 1.26
.87 31 .3884 .8258 -.09
6 3-4-1
|
|-------------------------------------------------------------------------------------------------------|
Size of Mantel-Haenszel slice = .100
DIF Specification defines the columns used to identify DIF classifications, using the selection rules.
Reading across the Table 30.1 columns:
KID CLASS identifies the CLASS of persons. KID is specified with PERSON=, e.g., the first CLASS is &quot;F&quot;
DIF MEASURE is the difficulty of this item for this class, with all else held constant, e.g., -5.24 is the local difficulty
for Class F of Item 4.
-5.24&gt; reports that this measure corresponds to an extreme maximum score. EXTRSCORE= controls
extreme score estimate.
5.31&lt; reports that this measure corresponds to an extreme minimum score. EXTRSCORE= controls
extreme score estimate.
DIF S.E. is the standard error of the DIF MEASURE
KID CLASS identifies the CLASS of persons, e.g., the second CLASS is &quot;M&quot;
DIF MEASURE is the difficulty of this item for this class, with all else held constant, e.g., -3.87 is the local difficulty
for Class M of Item 4.
DIF S.E. is the standard error of the second DIF MEASURE
DIF CONTRAST is the difference between the DIF MEASUREs, i.e., size of the DIF across the two classifications
of persons, e.g., 2.85 - 1.24 = 1.61. A positive DIF contrast indicates that the item is more difficult for the
left-hand-listed CLASS.
JOINT S.E. is the standard error of the DIF CONTRAST = sqrt(first DIF S.E.&sup2; + second DIF S.E.&sup2;), e.g., 1.13 =
sqrt(.89&sup2; + .70&sup2;)
t gives the DIF significance as a unit normal deviate = DIF CONTRAST / JOINT S.E. The t-test is a two-sided test
for the difference between two means (i.e., the estimates) based on the standard error of the means (i.e.,
the standard error of the estimates). The null hypothesis is that the two estimates are the same, except
for measurement error.
d.f. is the joint degrees of freedom. This is shown as the sum of the sizes of two classifications - 2 for the two
measure estimates, but this estimate of d.f. is somewhat high, so interpret the t-test conservatively, e.g.,
d.f. = (17 F + 17 M - 2) = 32.
Prob. is the probability of the reported t with the reported d.f., but interpret this conservatively. If you wish to make
a Bonferroni multiple-comparison correction, compare this Prob. with your chosen significance level, e.g.,
.05, divided by the number of entries in this Table.
MantelHanzel reports Mantel-Haenszel (1959) DIF test for dichotomies or Mantel (1963) for polytomies using
MHSLICE=
Prob. is the probability of observing these data (or worse) when there is no DIF. Reported when computable.
Size is an estimate of the DIF (scaled by USCALE=). Reported when computable. Otherwise +. and -. indicate
direction.
TAP Number is the item entry number. TAP is specifed by ITEM=
Name is the item label.
Males
Females
230
Item 13: +---------+---------+-+-------+-------+-+&gt;&gt; difficulty increases
-1
0
1.24
+2
2.85
DIF measure
+---------------+ = 1.61 DIF contrast
Table 30.2 is best for multiple comparisons, e.g., regions against the national average.
DIF specification is: DIF=$S9W1
+---------------------------------------------------------------------------------------------+
| KID
OBSERVATIONS
BASELINE
DIF
DIF
DIF
DIF DIF TAP
|
| CLASS
COUNT AVERAGE EXPECT MEASURE
SCORE MEASURE SIZE S.E.
t
Number Name
|
|---------------------------------------------------------------------------------------------|
| F
17
1.00
.96
-4.40
.04
-5.24 -.84&gt; 1.90 -.44
4 1-3-4
|
| M
17
.88
.92
-4.40
-.04
-3.87
.53
.90
.59
4 1-3-4
|
This displays a list of the local difficulty/ability estimates underlying the paired DIF analysis. These can be plotted
directly from the Plots menu.
DIF Specification defines the columns used to identify DIF classifications, using the selection rules.
KID CLASS identifies the CLASS of persons. KID is specified with PERSON=, e.g., the first CLASS is &quot;F&quot;
OBSERVATIONS are what are seen in the data
COUNT is the number of observations of the classification, e.g., 17 F persons responded to TAP item 4.
AVERAGE is the average observation on the classification, e.g., 1.00 is the p-value of item 4 for F persons.
COUNT * AVERAGE = total score of person class on the item
BASELINE is the prediction without DIF
EXPECT is the expected value of the average observation when there is no DIF, e.g., .96 is the expected p-value
for F without DIF.
MEASURE is the what the overall measure would be without DIF, e.g., -4.40 is the overall item difficulty of item 4.
DIF: Differential Item Functioning
DIF SCORE is the difference between the observed and the expected average observations, e.g., 1.00 - .96 = .04
DIF MEASURE is the item difficulty for this class, e.g., item 4 has a local difficulty of -5.24 for CLASS F.
Average of DIF measures across CLASS for an item is not the BASELINE MEASURE because score-tomeasure conversion is non-linear.
DIF SIZE is the difference between the difficulty for this class and the baseline difficulty, i.e., -4.40 - -5.24 = -.84,
item 4 is .84 logits easier for class F than expected.
DIF S.E. is the approximate standard error of the difference, e.g., 1.90 logits
DIF t is an approximate Student's t-test, DIF SIZE divided by the DIF S.E. with a little less than (COUNT-2)
degrees of freedom.
A probability is not reported for the t, because the computation is too inexact, but this Table provides a guide.
Table of the two-sided t distribution:
d.f.
p=.05 p=.01
d.f.
p=.05 p=.01
1
12.71 63.66
11 2.20
3.11
21
2
4.30
9.93
12 2.18
3.06
22
3
3.18
5.84
13 2.16
3.01
23
4
2.78
4.60
14 2.15
2.98
24
5
2.57
4.03
15 2.13
2.95
25
6
2.45
3.71
16 2.12
2.92
26
7
2.37
3.50
17 2.11
2.90
27
8
2.31
3.36
18 2.10
2.88
28
9
2.26
3.25
19 2.09
2.86
29
10 2.23
3.17
20 2.09
2.85
30
Inf.
1.96
2.58
249.
d.f.
p=.05
2.08
2.83
2.07
2.82
2.07
2.81
2.06
2.80
2.06
2.79
2.06
2.78
2.05
2.77
2.05
2.76
2.05
2.76
2.04
2.75
p=.01
Table 31 Differential person functioning DPF bias/interaction analysis
Table 31 supports person bias, Differential Person Functioning (DPF), i.e., interactions between individual
persons and classifications of items. Specify DPF= for classifying indicators in item labels. Use difficulty
stratification to look for non-uniform DPF using the selection rules.
Table 30 supports the investigation of item bias, Differential Item Functioning (DIF), i.e., interactions between
individual items and types of persons.
Table 33 reports bias or interactions between classifications of items and classifications of persons.
In these analyses, persons and items with extreme scores are excluded, because they do not exhibit
231
differential ability across items. For background discussion, see DIF and DPF considerations.
Example output:
Table 31.1
DPF class specification is: DPF=$S1W1
+----------------------------------------------------------------------------------------+
| TAP
DPF
DPF
TAP
DPF
DPF
DPF
JOINT
KID
|
| CLASS MEASURE S.E. CLASS MEASURE S.E. CONTRAST S.E.
t d.f. Prob. Number Name
|
|----------------------------------------------------------------------------------------|
| 1
-3.53 1.05 2
-2.70 1.65
-.83 1.95 -.42 11 .6801
1 Richard M|
| 1
-3.53 1.05 3
-2.53&gt; 2.18
-1.00 2.42 -.41 10 .6891
1 Richard M|
DPF Specification defines the columns used to identify Differential Person Function classifications, using the
selection rules.
TAP CLASS is the item class
DPF MEASURE is the ability of the person for this item class, with all else held constant.
DPF S.E. is the standard error of the measure
DPF CONTRAST is the difference in the person ability measures, i.e., size of the DPF, for the two classifications
of items.
JOINT S.E. is the standard error of the DPF CONTRAST
t gives the DPF significance as a Student's t-test. The t-test is a two-sided test for the difference between two
means (i.e., the estimates) based on the standard error of the means (i.e., the standard error of the
estimates). The null hypothesis is that the two estimates are the same, except for measurement error.
d.f. is the joint degrees of freedom. This is shown as the sum of the sizes of two classifications - 2 for the two
measure estimates, but this estimate of d.f. is somewhat high, so interpret the t-test conservatively.
Prob. is the probability of the reported t with the reported d.f., but interpret this conservatively. If you wish to make
a Bonferroni multiple-comparison correction, compare this Prob. with your chosen significance level, e.g.,
.05, divided by the number of entries in this Table.
-5.24&gt; reports that this measure corresponds to an extreme maximum score. EXTRSCORE= controls extreme
score estimate.
5.30&lt; reports that this measure corresponds to an extreme minimum score. EXTRSCORE= controls extreme
score estimate.
Table 31.2
+-----------------------------------------------------------------------------------------+
| TAP
OBSERVATIONS
BASELINE
DPF
DPF
DPF
DPF DPF KID
|
| CLASS
COUNT AVERAGE EXPECT MEASURE
SCORE MEASURE SIZE S.E.
t
Number Name
|
|-----------------------------------------------------------------------------------------|
| 1
11
.18
.23
-2.94
-.05
-3.53 -.59 1.05 -.56
1 Richard |
| 2
2
.50
.46
-2.94
.04
-2.70
.24 1.65
.15
1 Richard |
| 3
1
1.00
.61
-2.94
.39
-2.53
.41&gt; 2.18
.19
1 Richard |
| 1
11
.36
.39
-.26
-.03
-.77 -.51 1.35 -.38
2 Tracie |
| 2
2
1.00
.88
-.26
.12
-.55 -.29&gt; 2.09 -.14
2 Tracie |
This displays a list of the local difficulty/ability estimates underlying the paired DPF analysis. These can be
plotted directly from the Plots menu.
TAP CLASS: is the item class - these are the subtests (item classifications) for which differential person
functioning is to be investigated.
OBSERVATIONS:
COUNT is the number of observations of the classification.
AVERAGE is the average observation on the classification.
COUNT * AVERAGE = total score of person on the item class (subtest)
BASELINE:
EXPECT is the expected average value of the observations based on all items.
MEASURE is the overall ability measure based on all items
Average of DPF measures across CLASS for a person is not the BASELINE MEASURE because score232
to-measure conversion is non-linear.
DPF: Differential Person Function
SCORE is the difference between average of the observations observed and expected for this class
MEASURE is the local ability of the person for this class
SIZE is the difference between the local and overall ability of this person. &gt;&lt; indicate extreme scores.
S.E. is the approximate standard error of the difference.
t is an approximate Student's t-test would be the MEASURE divided by the S.E. with a little less than (COUNT-2)
degrees of freedom.
Probability is not reported, because the t-test is too inexact.
250.
Table 32 Control specifications
This gives the setting of every Winsteps control variable. It can be accessed as Table 32 from a control file, or
from the Output Files pull-down menu. It is written to a temporaty text file, but can be &quot;saved as&quot; to a permanent
file.
; Values of Control Specifications
; CONTROL FILE = C:\e\Ab6.0\bsteps\mrwe\data\kct.txt
; OUTPUT REPORT FILE = C:\e\Ab6.0\bsteps\mrwe\data\ZOU713ws.txt
; DATE AND TIME = Feb 11 23:29 2004
ALPHANUM =
ASCII = Y
BATCH = N
CATREF = 0
CFILE =
CHART = N
CLFILE =
CODES = 01
CONVERGE = E
....
251.
Table 33 DIF/DPF/DGF interactions by class
This Table identifies interactions between classifications of persons (identified by DIF=) and classifications of
items (identified by DPF=) using the column selection rules. Differential average classification-group performance
(DGF) is powerful when looking for latent classes among the persons. For more details, see Table 30 (DIF) and
Table 31 (DPF).
Table 33.1
CLASS-LEVEL BIAS/INTERACTIONS FOR DIF=$s9w1 AND DPF=$S1W1
+------------------------------------------------------------------------------+
| KID
DIF
DIF KID
DIF
DIF
DIF
JOINT
TAP
|
| CLASS
SIZE S.E. CLASS
SIZE S.E. CONTRAST S.E.
t d.f. Prob. CLASS |
|------------------------------------------------------------------------------|
| F
-.03
.28 M
.05
.27
-.08
.39
.20 372 .8407 1
|
| F
-.07
.58 M
.04
.55
-.11
.80
.14 66 .8906 2
|
| F
.55
.88 M
-.49
.90
1.04 1.25 -.83 32 .4114 3
|
+------------------------------------------------------------------------------+
This Table contrasts, for each item class, the size and significance of the Differential Item Functioning for pairs of
person classifications.
Table 33.2 - These can be plotted directly from the Plots menu.
CLASS-LEVEL BIAS/INTERACTIONS FOR DIF=$s9w1 AND DPF=$S1W1
+--------------------------------------------------------------------------+
| KID
OBSERVATIONS
BASELINE
DIF
DIF
DIF
DIF
TAP
|
| CLASS
COUNT AVERAGE EXPECT
SCORE
SIZE S.E.
t
CLASS |
|--------------------------------------------------------------------------|
| F
187
.42
.41
.00
-.03
.28 -.11 1
|
| M
187
.39
.39
.00
.05
.27
.17 1
|
| F
34
.85
.85
.01
-.07
.58 -.12 2
|
| M
34
.76
.77
.00
.04
.55
.08 2
|
| F
17
.88
.92
-.04
.55
.88
.63 3
|
| M
17
.88
.84
.04
-.49
.90 -.55 3
|
233
+--------------------------------------------------------------------------+
This show the local size of the DIF for each person class on each item class. The reported t value is approximate,
and should be interpreted conservatively.
Table 33.3
CLASS-LEVEL BIAS/INTERACTIONS FOR DIF=$s9w1 AND DPF=$S1W1
+------------------------------------------------------------------------------+
| TAP
DPF
DPF TAP
DPF
DPF
DPF
JOINT
KID
|
| CLASS
SIZE S.E. CLASS
SIZE S.E. CONTRAST S.E.
t d.f. Prob. CLASS |
|------------------------------------------------------------------------------|
| 1
.03
.28 2
.07
.58
-.04
.65 -.06 219 .9553 F
|
| 1
.03
.28 3
-.55
.88
.58
.92
.63 202 .5286 F
|
| 2
.07
.58 3
-.55
.88
.62 1.05
.59 49 .5602 F
|
| 1
-.05
.27 2
-.04
.55
.00
.61 -.01 219 .9939 M
|
| 1
-.05
.27 3
.49
.90
-.54
.94 -.58 202 .5643 M
|
| 2
-.04
.55 3
.49
.90
-.54 1.05 -.51 49 .6125 M
|
+------------------------------------------------------------------------------+
This Table contrasts, for each person class, the size and significance of the Differential Person Functioning for
pairs of item classifications.
Table 33.4 - These can be plotted directly from the Plots menu.
CLASS-LEVEL BIAS/INTERACTIONS FOR DIF=$s9w1 AND DPF=$S1W1
+--------------------------------------------------------------------------+
| TAP
OBSERVATIONS
BASELINE
DPF
DPF
DPF
DPF
KID
|
| CLASS
COUNT AVERAGE EXPECT
SCORE
SIZE S.E.
t
CLASS |
|--------------------------------------------------------------------------|
| 1
187
.42
.41
.00
.03
.28
.11 F
|
| 2
34
.85
.85
.01
.07
.58
.12 F
|
| 3
17
.88
.92
-.04
-.55
.88 -.63 F
|
| 1
187
.39
.39
.00
-.05
.27 -.17 M
|
| 2
34
.76
.77
.00
-.04
.55 -.08 M
|
| 3
17
.88
.84
.04
.49
.90
.55 M
|
+--------------------------------------------------------------------------+
This show the local size of the DP for each item class on each person class. The reported t value is approximate,
and should be interpreted conservatively.
252.
Table 34 Columnar statistical comparison and scatterplot
To automatically produce this Excel scatterplot of two sets of measures or fits statistics:
Select Compare Statistics on the Plots pull-down menu. If this is too big for your screen see Display too big.
234
Measures, standard errors, fit statistics indicate which statistic (or which field of the IFILE= or PFILE=) is to be
the basis of the comparison.
Display with columns generates a the line-printer graphical-columns plot. It is displayed as Table 34.
The first column is the Outfit Mean-Square of this analysis.
The third column is the Outfit Mean-Square of the Right File (exam12lopf.txt in this case)
The second column is the difference.
The fourth column is the identification, according to the current analysis.
Persons or items are matched and listed by Entry number.
+-----------------------------------------------------------------------------+
|
PERSON
| Outfit MnSq Difference | exam12lopf.txt
| File Compa|
|0
1
3|-2
0
2|0
1
3| NUM LABEL|
|-------------------+-------------------------+-------------------+-----------|
|
.
*
|
*
.
|
* .
| 1 21101 |
|
.
*
|
* .
|
.
*
| 2 21170 |
| *
.
|
.*
|
* .
| 3 21174 |
....
|
*.
|
*
|
*.
| 35 22693 |
+-----------------------------------------------------------------------------+
Display with Excel scatterplot initiates a graphical scatterplot plot. If the statistics being compared are both
measures, then a 95% confidence interval is shown. This plot can be edited with all Excel tools.
253.
Table 0 The title page
This page contains the authorship and version information. It appears in the Report Output File, but not on the
Output Table menu.
WINSTEPS is updated frequently. Please refer to the web page for current version numbers and recent
enhancements at www.winsteps.com
******************************************************************************
*
*
*
* * * * W I N S T E P S * * * *
*
*
--------------------------------*
*
*
*
- RASCH ANALYSIS FOR TWO-FACET MODELS *
*
*
*
PERSON, ITEM &amp; RESPONSE STRUCTURE MEASUREMENT AND FIT ANALYSIS
*
*
*
*
INQUIRE: WINSTEPS
*
*
PO BOX 811322, CHICAGO ILLINOIS 60681-1322
*
*
Tel. &amp; FAX (312) 264-2352
*
*
www.winsteps.com
*
*
*
*
COPYRIGHT (C) JOHN MICHAEL LINACRE, 1991-2002
*
*
AUGUST 6, 2002
VERSION 3.36 *
******************************************************************************
235
254.
Table 0.1 Analysis identification
This shows the Title, Control file, Output report file, date and time of this analysis.
If the output file name has the form ZOU???ws.txt, then it is a temporary file which will be erased when Winsteps
terminates. You can open this file from the &quot;Edit&quot; menu and save it as a permanent file.
TABLE 0.1 LIKING FOR SCIENCE (Wright &amp; Masters p. ZOU042ws.txt Oct 9 9:00 2002
-------------------------------------------------------------------------------TITLE=
CONTROL FILE:
OUTPUT FILE:
DATE:
LIKING FOR SCIENCE (Wright &amp; Masters p.18)
C:\WINSTEPS\sf.txt
C:\WINSTEPS\ZOU042ws.txt
Oct 9 9:00 2002
ACT DELETIONS: 2-3 10-15 17-19 21 24
76 PUPIL
Records Input.
255.
Table 0.2 Convergence report
(controlled by LCONV=, RCONV=, CONVERGE=, MPROX=, MJMLE=, CUTLO=, CUTHI=)
TABLE 0.2 LIKING FOR SCIENCE (Wright &amp; Masters p. ZOU042ws.txt Oct 9 9:00 2002
INPUT: 76 PUPILS, 25 ACTS
WINSTEPS 3.36
-------------------------------------------------------------------------------CONVERGENCE TABLE
+----------------------------------------------------------------------------+
|
PROX
ACTIVE COUNT
EXTREME 5 RANGE
MAX LOGIT CHANGE |
| ITERATION
PUPILS ACTS
CATS
PUPILS ACTS
MEASURES STRUCTURE|
|----------------------------------------------------------------------------|
|
1
76
25
3
3.59
1.62
3.1355
-.1229 |
|
2
74
12
3
4.03
1.90
.3862
-.5328 |
|
3
74
12
3
4.19
1.96
.1356
-.0783 |
WARNING: DATA MAY BE AMBIGUOUSLY CONNECTED INTO 6 SUBSETS, see Connection Ambiguities
|----------------------------------------------------------------------------|
|
JMLE
MAX SCORE
MAX LOGIT
LEAST CONVERGED
CATEGORY STRUCTURE|
| ITERATION
RESIDUAL*
CHANGE
PUPIL ACT
CAT
RESIDUAL
CHANGE |
|----------------------------------------------------------------------------|
|
1
-2.04
.2562
7
5*
2
-.72
.0003|
+----------------------------------------------------------------------------+
Standardized Residuals N(0,1) Mean: .03 S.D.: 1.24
Look for scores and residuals in last line to be close to 0, and standardized residuals to be close to mean 0.0,
S.D. 1.0.
The meanings of the columns are:
PROX
normal approximation algorithm - for quick initial estimates
ITERATION
number of times through your data to calculate estimates
ACTIVE COUNT
number of elements participating in the estimation process after elimination of deletions and
perfect/zero scores
PERSONS person parameters
ITEMS item parameters
CATS
rating scale categories - shows 2 for dichotomies
EXTREME 5 RANGE
PERSONS The current estimate of the spread between the average measure of the top 5 persons and the
average measure of the bottom 5 persons.
ITEMS The current estimate of the spread between the average measure of the top 5 items and the average
measure of the bottom 5 items.
MAX LOGIT CHANGE
MEASURES
maximum logit change in any person or item estimate. This i expected to decrease gradually
until convergence, i.e., less than LCONV=.
STRUCTURE maximum logit change in any structure measure estimate - for your information - need not be
236
as small as MEASURES.
JMLE
JMLE joint maximum likelihood estimation - for precise estimates
ITERATION
number of times through your data to calculate estimates
It is unusual for more than 100 iterations to be required
MAX SCORE RESIDUAL maximum score residual (difference between integral observed core and decimal
expected score) for any person or item estimate - used to compare with RCONV=. This number is expected to
decrease gradually until convergence acceptable.
*
indicates to which person or item the residual applies.
MAX LOGIT CHANGE maximum logit change in any person or item estimate - used to compare with LCONV=.
This number is expected to decrease gradually until convergence is acceptable.
LEAST CONVERGED element numbers are reported for the person, item and category farthest from meeting
the convergence criteria.
*
indicates whether the person or the item is farthest from convergence.
the CAT (category) may not be related to the ITEM to its left. See Table 3.2 for details of
unconverged categories.
CATEGORY RESIDUAL
maximum count residual (difference between integral observed count and decimal
expected count) for any response structure category - for your information. This number is expected to decrease
gradually. Values less than 0.5 have no substantive meaning.
STRUCTURE CHANGE maximum logit change in any structure calibration. Not used to decide convergence, but
only for your information. This number is expected to decrease gradually.
Standardized Residuals These are modeled to have a unit normal distribution. Gross departures from mean of
0.0 and standard deviation of 1.0 indicate that the data do not conform to the basic Rasch model specification that
randomness in the data be normally distributed.
256.
Table 0.3 Control file
This Table shows the control file used in the analysis. It includes the extra specifications and expands SPFILE=
commands. This is also appended to the LOGFILE=, when specified.
A complete listing of control variables is obtained using &quot;Control variable file=&quot; from the Output Files pull-down
menu, which is also Table 32.
&quot;Extra Specifications&quot; are listed after &amp;END.
TABLE 0.3 LIKING FOR SCIENCE (Wright &amp; Masters p. ZOU042ws.txt Oct 9 9:00 2002
INPUT: 76 PUPILS, 25 ACTS MEASURED: 75 PUPILS, 12 ACTS, 3 CATS
WINSTEPS 3.36
-------------------------------------------------------------------------------&amp;INST
TITLE='LIKING FOR SCIENCE +
+(Wright &amp; Masters p.18)'
ITEMS=ACT
PERSONS=PUPIL
ITEM1=1
NI=25
NAMLMP=20
XWIDE=2
NAME1=51
isubtot=$s1W1
psubtot=$s2W1
pweight=$s2w1
iweight=$s3w1
dif=$s3W1
dpf=$s4W1
CODES=000102
;demonstrates continuation line
237
; ISGROUPS=0
CFILE=*
00 dislike
01 neutral
02 like
*
idfile=*
13 GROW GARDEN
10 LISTEN TO BIRD SING
2 READ BOOKS ON ANIMALS
12 GO TO MUSEUM
21 WATCH BIRD MAKE NEST
18 GO ON PICNIC
24 FIND OUT WHAT FLOWERS LIVE ON
19 GO TO ZOO
15 READ ANIMAL STORIES
11 FIND WHERE ANIMAL LIVES
3 READ BOOKS ON PLANTS
14 LOOK AT PICTURES OF PLANTS
17 WATCH WHAT ANIMALS EAT
*
|
|
|
|
|
|
|
|
|
|
|
|
|
EXTRSC=0.3
; TABLES=00000011111111111111111111111111
; CURVES=111
CSV=Y
; ISGROUPS=0
; IFILE = SFIF.txt
; PFILE = SFPF.txt
; XFILE = SFXF.txt
; RFILE = SFRF.txt
; SFILE = SFSF.txt
; ISFILE = SFIS.txt
&amp;END
MJMLE=1
257.
Probability category curves
Select by clicking on &quot;Probability Cat. Curves&quot; or from the Graphs menu. If you don't see all this on your screen,
you may have your screen resolution set to 800 x 600 pixels. Try setting it to 1024 x 768. Windows &quot;Start&quot;,
&quot;Settings&quot;, &quot;Control Panel&quot;, &quot;Display&quot;, &quot;Settings&quot;, Move &quot;Screen resolution&quot; slider to the right.
Traceline: You can identify an individual traceline by single-left-clicking on it. Its description will then appear
below the plot. Click elsewhere on the plot to remove the selection indicators. You can remove a traceline
by double-left-clicking on it. Click on the command button, e.g., &quot;Probability Curves&quot;, to return the plot
to its initial appearance.
Adjust or fine-tune minimum or maximum enables you to change the x- and y-axis ranges.
Adjust number enables you to change the number of x- and y- tick marks and labels.
238
Copy Plot to Clipboard places this screen on the clipboard. Open a graphics program, such as Paint, and
paste in the image for editing. This only copies the part of the plot that is visible on your screen. Maximize
the chart window and increase your screen resolution if the entire plot is not visible. To increase screen
resolution: Windows &quot;Start&quot;, &quot;Settings&quot;, &quot;Control Panel&quot;, &quot;Display&quot;, &quot;Settings&quot; and move the &quot;Screen
resolution&quot; slider to the right.
Copy Plot Data to Clipboard places the plotted data on the clipboard. Use paste special to paste as a picture
meta-file, bitmap or as a text listing of the data points.
Next Curve takes you to the curves for the next grouping.
Previous Curve takes you to the curves for the previous grouping.
Click for Relative (Absolute) x-axis plots relative to difficulty of the current item or relative to the latent trait.
Select Curves enables you to jump to the set of curves you want to see by clicking on the list that is displayed.
Background color changes the background color behind the plot, or the color of the selected line on the plot.
Click on Close Box
in upper right corner to close.
This shows the probability of observing each ordered category according to the Rasch model. To identify a
category, click on it:
The caption can be clicked on and moved. &quot;2&quot; is the category score. &quot;25% independent&quot; is the category
description from CFILE= or CLFILE=.
To delete the line corresponding to a category, double-click on it:
For individual items, the horizontal scaling can be changed from relative to item difficulty to relative to the latent
variable by clicking on &quot;Click for Absolute x-axis&quot;:
239
258.
Empirical category curves
These are the empirical (data-describing) category curves. They are obtained by clicking on &quot;Empirical Cat.
Curves&quot; or from the Graphs menu. The width of each empirical interval can be adjusted by the &quot;Empirical Interval&quot;
control. The smoothing of the empirical curves by means of cubic splines is adjusted by the smoothing control.
259.
Expected score ICC
Select by clicking on &quot;Expected Score ICC&quot; or from the Graphs menu. Expected Score ICC plots the modelexpected item characteristic curve. This shows the Rasch-model prediction for each measure relative to item
difficulty Its shape is always ascending monotonic. The dashed lines indicate the Rasch-half-point thresholds
correspond to expected values of .5 score points. The intervals on the x-axis demarcated by dashed lines are the
zones within which the expected score rounds to each observed category. To remove the dashed lines, doubleclick on them.
260.
Empirical ICC
Select by clicking on &quot;Empirical ICC&quot; or from the Graphs menu. This shows the empirical (data-descriptive) item
characteristic curve. Each black &quot;x&quot; represents observations in an interval on the latent variable. The &quot;x&quot; is
positioned at the average rating (y-axis) at the average measure (x-axis) for observations close by. &quot;Close by&quot; is
set by the empirical slider beneath the plot. The blue lines are merely to aid the eye discern the trend. The curve
can be smoothed with the &quot;smoothing&quot; slider.
240
This shows the joint display of the expected and empirical ICCs. The boundary lines indicate the upper and lower
95% two-sided confidence intervals (interpreted vertically). When an empirical point lies outside of the
boundaries, then some unmodeled source of variance maybe present in the observations. Double-click on a line
on this plot to remove it from the display. The solid red &quot;model&quot; line is generated by the relevant Rasch model.
For a test of dichotomous items, these red curves will be the same for every item. The empirical blue line is
constructed from the observed frequencies along the variable, marked by x. The empirical (&quot;x&quot;) x- and ycoordinates are the means of the measures and ratings for observations in the interval. The upper green line
(and the lower grey line) are at 1.96 model standard errors above (and below) the model &quot;red line&quot;, i.e., form a
two-sided 95% confidence band. The distance of these lines from the red line is determined by the number of
observations in the interval, not by their fit.
261.
Empirical randomness
Select by clicking on &quot;Empirical Randomness&quot; or from the Graphs menu. Empirical intervals are set with the
&quot;Empirical Interval&quot; slider.
This displays the local value of the mean-square statistics. The Outfit mean-square statistic (standardized residual
chi-square divided by its degrees of freedom) is the red line. The Infit mean-square statistic (ratio of observed to
expected residual variance) is the blue line.
262.
Cumulative probabilities
241
Select by clicking on &quot;Cumulative Probabilities&quot; or from the Graphs menu. This shows the modeled category
probability curves accumulated so that the left-hand curve (red) is the probability of being observed in the lowest
category. The next curve (blue) is the probability of being observed in the lowest or next lowest category. And so
on to the right. The points of intersection between these curves and the 0.5 probability line are the RaschThurstone thresholds. The points at which being observed in this category (or below) and the category abover (or
higher) are equal. These curves are always in the order of the category scores.
263.
Item information function
Select by clicking on &quot;Item Information&quot; or from the Graphs menu. This shows the (Ronald A.) Fisher information
in responses made to items. It is the same as the binomial variance (dichotomies) or the multinomial variance
(polytomies).
264.
Category information function
Select by clicking on &quot;Category Information&quot; or from the Graphs menu. This shows the item information partitioned
according to the probability of observing each category.
265.
Conditional probability curves
Select by clicking on &quot;Conditional Probabilities&quot; or from the Graphs menu. Conditional probabilities of observing
242
adjacent categories. These are a series of Rasch dichotomous ogives. The intercepts with 0.5 probability are the
Rasch-Andrich thresholds. They can be disordered relative to the latent variable.
266.
Test characteristic curve
Select by clicking on &quot;Test CC&quot; or from the Graphs menu. This is the test characteristic curve, the score-tomeasure ogive for this set of items. It is always monotonic ascending. See also Table 20.
267.
Test information function
Select by clicking on &quot;Test Information&quot; or from the Graphs menu. This shows the Fisher information for the test
(set of items) on each point along the latent variable. The information is the inverse-square of the person measure
at that location of the variable. See also Table 20.
268.
Test randomness
Select by clicking on &quot;Test Randomness&quot; or from the Graphs menu. Empirical intervals are set with the &quot;Empirical
Interval&quot; slider. This displays the local value of the mean-square statistics. The Outfit mean-square statistic
(standardized residual chi-square divided by its degrees of freedom) is the red line. The Infit mean-square statistic
(ratio of observed to expected residual variance) is the blue line.
243
269.
Multiple item ICCs
Select by clicking on &quot;Multiple Item ICCs&quot; or from the Graphs menu. This enables the display of multiple model
and empirical ICCs on the same graph. Click on the &quot;Model&quot; and &quot;Empirical&quot; curves you wish to display. Click
again on your selection to clear it.
Displayed are the selected model and empirical ICCs.
270.
Compare statistics
From the Plots menu, this enables the simple graphical or tabular comparison of equivalent statistics from two
runs. External files must be in IFILE= or PFILE= format.
244
There are several decisions to make:
1. Do you want to plot person (row) or item (column) statistics?
2. Which statistic for the x-axis?
3. Which statistic for the y-axis?
4. Do you want to use the statistic from this analysis or from the PFILE= or IFILE= of another analysis?
5. Do you want to display in the statistics as Columns in a Table or as an Excel scatterplot or both?
If you are using the statistic from a PFILE= or IFILE= and Winsteps selects the wrong column, then identify the
correct column using the &quot;Statistic field number&quot; area.
When two measures are compared, then their standard errors are used to construct confidence bands:
Here the item calibrations in the current analysis are being compared with the item calibrations in file
IFILE=SFIF.txt from another analysis. This is the columnar output:
TABLE 34.1 An MCQ Test: administration was Comput ZOU630WS.TXT Apr 21 2:21 2006
INPUT: 30 STUDENTS 69 TOPICS MEASURED: 30 STUDENTS 69 TOPICS 2 CATS
3.60.2
-------------------------------------------------------------------------------+----------------------------------------------------------------------------------------------------+
|
Measures
|
Differences
|
Measures
| Comparison
|
|
|
|
SFIF.txt
|
|
|-4
1|-2
5|-3
2| NUM TOPIC
|
|-------------------+--------------------------+-------------------+---------------------------------|
|
*
.
|
.
*
|
*.
| 1 nl01 Month
|
|
. * | *
.
|
* .
| 2 nl02 Sign
|
|
*
.
|
.
*
|
.*
| 3 nl03 Phone number
|
| *
.
|
.
* |
.
* | 4 nl04 Ticket
|
|
*
.
|
.
*|
.
*| 5 nl05 building
|
|*
.
|
.
*
|
.*
| 6 nm01 student ticket
|
|
*
.
|
.
*
|
. *
| 7 nm02 menu
|
|
*
.
|
.
*
|
.
*
| 8 nm03 sweater
|
|
*
.
|
.
*
|
. *
| 9 nm04 Forbidden City
|
|
*
.
|
.
*
|
*
.
| 10 nm05 public place
|
|
*
.
|
.
*
|
* .
| 11 nm06 post office
|
|
*
.
|
.
*
|
*
.
| 12 nm07 sign on wall
|
|
*
.
|
*
|
*
.
| 13 nh01 supermarket
|
|
*
.
|
.
*
|
.*
| 14 nh02 advertisement
|
|
.*
|
* .
|
* .
| 15 nh03 vending machine
|
|
*
.
|
.
*
|
.*
| 16 nh04 outside store
|
|
*
|
*
|
*
| 17 nh05 stairway
|
|
*
.
|
* .
| *
.
| 18 nh06 gas station
|
|
*
.
|
* .
|
*
.
| 19 nh07 Taipei
|
|
*
.
|
.
*
|
.
* | 20 nh08 window at post office |
|
*
|
* .
|
* .
| 21 nh09 weather forecast
|
|
*
.
|
.
*
|
*
| 22 nh10 section of newspaper |
|
*
.
|
.
*
|
.
* | 23 nh11 exchange rate
|
|
*.
|
*
|
*.
| 24 il01 open
|
|
* .
|
.
*
|
.*
| 25 il02 vending machine
|
+----------------------------------------------------------------------------------------------------+
and the plotted output:
We are selecting only the first 4 characters of the item label, e.g., &quot;nl01&quot; and plotting only the Label:
245
The points are plotted by their labels. The squiggly lines are the confidence bands. They are not straight because
the standard errors of the points differ greatly. You can use the Excel functions to draw in straight lines by eye
and remove the plotted curved lines. In this plot, the dotted line is the empirical equivalence (best-fit) line. The
empirical identity line is shown in another plot by selecting the tab on the bottom of the Excel screen (green
arrow). You can edit the data points by selecting the tab labeled &quot;Z...&quot; (red arrow).
271.
Bubble charts
From the Plots menu, Bubble charts show measures and fit values graphically. They are featured in Bond &amp; Fox.
For successful operation, Excel must be available on your computer.
To produce these charts, Winsteps writes out the requisite values into a temporary file. Excel is then launched
automatically. It reads in the temporary file and follows instructions to create a bubble plot. The plot is displayed,
and the Excel worksheet becomes available for any alterations that are desired. The Excel worksheet may be
made permanent using &quot;Save As&quot;.
Selecting &quot;Bubble Chart&quot; from the Plots menu:
Here is the vertical-measure bubble-plot for the KCT data as it appears on your computer screen. Bubbles are
sized by their standard errors.
246
Or when it is formatted for printing portrait-orientation on legal-size paper, using Excel printer set-up:
And the horizontal-measure bubble-plot for the KCT data as it appears on your computer screen:
Bubble Resizing: The relative bubble sizes are set by the standard errors of the measures. But their overall
standard size is set by Excel. To change their overall sizes, right click on the edge of a circle. Then use Excel's
&quot;Format Data Series&quot;, &quot;Options&quot;, &quot;Scale bubble size to:&quot;
247
272.
Keyform plot
From the Plots menu, Keyforms are self-measuring and diagnosing forms, such as the KeyMath diagnostic
profile, and the KeyFIM. These are a powerful application of Rasch measurement for instantaneous use. These
are plotted in horizontal or vertical format using Excel - but be patient, Excel is somewhat slow to display them.
The keyform can be plotted with either horizontal or vertical orientation. In earlier version of Winsteps, these were
specified by KEYFORM=
The 7 columns in the Excel Worksheet are:
For points in the KeyForm:
COLUMN The horizontal location (x-axis) in the vertical layout or vertical location (y-axis) in the horizontal
layout.
MEASURE The measure (y-axis) in the vertical layout or (x-axis) in the horizontal
POINT-LABEL The value with which to label the point. Use the Excel addin at www.winsteps.com/ministep.htm
For column (row headings)
COLUMN The horizontal location (x-axis) in the vertical layout or vertical location (y-axis) in the horizontal
layout.
HEAD-MEASURE The top-of-column measure (y-axis) in the vertical layout or end-of-row (x-axis) in the
horizontal
ITEM-ID
The item number
ITEM-LABEL The item identifying label
Example: For the first 3 items of the &quot;Liking For Science&quot; Data
COLUMN MEASURE
1
1
1
1
1
-3.87
-2.79
-.42
1.94
3.02
2
2
2
2
2
-4.52
-3.44
-1.08
1.29
2.37
3
3
3
3
3
-1.94
-.86
1.50
3.87
4.95
5
5
5
5
5
5
5
-4.93
-3.28
-1.64
.04
1.59
3.23
5.03
7
9
7
9
7
9
7
9
7
9
-5
-5
-4
-4
-3
-3
-2
-2
-1
-1
POINT-LABEL
COLUMN HEAD-MEASURE
ITEM-ID ITEM-LABEL
1
6.00
1
WATCH BIRDS
0
1
2
2
6.00
2
READ BOOKS ON ANIMALS
0
1
2
3
6.00
3
READ BOOKS ON PLANTS
0
1
2
5
6.00
Raw Score
Raw Score
0
1
2
3
4
5
6
7
6.00
Measure Measure
9
6.00
S.E.
S.E.
-5
1.97
-4
1.32
-3
1.32
-2
1.29
-1
1.29
248
7
9
7
9
7
9
7
9
7
9
7
9
0
0
1
1
2
2
3
3
4
4
5
5
0
1.28
1
1.22
2
1.22
3
1.38
4
1.38
5
2.02
with Excel, produces vertical plots like: (These can be plotted directly from the Plots menu.)
with Excel, produces horizontal plots like: (These can be plotted directly from the Plots menu.)
273.
DIF Plot
From the Plots menu, DIF Plot 30 produces the plotted output corresponding to Table 30. It shows differential
item functioning (DIF).
249
In the DIF= box specify the column in the person label that identifies the DIF classification for each person.
Select what item identification is to be displayed on the the Excel plot. In this case, the item entry number. The
person classifications will be identified by their column codes.
There are three standard plots:
t-value
reports simple t-tests of the item DIF against the overall item difficulty
Relative Measure reports the size of the item DIF relative to the overall item difficulty
Local Measure
reports the difficulty of the item for each person classification
274.
DPF Plot
From the Plots menu, DPF Plot 31 produces the plotted output corresponding to Table 31. It shows differential
person functioning (DPF).
250
This brings up
In the DPF= box specify the column in the item label that identifies the DPF classification for each person.
Select what person identification is to be displayed on the the Excel plot. In this case, the person label. The item
classifications will be identified by their column codes.
There are three standard plots:
251
t-value
reports simple t-tests of the item DIF against the overall item difficulty
Relative Measure reports the size of the item DIF relative to the overall item difficulty
Local Measure
reports the difficulty of the item for each person classification
275.
DIF-DPF Plot
From the Plots menu, DIF-DPF Plot 33 produces the plotted output corresponding to Table 33. It shows
differential functioning between classes of items and classes of persons, DIF &amp; DPF, also called differential group
functioning (DGF).
This brings up
In the upper DIF= box specify the column in the person label that identifies the DIF classification for each person.
In the lower DPF= box specify the column in the item label that identifies the DPF classification for each person.
There are four standard plots:
Relative Measure-ip reports the relative measures of the item classes as points and the person classes as
columns
Relative Measure-pi reports the relative measure of the person classes as points and the item classes as columns
t-value-ip
reports simple t-tests of the item class measures against the overall measures for each person
class
t-value-pi
reports simple t-tests of the person class measures against the overall measures for each item
252
class
276.
Advice to novice analysts
In test construction, the rule is &quot;all items must be about the same thing, but then be as different as possible&quot;! The
central idea is that there is a latent variable which we are attempting to measure people on. The empirical
definition of the latent variable is the content of the items. Essentially, we should be able to summarize on the
items into a sentence which matches our intended definition of the latent variable. Latent variables can be very
broad, e.g., &quot;psychological state&quot; or &quot;educational achievement&quot;, or very narrow, e.g., &quot;degree of paranoia&quot; or
&quot;ability to do long division&quot;.
In other words, all items share something in common, but each item also brings in something that the others don't
have.
Of course, this never happens perfectly. So what we need is:
(a) all items to point in the same direction, so that a higher rating (or &quot;correct&quot; answer) on the item indicates more
of the latent variable. The first entry on the Diagnosis menu displays correlations. Items with negative correlations
probably need their scoring reversed with IVALUE=.
(b) what the items share overwhelms what they don't share
(c) what the items don't all share, i.e., what is different about each of the items, is unique to each item or shared
by only a few items.
What they all (or almost all) share, is usually thought of as the &quot;test construct&quot;, the &quot;major dimension&quot;, or the
&quot;Rasch dimension&quot;, or the &quot;first factor in the data&quot;. This is what test validation studies focus on. Evaluating or
confirming the nature of this construct.
What is unique to each item, or to clusters of only a few items, are &quot;subdimensions&quot;, &quot;secondary dimensions&quot;,
&quot;secondary contrasts&quot;, &quot;misfit to the Rasch dimension&quot;, etc. We are concerned to evaluate: (i) are they a threat to
scores/measures on the major dimension? (ii) do they manifest any useful information?
There are always as many contrasts in a test as there are items (less one). So how do we proceed?
(a) We want the first dimension to be much larger than all other dimensions, and for all items to have a large
positive loading on it. This is essentially what the point-biserial correlation tells us in a rough way, and Rasch
analysis tells us in a more exact way.
(b) We want so few items to load on each subdimension that we would not consider making that subdimension
into a separate instrument. In practice, we would need at least 5 items to load heavily on a contrast, maybe more,
before we would consider those items as a separate instrument. Then we crossplot and correlate scores or
measures on the subdimension against scores on the rest of the test to see its impact.
(c) When a contrast has 2 items or less heavily loading on it - it may be interesting, but it is only a wrinkle in this
test. For instance, when we look at a two item contrast, we may say, &quot;That is interesting, we could make a test of
items like these!&quot; But to make that test, we would need to write new items and collect more data. Its impact on
this test is obviously minimal.
In reporting your results, you would want to:
(a) Describe, and statistically support, what most items share: the test construct.
(b) Identify, describe and statistically support, sub-dimensions big enough to be split off as separate tests. Then
contrast scores/measures on those subdimensions with scores/measures on the rest of the test.
(c) Identify smaller sub-dimensions and speculate as to whether they could form the basis of useful new tests.
In all this, statistical techniques, like Rasch analysis and factor analysis, support your thinking, but do not do your
253
thinking for you!
In what you are doing, I suggest that you choose the simplest analytical technique that enables you to tell your
story, and certainly choose one you understand!
277.
Anchored estimation
Anchoring or fixing parameter estimates (measures) is done with IAFILE= for items, PAFILE= for persons, and
SAFILE= for response structures.
From the estimation perspective under JMLE, anchored and unanchored items appear exactly alike. The only
difference is that anchored values are not changed at the end of each estimation iteration, but unanchored
estimates are. JMLE converges when &quot;observed raw score = expected raw score based on the estimates&quot;. For
anchored values, this convergence criterion is never met, but the fit statistics etc. are computed and reported by
Winsteps as though the anchor value is the &quot;true&quot; parameter value. Convergence of the overall analysis is based
on the unanchored estimates.
Using pre-set &quot;anchor&quot; values to fix the measures of items (or persons) in order to equate the results of the
current analysis to those of other analyses is a form of &quot;common item&quot; (or &quot;common person&quot;) equating. Unlike
common-item equating methods in which all datasets contribute to determining the difficulties of the linking items,
the current anchored dataset has no influence on those values. Typically, the use of anchored items (or persons)
does not require the computation of equating or linking constants. During an anchored analysis, the person
measures are computed from the anchored item values. Those person measures are used to compute item
difficulties for all non-anchored items. Then all non-anchored item and person measures are fine-tuned until the
best possible overall set of measures is obtained. Discrepancies between the anchor values and the values that
would have been estimated from the current data can be reported as displacements. The standard errors
associated with the displacements can be used to compute approximate t-statistics to test the hypothesis that the
displacements are merely due to measurement error.
278.
Average measures, distractors and rating scales
The &quot;average measure&quot; for a category is the average ability of the people who respond in that category or to that
distractor (or distracter. The term &quot;distractor&quot; has been in use since at least 1934, and was perhaps originated by
Paul Horst in 1933). This is an empirical value. It is not a Rasch-model parameter.
The &quot;step difficulty&quot; (Rasch-Andrich threshold, step calibration, etc.) is an expression of the log-odds of being
observed in one or other of the adjacent categories. This is a model-based value. It is a Rasch-model parameter.
Our theory is that people who respond in higher categories (or to the correct MCQ option) should have higher
average measures. This is verified by &quot;average measure&quot;.
Often there is also a theory about the rating scale, such as &quot;each category in turn should be the most probable
one to be observed as one advances along the latent variable.&quot; If this is your theory, then the &quot;step difficulties&quot;
should also advance. But alternative theories can be employed. For instance, in order to increase item
discrimination one may deliberately over-categorize a rating scale - visual-analog scales are an example of this. A
typical visual analog-scale has 101 categories. If these functioned operationally according to the &quot;most probable&quot;
theory, it would take something like 100 logits to get from one end of the scale to the other.
The relationship between &quot;average measure&quot; and &quot;step difficulties&quot; or &quot;item difficulties&quot; is complex. It is
something like:
step difficulty = log ((count in lower category) / (count in higher category)) + (average of the measures across both
categories) - normalizer
normalized so that: sum(step calibrations) = 0
So that,
the higher the frequency of the higher category relative to the lower category, the lower (more negative) the step
calibration (and/or item difficulty)
254
and the higher the average of the person measures across both categories, the higher (more positive) the step
calibration (and/or item difficulty)
but the step calibrations are estimated as a set, so that the numerical relationship between a pair of categories is
influenced by their relationships with every other category. This has the useful consequence that even if a
category is not observed, it is still possible to construct a set of step calibrations for the rating scale as a whole.
Rules of Thumb:
In general, this is what we like to see:
(1) More than 10 observations per category (or the findings may be unstable, i.e., non-replicable)
(2) A smooth distribution of category frequencies. The frequency distribution is not jagged. Jaggedness can
indicate categories which are very narrow, perhaps category transitions have been defined to be categories. But
this is sample-distribution-dependent.
(3) Clearly advancing average measures.
(4) Average measures near their expected values.
(5) Observations fit of the observations with their categories: Outfit mean-squares near 1.0. Values much above
1.0 are much more problematic than values much below 1.0.
279.
Automating file selection
Use the Winsteps &quot;Batch&quot; pull-down menu to do this.
Assigning similar names to similar disk files can be automated using Batch commands.
For example, suppose you want to analyze your data file, and always have your output file have suffix &quot;O.TXT&quot;,
the PFILE have suffix &quot;.PF&quot; and the IFILE have suffix &quot;.IF&quot;. Key in your control file and data, say &quot;ANAL1&quot;,
omitting PFILE= and IFILE= control variables. Then key in the following Batch script file, called, say,
MYBATCH.BAT (for Windows-95 etc.) or MYBATCH.CMD (for Windows-2000 etc.), using your word processor,
saving the file as an ASCII or DOS text file:
REM
the MYBATCH.BAT batch file to automate WINSTEPS
START /w ..\WINSTEPS BATCH=YES %1 %1O.TXT PFILE=%1.PF IFILE=%1.IF
For WINSTEPS, specify BATCH=YES to close all windows and terminate the program when analysis is complete.
To execute this, type at the DOS prompt (or using the Winsteps &quot;Batch&quot; pull-down menu:
C:&gt; MYBATCH ANAL1 (Press Enter Key)
This outputs the tables in ANAL1O.TXT, PFILE= in ANAL1.PF and IFILE= in ANAL1.IF.
You can also edit the files WINBATCH.BAT or WINBATCH.CMD. These can be executed from the DOS prompt
or from the Winsteps Batch pull-down menu. See Running Batch Files.
280.
Batch mode example: Score Tables
Winsteps produces a Score Table in Table 20 which is for responses to every active item. You can use the Batch
File feature to automate the production of Score Tables for subtests of items.
(1) Do the calibration run:
title=&quot;Item calibration&quot;
ni=3
; these match your data file
item1=1
name1=1
codes=123
ISGROUPS=0
ifile= if.txt ; item calibrations
sfile= sf.txt ; rating (or partial credit) scale strucure
data=data.txt
&amp;end
255
......
END LABELS
(2) Set up the control file for the batch run:
; This is scon.txt
title=&quot;Produce score table&quot;
ni=3
; match your data file
item1=1
name1=1
codes=123
ISGROUPS=0
iafile = if.txt
; item anchor file
safile = sf.txt
; rating (or partial credit) scale structure anchor file
pafile=*
1 0
; dummy person measures
2 0
*
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
&amp;end
....
END LABELS
121.... ; two lines of dummy data - such that every item has a non-extreme score.
212....
(3) Set up the Batch (.bat or .cmd) file and run it. Use IDFILE=* to select the items you want (or ISELECT=)
rem this is score.cmd
del sc*.txt
start /w ..\winsteps sc.txt
start /w ..\winsteps sc.txt
start /w ..\winsteps sc.txt
start /w ..\winsteps sc.txt
start /w ..\winsteps sc.txt
start /w ..\winsteps sc.txt
start /w ..\winsteps sc.txt
.....
copy sc*.txt scores.txt
dummy
dummy
dummy
dummy
dummy
dummy
dummy
SCOREFILE=sc001.txt
SCOREFILE=sc010.txt
SCOREFILE=sc100.txt
SCOREFILE=sc011.txt
SCOREFILE=sc101.txt
SCOREFILE=sc110.txt
SCOREFILE=sc111.txt
batch=yes
batch=yes
batch=yes
batch=yes
batch=yes
batch=yes
batch=yes
idfile=*
idfile=*
idfile=*
idfile=*
idfile=*
idfile=*
idfile=*
+3
+2
+1
+2
+1
+1
+1
* title=item3
* title=item2
* title=item1
+3 * title=items23
+3 * title=item13
+2 * title=items12
+2 +3 * title=items123
(4) The Score Tables are in file scores.txt
281.
Biserial correlation
If the sample is normally distributed (i.e., conditions for the computation of the biserial exist), then to obtain the
biserial correlation from the point-biserial:
Biserial = Point-biserial * f(P-value)
Example: Specify PTBISERIAL=Yes and PVALUE=Yes. Display Table 14.
+-------------------------------------------------------------------------------------+
|ENTRY
RAW
MODEL|
INFIT | OUTFIT |PTBIS| P- |
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.|VALUE| TAP
|
|------------------------------------+----------+----------+-----+-----+--------------|
|
8
27
34
-2.35
.54| .59 -1.3| .43
-.2| .65| .77| 1-4-2-3
|
Point-biserial = .65. P-value = .77. Then, from the Table below, f(P-value) = 1.3861, so Biserial correlation = .65 *
1.39 = 0.90
Here is the Table of p-value and f(p-value).
p-va f(p-val) p-va f(p-val)
0.99 3.7335
0.01 3.7335
0.98 2.8914
0.02 2.8914
256
0.97 2.5072
0.03 2.5072
0.96 2.2741
0.04 2.2741
0.95 2.1139
0.05 2.1139
0.94 1.9940
0.06 1.9940
0.93 1.8998
0.07 1.8998
0.92 1.8244
0.08 1.8244
0.91 1.7622
0.09 1.7622
0.90 1.7094
0.10 1.7094
0.89 1.6643
0.11 1.6643
0.88 1.6248
0.12 1.6248
0.87 1.5901
0.13 1.5901
0.86 1.5588
0.14 1.5588
0.85 1.5312
0.15 1.5312
0.84 1.5068
0.16 1.5068
0.83 1.4841
0.17 1.4841
0.82 1.4641
0.18 1.4641
0.81 1.4455
0.19 1.4455
0.80 1.4286
0.20 1.4286
0.79 1.4133
0.21 1.4133
0.78 1.3990
0.22 1.3990
0.77 1.3861
0.23 1.3861
0.76 1.3737
0.24 1.3737
0.75 1.3625
0.25 1.3625
0.74 1.3521
0.26 1.3521
0.73 1.3429
0.27 1.3429
0.72 1.3339
0.28 1.3339
0.71 1.3256
0.29 1.3256
0.70 1.3180
0.30 1.3180
0.69 1.3109
0.31 1.3109
0.68 1.3045
0.32 1.3045
0.67 1.2986
0.33 1.2986
0.66 1.2929
0.34 1.2929
0.65 1.2877
0.35 1.2877
0.64 1.2831
0.36 1.2831
0.63 1.2786
0.37 1.2786
0.62 1.2746
0.38 1.2746
0.61 1.2712
0.39 1.2712
0.60 1.2682
0.40 1.2682
0.59 1.2650
0.41 1.2650
0.58 1.2626
0.42 1.2626
0.57 1.2604
0.43 1.2604
0.56 1.2586
0.44 1.2586
0.55 1.2569
0.45 1.2569
0.54 1.2557
0.46 1.2557
0.53 1.2546
0.47 1.2546
0.52 1.2540
0.48 1.2540
0.51 1.2535
0.49 1.2535
0.50 1.2534
0.50 1.2534
282.
Category boundaries and thresholds
Conceptualizing rating scales and partial-credit response structures for communication can be challenging. Rasch
measurement provides several approaches. Choose the one that is most meaningful for you.
Look at this excerpt of Table 3.2:
-----------------------------------------------------------------------------------|
DATA
| QUALITY CONTROL |
STEP
| EXPECTATION |.5 Cumul.|
| Category Counts Cum.| Avge Exp. OUTFIT|CALIBRATIONS | Measure at
|Probabil.|
|Score
Used
%
% | Meas Meas
MnSq |Measure S.E.|Category -0.5 |
at
|
-----------------------------------------------------------------------------------| 0
891
6%
6%| -.04
-.07 1.3 |
|( -1.12)
|
low
|
| 1
383
3%
9%|
.47
.51 1.3 | 1.07
.04|
-.31
-.74|
-.25 |
| 2
1017
7% 15%| 1.07
1.17
.8 | -.15
.04|
.30
-.01|
-.02 |
| 3
12683 85% 100%| 2.15
2.14 1.0 | -.91
.03|( 1.13)
.73|
.25 |
------------------------------------------------------------(Mean)---------(Median)-
Here at three ways of conceptualizing and communicating the transition, threshold, boundary between category 1
and category 2:
(1) Rasch-half-point thresholds. Someone at the boundard between &quot;1&quot; and &quot;2&quot; would have an expected rating of
257
1.5, or 1000 persons at the boundary between &quot;1&quot; and &quot;2&quot; would have an average rating of 1.5. This boundary is
the &quot;EXPECTATION Measure at 2 -0.5&quot; which is -.01 logits, the Rasch-half-point threshold. To illustrate this, use
the model item characteristic curve. The expected score ogive / model ICC (Table 21.2 - second on list in Graphs
menu). The CAT+.25, CAT-0.5, AT CAT, and CAT-.25 columns in the ISFILE= plot points on this ogive. The
expected score ogive relates most directly to the estimation of the Rasch parameters. Since it is only one line, it is
also convenient for summarizing performance at any point on the latent variable by one number. Crucial points
are the points on the variable corresponding to the lower category value + 0.5, i..e, more than the higher adjacent
category value - 0.5. These Rasch-half-point thresholds are &quot;average score thresholds&quot; or &quot;Rasch-ICC
thresholds&quot;.
(2) Rasch-Thurstone thresholds. Someone at the boundary between &quot;1&quot; and &quot;2&quot; would have a 50% chance of
being rated 1 or below, and a 50% chance of being rated 2 or above. This is the Rasch-Thurstone threshold of .02 logits. To illustrate this, use the cumulative probability curves. The cumulative probability curves (Table 21.3 and third on list in Graphs menu). The 50%PRB columns in the ISFILE= are the crucial points on these curves.
and are the Rasch-Thurstone thresholds, useful for identifying whether a person is most likely to respond below,
at or above a certain category.
(3) Rasch-Andrich thresholds. Someone at the boundary between &quot;1&quot; and &quot;2&quot; would have an equal chance of
being rated 1 or 2. This is the Rasch-Step Calibration (Rasch-Andrich Threshold) of -.15 logits. To illustrate this,
use the category probability curves. The probability curves (Table 21.1 - and top of list in Graphs menu). The
Structure MEASURE in the ISFILE= gives the point of equal probability between adjacent categories. The points
of highest probability of intermediate categories are given by the AT CAT values. These probability curves relate
most directly to the Rasch parameter values, also called Rasch-Andrich thresholds. They are at the intersection of
adjacent probability curves, and indicate when the probability of being observed in the higher category starts to
exceed that of being observed in the adjacent lower one. This considers the categories two at a time, but can lead
to misinference if there is Rasch-Andrich threshold disordering.
d) Empirical average measures. For any particular sample, there is the average ability of the people who scored
in any particular category of any particular item. This is the &quot;Average Measure&quot; reported in Table 3.2. This is
entirely sample-dependent. It is not reported in ISFILE=
283.
Column and classification selection and transformation
Selection: Several specifications require or allow the choice of columns in the item label, person label or data
record. They include DIF=, DPF=, IAFILE=, IMAP=, IPMATRIX=, ISORT=, ISUBTOT=, IWEIGHT=, PAFILE=,
PMAP=, PSORT=, PSUBTOT=, PWEIGHT=. There are several formats:
Specification = C number or $C number (can be followed by - number, E number, W number)
selects one or a block of columns in the person data record (or item label)
PWEIGHT = $C203W10
the person weighting is in column 203 of the data record, with a width of 10
columns.
This always works if the columns are within the person or item label. If the columns referenced are in the data
record, but outside the person label, the information may not be available.
Specification = number or $S number or S number
selects one column or a block of columns in the person or item label
DIF=3
the column containing the DIF classification identifed is the third column in the person label
PMAP=$S5 print on the person map, Table 16, the character in column 5 of the person label
Specification = number - number or number E number (also commencing with $S or S)
258
selects a block of columns from the person or item label
ISORT = 3 - 7 sort the items in Table 15 by the contents of columns 3 to 7 of the item labels.
Specification = number W number (also commencing with $S or S)
selects a block of columns from the person or item label
PSUBTOTAL = 12W3 subtotal the persons using classifiers starting column 12 with a width of 3
columns of the person label.
Specification = @Fieldname
selects one or a block of columns as specified by a prior @Fieldname= instruction
@AGEGROUPED = 2-3 ; the age group-classification is in columns 2-3 of the person label
PSUBTOTAL = @AGEGROUPED ; subtotal the persons by age classifier
Specification = measure or fit stratification, e.g., MA3
M = Measure
A = Ascending or D = Descending
1 or higher integer: number of strata
e.g., MA3 = Measures Ascending in 3 ability strata
F = Fit
I = Infit, O = Outfit
M = Mean-square, L = Log-mean-square, S = t standardized
A = Ascending or D = Descending
1 or higher integer: number of strata
e.g., FILD2 = Fit - Infit - Log-scaled - Descending -2 fit strata
Stratum for this value = Max(1+ Number of strata * (Value - Min)/(Max - Min), Number of strata)
Specification = selection + selection
+ signs can be used to concatenate selections
IMAP = 3-4 + 8W3 show in Table 12 columns 3, 4, 8, 9, 10 of the item label.
Specification = &quot;...&quot; or '...'
constants in quotes can be included with selected columns.
PMAP = 3 + &quot;/&quot; + 6 show in Table 15 column 3, a / and column 6 of the person label, e.g., F/3
Some specifications have several options, e.g.,
IAFILE = value can be one of the following:
IAFILE = 3-7
the item anchor values are in columns 3-7 of the item label
IAFILE = *
the item anchor values follow this specification in a list
IAFILE = file name the item anchor values are in the file called &quot;file name&quot;
IAFILE=, PAFILE=, etc., value is first checked for &quot;*&quot;. If not this, it is parsed as a selection. If not a valid selection,
it is parsed as a file name. Consequently, a mis-specified value may produce a &quot;file not found&quot; error.
Transformation: Selected codes can be transformed into other codes (of the same or fewer characters) for
Tables 27, 28, 30, 31, 33:
First produce the Table with the transformation box blank. Inspect the reported codes. Transform and combine
them using the transformation box in a second run. In this example, codes 1,A,B,C,D,E,F,S will all be converted to
1. 2,G,H,I,J,K,L,M,N,O,P,Q,R will all be converted to 2. T,U,V,W will all be converted to T. Codes X,Y,Z and any
others will be unchanged.
259
In each line in the transformation box, the code at the beginning (extreme left) of each line is the code into which it
and all other blank-separated codes on the line are transformed. Ranges of codes are indicated by -. To specify
a blank or hyphen as a code, place them in quotes: &quot; &quot; and &quot;-&quot;. Codes are matched to the transformation box
starting at the top left and line by line until a match is found, or there is no match.
284.
Comparing estimates with other Rasch software
There are many Rasch-specific software packages and IRT packages which can be configured for Rasch models.
Each implements particular estimation approaches and other assumptions or specifications about the estimates.
Comparing or combining measures across packages can be awkward. There are three main considerations:
(a) choice of origin or zero-point
(b) choice of user-scaling multiplier.
(c) handling of extreme (zero and perfect) scores.
Here is one approach:
Produce person measures from Winsteps and the other computer program on the same data set. For Winsteps
set USCALE=1 and UIMEAN=0.
Cross-plot the person measures with the Winsteps estimates on the x-axis. (This is preferential to comparing on
item estimates, because these are more parameterization-dependent.)
Draw a best-fit line through the measures, ignoring the measures for extreme scores.
The slope is the user-scaling multiplier to apply. You can do this with USCALE= slope.
The intercept is the correction for origin to apply when comparing measures. You can do this with UIMEAN= yaxis intercept.
The departure of extreme scores from the best-fit line requires adjustment. You can do this with EXTRSCORE=.
This may take multiple runs of Winsteps. If the measures for perfect scores are above the best-fit line, and
those for zero scores are below, then decrease EXTRSCORE= in 0.1 increments or less. If vice-versa, then
increase EXTRSCORE= in 0.1 increments or less.
With suitable choices of UIMEAN=, USCALE= and EXTRSCORE=, the crossplotted person measures should
approximate the identity line.
The item estimates are now as equivalent as they can be even if, due to different choice of parameterization or
estimation procedure, they appear very different.
You may notice scatter of the person measures around the identity line or obvious curvature. These could reflect
differential weighting of the items in a response string, the imposition of prior distributions, the choice of
approximation to the logistic function, the choice of parameterization of the Rasch model or other reasons.These
are generally specific to each software program and become an additional source of error when comparing
measures.
285.
Connection ambiguities
Winsteps attempts to estimate an individual measure for each person and item within one frame of reference.
Usually this happens. But there are exceptions. The data may not be &quot;well-conditioned&quot; (Fischer G.H., Molenaar,
I.W. (eds.) (1995) Rasch models: foundations, recent developments, and applications. New York: SpringerVerlag. p. 41-43).
Extreme scores (zero and perfect scores) imply measures that our beyond the current frame of reference.
Winsteps uses Bayesian logic to provide measures corresponding to those scores.
More awkward situations are shown in this data set. It is Examsubs.txt
Title = &quot;Example of subset reporting&quot;
Name1 = 1
Item1 = 9
NI = 10
&amp;End
Item 1 dropped as extreme perfect score
Item 2 in subset 2
260
Item 3 in subset 2
Item 4 in subset 3
Item 5 in subset 3
Item 6 has a Guttman
Item 7 in subset 5
Item 8 in subset 5
Item 9 in subset 6
Item 10 in subset 6
END LABELS
Alf
100000
Ben
101001
Carl
110001
David
111011
Edward 111101
Frank
011
George
001
Henry
010
Ivan
01
Jack
10
pattern: item subset 1
;
;
;
;
;
;
;
;
;
;
drops with extreme score
subset 2
subset 2
subset 3
subset 3
subset 4
subset 5
subset 5
subset 6
subset 6
There are 10 items. The first item is answered correctly by all who responded to it. So it is estimated as extreme
and dropped from further analysis. Then the first person Alf responds incorrectly to all non-extreme items and is
dropped.
After eliminating Item 1 and Alf,
Subset 1: Item 6 has a Guttman pattern. It distinguishes between those who succeeded on it from those who
failed, with no contradiction to that distinction in the data. So there is an unknown logit distance between those
who succeeded on Item 6 and those who failed on it. Consequently the difficulty of Item 6 is uncertain.
The remaining subsets have measures that can be estimated within the subset, but have unknown distance from
the persons and items in the other subsets.
Subset 2: Items 2, 3 and Ben, Carl.
Subset 3: Items 4, 5 and David, Edward.
Subset 4: Frank
Subset 5: Items 7,8 and George, Henry
Subset 6: Items 9, 10 and Ivan, Jack
Under these circumstance, Winsteps reports one of an infinite number of possible solutions. Fit statistics
and standard errors are usually correct. Reliability coefficients are accidental. Measure comparisons within
subsets are correct. Across-subset measure comparisions are accidental.
A solution would be to anchor two equivalent items (or two equivalent persons) in the different subsets to the
same values - or juggle the anchor values to make the mean of each subset the same (or whatever). Or else do
separate analyses. Or construct a real or dummy data records which include 0 &amp; 1 responses to all items.
This data set causes Winsteps to report, near the top of the iteration screen, and in the output file:
WARNING:
SUBSET 1
SUBSET 2
SUBSET 3
SUBSET 4
SUBSET 5
SUBSET 6
DATA
OF 1
OF 2
OF 2
OF 1
OF 2
OF 2
MAY BE AMBIGUOUSLY CONNECTED INTO 6 SUBSETS
ITEMS includes ITEM 6: Item 6
PERSONS includes ITEM 2 and PERSON 2: Ben
PERSONS includes ITEM 4 and PERSON 4: David
PERSONS includes PERSON 6: Frank
PERSONS includes ITEM 7 and PERSON 7: George
PERSONS includes ITEM 9 and PERSON 9: Ivan
Winsteps reports an example entry number and person name from each subset, so that you can compare their
response strings.
286.
Convergence considerations
For early runs, set the convergence criteria loosely, or use Ctrl+F to stop the iterations.
If in doubt, set the convergence criteria very tightly for your final report, e.g.,
CONVERGE=B ; both LCONV= and RCONV= apply
LCONV= .0001 ; largest logit change .0001 logits
RCONV= .01 ; largest score residual .01 score points
MJMLE= 0 ; unlimited JMLE iterations
261
and be prepared to stop the iterations manually, if necessary, using &quot;Finish Iterating&quot;, Ctrl+F, on the File pull-down
menu
Remember that convergence criteria tighter than the reported standard error of a measure are somewhat of a
numerical fiction.
The Rasch model is non-linear. This means that estimates cannot be obtained immediately and exactly, as can
be done with the solution of simultaneous linear equations. Instead, estimates are obtained by means of a series
of guesses at the correct answer.
The initial guess made by Winsteps is that all items are equally difficult and all persons equally able. The
expected responses based on this guess are compared with the data. Some persons have performed better than
expected, some worse. Some items are harder than expected, some easier. New guesses, i.e., estimates, are
made of the person abilities and item difficulties, and also the rating (or partial credit) scale structures where
relevant.
The data are again examined. This is an &quot;iteration&quot; through the data. Expectations are compared with
observations, and revised estimates are computed.
This process continues until the change of the estimates is smaller than specified in LCONV=, or the biggest
difference between the marginal &quot;total&quot; scores and the expected scores is smaller than specified in RCONV=.
The precise stopping rule is controlled by CONVERGE=. When the estimates are good enough, the iterative
process has &quot;converged&quot;. Then iteration stops. Fit statistics are computed and the reporting process begins.
There are standard convergence criteria which are suitable for most small and medium-sized complete data sets.
LCONV= is harder to satisfy for small complete data sets and many sparse data designs, RCONV= for large
complete data sets.
Anchored analyses
Anchor values always misalign somewhat with the current dataset unless they are estimated from it. Thus, the
maximum residuals can never reduce to zero. Convergence occurs when the maximum logit change is too small
to be meaningful. Accordingly, RCONV= is unproductive and only LCONV= is useful. Suggested specifications
are:
CONVERGE = L
; only LCONV is operative
LCONV = .005
; smaller than visible in standard, two decimal, output.
Missing data
For some data designs much tighter criteria are required, particularly linked designs with large amounts of missing
data. For instance, in a vertical equating design of 5 linked tests, standard convergence occurred after 85
iterations. Tightening the convergence criteria, i.e., using smaller values of LCONV= and RCONV=, convergence
occurred after 400 iterations. Further iterations made no difference as the limit of mathematical precision in the
computer for this data set had been reached. The plot shows a comparison of the item difficulties for the 5 linked
tests estimated according to the standard and strict convergence criteria.
CONVERGE = B
LCONV = .001
; 10 time stricter than usual
RCONV = .01
; 10 times stricter than usual
Note that convergence may take many iterations, and may require manual intervention to occur: Ctrl+F.
262
287.
Decimal and percentage data
Winsteps analyzes ordinal variables expressed as integers, cardinal numbers, in the range 0-254, i.e., 255
ordered categories.
Percentage observations:
Observations may be presented for Rasch analysis in the form of percentages in the range 0-100. These are
straightforward computationally but are often awkward in other respects.
A typical specification is
XWIDE = 3
CODES = &quot; 0
+ 20
+ 40
+ 60
+ 80
STKEEP = Yes
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
41 42 43 44 45 46 47 48 49 50 51 52 53 54 55
61 62 63 64 65 66 67 68 69 70 71 72 73 74 75
81 82 83 84 85 86 87 88 89 90 91 92 93 94 95
; to keep intermediate unobserved categories
16
36
56
76
96
17
37
57
77
97
18
38
58
78
98
19+
39+
59+
79+
99100&quot;
Since it is unlikely that all percentages will be observed, the rating (or partial credit) scale structure will be difficult
to estimate. Since it is even more unlikely that there will be at least 10 observations of each percentage value, the
structure will be unstable across similar datasets.
It is usually better from a measurement perspective (increased person &quot;test&quot; reliability, increased stability) to
collapse percentages into shorter rating (or partial credit) scales, e.g., 0-10, using IREFER= and IVALUE= or
NEWSCORE=.
Decimal observations:
When observations are reported in fractional or decimal form, e.g., 2.5 or 3.9, multiply them by suitable
multipliers, e.g., 2 or 10, to bring them into exact integer form.
Specify STKEEP=NO, if the range of observed integer categories includes integers that cannot be observed.
Continuous and percentage observations:
These are of two forms:
(a) Very rarely, observations are already in the linear, continuous form of a Rasch variable. Since these are in the
form of the measures produced by Winsteps, they can be compared and combined with Rasch measures using
standard statistical techniques, in the same way that weight and height are analyzed.
(b) Observations are continuous or percentages, but they are not (or may not be) linear in the local Rasch
context. Examples are &quot;time to perform a task&quot;, &quot;weight lifted with the left hand&quot;. Though time and weight are
reported in linear units, e.g., seconds and grams, their implications in the specific context is unlikely to be linear.
&quot;Continuous&quot; data are an illusion. All data are discrete at some level. A major difficulty with continuous data is
determining the precision of the data for this application. This indicates how big a change in the observed data
constitutes a meaningful difference. For instance, time measured to .001 seconds is statistically meaningless in
the Le Mans 24-hour car race - even though it may decide the winner!
To analyze these forms of data, segment them into ranges of observably different values. Identify each segment
with a category number, and analyze these categories as rating scales. It is best to start with a few, very wide
segments. If these produce good fit, then narrow the segments until no more statistical improvement is evident.
The general rule is: if the data analysis is successful when the data are stratified into a few levels, then it may be
successful if the data are stratified into more levels. If the analysis is not successful at a few levels, then more
levels will merely be more chaotic. Signs of increasing chaos are increasing misfit, categories &quot;average
measures&quot; no longer advancing, and a reduction in the sample &quot;test&quot; reliability.
May I suggest that you start by stratifying your data into 2 levels? (You can use Excel to do this.) Then
analyze the resulting the 2 category data. Is a meaningful variable constructed? If the analysis is successful (e.g.,
average measures per category advance with reasonable fit and sample reliability), you could try stratifying into
more levels.
Example: My dataset contains negative numbers such as &quot;-1.60&quot;, as well as postive numbers such as &quot;2.43&quot;. The
263
range of potential responses is -100.00 to +100.00.
Winsteps expects integer data, where each advancing integer indicates one qualitatively higher level of
performance (or whatever) on the latent variable. The maximum number of levels is 0-254. There are numerous
ways in which data can be recoded. On is to use Excel. Read your data file into Excel. Its &quot;Text to columns&quot;
feature in the &quot;Data&quot; menu may be useful. Then apply a transformation to the responses, for instance,
recoded response = integer ( (observed response - minimum response)*100 / (maximum response - minimum
response) )
This yields integer data in the range 0-100, i.e., 101 levels. Set the Excel column width, and &quot;Save As&quot; the Excel
file in &quot;.prn&quot; (formatted text) format. Or you can do the same thing in SAS or SPSS and then use the Winsteps
SAS/SPSS menu.
288.
Dependency and unidimensionality
Question: To calibrate item difficulty, I am using data from 75 subjects. Most of the subjects have been tested
repeatedly, between two and 9 times each. The reason for this was that I assumed that by training and time (with
natural development) the subjects ability was different between different testing situations. Now the referee has
asked me to verify that &quot;the requirement of local independence is not breached&quot;. How can I check this?
Unidimensionality can be violated in many different ways. If you run all known statistical tests to check for
violations (even with your subjects tested only once), your data would undoubtedly fail some of them - (for
technical details of some of these tests see Fischer &amp; Molenaar, &quot;Rasch Models&quot;, chapter 5.) Consequently, the
question is not &quot;are my data perfectly unidimensional&quot; - because they aren't. The question becomes &quot;Is the lack of
unidimensionality in my data sufficiently large to threaten the validity of my results?&quot;
Imagine that you accidentally entered all your data twice. Then you know there is a lack of local independence.
What would happen? Here is what happened when I did this with the dataset exam12lo.txt:
Data in once:
SUMMARY OF 35 MEASURED PERSONS
+-----------------------------------------------------------------------------+
|
RAW
MODEL
INFIT
OUTFIT
|
|
SCORE
COUNT
MEASURE
ERROR
MNSQ
ZSTD
MNSQ
ZSTD |
|-----------------------------------------------------------------------------|
| MEAN
38.2
13.0
-.18
.32
1.01
-.1
1.02
.0 |
| S.D.
10.1
.0
.99
.06
.56
1.4
.57
1.3 |
| MAX.
54.0
13.0
1.44
.59
2.36
2.9
2.28
2.5 |
| MIN.
16.0
13.0
-2.92
.29
.23
-2.9
.24
-2.3 |
|-----------------------------------------------------------------------------|
| REAL RMSE
.36 ADJ.SD
.92 SEPARATION 2.55 PERSON RELIABILITY .87 |
|MODEL RMSE
.33 ADJ.SD
.94 SEPARATION 2.85 PERSON RELIABILITY .89 |
| S.E. OF PERSON MEAN = .17
|
+-----------------------------------------------------------------------------+
PERSON RAW SCORE-TO-MEASURE CORRELATION = .99
CRONBACH ALPHA (KR-20) PERSON RAW SCORE RELIABILITY = .89
SUMMARY OF 13 MEASURED ITEMS
+-----------------------------------------------------------------------------+
|
RAW
MODEL
INFIT
OUTFIT
|
|
SCORE
COUNT
MEASURE
ERROR
MNSQ
ZSTD
MNSQ
ZSTD |
|-----------------------------------------------------------------------------|
| MEAN
102.9
35.0
.00
.20
1.08
-.2
1.02
-.2 |
| S.D.
23.6
.0
.93
.03
.58
2.3
.53
2.0 |
| MAX.
145.0
35.0
2.45
.31
2.16
3.9
2.42
4.3 |
| MIN.
46.0
35.0
-1.65
.18
.31
-4.2
.39
-3.3 |
|-----------------------------------------------------------------------------|
| REAL RMSE
.24 ADJ.SD
.90 SEPARATION 3.81 ITEM
RELIABILITY .94 |
|MODEL RMSE
.20 ADJ.SD
.91 SEPARATION 4.53 ITEM
RELIABILITY .95 |
| S.E. OF ITEM MEAN = .27
|
+-----------------------------------------------------------------------------+
Data in twice:
SUMMARY OF 70 MEASURED PERSONS
+-----------------------------------------------------------------------------+
264
|
RAW
MODEL
INFIT
OUTFIT
|
|
SCORE
COUNT
MEASURE
ERROR
MNSQ
ZSTD
MNSQ
ZSTD |
|-----------------------------------------------------------------------------|
| MEAN
38.2
13.0
-.18
.32
1.01
-.1
1.02
.0 |
| S.D.
10.1
.0
.99
.06
.56
1.4
.57
1.3 |
| MAX.
54.0
13.0
1.44
.59
2.36
2.9
2.28
2.5 |
| MIN.
16.0
13.0
-2.92
.29
.23
-2.9
.24
-2.3 |
|-----------------------------------------------------------------------------|
| REAL RMSE
.36 ADJ.SD
.92 SEPARATION 2.55 PERSON RELIABILITY .87 |
|MODEL RMSE
.33 ADJ.SD
.94 SEPARATION 2.85 PERSON RELIABILITY .89 |
| S.E. OF PERSON MEAN = .12
|
+-----------------------------------------------------------------------------+
PERSON RAW SCORE-TO-MEASURE CORRELATION = .99
CRONBACH ALPHA (KR-20) PERSON RAW SCORE RELIABILITY = .89
SUMMARY OF 13 MEASURED ITEMS
+-----------------------------------------------------------------------------+
|
RAW
MODEL
INFIT
OUTFIT
|
|
SCORE
COUNT
MEASURE
ERROR
MNSQ
ZSTD
MNSQ
ZSTD |
|-----------------------------------------------------------------------------|
| MEAN
205.8
70.0
.00
.14
1.08
-.3
1.02
-.4 |
| S.D.
47.2
.0
.93
.02
.58
3.2
.53
2.9 |
| MAX.
290.0
70.0
2.45
.22
2.16
5.4
2.42
6.1 |
| MIN.
92.0
70.0
-1.65
.13
.31
-6.0
.39
-4.7 |
|-----------------------------------------------------------------------------|
| REAL RMSE
.17 ADJ.SD
.92 SEPARATION 5.48 ITEM
RELIABILITY .97 |
|MODEL RMSE
.14 ADJ.SD
.92 SEPARATION 6.48 ITEM
RELIABILITY .98 |
| S.E. OF ITEM MEAN = .27
|
+-----------------------------------------------------------------------------+
There is almost no difference in the person report. The biggest impact the lack of local independence has in this
situation is to make the item standard errors too small. Consequently you might report item results as statistically
significant that aren't.
So, with your current data, you could adjust the size of the standard errors to their biggest &quot;worst case&quot; size:
Compute k = number of observations in your data / number of observations if each person had only been tested
once
Adjusted standard error = reported standard error * sqrt (k).
This would also affect Reliability computations:
Adjusted separation =reported separation / sqrt(k)
Adjusted Reliability = Rel. / ( k + Rel. - Rel.*k) = Adj.Sep**2 / (1+Adj.Sep.**2)
The size of the mean-square fit statistics does not change, but you would also need to adjust the size of the t
standardized fit statistics (if you use them). This is more complicated. It is probably easiest to read them off the
plot from Rasch Measurement Transactions 17:1 shown below.
Look at your current mean-square and significance. Find the point on the plot. Go down to the x-axis. Divide the
value there by k. Go to the same mean-square value contour. The &quot;worst case&quot; lower statistical significance value
is on the y-axis.
Another noticeable aspect of your current data could be that there are misfitting subjects who were tested 9 times,
while fitting persons are tested only twice. This would introduce a small distortion into the measurement system.
So, arrange all the Tables in fit order, and look at each end, do some subjects appear numerous times near the
end of a Table? If so, drop out those subjects and compare item calibrations with and without those subjects. If
there is no meaningful difference, then those subjects are merely at the ends of the probabilistic range predicted
by the Rasch model.
265
289.
Dichotomous mean-square fit statistics
For a general introduction, see Diagnosing Misfit
Responses:
Diagnosis
INFIT OUTFIT
Easy--Items--Hard
Pattern
MnSq
MnSq
-----------------------------------------------------111|0110110100|000 Modelled/Ideal
1.1
1.0
000|0000011111|111 Miscode
4.3
12.6
011|1111110000|000 Carelessness/Sleeping 1.0
3.8
111|1111000000|001 Lucky Guessing
1.0
3.8
101|0101010101|010 Response set/Miskey
2.3
4.0
111|1000011110|000 Special knowledge
1.3
0.9
0.3
111|1111100000|000 Guttman/Deterministic 0.5
111|1010110010|000 Imputed outliers *
1.0
0.6
-----------------------------------------------------Right|Transition|Wrong
Expectation: 1.0
Overall pattern:
high - low - high
OUTFIT sensitive to outlying observations
&gt;&gt;1.0 unexpected outliers
&lt;&lt;1.0 overly predictable outliers
low - high - low
INFIT sensitive to pattern of inlying observations
&gt;&gt;1.0 disturbed pattern
&lt;&lt;1.0 Guttman pattern
* as when a tailored test is filled out by imputing all &quot;right&quot; response to easier items and all &quot;wrong&quot; to harder
items.
The exact details of these computations have been lost, but the items appear to be uniformly distributed about 0.4
logits apart, extracted from Linacre, Wright (1994) Rasch Measurement Transactions 8:2 p. 360
The Z-score standardized Student's-t statistics report, as unit normal deviates, how likely it is to observe the
reported mean-square values, when the data fit the model. The term Z-score is used of a t-test result when either
the t-test value has effectively infinite degrees of freedom (i.e., approximates a unit normal value) or the Student's
t-distribution value has been adjusted to a unit normal value.
290.
DIF - DPF - bias - interactions concepts
Computation
The DIF (differential item functioning) or DPF (differential person functioning) analysis proceeds with all items and
persons, except the item or person currently targeted, anchored at the measures from the main analysis
266
(estimated from all persons and items, including the currently targeted ones). The item or person measure for the
current classification is then computed, along with its S.E. Mathematically, it is unlikely that no bias effects will be
observed, or that bias sizes will cancel out exactly. The DIF contrast in Table 30 and 31 is the difference
between the DIF sizes, and is a log-odds estimate, equivalent to a Mantel-Haenszel DIF size. The t is the DIF
contrast divided by the joint S.E. of the two DIF measures. It is equivalent to the Mantel-Haenszel significance
test, but has the advantage of allowing missing data. This analysis is the same for all item types supported by
Winsteps (dichotomies, rating (or partial credit) scales, etc.).
To replicate this with Winsteps yourself:
From a run of all the data, produce a PFILE=pf.txt and a SFILE=sf.txt
Then for each person classification of interest:
PAFILE=pf.txt
SAFILE=sf.txt
PSELECT=?????X ; to select only the person classification of interest
IFILE = X.txt
; item difficulties for person classification on interest
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
Do this for each class.
The IFILE= values should match the values shown in Table 30.2
To graph the ICCs for different DIF classes on the same plot, see DIF item characteristic curves.
Classification sizes
There is no minimum size, but the smaller the classification size (also called reference groups and focal groups),
the less sensitive the DIF test is statistically. Generally, results produced by classifications sizes of less than 30
are too much influenced be idiosyncratic behavior to be considered dependable.
Effect of imprecision in person or item estimates
This computation treats the person measures (for DIF) or the item measures (for DPF) as point estimates (i.e.,
exact). You can inflate the reported standard errors to allow for the imprecision in those measures. Formula 29 of
Wright and Panchapakesan (1969), www.rasch.org/memo46.htm, applies. You will see there that, for
dichotomies, the most by which imprecision in the baseline measures can inflate the variance is 25%. So, if you
multiply the DIF/DPF point estimate S.E. by sqrt(1.25) = 1.12 (and divide the t by 1.12), then you will be as
conservative as possible in computing the DIF/DPF significance.
Impact on Person/Item Measurement
Unless DIF/DPF is large and mostly in one direction, the impact of DIF/DPF on person/item measurement is
generally small. Wright &amp; Douglas (1976) Rasch Item Analysis by Hand. &quot;In other work we have found that when
[test length] is greater than 20, random values of [item calibration misestimation] as high as 0.50 have negligible
effects on measurement.&quot;
Wright &amp; Douglas (1975) Best Test Design and Self-Tailored Testing. &quot;They allow the test designer to incur item
discrepancies, that is item calibration errors, as large as 1.0. This may appear unnecessarily generous, since it
permits use of an item of difficulty 2.0, say, when the design calls for 1.0, but it is offered as an upper limit
because we found a large area of the test design domain to be exceptionally robust with respect to independent
item discrepancies.&quot;
DIF/DPF statistical significance
Table 30.1 shows pair-wise test of the statistical significance of DIF across classes. Table 30.2 shows statistical
significance of DIF for a class against the average difficulty. A statistical test for DIF for multiple classes on one
item is a &quot;fixed effects&quot; chi-square of homogeneity. For L measures, Di, with standard errors SEi, a test of the
hypothesis that all L measures are statistically equivalent to one common &quot;fixed effect&quot; apart from measurement
error is a chi-square statistics with L-1 d.f. where p&gt;.05 (or &gt;.01) indicates statisticaly equivalent estimates.
267
Non-Uniform DIF or DPF
To investigate this with the Winsteps, include in the item or person label a stratification variable, indicating, low,
middle or high performers (or item difficulties). Use this is the classification variable for DIF= or DPF=. A graphical
approach is to PSELECT= the people you want, and then draw their empirical ICCs for items. Several of these
can be combined into one plot by copying the plotted data and pasting to Excel, then using the Excel graphing
function.
The Mathematics of Winsteps DIF and DPF Estimation
The DIF and DPF are estimated as minor effects in a logit-linear procedure. The major effects are the person
abilities, item difficulties, and rating scale structures. The approach in Winsteps parallels the use of logit models in
Mellenbergh, G. J. (1982). Contingency table models for assessing item bias. Journal of Educational Statistics, 7,
105-107; Van Der Flier, H., Mellenbergh, G. J., Ad&egrave;r, H. J. &amp; Wijn, M. (1984). An iterative item bias detection
method. Journal of Educational Measurement, 21, 131-145; Kok, F. G., Mellenbergh, G. J. &amp; Van Der Flier, H.
(1985). Detecting experimentally induced item bias using the iterative logit method. Journal of Educational
Measurement, 22, 295-303.
Algebraically, the general model is in two stages:
Stage 1:
Log ( Pnij / Pni(j-1) ) = Bn - Dgi - Fgj
Where Bn is the ability of person n, Dgi is the difficulty of person i in classification g, Fgj is the Rasch-Andrich
threshold measure of category j relative to category j-1 for items in item-grouping g.
For the Rasch dichotomous model, all items are in the same item-grouping (so that g is omitted), and there are
only two categories, with F1=0.
For the Andrich rating-scale model, all items are in the same item-grouping (so that g is omitted), and there are
more than two categories, with sum(Fj)=0.
For the Masters' partial-credit model, each item is in its own item-grouping (g=i), and there are more than two
categories, with sum(Fij)=0. To reparameterize into the conventional partial-credit model formulation, Di + Fij =
Dij.
Estimates of bn, dgi and fgj are obtained.
Stage 2:
Table 30: For person-subsample (ps) DIF: Log ( Pnij / Pni(j-1) ) = bn - dgi - fgj - DIF(ps)i
Table 31: For item-subsample (is) DPF: Log ( Pnij / Pni(j-1) ) = bn - dgi - fgj + DPF(is)n
Table 33: For person-subsample item-subsample (ps)(is) DIPF: Log ( Pnij / Pni(j-1) ) = bn - dgi - fgj +
DIPF(ps)(is)
Estimates of bn, dgi and fgj anchored (fixed) from stage 1. The estimates of DIF, DPF or DIPF are the maximum
likelihood estimates for which the marginal residuals for the subsamples from the stage 1 analysis are the
sufficient statistics. All these computations are as sample-distribution-free as is statistically possible, except when
the subsampling is based on the sample-distribution (e.g., when persons are stratified into subsamples according
to their ability estimates.)
Different forms of DIF detection
A cross-plot of item difficulties derived from independent runs of the focal and reference classifying-groups, is
basically reporting &quot;Is the instrument working differently for the two sample classifications?&quot;, and, if so, &quot;Where
are the most conspicuous differences?&quot; In the old days, when much analysis was done by hand, this would
identify which items to choose for more explicitly constructed DIF tests, such as Mantel-Haenszel. From these
plots we can get approximate DIF t-tests. This approach is obviously useful - maybe more useful than the item-byitem DIF tests. But it allows DIF in an item to change the person measures, and to alter the difficulties of other
items and to change the rating (or partial credit) scale structure. To apply this &quot;differential test functioning&quot;
approach to DIF detection, perform independent analyses of each sample class, produce IFILE= and cross-plot
the measures using the Compare Statistics plot.
But, it is the item-by-item DIF tests that have traditional support. So, for these, we need to hold everything else
268
constant while we examine the DIF of each item. This is what Mantel-Haenszel does (using person raw
scores), or the Winsteps DIF Table does (using person measures).
The Winsteps DIF table is equivalent to doing:
(a) The joint run of all person classifications, producing anchor values for person abilities and rating (or partial
credit) scale structure.
(b) The classification A run with person abilities and rating (or partial credit) scale structure anchored to produce
classification A item difficulties.
(c) The classification B run with person abilities and rating (or partial credit) scale structure anchored to produce
classification B item difficulties.
(d) Pairwise item difficulty difference t-tests between the two sets of item difficulties (for classifications A and B).
Lord's Chi-square DIF method takes a different approach, automatically looking for a core of stable items, but it
is accident-prone and appears to overdetect DIF. In particular, if items were slightly biased, 50% against boys
and 50% against girls, it would be accidental which set of items would be reported as &quot;unbiased&quot; and which as
&quot;biased&quot;.
Mantel-Haenszel method. See MHSLICE=.
ANOVA method. This can be facilitated by Winsteps.
(1) Identify the relevant demographic variable in the person label, and set ITEM1= at the variable, and
NAMLEN=1.
(2) Perform a standard Winsteps analysis
(3) Use USCALE=, UMEAN= and UDECIM= to transform the person measures into convenient &quot;class intervals&quot;:
integers with lowest value 1, and highest value 10 for 10 class intervals.
(4) Write out an XFILE= selecting only:
person measure (class interval)
standardized residual
person label (demographic variable)
(5) Read this file into your statistics software.
(6) Transform the demographic variable into 1 and 2.
(7) Perform the &quot;fully randomized&quot; ANOVA with standardized residual as the dependent variable, and person
measure and person label as the independent variables.
291.
DIF item characteristic curves
Here is a way to show the item characteristic curves (ICCs) for different DIF classifications on the same plot:
TITLE='DATA SET UP FOR MULTIPLE PERSON CLASSIFICATIONS (DIF GROUPS)'
NAME1=1
@DIF = $S1W1
; DIF classification in column 1 of Person Label. Codes are 1 and 2.
ITEM1=3
NI = 32 ; 10 FOR MAIN ANALYSIS: 10 FOR EACH GROUP + 2 BLANKS
CODES=01
IWEIGHT=*
11-32 0 ; WEIGHTED ZERO: FOR DIF ICC GRAPHS ONLY
*
; ISELECT=0
; enter this in Specification pull-down menu to report 1st 10 items only
&amp;END
0-01
; original 10 items
0-02
0-03
0-04
0-05
0-06
0-07
0-08
0-09
0-10
BLANK
1-01 ; 10 items (12-21), weighted 0 for classification 1
1-02
1-03
1-04
1-05
269
1-06
1-07
1-08
1-09
1-10
BLANK
2-01 ; 10 items (23-32), weighted 0 to classification 2
2-02
2-03
2-04
2-05
2-06
2-07
2-08
2-09
2-10
END LABELS
1 0111001100 0111001100
; each item 3 times: do this with a rectangular copy
1 1111111000 1111111000
; first 10 items for the analysis
1 1101011101 1101011101
; second 10 items, weighted 0, for classification 1
1 1101101100 1101101100
; third 10 items, weighted 0, for classification 2
1 0110000100 0110000100
1 1010110100 1010110100
1 1111010110 1111010110
1 1111110100 1111110100
1 0111110000 0111110000
1 1111110110 1111110110
1 0000010000 0000010000
1 1101010010 1101010010
1 1001001100 1001001100
1 1011111001 1011111001
1 0100000000 0100000000
2 1111111100
1111111100 ; rectangular copies can be done in Word
2 1111111100
1111111100
2 1111110100
1111110100
2 1110000000
1110000000
2 0000010000
0000010000
2 0100000000
0100000000
2 1000000000
1000000000
2 0111110010
0111110010
2 1111100110
1111100110
2 0100101000
0100101000
Specifying the items on the Multiple ICCs screen produces:
The same data layout can be obtained without a rectangular copy as follows:
270
TITLE='DATA SET UP FOR MULTIPLE PERSON CLASSIFICATIONS (DIF GROUPS)'
NAME1=1
@DIF = $S1W1
; DIF classification in column 1 of Person Label. Codes are 1 and 2.
ITEM1=3
NI = 32 ; 10 FOR MAIN ANALYSIS: 10 FOR EACH GROUP + 2 BLANKS
CODES=01
IWEIGHT=*
11-32 0 ; WEIGHTED ZERO: FOR DIF ICC GRAPHS ONLY
*
; ISELECT=0
; enter this in Specification pull-down menu to report 1st 10 items only
; Format the data into 3 columns:
FORMAT = (13A,T3,11A,T3,10A) ; this copies the items 3 times.
EDFILE = *
1-15 23-32 X ; blank out top right: persons 1-15, items 23-32
16-25 12-21 X ; blank out bottom center: persons 16-25, item 12-21
*
RFILE = reformat.txt ; look at to check that the data looks like the example above
&amp;END
0-01
; original 10 items
0-02
0-03
0-04
0-05
0-06
0-07
0-08
0-09
0-10
BLANK
1-01 ; 10 items (12-21), weighted 0 for classification 1
1-02
1-03
1-04
1-05
1-06
1-07
1-08
1-09
1-10
BLANK
2-01 ; 10 items (23-32), weighted 0 to classification 2
2-02
2-03
2-04
2-05
2-06
2-07
2-08
2-09
2-10
END LABELS
1 0111001100 ; original data
1 1111111000
1 1101011101
1 1101101100
1 0110000100
1 1010110100
1 1111010110
1 1111110100
1 0111110000
1 1111110110
1 0000010000
1 1101010010
1 1001001100
1 1011111001
1 0100000000
2 1111111100
2 1111111100
2 1111110100
2 1110000000
2 0000010000
2 0100000000
2 1000000000
2 0111110010
2 1111100110
271
2 0100101000
292.
Dimensionality: contrasts &amp; variances
Please do not interpret Rasch-residual-based Principal Components Analysis (PCAR) as a usual factor
analysis. These components show contrasts between opposing factors, not loadings on one factor.
Criteria have yet to be established for when a deviation becomes a dimension. So PCA is indicative, but not
definitive, about secondary dimensions.
Example from Table 23.0 from Example0.txt:
STANDARDIZED RESIDUAL VARIANCE SCREE PLOT
Table of STANDARDIZED RESIDUAL variance (in Eigenvalue units)
Empirical
Modeled
Total variance in observations
=
127.9 100.0%
100.0%
Variance explained by measures
=
102.9 80.5%
82.0%
Unexplained variance (total)
=
25.0 19.5% 100.0%
18.0%
Unexplned variance in 1st contrast =
4.6
3.6% 18.5%
Unexplned variance in 2nd contrast =
2.9
2.3% 11.8%
Unexplned variance in 3rd contrast =
2.3
1.8%
9.2%
Unexplned variance in 4th contrast =
1.7
1.4%
6.9%
Unexplned variance in 5th contrast =
1.6
1.3%
6.5%
The Rasch dimension explains 80.5% of the variance in the data: good! The largest secondary dimension, &quot;the
first contrast in the residuals&quot; explains 3.6% of the variance - somewhat greater than around 1% that would be
observed in data like these simulated to fit the Rasch model. Check this by using the SIMUL= option in Winsteps
to simulate a Rasch-fitting dataset with same characteristics as this dataset. Then produce this Table for it. Also
see: www.rasch.org/rmt/rmt191h.htm
The eigenvalue of the biggest residual contrast is 4.6 - this indicates that it has the strength of about 5 items,
somewhat bigger than the strength of two items, the smallest amount that could be considered a &quot;dimension&quot;.
Contrast the items at the top and bottom of the first PCAR plot to identify what this secondary dimension reflects.
Rules of Thumb:
Variance explained by measures &gt; 60% is good.
Unexplained variance explained by 1st contrast (size) &lt; 3.0 is good.
Unexplained variance explained by 1st contrast &lt; 5% is good.
But there are plenty of exceptions ....
Analytical Note:
Winsteps performs an unrotated &quot;principal components&quot; factor analysis. (using Hotelling's terminology). If you
would like to rotate axes, have oblique axes, or perform a &quot;common factor&quot; factor analysis of the residuals,
Winsteps can write out the matrices of residual item (or person) correlations, see the &quot;Output Files&quot; pull down
menu or ICORFIL= and PCORFIL=. You can import these into any statistics software package.
The purpose of PCA of residuals is not to construct variables (as it is with &quot;common factor&quot; analysis), but to
explain variance. First off, we are looking for the contrast in the residuals that explains the most variance. If this
contrast is at the &quot;noise&quot; level, then we have no shared second dimension. If it does, then this contrast is the
&quot;second&quot; dimension in the data. (The Rasch dimension is hypothesized to be the first). Similarly we look for a
third dimension, etc. Rotation, oblique axes, the &quot;common factor&quot; approach, all reapportion variance, usually in
an attempt to make the factor structure more clearly align with the items, but, in so doing, the actual variance
structure and dimensionality of the data is masked.
In Rasch analysis, we are trying to do the opposite of what is usually happening in factor analysis. In Rasch
analysis of residuals, we want not to find contrasts, and, if we do, we want to find the least number of contrasts
above the noise level, each, in turn, explaining as much variance as possible. This is exactly what unrotated
PCA does.
In conventional factor analysis of observations, we are hoping desperately to find shared factors, and to assign
272
the items to them as clearly and meaningfully as possible. In this endeavor, we use a whole toolbox of rotations,
obliquenesses and choices of diagonal self-correlations (i.e., the &quot;common factor&quot; approach).
But, different analysts have different aims, and so Winsteps provides the matrix of residual correlations to enable
the analyst to perform whatever factor analysis is desired!
The Rasch Model: Expected values, Model Variances, and Standardized Residuals
The Rasch model constructs linear measures from ordinal observations. It uses disordering of the observations
across persons and items to construct the linear frame of reference. Perfectly ordered observations would accord
with the ideal model of Louis Guttman, but lack information as to the distances involved.
Since the Rasch model uses disordering in the data to construct distances, it predicts that this disordering will
have a particular ideal form. Of course, empirical data never exactly accord with this ideal, so a major focus of
Rasch fit analysis is to discover where and in what ways the disordering departs from the ideal. If the departures
have substantive implications, then they may indicate that the quality of the measures is compromised.
A typical Rasch model is:
log (Pnik / Pni(k-1) ) = Bn - Di - Fk
where
Pnik = the probability that person n on item i is observed in category k, where k=0,m
Pni(k-1) = the probability that person n on item i is observed in category k-1
Bn = the ability measure of person n
Di = the difficulty measure of item i
Fk = the structure calibration from category k-1 to category k
This predicts the observation Xni. Then
Xni = Eni &plusmn; sqrt(Vni)
where
Eni = sum (kPnik) for k=0,m.
This is the expected value of the observation.
Vni = sum (k&sup2;Pnik) - (Eni)&sup2; for k=0,m.
This is the model variance of the observation about its expectation, i.e., the predicted randomness in the
data.
The Rasch model is based on the specification of &quot;local independence&quot;. This asserts that, after the contribution of
the measures to the data has been removed, all that will be left is random, normally distributed. noise. This
implies that when a residual, (Xni - Eni), is divided by its model standard deviation, it will have the characteristics
of being sampled from a unit normal distribution. That is:
(Xni - Eni) / sqrt (Vni), the standardized residual of an observation, is specified to be N(0,1)
The bias in a measure estimate due to the misfit in an observation approximates (Xni - Eni) * S.E.&sup2;(measure)
Principal Components Analysis of Residuals
&quot;Principal Component Analysis (PCA) is a powerful technique for extracting structure from possibly highdimensional data sets. It is readily performed by solving an eigenvalue problem, or by using iterative algorithms
which estimate principal components [as in Winsteps]. ... some of the classical papers are due to Pearson (1901);
Hotelling (1933); ... PCA is an orthogonal transformation of the coordinate system in which we describe our data.
The new coordinate values by which we represent the data are called principal components. It is often the case
that a small number of principal components is sufficient to account for most of the structure in the data. These
are sometimes called factors or latent variables of the data.&quot; (Sch&ouml;lkopf, D., Smola A.J., M&uuml;ller K.-R., 1999,
Kernel Principal Component Analysis, in Sch&ouml;lkopf at al. &quot;Advances in Kernel Methods&quot;, London: MIT Press).
Pearson, K. (1901) On lines and planes of closest fit to points in space. Philosophical Magazine, 2:559572.
Hotelling, H. (1933) Analysis of a complex of statistical variables into principal components. Journal of
Educational Psychology, 24:417-441, 498-520.
The standardized residuals are modeled to have unit normal distributions which are independent and so
273
uncorrelated. A PCA of Rasch standardized residuals should look like a PCA of random normal deviates.
Simulation studies indicate that the largest component would have an eigenvalue of about 1.4 and they get
smaller from there. But there is usually something else going on in the data, so, since we are looking at residuals,
each component contrasts deviations in one direction (&quot;positive loading&quot;) against deviation in the other direction
(&quot;negative loading&quot;). As always with factor analysis, positive and negative loading directions are arbitrary. Each
component in the residuals only has substantive meaning when its two ends are contrasted. This is a little
different from PCA of raw observations where the component is thought of as capturing the &quot;thing&quot;.
Loadings are plotted against Rasch measures because deviation in the data from the Rasch model is often not
uniform along the variable (which is actually the &quot;first&quot; dimension). It can be localized in easy or hard items, high
or low ability people. The Wright and Masters &quot;Liking for Science&quot; data is an excellent example of this.
Total, Explained and Unexplained Variances
The decomposition of the total variance in the data set proceeds as follows for the standardized residual,
PRCOMP=S and raw score residual PRCOMP=R, option.
(i) The average person ability measure, b, and the average item difficulty measure, d, are computed.
(ii) The expected response, Ebd, by a person of average ability measure to an item of average difficulty measure
is computed. (If there are multiple rating or partial credit scales, then this is done for each rating or partial credit
scale.)
(iii) Each observed interaction of person n, of estimated measure Bn, with item i, of estimated measure Di,
produces an observation Xni, with an expected value, Eni, and model variance, Vni.
The raw-score residual, Zni, of each Xni is Zni = Xni-Eni.
The standardized residual, Zni, of each Xni is Zni = (Xni-Eni)/sqrt(Vni).
Empirically:
(iv) The piece of the observation available for explanation by Bn and Di is approximately Xni - Ebd.
In raw-score residual units, this is Cni = Xni-Ebd
In standardized residual units, this is Cni = (Xni-Ebd)/sqrt(Vni)
The total variance sum-of-squares in the data set available for explanation by the measures is: VAvailable =
sum(Cni&sup2;)
(v) The total variance sum of squares predicted to be unexplained by the measures is: VUnexplained = sum(Zni&sup2;)
(vi) The total variance sum of squares explained by the measures is: VExplained = VAvailable - VUnexplained
If VEXplained is negative, see below.
Under model conditions:
(viii)
The total variance sum of squares explained by the measures is:
Raw-score residuals: VMexplained = sum((Eni-Ebd)&sup2;)
Standardized residuals: VMexplained = sum((Eni-Ebd)&sup2;/Vni)
(ix) The total variance sum of squares predicted to be unexplained by the measures is:
Raw score residuals: VMunexplained = sum(Vni)
Standardized residuals: VMunexplained = sum(Vni/Vni) = sum(1)
x) total variance sum-of-squares in the data set predicted to be available for explanation by the measures is:
VMAvailable = VMexplained + VMUnexplained
Negative Variance Explained
Table of STANDARDIZED RESIDUAL variance (in Eigenvalue units)
Total variance in observations = 20.3 100.0%
Variance explained by measures = -23.7 -116.2%
According to this Table, the variance explained by the measures is less than the theoretical minimum of 0.00. This
&quot;negative variance&quot; arises when there is unmodeled covariance in the data. In Rasch situations this happens
when the randomness in the data, though normally distributed when considered overall, is skewed when
partitioned by measure difference. A likely explanation is that some items are reverse-coded. Check that all
274
correlations are positive by viewing the Diagnosis Menu, Table A. If necessary, use IREFER= to recode items. If
there is no obvious explanation, please email your control and data file to www.winsteps.com
Principal Components Analysis of Standardized Residuals
(i) The standardized residuals for all observations are computed. Missing observations are imputed to have a
standardized residual of 0, i.e., to fit the model.
(ii) Correlation matrices of standardized residuals across items and across persons are computed. The
correlations furthest from 0 (uncorrelated) are reported in Tables 23.99 and 24.99.
(iii) In order to test the specification that the standardized residuals are uncorrelated, it is asserted that all
randomness in the data is shared across the items and persons. This is done by placing 1's in the main diagonal
of the correlation matrix. This accords with the &quot;Principal Components&quot; approach to Factor Analysis. (&quot;General&quot;
Factor Analysis attempts to estimate what proportion of the variance is shared across items and persons, and
reduces the diagonal values from 1's accordingly. This approach contradicts our purpose here.)
(iv) The correlation matrices are decomposed. In principal, if there are L items (or N persons), and they are
locally independent, then there are L item components (or N person components) each of size (i.e., eigenvalue) 1,
the value in the main diagonal. But there are expected to be random fluctuations in the structure of the
randomness. However, eigenvalues of less than 2 indicate that the implied substructure (dimension) in these
data has less than the strength of 2 items (or 2 persons), and so, however powerful it may be diagnostically, it has
little strength in these data.
(v) If items (or persons) do have commonalities beyond those predicted by the Rasch model, then these may
appear as shared fluctuations in their residuals. These will inflate the correlations between those items (or
persons) and result in components with eigenvalues greater than 1. The largest of these components is shown in
Table 23.2 and 24.3, and sequentially smaller ones in later subtables.
(vi) In the Principal Components Analysis, the total variance is expressed as the sum of cells along the main
diagonal, which is the number of items, L, (or number of persons, N). This corresponds to the total unexplained
variance in the dataset, VUnexplained.
(vii) The variance explained by the current contrast is its eigenvalue.
Example: Item Decomposition
From Table 23.2: The Principal Components decomposition of the standardized residuals for the items, correlated
across persons. Winsteps reports:
Table of STANDARDIZED RESIDUAL variance (in Eigenvalue units)
Empirical
Modeled
Total variance in observations
=
1452.0 100.0% 100.0%
Variance explained by measures
=
1438.0
99.0%
98.6%
Unexplained variance (total)
=
14.0
1.0%
1.4%
Unexpl var explained by 1st contrast =
2.7
.2%
The first contrast has an eigenvalue size of 2.7 This corresponds to 2.7 items.
There are 14 active items, so that the total unexplained variance in the correlation matrix is 14 units.
The &quot;Modeled&quot; column shows what this would have looked like if these data fit the model exactly.
Conclusion: Though this contrast has the strength of 3 items, and so might be independently constructed from
these data, its strength is so small that it is barely a ripple on the total measurement structure.
Caution: The 1st contrast may be an extra dimension, or it may be a local change in the intensity of this
dimension:
Table of STANDARDIZED RESIDUAL variance (in Eigenvalue units)
Empirical
Modeled
Total variance in observations
=
97.1 100.0%
100.0%
Variance explained by measures
=
58.1 59.8%
59.0%
Unexplained variance (total)
=
39.0 40.2% 100.0%
41.0%
275
Unexpl var explained by 1st contrast =
.7
F
A
C
T
O
R
.6
.5
.4
.3
1
.2
L
O .1
A
D .0
I
N -.1
G
-.2
-.3
2.8
2.9%
-3
-2
-1
0
1
2
3
++----------+----------+----------+----------+----------+----------++
|
|
A
|
+
|
+
|
| B
C
|
+
|
+
|
|
D
|
+
E
|
+
|
|
|
+
|
+
|
|
|
+
|
+
|
F
|
|
+
|
+
|
|
G
|
+
H
|
+
|
L
I |
J K
|
+----------M-----------------N----|-O-------------------------------+
|
T S
QR
| P
s
|
+
p
o|
q
r
+
|
|
l
m
n k
|
+
i
g
|
jh
+
| f
c
e
|b
d
|
+
|
+
|
a |
|
++----------+----------+----------+----------+----------+----------++
-3
-2
-1
0
1
2
3
Item MEASURE
7.2%
COUNT
1
2
1
1
1
1
1
4
3
6
4
4
4
5
1
+----------------------------------------|
|
|
INFIT OUTFIT| ENTRY
|contrast|LOADING|MEASURE MNSQ MNSQ |NUMBER
|------+-------+-------------------+-----| 1
|
.74 |
.71 .78 .70 |A
26
| 1
|
.65 |
.26 .79 .68 |B
23
| 1
|
.64 |
1.34 .87 .84 |C
25
| 1
|
.56 |
.64 .85 .80 |D
24
| 1
|
.51 |
-.85 .84 .60 |E
22
The first contrast comprises items A-E. But their mean-squares are all less than 1.0, indicating they do not
contradict the Rasch variable, but are rather too predictable. They appear to represent a local intensification of the
Rasch dimension, rather than a contradictory dimension.
Comparison with Rasch-fitting data
Winsteps makes it easy to compare empirical PCA results with the results for an equivalent Rasch-fitting data set.
From the Output Files menu, make a &quot;Simulated Data&quot; file, call it, say, test.txt
From the Files menu, Restart Winsteps. Under &quot;Extra specifications&quot;, type in &quot;data=test.txt&quot;.
Exactly the same analysis is performed, but with Rasch-fitting data. Look at the Dimensionality table:
Table of STANDARDIZED RESIDUAL variance (in Eigenvalue units)
Empirical
Modeled
Total variance in observations
=
576.8 100.0% 100.0%
Variance explained by measures
=
562.8
97.6%
97.1%
Unexplained variance (total)
=
14.0
2.4%
2.9%
Unexpl var explained by 1st contrast =
2.2
.4%
Repeat this process several times, simulating a new dataset each time. If they all look like this, we can conclude
that the value of 2.7 for the 1st contrast in the residuals is negligibly bigger than the 2.2 expected by chance.
General Advice
A question here is &quot;how big is big&quot;? Much depends on what you are looking for. If you expect your instrument to
have a wide spread of items and a wide spread of persons, then your measures should explain most of the
variance. But if your items are of almost equal difficulty (as recommended, for instance, by G-Theory) and your
persons are of similar ability (e.g., hospital nurses at the end of their training) then the measures will only explain
a small amount of the variance.
Ben Wright recommends that the analyst split the test into two halves, assigning the items, top vs. bottom of the
first component in the residuals. Measure the persons on both halves of the test. Cross-plot the person
measures. If the plot would lead you to different conclusions about the persons depending on test half, then there
is a multidimensionality. If the plot is just a fuzzy straight line, then there is one, perhaps somewhat vague,
dimension.
276
Rules of Thumb
&quot;Reliability&quot; (= Reproducibility) is &quot;True&quot; variance divided by Observed variance. If an acceptable, &quot;test reliability&quot;
(i.e., reproducibility of this sample of person measures on these items) is 0.8, then an acceptable Rasch &quot;data
reliability&quot; is also 0.8, i.e., &quot;variance explained by measures&quot; is 4 times &quot;total unexplained variance&quot;.
In the unexplained variance, a &quot;secondary dimension&quot; must have the strength of at least 3 items, so if the first first
contrast has &quot;units&quot; (i.e., eigenvalue) less than 3 (for a reasonable length test) then the test is probably
unidimensional. (Of course, individual items can still misfit).
Negative variance can occur when the unexpectedness in the data is not random. An example is people who
flatline an attitude survey. Their unexpected responses are always biased towards one category of the rating (or
partial credit) scale.
Simulation studies indicate that eigenvalues less than 1.4 are at the random level. Smith RM, Miao CY (1994)
Assessing unidimensionality for Rasch measurement. Chapter 18 in M. Wilson (Ed.) Objective Measurement:
Theory into Practice. Vol. 2. Norwood NJ: Ablex.) On occasion, values as high as 2.0 are at the random level.
(Critical Eigenvalue Sizes in Standardized Residual Principal Components Analysis, Ra&icirc;che G., Rasch
Measurement Transactions, 2005, 19:1 p. 1012.
293.
Dimensionality: when is a test multidimensional?
For more discussion see dimensionality and contrasts.
&quot;I can not understand the residual contrast analysis you explained. For example, in Winsteps, it gave me the five
contrasts' eigenvalues: 3.1, 2.4, 1.9, 1.6, 1.4. (I have 26 items in this data). The result is the same as when I put
the residuals into SPSS.&quot;
Reply:
Unidimensionality is never perfect. It is always approximate. The Rasch model constructs from the data
parameter estimates along the unidimensional latent variable that best concurs with the data. But, though the
Rasch measures are always unidimensional and linear, their concurrence with the data is never perfect.
Imperfection results from multi-dimensionality in the data and other causes of misfit.
Multidimensionality always exists to a lesser or greater extent. The vital question is: &quot;Is the multi-dimensionality in
the data big enough to merit dividing the items into separate tests, or constructing new tests, one for each
dimension?&quot;
In your example, the first contrast has eigenvalue of 3.1. This means that the contrast between the strongly
positively loading items and the strongly negatively loading items on the first contrast in the residuals has the
strength of about 3 items. Since positive and negative loading is arbitrary, you must look at the items at the top
and the bottom of the contrast plot. Are those items substantively different? Are they so different that they merit
the construction of two separate tests?
It may be that two or three off-dimension items have been included in your 26 item instrument and should be
dropped. But this is unusual for a carefully developed instrument. It is more likely that you have a &quot;fuzzy&quot; or
&quot;broad&quot; dimension, like mathematics. Mathematics includes arithmetic, algebra, geometry and word problems.
Sometimes we want a &quot;geometry test&quot;. But, for most purposes, we want a &quot;math test&quot;.
If in doubt, split your 26 items into two subtests, based on +ve and -ve loadings on the first residual contrast.
Measure everyone on the two subtests. Cross-plot the measures. What is their correlation? Do you see two
versions of the same story about the persons, or are they different stories? Which people are off-diagonal? Is that
important? If only a few people are noticeably off-diagonal, or off-diagonal deviance would not lead to any action,
then you have a substantively unidimensional test. A straightforward way to obtain the correlation is to write out a
PFILE= output file for each subtest. Read the measures into EXCEL and have it produce their Pearson productmoment correlation. If R1 and R2 are the reliabilities of the two subtests, and C is the correlation of their ability
estimates reported by Excel, then their latent (error-disattenuated) correlation approximates C / sqrt (R1*R2). If
this approaches 1.0, then the two subtests are statistically telling the same story. But you may have a
&quot;Fahrenheit-Celsius&quot; equating situation if the best fit line on the plot departs from a unit slope.
277
You can do a similar investigation for the second contrast of size 2.4, and third of size 1.9, but each time the
motivation for doing more than dropping an off-dimension item or two becomes weaker. Since random data can
have eigenvalues of size 1.4, there is little motivation to look at your 5th contrast.
294.
Disjoint strings of responses
When the responses are not arranged in one continuous string in the record, instruct Winsteps to skip over or
ignore the gaps.
Example: The 18 item string is in columns 40 to 49 and then 53 to 60 of your data file. The person-id is in
columns 11-30. Data look like:
xxxxxxxxxxPocahontas Smith, Jrxxxxxxxxx1001001110xxx11001110
Method a: Delete unwanted &quot;items&quot; in columns 50, 51, 52 using an item delete file, IDFILE=.
NAME1 =11
in original record
NAMLEN=20 length in original record
ITEM1 =40 in original record
NI =21
include deleted items
IDFILE =DEL5052 file of deletions
The contents of DEL5052 are:
11 - 13
Cols 50-52 are items 11-13
Method b: Rescore &quot;items&quot; in columns 50, 51, 52 as missing values with RESCORE=.
NAME1 =11
in original record
NAMLEN=20
ITEM1 =40
NI =21
include rescored items
RESCORE=000000000011100000000 rescore 50-52
CODES =01
(the standard)
NEWSCORE=XX
non-numerics specify &quot;missing&quot;
Method c: Make the items form one continuous string in a new record created with FORMAT=. Then the item
string starts in the 21st column of the new record. Reformatted record looks like: Pocahontas Smith,
Jr100100111011001110
FORMAT=(T11,20A,T40,10A,T53,8A) reformatting
NAME1 =1
in the formatted record
ITEM1 =21
in the formatted record
NI =18
the actual number of items
295.
Disordered rating categories
There is considerable debate in the Rasch community about the status of rating (or partial credit) scales and
polytomies which exhibit &quot;disorder&quot;. Look at Table 3.2, distractor/option analysis. Two types of disorder have been
noticed:
(i) Disorder in the &quot;average measures&quot; of the categories can imply disorder in the category definitions.
In this example, from Linacre, J.M. (1999) Category Disordering vs. Step Disordering, Rasch Measurement
Transactions 13:1 p. 675, &quot;FIM™ Level&quot; categories have been deliberately disordered in the data. It is seen that
278
this results in disordering of the &quot;average measures&quot; or &quot;observed averages&quot;, the average abilities of the people
observed in each category, and also large mean-square fit statistics. The &quot;scale structure measures&quot;, also called
&quot;step calibrations&quot;, &quot;step difficulties&quot;, &quot;step measures&quot;, &quot;Rasch-Andrich thresholds&quot;, &quot;deltas&quot;, &quot;taus&quot;, etc., remain
ordered.
(ii) Disorder in the &quot;step calibrations&quot; or &quot;disordered Rasch-Andrich thresholds&quot; implies less frequently
observed intermediate categories, i.e., that they correspond to narrow intervals on the latent variable.
In this example, the FIM categories are correctly ordered, but the frequency of level 2 has been reduced by
removing some observations from the data. Average measures and fit statistics remain well behaved. The
disordering in the &quot;step calibrations&quot; now reflects the relative infrequency of category 2. This infrequency is
pictured in plot of probability curves which shows that category 2 is never a modal category in these data. The
step calibration values do not indicate whether measurement would be improved by collapsing levels 1 and 2, or
collapsing levels 2 and 3, relative to leaving the categories as they stand.
296.
Displacement measures
DISPLACE column should only appear with anchored or TARGET= runs. Otherwise its appearance
indicates lack of convergence.
The DISPLACE value is the size of the change in the parameter estimate that would be observed in the next
estimation iteration if this parameter was free (unanchored) and all other parameter estimates were anchored at
their current values.
For a parameter (item or person) that is anchored in the main estimation, DISPLACE indicates the size of
disagreement between an estimate based on the current data and the anchor value.
For an unanchored item, if the DISPLACE value is large enough to be of concern, then the convergence criteria
are not tight enough LCONV=, RCONV=, CONVERGE=, MUCON=
279
It is calculated using Newton-Raphson estimation.
Person: DISPLACE = (observed marginal score - expected marginal score)/(model variance of the marginal
score)
Item: DISPLACE = - (observed marginal score - expected marginal score)/(model variance of the marginal
score)
DISPLACE approximates the displacement of the estimate away from the statistically better value which would
result from the best fit of your data to the model. Each DISPLACE value is computed as though all other
parameter estimates are exact. Only meaningfully large values are displayed. They indicate lack of
convergence, or the presence of anchored or targeted values. The best fit value can be approximated by adding
the displacement to the reported measure or calibration. It is computed as:
DISPLACE = (observed score - expected score based on reported measure) / (Rasch-model-derived score
variance).
This value is the Newton-Raphson adjustment to the reported measure to obtain the measure estimated from the
current data. In BTD, p. 64, equation 3.7.11: di(j) is the anchor value, di(j+1) is the value estimated from the
current data, and di(j+1) - di(j) is the displacement, given by the right-hand term of the estimation equation, also in
step 6 of www.rasch.org/rmt/rmt102t.htm. In RSA, p. 77, equation 4.4.6, di(t) is the anchor value, di(t+1) is the
value estimated from the current data, and di(t+1) - di(t) is the displacement, given by the right-hand term of the
estimation equation, also in step 6 of www.rasch.org/rmt/rmt122q.htm
Standard Error of the Displacement Measure
+----------------------------------------------------------------------------------------+
|ENTRY
RAW
MODEL|
INFIT | OUTFIT |PTMEA|
|
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.|DISPLACE| TAP
|
|------------------------------------+----------+----------+-----+--------+--------------|
|
3
35
35
2.00A
.74| .69
-.6| .22
.5| .00|
-3.90| 1-2-4
|
Since the reported &quot;measure&quot; is treated as a constant when &quot;displacement&quot; is computed, the S.E. of the reported
&quot;measure&quot; actually is the same as the S.E. of the displacement. The DISPLACE column shows the displacement
in the same units as the MEASURE. This is logits when USCALE=1, the default. If the anchored measure value is
considered to be exact, i.e., a point-estimate, then the S.E. standard error column indicates the standard error of
the displacement. The statistical significance of the Displacement is given by
t = Displacement / S.E.(Displacement) with approximately COUNT degrees of freedom.
This evaluates how likely the reported size of the displacement is, if its &quot;true&quot; size is zero. But both the
displacements and their standard errors are estimates, so the t-value may be slightly mis-estimated.
Consequently allow for a margin of error when interpreting the t-values.
If the anchored measure value has a standard error obtained from a different data set, then the standard error of
the displacement is:
S.E. (Displacement) = Sqrt(S.E.&sup2; + S.E.(anchor value from original data)&sup2; )
When does large displacement indicate that an item or person should be unanchored or omitted?
This depends on your purpose. If you are anchoring items in order to measure three additional people to add to
your measured database of thousands, then item displacement doesn't matter.
Anchor values should be validated before they are used. Do two analyses:
(a) with no items anchored (i.e., all items floating), produce person and item measures.
(b) with anchored items anchored, produce person and item measures.
Then cross-plot the item difficulties for the two runs, and also the person measures. The person measures will
usually form an almost straight line.
For the item difficulties, unanchored items will form a straight-line. Some anchored items may be noticeably off
the line. These are candidates for dropping as anchors. The effect of dropping or unanchoring a &quot;displaced&quot;
anchor item is to realign the person measures by roughly (displacement / (number of remaining anchored items)).
Random displacements of less than 0.5 logits are unlikely to have much impact in a test instrument.
280
&quot;In other work we have found that when [test length] is greater than 20, random values of [discrepancies
in item calibration] as high as 0.50 [logits] have negligible effects on measurement.&quot; ( Wright &amp; Douglas,
1976, &quot;Rasch Item Analysis by Hand&quot;)
&quot;They allow the test designer to incur item discrepancies, that is item calibration errors, as large as 1.0
[logit]. This may appear unnecessarily generous, since it permits use of an item of difficulty 2.0, say, when
the design calls for 1.0, but it is offered as an upper limit because we found a large area of the test design
domain to be exceptionally robust with respect to independent item discrepancies.&quot; (Wright &amp; Douglas,
1975, &quot;Best Test Design and Self-Tailored Testing.&quot;)
Most DIF work seems to be done by statisticians with little interest in, and often no access to, the substantive
material. So they have no qualitative criteria on which to base their DIF acceptance/rejection decisions. The
result is that the number of items with DIF is grossly over-reported (Hills J.R. (1989) Screening for potentially
biased items in testing programs. Educational Measurement: Issues and practice. 8(4) pp. 5-11).
297.
Edit Initial Settings (Winsteps.ini) file
If program hangs during &quot;Constructing Winsteps.ini ...&quot; then see Initialization Fails
To change the Winsteps standard starting directory from a short-cut icon:
Right-click the short-cut to Winsteps.exe (this may be in your \windows\start menu directory)
Click on &quot;Properties&quot;
Select &quot;Shortcut&quot;
Type the desired directory into &quot;Start in&quot;
Click on &quot;Apply&quot;
Altering the Initial Settings
1) Pull-down the &quot;Edit&quot; menu
281
2) Click on &quot;Edit initial settings&quot;
3) The Settings display. If this is too big for your screen see Display too big.
Winsteps.ini is the name of a file in your Winsteps folder which maintains various settings for Winsteps.
It is a Text file. Here are the typical contents:
Editor=&quot;C:\Program Files\Windows NT\Accessories\wordpad.exe&quot;
Temporary directory=&quot;C:\DOCUME~1\Mike\LOCALS~1\Temp\&quot;
Filter=All Files (*.*)
Excel=C:\Program Files\Microsoft Office\Office\EXCEL.EXE
SPSS=C:\WINDOWS\system32\NOTEPAD.EXE
Reportprompt=Y
Extraask=Y
Closeonexit=Ask
Welcome=Yes
OFOPTIONS=NTPYYYN
XFIELDS=&quot;XXXXXXXXXXXXXXXXXXXXXXXXXXXX&quot;
SIGNS=;;,,..
(1) Editor path: Editor=&quot;C:\Program Files\Windows NT\Accessories\wordpad.exe&quot;
This specifies the word processor or text editor that is to be used by Winsteps - WordPad or your own text editor.
If Output Tables do not display, this may not be a working version of WordPad, see Changing Wordpad
(2) Excel path: Excel=C:\Program Files\Microsoft Office\Office\EXCEL.EXE
This provides a fast link to Excel from the File pull-down menu. The path to any program can be placed here. If
Winsteps does not find Excel automatically, please go to the Windows &quot;Start&quot; menu, do a &quot;Find&quot; or &quot;Search&quot; for
&quot;Files and Folders&quot; to locate it, and enter its path here. It is not part of Winsteps, and may not be present on your
computer.
(3) SPSS path: SPSS=C:\SPSS.EXE
This provides a fast link to SPSS from the File pull-down menu. The path to any program can be placed here. If
Winsteps does not find SPSS automatically, please go to the Windows &quot;Start&quot; menu, do a &quot;Find&quot; or &quot;Search&quot; for
&quot;Files and Folders&quot; to locate it, and enter its path here. It is not part of Winsteps, and may not be installed on your
computer.
(4) Filter for file selection: Filter=All Files (*.*)
This is the selection for the Files dialog box, used when setting up or saving files.
If everything is in Text files, then specify: Filter=Text Files (*.txt)
(5) Temporary directory for work files: Temporary directory=&quot;C:\Temp\&quot;
Temporary Output and Table files are placed:
(a) In the same directory as your Input file (if possible)
or (b) in the &quot;Temporary directory&quot;, which is ordinarily &quot;C:\TEMP&quot;
Other temporary files are placed in the &quot;Temporary directory.&quot;
Files ending &quot;....ws.txt&quot; can be deleted when Winsteps is not running.
(6) Prompt for output file name: Reportprompt=No specifies that your standard report output file will be a
temporary file.
Reportprompt=Yes produces a prompt for a file name on the screen. You can always view the Report output file
from the Edit menu, and save it as another file.
(7) Prompt for Extra Specifications: Extraask=No specifies that there will be no extra specifications to be
entered.
(8) Show &quot;Welcome&quot; Help? Welcome=Yes displays the &quot;Easy Start&quot; help message. Welcome=No does not.
(9) Close output windows on exit? Closeonexit=Ask
This choice can be overridden when Winsteps stops.
(10) Character that starts .... SIGNS=;;,,..
This enables the processing of files in accordance with international usages. For instance, it may be more
282
convenient to output files with decimal commas, use semicolons as separators, and indicate comments with
exclamation marks:
(11) OFOPTIONS=, XFIELDS= and other settings are internal switches. Deleting them does no harm. They will
be automatically reset.
OK: Click on this to accept the settings. Some settings go into effect the next time your start Winsteps.
Cancel: Do not change the settings.
Help: Displays this page of the Help file
To reset all settings back to their standard values, find Winsteps.ini in your Winsteps folder and delete it. Standard
values will be instituted next time you start Winsteps.
298.
Equating and linking tests
Test Equating and linking are usually straightforward with Winsteps, but do require clerical care. The more
thought is put into test construction and data collection, the easier the equating will be.
Imagine that Test A (the more definitive test, if there is one) has been given to one sample of persons, and Test B
to another. It is now desired to put all the items together into one item hierarchy, and to produce one set of
measures encompassing all the persons.
Initially, analyze each test separately. Go down the &quot;Diagnosis&quot; pull-down menu. If the tests don't make sense
separately, they won't make sense together.
There are several equating methods which work well with Winsteps. Test equating is discussed in Bond &amp; Fox
&quot;Applying the Rasch model&quot;, and earlier in Wright &amp; Stone, &quot;Best Test Design&quot;, George Ingebo &quot;Probability in the
Measure of Achievement&quot; - all available from www.rasch.org/books.htm
Concurrent or One-step Equating
All the data are entered into one big array. This is convenient but has its hazards. Off-target items can introduce
noise and skew the equating process, CUTLO= and CUTHI= may remedy targeting deficiencies. Linking designs
forming long chains require much tighter than usual convergence criteria. Always cross-check results with those
obtained by one of the other equating methods.
Common Item Equating
This is the best and easiest equating method. The two tests share items in common, preferably at least 5 spread
out across the difficulty continuum.
Step 1. From the separate analyses, crossplot the difficulties of the common items, with Test B on the y-axis and
Test A on the x-axis. The slope of the best-fit line i.e., the line though the point at the means of the common items
and through the (mean + 1 S.D.) point should have slope near 1.0. If it does, then the intercept of the line with the
x-axis is the equating constant.
First approximation: Test B measures in the Test A frame of reference = Test B measure + x-axis intercept.
Step 2. Examine the scatterplot. Points far away from the best fit line indicate items that have behaved differently
on the two occasions. You may wish to consider these to be no longer common items. Drop the items from the
plot and redraw the best fit line. Items may be off the diagonal, or exhibiting large misfit because they are offtarget to the current sample. This is a hazard of vertical equating. CUTLO= and CUTHI= may remedy targeting
deficiencies.
283
Step 3a.
If the best-fit slope remains far from 1.0, then there is something systematically different about Test A
and Test B. You must do &quot;Celsius - Fahrenheit&quot; equating. Test A remains as it stands.
The slope of the best fit is: slope = (S.D. of Test B common items) / (S.D. of Test A common items)
Include in the Test B control file:
USCALE = the value of 1/slope
UMEAN = the value of the x-intercept
and reanalyze Test B. Test B is now in the Test A frame of reference, and the person measures from Test A and
Test B can be reported together.
Step 3b.
The best-fit slope is near to 1.0. Suppose that Test A is the &quot;benchmark&quot; test. Then we do not want
responses to Test B to change the results of Test A.
From a Test A analysis produce IFILE= and SFILE= (if there are rating or partial credit scales).
Edit the IFILE= and SFILE= to match Test B item numbers and rating (or partial credit) scale.
Use them as an IAFILE= and SAFILE= in a Test B analysis.
Test B is now in the same frame of reference as Test A, so the person measures and item difficulties can be
reported together
Step 3c.
The best-fit slope is near to 1.0. Test A and Test B have equal status, and you want to use both to
define the common items.
Use the MFORMS= command to combine the data files for Test A and Test B into one analysis. The results of
that analysis will have Test A and Test B items and persons reported together.
Items
-----------||||||||||||
||||||||||||
||||||||||||
||||||||||||
||||||||||||
||||||||||||
------------ -----------|
|
| ||||||||||||
|
|
| ||||||||||||
|
|
| ||||||||||||
|
|
| ||||||||||||
|
|
| ||||||||||||
|
|
| ||||||||||||
- ------------
Test A Persons
Test B Persons
Partial Credit items
&quot;Partial credit&quot; values are much less stable than dichotomies. Rather than trying to equate across the whole
partial credit structure, one usually needs to assert that, for each item, a particular &quot;threshold&quot; or &quot;step&quot; is the
critical one for equating purposes. Then use the difficulties of those thresholds for equating. This relevant
threshold for an item is usually the transition point between the two most frequently observed categories - the
Rasch-Andrich threshold - and so the most stable point in the partial credit structure.
Stocking and Lord iterative procedure
The Stocking and Lord (1983) present an iterative common-item procedure in which items exhibiting DIF across
tests are dropped from the link until no items exhibiting inter-test DIF remain. A known hazard is that if the DIF
distribution is skewed, the procedure trims the longer tail and the equating will be biased. To implement the
Stocking and Lord procedure in Winsteps, code each person (in the person id label) according to which test form
was taken. Then request a DIF analysis of item x person-test-code (Table 30). Drop items exhibiting DIF from the
link, by coding them as different items in different tests.
Stocking and Lord (1983) Developing a common metric in item response theory. Applied Psychological
Measurement 7:201-210.
Common Person Equating
Some persons have taken both tests, preferably at least 5 spread out across the ability continuum.
Step 1. From the separate analyses, crossplot the abilities of the common persons, with Test B on the y-axis and
Test A on the x-axis. The slope of the best-fit line i.e., the line though the point at the means of the common
284
persons and through the (mean + 1 S.D.) point should have slope near 1.0. If it does, then the intercept of the
line with the x-axis is the equating constant.
First approximation: Test B measures in the Test A frame of reference = Test B measure + x-axis intercept.
Step 2. Examine the scatterplot. Points far away from the best fit line indicate persons that have behaved
differently on the two occasions. You may wish to consider these to be no longer common persons. Drop the
persons from the plot and redraw the best fit line.
Step 3a.
If the best-fit slope remains far from 1.0, then there is something systematically different about Test A
and Test B. You must do &quot;Celsius - Fahrenheit&quot; equating. Test A remains as it stands.
The slope of the best fit is: slope = (S.D. of Test B common persons) / (S.D. of Test A common persons)
Include in the Test B control file:
USCALE = the value of 1/slope
UMEAN = the value of the x-intercept
and reanalyze Test B. Test B is now in the Test A frame of reference, and the person measures from Test A and
Test B can be reported together.
Step 3b.
The best-fit slope is near to 1.0. Suppose that Test A is the &quot;benchmark&quot; test. Then we do not want
responses to Test B to change the results of Test A.
From a Test A analysis produce PFILE=
Edit the PFILE= to match Test B person numbers
Use it as a PAFILE= in a Test B analysis.
Test B is now in the same frame of reference as Test A, so the person measures and person difficulties can be
reported together
Step 3c.
The best-fit slope is near to 1.0. Test A and Test B have equal status, and you want to use both to
define the common persons.
Use your text editor or word processor to append the common persons' Test B responses after their Test A ones,
as in the desing below. Then put the rest of the Test B responses after the Test A responses, but aligned in
columns with the commonpersons's Test B responses. Perform an analysis of the combined data set. The results
of that analysis will have Test A and Test B persons and persons reported together.
Test A items
Test B items
|-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------||-----------------------------|
|-----------------------------|
|-----------------------------||-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------||-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------||-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------|
|-----------------------------|
Common Person
Common Person
Common Person
Common Person
Virtual Equating of Test Forms
The two tests share no items or persons in common, but the items cover similar material.
Step 1. Identify pairs of items of similar content and difficulty in the two tests. Be generous about interpreting
&quot;similar&quot; at this stage.
285
Steps 2-4: simple: The two item hierarchies (Table 1 using short clear item labels) are printed and compared,
equivalent items are identified. The sheets of paper are moved vertically relative to each other until the overall
hierarchy makes the most sense. The value on Test A corresponding to the zero on Test B is the UMEAN= value
to use for Test B. If the item spacing on one test appear expanded or compressed relative to the other test, use
USCALE= to compensate.
Or:
Step 2. From the separate analyses, crossplot the difficulties of the pairs of items, with Test B on the y-axis and
Test A on the x-axis. The slope of the best-fit line i.e., the line though the point at the means of the common items
and through the (mean + 1 S.D.) point should have slope near 1.0. If it does, then the intercept of the line with the
x-axis is the equating constant.
First approximation: Test B measures in the Test A frame of reference = Test B measure + x-axis intercept.
Step 3. Examine the scatterplot. Points far away from the best fit line indicate items that are not good pairs. You
may wish to consider these to be no longer paired. Drop the items from the plot and redraw the best fit line.
Step 4. The slope of the best fit is: slope = (S.D. of Test B common items) / (S.D. of Test A common items)
Include in the Test B control file:
USCALE = the value of 1/slope
UMEAN = the value of the x-intercept
and reanalyze Test B. Test B is now in the Test A frame of reference, and the person and item measures from
Test A and Test B can be reported together.
Random Equivalence Equating
The samples of persons who took both tests are believed to be randomly equivalent. Or, less commonly, the
samples of items in the tests are are believed to be randomly equivalent.
Step 1. From the separate analyses of Test A and Test B, obtain the means and sample standard deviations of
the two person samples (including extreme scores).
Step 2. To bring Test B into the frame of reference of Test A, adjust by the difference between the means of the
person samples and user-rescale by the ratio of their sample standard deviations.
Include in the Test B control file:
USCALE = value of (S.D. person sample for Test A) / (S.D. person sample for Test B)
UMEAN = value of (mean for Test A) - (mean for Test B * USCALE)
and reanalyze Test B.
Check: Test B should now report the same sample mean and sample standard deviation as Test A.
Test B is now in the Test A frame of reference, and the person measures from Test A and Test B can be reported
together.
Linking Tests with Common Items
Here is an example:
A. The first test (50 items, 1,000 students)
B. The second test (60 items, 1,000 students)
C. A linking test (20 items from the first test, 25 from the second test, 250 students)
Here is a typical Rasch approach. It is equivalent to applying the &quot;common item&quot; linking method twice.
(a) Rasch analyze each test separately to verify that all is correct.
(b) Cross-plot the item difficulties for the 20 common items between the first test and the linking test. Verify that
the link items are on a statistical trend line parallel to the identity line. Omit from the list of linking items, any items
that have clearly changed relative difficulty. If the slope of the trend line is not parallel to the identity line (45
degrees), then the test discrimination has changed. The test linking will use a &quot;best fit to trend line&quot; conversion:
286
Corrected measure on test 2 in test 1 frame-of-reference =
((observed measure on test 2 - mean measure of test 2 link items)*(SD of test 1 link items)/(SD of test 1 link
items))
+ mean measure of test 1 link items
(c) Cross-plot the item difficulties for the 25 common items between the second test and the linking test. Repeat
(b).
(d1) If both trend lines are approximately parallel to the identity line, than all three tests are equally discriminating,
and the simplest equating is &quot;concurrent&quot;. Put all 3 tests in one analysis.You can use the MFORMS= command to
put all items into one analysis. You can also selectively delete items using the Specification pull-down menu in
order to construct measure-to-raw score conversion tables for each test, if necessary.
Or you can use a direct arithmetical adjustment to the measures based on the mean differences of the common
items: www.rasch.org/memo42.htm &quot;Linking tests&quot;.
(d2) If best-fit trend lines are not parallel to the identity line, then tests have different discriminations. Equate the
first test to the linking test, and then the linking test to the second test, using the &quot;best fit to trend line&quot; conversion,
shown in (b) above. You can also apply the &quot;best fit to trend&quot; conversion to Table 20 to convert every possible raw
score.
299.
Estimation bias correction - warnings
At least two sources of estimation error are reported in the literature.
An &quot;estimation bias&quot; - this is usually negligibly small after the administration of 10 dichotomous items (and fewer
rating scale items). Its size depends on the probability of observing extreme score vectors. For a two item test, the
item measure differences are twice their theoretical values, reducing as test length increases. This can be
corrected. STBIAS= does this approximately, but is only required if exact probability inferences are to be
made from logit measure differences. The experimental XMLE= option in Winsteps does this more exactly. But
these corrections have their drawbacks.
A &quot;statistical inflation&quot;. Since error variance always adds to observed variance, individual measures are always
reported to be further apart (on average) than they really are. This cannot be corrected, in general, at an
individual- measure level, because, for any particular measurement it cannot be known to what extent that
measurement is biased by measurement error. However, if it is hypothesized that the persons, for instance, follow
a normal distribution of known mean and standard deviation, this can be imposed on the estimates (as in MMLE)
and the global effects of the estimate dispersion inflation removed. This is done in some other Rasch estimation
software.
Estimation Bias
All Rasch estimation methods have some amount of estimation bias (which has no relationship with demographic
bias). The estimation algorithm used by Winsteps, JMLE, has a slight bias in measures estimated from most
datasets. The effect of the bias is to spread out the measures more widely than the data indicate. In practice, a
test of more than 20 dichotomous items administered to a reasonably large sample will produce measures with
inconsequential estimation bias. Estimation bias is only of concern when exact probabilistic inferences are to be
made from short tests or small samples. Ben Wright opted for JMLE in the late 1960's because users were rarely
concerned about such exact inferences, but they were concerned to obtain speedy, robust, verifiable results from
messy data sets with unknown latent parameter distributions. Both of the identifiable sources of error are reduced
by giving longer tests to bigger samples. With short tests, or small samples, other threats to validity tend to be of
greater concern than the inflationary ones.
If estimation bias would be observed even with an infinitely large sample (which it would be with JMLE), then the
estimation method is labeled &quot;statistically inconsistent&quot; (even though the estimates are predictable and logical).
This sounds alarming but the inconsistency is usually inconsequential, or can be easily corrected in the unlikely
event that it does have substantive consequences.
The JMLE joint likelihood estimation algorithm produces estimates that have a usually small statistical bias. This
bias increases the spread of measures and calibrations, but usually less than the standard error of measurement.
287
The bias quickly becomes insignificantly small as the number of persons and items increases. The reason that
JMLE is statistically inconsistent under some conditions, and noticeably biased for short tests or small samples,
is that it includes the possibility of extreme scores in the estimation space, but cannot actually estimate them.
Inconsistency doesn't really matter, because it asks &quot;if we have infinite data, will the estimation method produce
the correct answer?&quot; Estimation bias, also called statistical bias, is more important because it asks &quot;How near to
correct are the estimates with finite data?&quot; In practice, JMLE bias is smaller than the other sources of noise in the
data. See Ben Wright's comments at www.rasch.org/memo45.htm
For paired comparisons and very short tests, estimation can double the apparent spread of the measures,
artificially inflating test reliability. This can be eliminated by specifying PAIRED=YES.
Correcting for bias may be helpful when it is desired to draw exact probabilistic inferences for small, complete
datasets without anchoring.
Correcting for bias may be misleading, or may be supressed by Winsteps, in the presence of missing data or
anchored persons or items.
Bias correction can produce apparently inconsistent measures if bias-corrected measures, estimated from an
unanchored analysis, are then used to anchor that same dataset.
Estimation correction methods:
STBIAS=YES implements a variant of the simple bias correction proposed in Wright, B.D. and Douglas, G.A.
(1977). Best procedures for sample-free item analysis. Applied Psychological Measurement, 1, 281-294. With
large samples, a useful correction for bias is to multiply the estimated measures by (L-1)/L, where L is the smaller
of the average person or item response count, so, for paired comparisons, multiply by 0.5. This is done
automatically when PAIRED=YES.
XMLE=YES implements an experimental more sophisticated bias correction.
Other Rasch programs may or may not attempt to correct for estimation bias. When comparing results from other
programs, try both STBIAS=Y and STBIAS=N to find the closest match. See also XMLE=
Estimation methods with less bias under sum circumstances include CMLE and MMLE, but these have other
limitations or restrictions which are deemed to outweigh their benefits for most uses.
Technical information:
Statistical estimation bias correction with JMLE is relevant when you wish to make exact probabilistic statements
about differences between measures for short tests or small samples. The (L-1)/L correction applies to items on
short dichotomous tests with large samples, where L is the number of non-extreme items on a test. For long
dichotomous tests with small samples, the correction to person measures would be (N-1)/N. Consequently
Winsteps uses a bias correction on dichotomous tests for items of (L-1)/L and for persons of (N-1)/N
The reason for this correction is because the sample space does not match the estimation space. The difference
is extreme score vectors. Estimation bias manifests itself as estimated measures which are more dispersed than
the unbiased measures. The less likely an extreme score vector, the smaller the correction to eliminate bias.
Extreme score vectors are less likely with polytomies than with dichotomies so the bias correction is smaller. For
example, if an instrument uses a rating scale with m categories, then Winsteps corrects the item measures by (m1)(L-1)/((m-1)(L-1)+1) and person measures by (m-1)(N-1)/((m-1)(N-1)+1) - but these are rough approximations.
Since the mathematics of bias correction is complex and involves person and item parameter distributions,
dichotomies or polytomies, anchored parameter values, person and item weighting, and missing data patterns, I
have devised an experimental algorithm, XMLE, which attempts to do an essentially exact adjustment for bias
under all conditions. You can activate this in Winsteps with XMLE=Yes. You can compare results of an
XMLE=YES run with those of a standard run with STBIAS=NO, to see the size of the bias with your data. This is
more exact than STBIAS=YES. It is experimental because the algorithm has yet to be checked for its behavior
under many conditions, and so may malfunction - (it definitely does with extreme scores, for which reasonable
bias-corrected estimates have yet to be devised).
288
With most Rasch software using CMLE, PMLE or MMLE bias correction of item measures is not done because
the estimation bias in the item difficulties is generally very small. Bias correction of person abilities is not done
though estimation bias exists.
Interaction terms are computed in an artificial situation in which the abilities and difficulties estimates are treated
as known. Estimation bias is a minor effect in the interaction estimates. It would tend to increase very slightly the
probability that differences between interaction estimates are reported as significant. So this is another reason to
interpret DIF tests conservatively. If the number of relevant observations for an interaction term is big enough for
the DIF effect to be regarded as real, and not a sampling accident, then the estimation bias will be very small. In
the worst case, the multiplier would be of the order of (C-1)/C where C is the number of relevant observations.
Comparing Estimates
Bigsteps and Winsteps should produce the same estimates when
(a) they are run with very tight convergence criteria, e.g.,
RCONV=.00001
LCONV=.00001
MUCON=0
(b) they have the same statistical bias adjustment
STBIAS=YES ; estimates will be wider spread
or
STBIAS=NO ; estimates will be narrower
(c) they have the same extreme score adjustment
EXTRSC=0.5
The item estimates in BTD were produced with statistical bias adjustment, but with convergence criteria that
would be considered loose today. Tighter convergence produces a wider logit spread. So the BTD item estimates
are slightly more central than Winsteps or Bigsteps.
Winsteps and Bigsteps are designed to be symmetric. Transpose persons and items, and the only change is the
sign of the estimates and an adjustment for local origin. The output reported in BTD (and by most modern Rasch
programs) is not symmetric. So the person measure estimates in BTD are somewhat different.
300.
Estimation methods: JMLE, PROX, XMLE
Winsteps implements three methods of estimating Rasch parameters from ordered qualitative observations:
JMLE, PROX and XMLE. Estimates of the Rasch measures are obtained by iterating through the data. Initially all
unanchored parameter estimates (measures) are set to zero. Then the PROX method is employed to obtain
rough estimates. Each iteration through the data improves the PROX estimates until they are usefully good. Then
those PROX estimates are the initial estimates for JMLE which fine-tunes them, again by iterating through the
data, in order to obtain the final JMLE estimates. The iterative process ceases when the convergence criteria are
met. These are set by MJMLE=, CONVERGE=, LCONV= and RCONV=. Depending on the data design, this
process can take hundreds of iterations (Convergence: Statistics or Substance?). When only rough estimates are
needed, force convergence by pressing Ctrl+F or by selecting &quot;Finish iterating&quot; on the File pull-down menu.
Extreme scores: (perfect, maximum possible scores, and zero, minimum possible scores) are dropped from the
main estimation procedure. Their measures are estimated separately using EXTRSC=.
Missing data: most Rasch estimation methods do not require that missing data be imputed, or that there be
case-wise or list-wise omission of data records with missing data. For datasets that accord with the Rasch model,
missing data lower the precision of the measures and lessen the sensitivity of the fit statistics, but do not bias the
measure estimates.
Likelihood: Using the current parameter estimates (Rasch measures), the probability of observing each data
point is computed, assuming the data fit the model. The probabilities of all the data points are multiplied together
289
to obtain the likelihood of the entire data set. The parameter estimates are then improved (in accordance with the
estimation method) and a new likelihood for the data is obtained. The values of the parameters for which the
likelihood of the data has its maximum are the &quot;maximum likelihood estimates&quot; (Ronald A. Fisher, 1922).
JMLE &quot;Joint Maximum Likelihood Estimation&quot; is also called UCON, &quot;Unconditional maximum likelihood
estimation&quot;. It was devised by Wright &amp; Panchapakesan. In this formulation, the estimate of the Rasch
parameter (for which the observed data are most likely, assuming those data fit the Rasch model) occurs when
the observed raw score for the parameter matches the expected raw score. &quot;Joint&quot; means that the estimates for
the persons (rows) and items (columns) and rating scale structures (if any) of the data matrix are obtained
simultaneously. The iterative estimation process is described at Iteration.
Advantages - these are implementation dependent, and are implemented in Winsteps:
(1) independence from specific person and item distributional forms.
(2) flexibility with missing data
(3) the ability to analyze test lengths and sample sizes of any size
(4) symmetrical analysis of person and item parameters so that transposing rows and columns does not change
the estimates
(5) flexibility with person, item and rating scale structure anchor values
(6) flexibility to include different variants of the Rasch model in the same analysis (dichotomous, rating scale,
partial credit, etc.)
(7) unobserved intermediate categories of rating scales can be maintained in the estimation with exact
probabilities.
(8) all non-extreme score estimable (after elimination of extreme scores and rarely-observed Guttman subsets)
(9) all persons with the same total raw scores on the same items have the same measures; all items with the
same raw scores across the same persons have the same measures.
Disadvantages:
(11) measures for extreme (zero, perfect) scores for persons or items require post-hoc estimation.
(12) estimates are statistically inconsistent
(13) estimation bias, particularly with small samples or short tests, inflates the logit distance between estimates.
(14) chi-squares reported for fit tests (particularly global fit tests) may be somewhat inflated, exaggerating misfit to
the Rasch model.
Comment on (8): An on-going debate is whether measures should be adjusted up or down based on the misfit in
response patterns. With conventional test scoring and Rasch JMLE, a lucky guess counts as a correct answer
exactly like any other correct answer. Unexpected responses can be identified by fit statistics. With the threeparameter-logistic item-response-theory (3-PL IRT) model, the score value of an unexpected correct answer is
diminished whether it is a lucky guess or due to special knowledge. In Winsteps, responses to off-target items
(the locations of lucky guesses and careless mistakes) can be trimmed with CUTLO= and CUTHI=, or be
diminished using TARGET=Yes.
Comment on (13): JMLE exhibits some estimation bias in small data sets (for reasons, see XMLE below), but this
rarely exceeds the precision (model standard error of measurement, SEM) of the measures. Estimation bias is
only of concern when exact probabilistic inferences are to be made from short tests or small samples. It can be
exactly corrected for paired-comparison data with PAIRED=Yes. For other data, It can be approximately corrected
with STBIAS=Yes, but, in practice, this is not necessary (and sometimes not advisable).
PROX is the Normal Approximation Algorithm devised of Cohen (1979). This algorithm capitalizes on the similar
shapes of the logistic and normal ogives. It models both the persons and the items to be normally distributed. The
variant of PROX implemented in Winsteps allows missing data. The form of the estimation equations is:
Ability of person = Mean difficulty of items encountered +
log ( (observed score - minimum possible score on items encountered) /
(maximum possible score on items encountered - observed score) )
* square-root ( 1 + (variance of difficulty of items encountered) / 2.9 )
In Winsteps, PROX iterations cease when the variance of the items encountered does not increase substantially
from one iteration to the next.
Advantages - these are implementation dependent, and are implemented in Winsteps:
290
(2)-(9) of JMLE
Computationally the fastest estimation method.
Disadvantages
(1) Person and item measures assumed to be normally distributed.
(11)-(14) of JMLE
Other estimation methods in common use (but not implemented in Winsteps):
Gaussian least-squares finds the Rasch parameter values which minimize the overall difference between the
observations and their expectations, Sum((Xni - Eni)^2) where the sum is overall all observations, Xni is the
observation when person encounters item i, and Eni is the expected value of the observation according to the
current Rasch parameter estimates. For Effectively, off-target observations are down-weighted, similar to
TARGET=Yes in Winsteps.
Minimum chi-square finds the Rasch parameter values which minimize the overall statistical misfit of the data to
the model, Sum((Xni - Eni)^2 / Vni) where Vni is the modeled binomial or multinomial variance of the observation
around its expectation. Effectively off-target observations are up-weighted to make them less improbable.
Gaussian least-squares and Minimum chi-square:
Advantages - these are implementation dependent:
(1)-(8) All those of JMLE.
Disadvantages:
(9) persons with the same total raw scores on the same items generally have different measures; items with the
same raw scores across the same persons generally have different measures.
(11)-(13) of JMLE
(14) global fit tests uncertain.
CMLE. Conditional maximum likelihood estimation. Item difficulties are structural parameters. Person abilities are
incidental parameters, conditioned out for item difficulty estimation by means of their raw scores. The item
difficulty estimates are those that maximize the likelihood of the data given the person raw scores and assuming
the data fit the model. The item difficulties are then used for person ability estimation using a JMLE approach.
Advantages - these are implementation dependent:
(1), (6)-(9) of JMLE
(3) the ability to analyze person sample sizes of any size
(5) flexibility with item and rating scale structure anchor values
(12) statistically-consistent item estimates
(13) minimally estimation-biased item estimates
(14) exact global fit statistics
Disadvantages:
(2) limited flexibility with missing data
(3) test length severely limited by mathematical precision of the computer
(4) asymmetric analysis of person and item parameters so that transposing rows and columns changes the
estimates
(5) no person anchor values
(11) of JMLE
(13) estimation-biased of person estimates small but uncertain
MMLE. Marginal maximum likelihood estimation. Item difficulties are structural parameters. Person abilities are
incidental parameters, integrated out for item difficulty estimation by imputing a person measure distribution. The
item difficulties are then used for person ability estimation using a JMLE approach.
Advantages - these are implementation dependent:
(3), (6)-(9) of JMLE
(1) independence from specific item distributional forms.
(2) flexibility with missing data extends to minimal length person response strings
291
(5) flexibility with item and rating scale structure anchor values
(11) extreme (zero, perfect) scores for persons are used for item estimation.
(12) statistically-consistent item estimates
(13) minimally estimation-biased item estimates
(14) exact global fit statistics
Disadvantages:
(1) specific person distribution required
(4) asymmetric analysis of person and item parameters so that transposing rows and columns changes the
estimates
(5) no person anchor values
(11) measures for extreme (zero, perfect) scores for specific persons or items require post-hoc estimation.
(13) estimation-biased of person estimates small but uncertain
PMLE. Pairwise maximum likelihood estimation. Person abilities are incidental parameters, conditioned out for
item difficulty estimation by means of pairing equivalent person observations. The item difficulties are then used
for person ability estimation usinga JMLE approach.
Advantages - these are implementation dependent:
(1), (3), (6), (7) of JMLE
(5) flexibility with item and rating scale structure anchor values
(8) all persons with the same total raw scores on the same items have the same measure
(12) statistically-consistent item estimates
Disadvantages:
(11) of JMLE
(2) reduced flexibility with missing data
(4) asymmetric analysis of person and item parameters so that transposing rows and columns changes the
estimates
(5) no person anchor values
(8) items with the same total raw scores across the same persons generally have different measures.
(13) estimation-biased or item and person estimates small but uncertain
(14) global fit tests uncertain.
(15) uneven use of data in estimation renders standard errors and estimates less secure
WMLE. Warm's (1989) Weighted Maximum Likelihood Estimation. Standard MLE estimates are the maximum
values of the likelihood function and so statistical modes. Warm shows that the likelihood function is skewed,
leading to an additional source of estimation bias. The mean likelihood estimate is less biased. Warm suggests an
unbiasing correction that can be applied, in principle, to any MLE method, but there are computational constraints.
Even when feasible, this fine tuning appears to be less than the relevant standard errors and have no practical
benefit. It is not currently implemented in Winsteps.
XMLE, &quot;Exclusory Maximum Likelihood Estimation&quot;, implements Linacre's (1989) XCON algorithm in Winsteps.
Statistical &quot;consistency&quot; is the property that an estimation method will yield the &quot;true&quot; value of a parameter when
there is infinite data. Statistical &quot;estimation bias&quot; is the degree to which an estimate differs from its &quot;true&quot; value
with a finite amount of data. JMLE is statistically inconsistent under some conditions, and noticeably estimationbiased for short tests or small samples, because it includes the possibility of extreme scores in the estimation
space, but cannot estimate them. The XMLE algorithm removes the possibility of extreme response vectors from
the estimation space, to a first approximation. This makes XMLE consistent, and much less estiamtion-biased
than JMLE. In fact XMLE is even less biased than CMLE for small samples, this is because CMLE only
eliminates the possibility of extreme person response vectors, not the possibility of extreme item response
vectors.
XMLE and JMLE use the same estimation methods. The difference is in the probability terms used in the
estimation equations.
For JMLE, for the dichotomous case,
loge(Pni1 / Pni0 ) = Bn - Di
where Pni1 is the probability that person n succeeds on item i. For XMLE,
Rni1 = Pni1 - Product(Pmi1) - Product(Pnj1) + Product(Pmi1) * Product(Pnj1)
292
where m =1,N and j=1,L, so that Product(Pmi1) is the likelihood of a perfect-score for person n, and Product(Pnj1)
is the likelihood of the sample all succeeding on item i. Similarly,
Rni0 = Pni0 - Product(Pmi0) - Product(Pnj0) + Product(Pmi0) * Product(Pnj0)
So the JMLE estimation equation for person n or item i is based on
Expected raw score = Sum(Rni1/(Rni1+Rni0)) for i or n
Example: Consider a two-item dichotomous test. Possible person scores are 0, 1, 2. Person scores of 0 and 2 are
dropped from estimation as extreme. The remaining very large sample of N persons all score 1 success so all
have the same measure, 0 logits for convenience. Twice as many successes are observed on item 1 as item 2.
Under these conditions, in the estimation sample, a success on item 1 requires a failure on item 2 and vice-versa.
So, according to the Rasch model, the logit distance between item 1 and item 2 = log (frequency of success on
item 1 / frequency of success on item 2) = log(2). And the expected score on item 1 is 2/3 and on item 2 is 1/3.
JMLE considers observations of item 1 and item 2 to be independent of the total raw score, and computes the
distance between item 1 and item 2 = log (frequency of success on item 1 / frequency of failure on item 1) - log
(frequency of success on item 2 / frequency of failure 2) = log(2/1) - log(1/2) = 2 log(2), i.e., twice that of the direct
Rasch model. This is the worst case of JMLE estimation bias and occurs with pairwise comparison data. For such
data, this estimation-bias of 2 can be automatically corrected with PAIRED=Yes. As test length increases, the
bias reduces and is considered to be non-consequential for test lengths of 10 items or more, Wright's Memo 45.
For XMLE, let's assume the item difficulties are -0.5 * log(2) and 0.5 * log(2). Set C = square-root(2). Then Pn11 =
Pn20 = C / (1+C) and Pn10 = Pn21 = 1 / (1+C). Rn11 = C/(1+C) - C/(1+C) * 1/(1+C) - 0 + 0 (due to very large
sample) = (C/(1+C))^2 and Rn10 = 1/(+C) - 1/(1+C) * C/(1+C) - 0 + 0 (again due to very large sample) =
(1/(1+C))^2. Then expected score for a person on item 1 = Rn11/(Rn11+Rn10) = (C/(1+C))^2 / ( ((C/(1+C))^2 +
(C/(1+C))^2 ) = C^2 / (C^2 + 1) = 2/3. Similarly, Rn21 = 1/3 - as required. And the logit distance between item 1
and item 2 is 0.5 * log(2) - 0.5 * log(2) = log(2) as required. There is no estimation bias in this example.
Considerations with XMLE=YES include:
(1) Anchoring values changes the XMLE probabilities. Consequently, measures from a Table 20 score table do
not match measures from the estimation run. Consequently, it may be necessary to estimate item calibrations
with XMLE=YES. Then anchor the items and perform XMLE=NO.
(2) Items and persons with extreme (zero and perfect) scores are deleted from the analysis.
(3) For particular data structures, measures for finite scores may not be calculable.
Advantages - these are implementation dependent, and are implemented in Winsteps:
(1)-(8) of JMLE
(12) estimates are statistically consistent
(13) estimation bias is small
Disadvantages:
(11) measures for extreme (zero, perfect) scores for persons or items require post-hoc estimation, and even then
may not be estimable
(14) global fit tests uncertain
Cohen Leslie. (1979) Approximate Expressions for Parameter Estimates in the Rasch Model, The British Journal
of Mathematical and Statistical Psychology, 32, 113-120
Fisher R.A. On the mathematical foundations of theoretical statistics. Proc. Roy. Soc. 1922 Vol. CCXXII p. 309368
Warm T.A. (1989). Weighted likelihood estimation of ability in item response theory. Psychometrika, 54, 427-450
301.
Exact Match: OBS% and EXP%
+---------------------------------------------------------------------------------------------------+
|ENTRY
RAW
MODEL|
INFIT | OUTFIT |PTMEA|EXACT MATCH|
|
|NUMBER SCORE COUNT MEASURE S.E. |MNSQ ZSTD|MNSQ ZSTD|CORR.| OBS% EXP%| KID
|
|------------------------------------+----------+----------+-----+-----------+----------------------|
|
72
14
25
-1.32
.37|2.02
2.9|5.16
5.7|A .04| 60.0 65.8| JACKSON, SOLOMON
|
293
Suppose your dataset consists of observations, {Xni}, of person n on item i. Based on the Rasch parameters
(measures), there is an expected value Eni corresponding to each observation Xni. Eni is obtained by a
calculation from the Rasch model.
When the absolute value of (Xni-Eni) is less than 0.5 then the observed data point is within 0.5 score points of its
expected value, so the match is the closest possible. Thus, across all observations of item i,
Count ( (Xni-Eni) &lt;0.5 ) = A - these observations are of the closest categories to their expectations
Count ( (Xni-Eni) =0.5 ) = B - these observations are on the borderline of matching their expectations
Count ( (Xni-Eni) &gt;0.5 ) = C - these observations are at least one category away from their expectations
So that A+B+C = Count (Xni)
OBS% = Observed % = 100 * ( A + B/2 ) / ( A+B+C )
Each possible value of Xni has a probability according to the Rasch model. Based on these, the expected value of
OBS% can be computed, this is the EXP%. So, if the possible values of Xni are j=0,1,2,...,m, with probabilities
Pnij, then
A = sum ( ((j-Eni)&lt;0.5 )*Pnij )
B = sum ( ((j-Eni)=0.5 )*Pnij )
C = sum ( ((j-Eni)&gt;0.5 )*Pnij )
So that A+B+C = Count (Xni)
EXP% = Expected % = 100 * ( A + B/2 ) / ( A+B+C )
If OBS%&lt;EXP% then the local data are more random than the model predicts.
If OBS%&gt;EXP% then the local data are more predictable than the model predicts.
302.
Exporting Tables to EXCEL
Exporting Winsteps Tables to Excel is easy.
Produce the Winsteps Table using the &quot;Output Tables&quot; pull-down menu.
&quot;Select&quot; with the mouse the part you want (Usually column heading lines to the bottom of the Table)
then right-click &quot;Copy&quot;
Start Excel (e.g., from the Winsteps Files menu)
Paste into Excel - top left cell. Everything will go into the first column.
Under &quot;Data&quot; go &quot;Text to columns&quot;
Excel usually gets everything exactly right, so that each Winsteps table column is in a separate Excel column.
Done!
303.
Extra Specifications prompt
WINSTEPS expects to find the control variables in your control file. You may, however, specify one or more
control variables on the &quot;Extra specifications&quot; line. These variables supersede instructions in the control file. This
is useful for making temporary changes to the control variables. There are special rules for blanks (see
Exampe 3). You can turn off the Extra Specifications prompt from the Edit Initial Settings menu.
Example 1: You want to verify that your data is correctly formatted, so you only want to do one JMLE iteration this
time, i.e., you want to set MJMLE=1 for this run only:
Please enter name of WINSTEPS control file: SF.TXT(Enter)
Please enter name of report output file: SFO.TXT(Enter)
294
Extra specifications? (e.g., MJMLE =1), or press Enter:
MJMLE=1(Enter)
Note:
Extra specifications? (e.g., MJMLE=1), or press Enter:
MJMLE = 1(Enter)
is invalid because there are blanks in MJMLE = 1.
Example 2: You want to produce the fit plot in Table 4 with specially chosen ranges on the axes:
Please enter name of WINSTEPS control file: SF.TXT(Enter)
Please enter name of report output file: SFO.TXT(Enter)
Extra specifications? (e.g., MJMLE=1), or press Enter:
TABLES=0001 MRANGE=3 FRANGE=4(Enter)
Example 3: To put blanks in an Extra Specification. Put the whole specification within &quot; &quot; (double quotes).
Put the argument within ' ' (single quotes). E.g., You want the title TITLE= to be: Analysis B, and UMEAN=50.
Please enter name of WINSTEPS control file: SF.TXT(Enter)
Please enter name of report output file: SFO.TXT(Enter)
Extra specifications? (e.g., MJMLE=1), or press Enter:
&quot;Title = 'Analysis B' &quot; UMEAN=50
304.
Extreme scores: what happens
Extreme scores are the lowest and highest possible scores for persons on items, or for items by persons. They
include zero and perfect scores. They are shown in the Tables as MINIMUM ESTIMATE MEASURE and
MAXIMUM ESTIMATE MEASURE.
Mathematically, they correspond to infinite or indefinite measures on the latent variable and so are not directly
estimable. Accordingly persons or items with extreme scores are dropped for the duration of the measurement
estimation process. The extreme persons are dropped casewise. The extreme items are dropped listwise.
Sometimes the effect of dropping extreme items and persons is to make other items and persons extreme. If so,
these are also dropped. If the data have a Guttman pattern, ultimately all items and persons are dropped and the
measures for that data set are reported as inestimable.
After the measures of all non-extreme items and persons have been estimated, then the extreme scores are
reinstated. Reasonable extreme measures are imputed for them (using a Bayesian approach), so that all persons
and items have measures.
See Extremescore=
305.
Global fit statistics
Winsteps reports global fit statistics and approximate global log-likelihood chi-square statistic in Table 3.1. The
variance tables report the relative sizes of explained and unexplained variances.
The chi-square value is approximate. It is based on the current reported estimates which may depart noticeably
from the &quot;true&quot; maximum likelihood estimates for these data. The degrees of freedom are the number of
datapoints used in the free estimation (i.e., excluding missing data, data in extreme scores, etc.) less the number
of free parameters. The number of free parameters is the least number of parameters from which all the Rasch
expected observations could be constructed. For complete data, this is the lower of the number of different
observed person marginal scores or the number of different observed item marginal scores, less one for
identifying the local origin.
If you wish to compute your own global (or any other) fit test, the response-level probabilities, residuals etc. are
reported in the XFILE=. For instance, for a global fit test, you could add up all the log-probabilities. Then chisquare estimate = - 2 * log-probability. A different chi-square estimate is the sum of squared-standardized
residuals. You can count up the number of free parameters. For complete dichotomous data, it is usually the
minimum of (number of different person marginal raw scores, number of different item marginal scores) - 1.
295
Deviance statistics are more trustworthy. They are the difference between the chi-squares of two analyses, with
d.f. of the difference between the number of free parameters estimated.
The Rasch model is an idealization, never achieved by real data. Accordingly, given enough data, we expect to
see statistically significant misfit the model. If the current data do not misfit, we merely have to collect more data,
and they will! In essence, the null hypothesis of this significance test is the wrong one! We learn nothing from
testing the hypothesis, &quot;Do the data fit the model (perfectly)?&quot; Or, as usually expressed in social science, &quot;Does
the model fit the data (perfectly)?&quot; Perfection is never obtained in empirical data. What we really want to test is
the hypothesis &quot;Do the data fit the model usefully?&quot; And, if not, where is the misfit, and what is it? Is it big
enough in size (not &quot;statistical significance&quot;) to cause trouble? This is the approach used in much of industrial
quality-control, and also in Winsteps.
306.
GradeMap interface
To produce this plot from GradeMap, an analysis and graphical program distributed in association with the book
&quot;Constructing Measures: An Item Response Modeling Approach&quot; by Mark Wilson (2004) Mahwah NJ: Lawrence
Erlbaum Associates. bearcenter.berkeley.edu/GradeMap/
The GradeMap option on the Output Files menu displays this box which enables a simple conversion of Winsteps
control and data files into GradeMap format. These files can then be imported by GradeMap - see the GradeMap
User Guide. The files are displayed after they are created by Winsteps, so that you can edit them if you wish.
Excel is used if it is available. If changes are made, save the file in tab-separated .txt format - which is expected
by GradeMap. Here is a Model Specification File from Example0.txt.
296
Example GradeMap Dialog:
Winsteps: Create Model Specification file
Winsteps: Create Student Data file
Winsteps: Launch GradeMap
GradeMap screen displays:
User name: admin
Password: bear
Menu: File
Menu: New Project
Remove: Yes
Menu: System
Menu: Import Model Specification
Import data: (your Model Specification file, e.g., lfsitems.txt)
Menu: File
Menu: Import Student Data
Import Student Data: (your Student Data file, e.g., lfschildren.txt)
Menu: View
Menu: Select Item Set
Click on: T
Click on: OK
Menu: Estimation Tasks
Menu: Compute
Accept: Yes
EAP: No
Menu: Reports &amp; Maps etc.
Be patient! GradeMap operates slowly with large files.
307.
Guttman patterns
Psychometrician Louis Guttman (1916-1987) perceived the ideal test to be one in which a person succeeds on all
the items upto a certain difficulty, and then fails on all the items above that difficulty. When persons and items are
ordered by raw score, this produces a data set with a &quot;Guttman pattern&quot;. This is data is not analyzable in the
usual way by Rasch analysis, because each person or item in turn becomes an extreme score. Here is a Guttman
pattern with dichotomous data:
Easy-&gt;Hard items (columns)
1111111 Most able person (rows)
1111110
1111100
297
1111100
1100000
1100000
1000000
0000000 Least able person
It is sometimes useful to make this type of data estimable by adding a dummy reversed-Guttman record:
Easy-&gt;Hard items (columns)
11111110 Most able person (rows)
11111100
11111000
11111000
11000000
11000000
10000000
00000000 Least able person
00000001 &lt; Dummy person record
^ Dummy item record
or by anchoring the most extreme items (or persons) a conveniently long distance apart, e.g., 10 logits:
PAFILE=*
1 10 ; anchor the first (highest score) person at 10 logits
8 0 ; anchor the last (lowest score) person at 0 logits
*
&amp;END
END LABELS
1111111 Most able person (rows)
1111110
1111100
1111100
1100000
1100000
1000000
0000000 Least able person
308.
Half-rounding
Rounding occurs when a number, such as 5.032 is to be displayed with only two decimal places, e.g., as 5.03.
The general rule followed by Winsteps is to round to the nearest displayable value. Examples:
5.034 rounds to 5.03
5.036 rounds to 5.04
-5.034 rounds to -5.03
-5.036 rounds to -5.04
Rounding errors may arise with 5.035 and -5.035. Winsteps intends to round these away from zero to 5.04 and 5.04.
In practice, the computer arithmetic sometimes loses precision due to hardware limitations, so that 5.035
becomes an internal number like 5.034999997 - a number which is the same computationally as 5.035. But this
value half-rounds to 5.03. This behavior is impossible to predict, because most displayed numbers are the result
of a long chain of internal computations, .
Increasing the value of UDECIM= may display more decimal places, and so better display the numerical results.
In recent versions, Winsteps adds .0000005 to positive numbers, and subtracts that amount from negative
numbers, in order that .005 will almost always round to .01.
298
309.
How big an analysis can I do?
WINSTEPS will operate with a minimum of two observations per item or person, or even 1 observation with
anchoring. This is useful for producing score tables with anchored items and dummy response strings. For
statistically stable measures to be estimated, 30 observations per element are needed:
www.rasch.org/rmt/rmt74m.htm
The upper limit of WINSTEPS is 10,000,000+ persons. Winsteps can analyze 30,000+ items.
There can be up to 255 ordinally numbered categories per item (polytomies, rating scales, partial credit etc.).
The wides logit range of measures that maintains computational precision is 90 logits, but useful results can be
reported with a measure range up to 700 logits wide.
310.
How long will an analysis take?
A PC with a math co-processor processes about 1,000,000 observations per minute. Most analyses have reached
convergence within 20 iterations, so a rule of thumb is:
length of analysis in minutes = (number of persons)*(length of test)*2/100,000
311.
Information - item and test
Fisher information is the amount of information data provide about a parameter. For item information, this is the
amount of information the response to an item provides about a person parameter. For test information, this is the
amount of information all of the items encountered by a person provide about a person parameter.
All dichotomous items have the same Rasch-derive item information. For a response of probability p, the item
information is the binomial variance of the probability, p(1-p). This function is:
For polytomous items, the information function depends on the rating scale structure. The better targeted the item
is on the person the more Fisher information the item provides about the person parameter. This motivates some
item selection techniques for computer-adaptive testing.
The test information function is the sum of the item information functions. The standard error of a measure is the
inverse square-root of the test information at the location of the maximum-likelihood parameter estimate (or
higher). Here is the test information function for the Knox Cube Test in Example 1.
299
.
Test (sample) reliability is a summary of the interaction between the sample distribution and the test information
function.
312.
Item difficulty: definition
As modeled in Winsteps, the difficulty (challenge, easiness, etc.) of an item (task, prompt, etc.) is the point on the
latent variable (unidimensional continuum) at which the highest and lowest category have equal probability of
being observed.
For a dichotomous item, this is the point at which each category has a 50% probability of being observed.
For a Rasch-Andrich rating-scale item, this definition implies that the sum of the rating-scale-structure measures
sum to zero relative to the item difficulty, i.e., the sum of the Rasch-Andrich thresholds is zero, i.e., sum(Fj) = 0.
For a Masters partial-credit item, this definition implies that the item difficulty is the average of the difficulties of the
Rasch-Masters thresholds for the item, i.e., Di = average (Dij), so that reparameterizing, Dij = Di + Fj, then
sum(Fij) = 0 for each item i.
313.
Item discrimination or slope estimation
The Rasch model specifies that item discrimination, also called the item slope, be uniform across items. This
supports additivity and construct stability. Winsteps estimates what the item discrimination parameter would have
been if it had been parameterized. The Rasch slope is the average discrimination of all the items. It is not the
mean of the individual slopes because discrimination parameters are non-linear. Mathematically, the average
slope is set at 1.0 when the Rasch model is formulated in logits, or 1.70 when it is formulated in probits (as 2-PL
and 3-PL usually are). 0.59 is the conversion from logits to probits.
The empirical discrimination is computed after first computing and anchoring the Rasch measures. In a post-hoc
analysis, a discrimination parameter, ai, is estimated for each item. The estimation model is of the form:
This has the appearance of a 2-PL IRT or &quot;Generalized Partial Credit&quot; model, but differs because the
discrimination or slope parameter is not used in the estimation of the other parameters. The reported values of
item discrimination, DISCR, are a first approximation to the precise value of ai obtained from the NewtonRaphson estimation equation:
The possible range of ai is -∞ to +∞, where +∞ corresponds to a Guttman data pattern (perfect discrimination)
and -∞ to a reversed Guttman pattern. Rasch estimation usually forces the average item discrimination to be near
1.0. Consequently an estimated discrimination of 1.0 accords with Rasch model expectations. Values greater
than 1.0 indicate over-discrimination, and values less than 1.0 indicate under-discrimination. Over-discrimination
is thought to be beneficial in many raw-score and IRT item analyses. High discrimination usually corresponds to
300
low MNSQ values, and low discrimination with high MNSQ values.
From an informal simulation study, Edward Wolfe reports Winsteps discrimination to have a .88 correlation with
the generating slope parameters for a 2-PL dataset. BILOG has a .95 correlation.
Table 29.1 allows you to estimate the empirical item discrimination, at least as well as a 2-PL IRT computer
program. This is because 2-PL discrimination estimation is degraded by the imputation of a person distribution
and constraints on discrimination values. It is also skewed by accidental outliers which your eye can disregard.
When Discrimination=Yes. exact computation is done in the measure tables.
In Table 29.1 draw in the line that, to your eye, matches the central slope of the empirical item characteristic curve
(ICC).
Estimate the logit distance from where the line intercepts the .0 score value to where it intercepts the 1.0 score
value (for dichotomies). The logit distance here is about 4.0 logits.
Use the central logit measure to logit discrimination line in this nomogram to estimate discrimination. In this
nomogram, a logit distance of 4.0 logits, corresponds to a logit discrimination of 1.0, in accordance with model
prediction. Steeper slopes, i.e., higher discriminations, correspond to shorter distances.
301
314.
Iterations - PROX &amp; JMLE
The Rasch model formulates a non-linear relationship between non-linear raw scores and linear measures. So,
estimating measures from scores requires a non-linear process. This is performed by means of iteration. Two
estimation methods are used, PROX and JMLE.
In Winsteps, initially every person is estimated to have the same ability measure at the origin of the measurement
scale. Each item is estimated to have the same difficulty measure, also at the origin of the measurement scale.
Each rating scale structure parameter, Rasch-Andrich threshold, is also estimated to be 0.
In Winsteps, the first phase of estimation uses the PROX (normal approximation) estimation algorithm. This takes
the initial set of estimates and produces revised estimates:
Bn = &micro; n +
1 + σ n / 2 . 9 log e (R n / (N n − R n ))
2
where Bn is the revised ability estimate for person n, &micro;n is the mean difficulty of the items encountered by person
n, and σn is the standard deviation of those item difficulties. Rn is the observed raw score for person n and Nn is
a perfect score on those same items. Similarly, for the items,
Di = &micro;i −
2
1 + σ i / 2 . 9 log
e
(R i / (N i
− R i ))
where Di is the revised difficulty estimate for item i, &micro;i is the mean ability of the persons encountering by item i,
and σi is the standard deviation of those person abilities. Ri is the observed raw score on item i and Ni is a perfect
score by those same persons.
To update these PROX estimates, Winsteps traverses the data computing the values of all the terms on the rightside of the estimation equations. This traversal is called an &quot;iteration&quot;. When the increase in the range of the
person or item measures is smaller than 0.5 logits, or when MPROX= is reached, iteration ceases.
Initial estimates of the Rasch-Andrich threshold between category k and category k-1 are obtained from log
(observed frequency of category k-1 / observed frequency of category k) normalized to sum to zero across the
thresholds of a rating scale.
The PROX estimates become the starting values for JMLE (Joint Maximum Likelihood Estimation). Using these
person, item and rating scale structure estimates, Winsteps computes the expected value, according to the Rasch
model, corresponding to each observation in term. After iterating through the entire data set, the marginal sums of
these expected values, the person expected raw scores and the item expected raw scores, are compared with
their observed (empirical) values. If a person's expected raw score is less than that person's observed raw score,
then the ability estimate raised. If the person's expected raw score is greater than the observed score, then the
ability estimate is lowered. For items, if the expected raw score is less than the observed score, then the difficulty
estimate is lowered. If the item's expected raw score is greater than the observed score, then the difficulty
estimate is raised.
The estimation equations for JMLE are derived in RSA, where Newton-Raphson iteration is employed.
y' = y + (observed score - Rasch expected score based on current estimates)/(modeled variance)
where y = a current estimated person measure and y' is the improved estimate.
Newton-Raphson estimation has proved unstable with sparse data sets and also with rating scales which have
alternating very high and very low frequency categories. Accordingly, Winsteps implements a more robust
proportional-curve-fitting algorithm to produce JMLE estimates. The relationship between raw scores and
measures is always monotonic, so the characteristic curve for each person or item parameter is modeled to have
the local form of a logistic ogive:
y = a * log( (x-l)/(h-x) ) + c
where y = an estimated measure, a = slope of the ogive, x = a raw score, l = the known minimum possible raw
score for the parameter, h = the known maximum possible raw score for the parameter, c = location of ogive
relative to local origin.
Values of x are obtained form the current estimated measure y and a nearby measure (y + d). From these, a and
c are estimated. The revised measure y' is obtained by evaluating the equation using the observed raw score as
the value of x. In the plot below for Example 1, the current estimate, y, is -3 logits, a nearby estimate, y+d, is -2
logits. These both estimate raw scores on the currently-estimated test characteristic curve (TCC, the remainder of
302
which is not yet known). The violet line is the logistic ogive going through these two known points. It is close to the
putative TCC. The observed score of &quot;5&quot; is then found on the logistic ogive and an improved estimate is obtained.
After all the person and item estimates are improved, the estimated TCC changes and this estimation process is
repeated by performing another iteration through the data.
For the rating scale structure, the estimate, yk, for Rasch-Andrich threshold k is improved by
yk' = yk - log ( observed count category k / observed count category k-1)
+ log (estimated count category k / estimated count category k-1)
When the various convergence criteria are satisfied, iteration ceases and the final estimates are obtained. These
are used in computing fit statistics.
Example:
Here is the iteration Table for example0.txt:
CONVERGENCE TABLE
+----------------------------------------------------------------------------+
|
PROX
ACTIVE COUNT
EXTREME 5 RANGE
MAX LOGIT CHANGE |
| ITERATION
KIDS
ACTS
CATS
KIDS
ACTS
MEASURES STRUCTURE|
|----------------------------------------------------------------------------|
|
1
75
25
3
3.78
3.20
3.8918
.0740 |
|
2
74
25
3
4.59
3.71
.8258
-.6158 |
|
3
74
25
3
4.83
3.92
.2511
-.1074 |
|----------------------------------------------------------------------------|
|
JMLE
MAX SCORE
MAX LOGIT
LEAST CONVERGED
CATEGORY STRUCTURE|
| ITERATION
RESIDUAL*
CHANGE
KID
ACT
CAT
RESIDUAL
CHANGE |
|----------------------------------------------------------------------------|
|
1
2.84
-.1955
60
22*
2
21.44
.0076|
|
2
.71
-.0335
53
15*
0
-5.89
.0144|
|
3
-.43
.0293
53
5*
1
3.48
.0101|
|
4
.32
.0235
18
11*
1
2.71
.0079|
|
5
.24
.0184
18
11*
0
-2.09
.0060|
|
6
.19
.0141
18
11*
0
-1.63
.0045|
|
7
.14
.0108
18
11*
0
-1.25
.0035|
|
8
.11
.0082
18
11*
0
-.96
.0026|
|
9
.08
.0062
18
11*
0
-.73
.0020|
|
10
.06
.0048
18
11*
0
-.56
.0015|
+----------------------------------------------------------------------------+
In the top section of the Convergence Table are reported the number of active persons, items and categories. The
range of item and person measures at the end of easch PROX iteration is shown, also the biggest change in any
person or item, and in any Rasch-Andrich threshold. PROX iteration ceases with iteration 3 because the &quot;KIDS&quot;
(persons) and &quot;ACTS&quot; (items) range has increased by less than 0.5 logits.
In the lower section, for each JMLE iteration, the maximum score residual, the biggest difference between any
303
observed and expected marginal score is shown. Also the biggest change in any measure. Iteration ceases when
the values, in iteration 10, are less than the convergence criteria.
315.
Local Dependence
In some data designs, data are collected from the same persons more than once, or observations are collected
on equivalent items. Consequently there is reason to suspect that local dependence exists in the data. What is its
impact on the Rasch measures?
Local dependence usually squeezes or stretches the logit measures, but does not usually change cut-points
much when they are expressed in raw-score terms.
Here is an experiment to determine whether local dependence is a problem.
Assuming that data from the same persons may be a problem, select from your cases one of each different
response string. This will make the data as heterogeneous as possible. Perform an analysis of this data set and
see if that changes your conclusions markedly. If it does, then local dependence may be a concern. If it doesn't
then local dependence is having no substantive impact.
Using Excel, a method of obtaining only one of each different response string:
0. Import the data into excel as a &quot;character&quot; column
1. from the Excel data pull down menu choose -&gt; filter -&gt; advanced filter
2. under &quot;action&quot; choose &quot;copy to another location&quot;
3. click &quot;list range&quot; and highlight the range of element numbers - if you want the whole column click on the letter at
the top of the column
4. click &quot;copy to&quot; and choose an empty column, e.g., column J.
5. click &quot;unique records only&quot;
6. click &quot;OK&quot;
7. look at column J. The data are unique.
316.
Logit and probit
When USCALE=1 (or USCALE= is omitted), measures are reported in logits. When USCALE=0.59, measures are
reported in approximated probits.
Logit: A logit (log-odds unit) is a unit of interval measurement which is well-defined within the context of a single
homogeneous test. When logit measures are compared between tests, their probabilistic meaning is maintained
but their substantive meanings may differ. This is often the case when two tests of the same construct contain
items of different types. Consequently, logit measures underlying different tests must be equated before the
measures can be meaningfully compared. This situation is parallel to that in Physics when some temperatures
are measured in degrees Fahrenheit, some in Celsius, and others in Kelvin.
As a first step in the equating process, plot the pairs of measures obtained for the same elements (e.g., persons)
from the two tests. You can use this plot to make a quick estimate of the nature of the relationship between the
two logit measurement frameworks. If the relationship is not close to linear, the two tests may not be measuring
the same thing.
Logarithms: In Rasch measurement all logarithms, &quot;log&quot;, are &quot;natural&quot; or &quot;Napierian&quot;, sometime abbreviated
elsewhere as &quot;ln&quot;. &quot;Logarithms to the base 10&quot; are written log10. Logits to the base 10 are called &quot;lods&quot;.
Logit-to-Probability Conversion Table
Logit difference between ability measure and item calibration &amp; Probability of success on a dichotomous item
5.0
4.6
4.0
3.0
2.2
2.0
99%
99%
98%
95%
90%
88%
-5.0
-4.6
-4.0
-3.0
-2.2
-2.0
1%
1%
2%
5%
10%
12%
304
0
1.4
1.1
1.0
0.8
0.5
0.4
0.2
0.1
50%
80%
75%
73%
70%
62%
60%
55%
52%
0 50%
-1.4
-1.1
-1.0
-0.8
-0.5
-0.4
-0.2
-0.1
20%
25%
27%
30%
38%
40%
45%
48%
Example with dichotomous data:
In Table 1, it is the distance between each person and each item which determines the probability.
MEASURE | MEASURE
&lt;more&gt; ------------------ PERSONS -+- ITEMS -------------------- &lt;rare&gt;
5.0
+
5.0
|
|
| XXX (items difficult for persons) &lt;- 4.5 logits
|
|
4.0
+
4.0
|
XX | &lt;- 3.7 logits
|
|
|
3.0
+
3.0
The two persons are at 3.7 logits. The three items are at 4.5 logits. The difference is 3.7 - 4.5 = -0.8 logits. From
the logit table above, this is predicted as 30% probability of success for persons like these on items like these.
Inference with Logits
Logit distances such as 1.2 logits are exactly correct for individual dichotomous items. 1.2 logits is the distance
between 50% success and 80% success.
Logit distances are also exactly correct for describing the relative performance on adjacent categories of a rating
scale, e.g, if in a Likert Scale, &quot;Agree&quot; and &quot;Strongly Agree&quot; are equally likely to be observed at a point on the
latent variable, then 1.2 logits higher, &quot;Strongly Agree&quot; is likely to be observed 8 times, for every 2 times that
&quot;Agree&quot; is observed.
For sets of dichtomous items, or performance on a rating scale item considered as a whole, the direct
interpretation of logits no longer applies. The mathematics of a probabilistic interpretation under these
circumstances is complex and rarely worth the effort to perform. Under these conditions, logits are usually only of
305
mathematical value for the computation of fit statistics - if you wish to compute your own.
Different tests usually have different probabilistic structures, so that interpretation of logits across tests are not the
same as interpretation of logits within tests. This is why test equating is necessary.
Logits and Probits
Logits are the &quot;natural&quot; unit for the logistic ogive. Probits are the &quot;natural&quot; units for the unit normal cumulative
distribution function, the &quot;normal&quot; ogive. Many statisticians are more familiar with the normal ogive, and prefer to
work in probits. The normal ogive and the logistic ogive are similar, and a conversion of 1.7 approximately aligns
them.
When the measurement units are probits, the dichotomous Rasch model is written:
log ( P / (1-P) ) = 1.7 * ( B -D )
To have the measures reported in probits, set USCALE = 0.59 = 1/1.7
Some History
Around 1940, researchers focussed on the &quot;normal ogive model&quot;. This was an IRT model, computed on the basis
that the person sample has a unit normal distribution N(0,1).
The &quot;normal ogive&quot; model is:
Probit (P) = theta - Di
where theta is a distribution, not an individual person.
But the normal ogive is difficult to compute. So they approximated the normal ogive (in probit units) with the much
simpler-to-compute logistic ogive (in logit units). The approximate relationship is: logit = 1.7 probit.
IRT philosophy is still based on the N(0,1) sample distribution, and so a 1-PL IRT model is:
(theta - Di)
where theta represents a sample distribution. Di is the &quot;one parameter&quot;.
log(P/(1-P)) = 1.7
The Rasch model takes a different approach. It does not assume any particular sample or item distribution. It
uses the logistic ogive because of its mathematical properties, not because of its similarity to the cumulative
normal ogive.
The Rasch model parameterizes each person individually, Bn. As a reference point it does not use the person
mean (norm referencing). Instead it conventionally uses the item mean (criterion referencing). In the Rasch model
there is no imputation of a normal distribution to the sample, so probits are not considered.
The Rasch model is:
log(P/(1-P)) = Bn - Di
Much IRT literature asserts that 1-PL = Rasch model. This is misleading. The mathematical equations can look
similar, but their motivation is entirely different.
If you want to approximate the &quot;normal ogive IRT model&quot; with Rasch software, then
(a) adjust the person measures so the person mean = 0: UPMEAN=0
(b) adjust the user-scaling: probits = logits/1.7: USCALE=0.59
After this, the sample may come close to having an N(0,1) sample distribution - but not usually! So you can force
S.D. = 1 unit, by setting USCALE = 1 / person S.D.
317.
Misfit diagnosis: infit outfit mean-square standardized
What do Infit Mean-square, Outfit Mean-square, Infit Zstd (z-standardized), Outfit Zstd (z-standardized)
mean?
Outfit: outlier-sensitive fit statistic. This is based on the conventional chi-square statistic. This is more sensitive to
unexpected observations made persons on items that are relatively very easy or very hard for them (and viceversa).
306
Infit: inlier-pattern-sensitive fit statistic. This is based on the chi-square statistic with each observation weighted
by its statistical information (model variance). This is more sensitive to unexpected patterns of observations made
persons on items that are roughly targeted on them (and vice-versa).
Mean-square: this is the chi-square statistic divided by its degrees of freedom. Consequently its expected value
is close to 1.0. Values greater than 1.0 (underfit) indicate unmodelled noise or other source of variance in the data
- these degrade measurement. Values less than 1.0 (overfit) indicate that the model predicts the data too well causing summary statistics, such as reliability statistics, to report inflated statistics. See further dichotomous and
polytomous mean-square statistics.
Example of computation:
The outfit mean-square is the accumulation of squared-standardized-residuals divided by their count, which is
their expectation. The infit mean-square is the accumulation of information-weighted residuals divided by their
expectation. The information an observation is its model variance. For dichotomies, this is the binomial variance =
p(1-p)
Oufit mean-square = sum ( observed-residual**2 / model variance ) / count
Infit mean-square = (observed information-weighted residual variance) / (modeled information-weighted residual
variance).
Outlying observations have smaller information and so have less information than on-target observations. If all
observations have the same amount of information, then the information cancels out. Then Infit mean-square =
Outfit mean-square.
For dichotomous data. Two observations: Model p=0.5, observed=1. Model p=0.25, observed =1.
Outfit mean-square = sum ( (obs-exp)**2 / model variance ) / (count of observations) = ((1-0.5)**2/(0.5*0.5) +
(1-0.25)**2/(0.25*0.75))/2 = (1 + 3)/2 = 2
Infit mean-square = sum ( (obs-exp)**2 ) / sum(model variance ) = ((1-0.5)**2 + (1-0.25)**2) /((0.5*0.5) +
(0.25*0.75)) = (0.25 + 0.56)/(0.25 +0.19) = 1.84. The off-target observation has less influence.
Z-Standardized: these report the statistical significance (probability) of the chi-square (mean-square) statistics
occurring by chance when the data fit the Rasch model. The values reported are unit-normal deviates, in which
.05% 2-sided significance corresponds to 1.96. Overfit is reported with negative values. These are also called &quot;tstatistics&quot; reported with infinite degrees of freedom.
General rules:
First, investigate negative point-measure or point-biserial correlations. Look at the Distractor Tables, 10.3.
Remedy miskeys, data entry errors, etc.
Then, the general rule is Investigate outfit before infit, mean-square before t standardized, high values before low
values.
There is an asymmetry in the implications of out-of-range high and low mean-squares (or positive and negative tstatistics). High mean-squares (or positive t-statistics) are a much greater threat to validity than low mean-squares
(or negative fit statistics).
Poor fit does not mean that the Rasch measures (parameter estimates) aren't linear. The Rasch model forces its
estimates to approximate linearity. Misfit means that the reported estimates, though effectively linear, provide a
distorted picture of the data.
High outfit mean-squares may be the result of a few random responses by low performers. If so, drop with
PDFILE= these performers when doing item analysis, or use EDFILE= to change those response to missing.
High infit mean-squares indicate that the items are mis-performing for the people on whom the items are targeted.
This is a bigger threat to validity, but more difficult to diagnose than high outfit.
Mean-squares show the size of the randomness, i.e., the amount of distortion of the measurement system. 1.0
are their expected values. Values less than 1.0 indicate observations are too predictable (redundancy, model
overfit). Values greater than 1.0 indicate unpredictability (unmodeled noise, model underfit). Mean-squares
307
usually vaerage to 1.0, so if there are high values, there must also be low ones. Examine the high ones first, and
temporarily remove them from the analysis if necessary, before investigating the low ones.
Zstd are t-tests of the hypotheses &quot;do the data fit the model (perfectly)?&quot; ZSTD (standardized as a z-score) is
used of a t-test result when either the t-test value has effectively infinite degrees of freedom (i.e., approximates a
unit normal value) or the Student's t-distribution value has been adjusted to a unit normal value. They show the
improbability (significance). 0.0 are their expected values. Less than 0.0 indicate too predictable. More than 0.0
indicates lack of predictability. If mean-squares are acceptable, then Zstd can be ignored. They are truncated
towards 0, so that 1.00 to 1.99 is reported as 1. So a value of 2 means 2.00 to 2.99, i.e., at least 2. For exact
values, see Output Files. If the test involves less than 30 observations, it is probably too insensitive, i.e.,
&quot;everything fits&quot;. If there are more than 300 observations, it is probably too sensitive, i.e., &quot;everything misfits&quot;.
The Wilson-Hilferty cube root transformation converts the mean-square statistics to the normally-distributed zstandardized ones. For more information, please see Patel's &quot;Handbook of the Normal Distribution&quot; or
www.rasch.org/rmt/rmt162g.htm.
Anchored runs:
Anchor values may not exactly accord with the current data. To the extent that they don't, they fit statistics may
be misleading. Anchor values that are too central for the current data tend to make the data appear to fit too well.
Anchor values that are too extreme for the current data tend to make the data appear noisy.
Interpretation of parameter-level mean-square fit statistics:
&gt;2.0
Distorts or degrades the measurement system.
1.5 - 2.0
Unproductive for construction of measurement, but not degrading.
0.5 - 1.5
Productive for measurement.
&lt;0.5
Less productive for measurement, but not degrading. May produce misleadingly good reliabilities and
separations.
In general, mean-squares near 1.0 indicate little distortion of the measurement system, regardless of the Zstd
value.
Evaluate high mean-squares before low ones, because the average mean-square is usually forced to be near 1.0.
Outfit mean-squares: influenced by outliers. Usually easy to diagnose and remedy. Less threat to
measurement.
Infit mean-squares: influenced by response patterns. Usually hard to diagnose and remedy. Greater threat to
measurement.
Extreme scores always fit the Rasch model exactly, so they are omitted from the computation of fit statistics. If
an extreme score has an anchored measure, then that measure is included in the fit statistic computations.
Question: Does it mean that these mean-square values, &gt;2 etc, are not sample size dependent?
Answer: Correct as a general rule-of-thumb. The mean-squares are already corrected for sample size: they are
the chi-squares divided by their degrees of freedom, i.e., sample size. The mean-squares answer &quot;how big is the
impact of the misfit&quot;. The t-statistics answer &quot;how unlikely to be observed when the data fit the model.&quot; We
eagerly await the theoretician who devises a statistical test for the hypothesis &quot;the data fit the Rasch model
usefully.&quot; (as opposed to the current tests for perfectly).
Question: Is this contradicting the usual statistical advice about model-data fit?
Statisticians are usually concerned with &quot;how likely are these data to be observed, assuming they accord with the
model?&quot; If it is too unlikely (i.e., significant misfit), then the verdict is &quot;these data don't accord with the model.&quot;
The practical concern is: &quot;In the imperfect empirical world, data never exactly accord with the Rasch model, but
do these data deviate seriously enough for the Rasch measures to be problematic?&quot; The builder of my house
followed the same rule (regarding Pythagoras theorem) when building my bathroom. It looked like the walls were
square enough for practical purposes. Some years later, I installed a full-length rectangular mirror - then I
discovered that the walls were not quite square enough for my purposes (so I had to do some cosmetic
adjustments) - so there is always a judgment call. The table of mean-squares is my judgment call as a &quot;builder of
Rasch measures&quot;.
308
Ben Wright's Infit and Outfit statistics (e.g., RSA, p. 100) are initially computed as mean-square statistics (i.e., chisquare statistics divided by their degrees of freedom). Their likelihood (significance) is then computed. This could
be done directly from chi-square tables, but the convention is to report them as unit normal deviates (i.e., tstatistics corrected for their degrees for freedom). I prefer to call them z-statistics, but the Rasch literature has
come to call them t-statistics, so now I do to. It is confusing because they are not strictly Student t-statistics (for
which one needs to know the degrees of freedom).
The relationship between mean-square and z-standardized t-statistics is shown in this plot. Basically, the
standardized statistics are insensitive to misfit with less than 30 observations and overly sensitive to misfit when
there are more than 300 observations.
309
318.
Missing data
One of Ben Wright's requirements for valid measurement, derived from the work of L.L. Thurstone, is that
&quot;Missing data must not matter.&quot; Of course, missing data always matters in the sense that it lessens the amount of
statistical information available for the construction and quality-control of measures. Further, if the missing data,
intentionally or unintentionally, skew the measures (e.g., incorrect answers are coded as &quot;missing responses&quot;),
then missing data definitely do matter. But generally, missing data are missing essentially at random (by design or
accident) or in some way that will have minimal impact on the estimated measures (e.g., adaptive tests).
Winsteps does not require complete data in order to make estimates. One reason that Winsteps uses JMLE is
that it is very flexible as regards estimable data structures. For each parameter (person, item or Rasch-Andrich
threshold) there are sufficient statistics: the marginal raw scores and counts of the non-missing observations.
During Winsteps estimation, the observed marginal counts and the observed and expected marginal scores are
computed from the same set of non-missing observations. Missing data are skipped over in these additions.
When required, Winsteps can compute an expected value for every observation (present or missing) for which the
item and person estimates are known.
The basic estimation algorithm used by Winsteps is:
Improved parameter estimate = current parameter estimate
+ (observed marginal score - expected marginal score) / (modeled variance of the expected marginal
score)
The observed and expected marginal scores are obtained by summing across the non-missing data. The
expected score and its variance are obtained by Rasch estimation using the current set of parameter estimates,
see RSA.
If data are missing, or observations are made, in such a way that measures cannot be constructed
unambiguously in one frame of reference, then the message
WARNING: DATA MAY BE AMBIGUOUSLY CONNECTED INTO nnn SUBSETS
is displayed on the Iteration screen to warn of ambiguous connection.
319.
Mixed and Saltus models
Rasch models are grounded in the concept of the unidimensional latent variable, i.e., the items defining the latent
variable operate in the same way for all members of the target population. Of course, this is a fiction. But reality
can often be made to cooperate.
But there are occasions when a population is comprised of different classes of persons with the items comprising
a different latent variable for each class. The classes are called &quot;Latent Classes&quot;.
Standard Rasch &quot;latent trait&quot; models can be extended to allow for latent classes. These are called &quot;Mixture
Models&quot; (Rost, 1990). The Saltus model (Mark Wilson, 1989) is a mixed model in which segments of items are
modeled to shift their difficulties together, and by the same amount, for different latent classes. In these models,
310
the different latent variables are defined by item diffiuclties, but individual respondents are not assigned to a
particular class, but rather the probability that each respondent beyonds to each class is reported.
Winsteps does not do a mixture or Saltus analysis directly, but it can provide much of the same information, and
also can indicate whether a more rigorous latent class analysis is likely to be productive.
Here is an approach:
Step 1. Identify meaningful potential respondent classes, e.g., male/female, high/low performers. The Winsteps
PCA analysis (e.g., Table 24.4) may help identify potential classes.
Step 2. Mark in the person label the class codes.The Microsoft Word &quot;rectangle copy&quot; function may be useful.
High and low performers do not need to be flagged, instead the MA2 function can be used.
Step 3. Perform DIF analysis based on the class codes. Items displaying strong DIF may be exhibiting classrelated behavior.
Step 4. Flag the the items by class in the item item identification.
Step 5. Look for item-classification by person-classification interactions (differential classification-grouped
functioning, DGF, Table 33). These would approximate the Saltus findings.
Rost, J&uuml;rgen. (1990) Rasch Models in Latent Classes: An Integration of Two Approaches to Item Analysis,
Applied Psychological Measurement, 14, 271-282
Wilson, M. (1989). Saltus: A psychometric model for discontinuity in cognitive development. Psychological
Bulletin, 105, 276-289.
320.
Multiple t-tests
Question: Winsteps Tables report many t-tests. Should Bonferroni adjustments for multiple comparisons be
made?
Reply: It depends on how you are conducting the t-tests. For instance, in Table 30.1. If your hypothesis (before
examining any data) is &quot;there is no DIF for this CLASS in comparison to that CLASS on this item&quot;, then the
reported probabilities are correct.
If you have 20 items, then one is expected to fail the p ≤ .05 rule. So if your hypothesis (before examining any
data) is &quot;there is no DIF in this set of items for any CLASS&quot;, then adjust individual t-test probabilities accordingly.
In general, we do not consider the rejection of a hypothesis test to be &quot;substantively significant&quot;, unless it is both
very unlikely (i.e., statistically significant) and reflets a discrepancy large enough to matter (i.e., to change some
decision). If so, even if there is only one such result in a large data set, we may want to take action. This is much
like sitting on the proverbial needle in a haystack. We take action to remove the needle from the haystack, even
though statistical theory says, &quot;given a big enough haystack, there will probably always be a needle in it
somewhere.&quot;
A strict Bonferroni correction for n multiple significance tests at joint level α is α /n for each single test. This
accepts or rejects the entire set of multiple tests. In an example of a 100 item test with 20 bad items (.005 &lt; p &lt;
.01), the threshold values for cut-off with p &lt;= .05 would be: 0.0005, so that the entire set of items is accepted.
Benjamini and Hochberg (1995) suggest that an incremental application of Bonferroni correction overcomes some
of its drawbacks. Here is their procedure:
i) Perform the n single significance tests.
ii) Number them in ascending order by probability P(i) where i=1,n in order.
iii) Identify k, the largest value if i for which P(i) = α * i/n
iv) Reject the null hypothesis for i = 1, k
In an example of a 100 item test with 20 bad items (.005 &lt; p &lt; .01), the threshold values for cut-off with p &lt;= .05
311
would be: 0.0005 for the 1st item, .005 for the 10th item, .01 for the 20th item, .015 for the 30th item. So that k
would be at least 20 and perhaps more. All 20 bad items have been flagged for rejection.
Benjamini Y. &amp; Hochberg Y. (1995) Controlling the false discovery rate: a practical and powerful approach to
multiple testing. Journal of the Royal Statistical Society B, 57,1, 289-300.
321.
Null or unobserved categories
There are two types of unobserved or null categories: structural zeroes and incidental/sampling zeroes.
Structural null categories occur when rating scale categories are number 10, 20, 30,… instead of 1,2,3. To force
Winsteps to eliminate non-existent categories 11, 12, 13, either rescore the data IVALUE= or specify
STKEEP=NO.
For intermediate incidental null zeroes, imagine this scenario: The Wright &amp; Masters &quot;Liking for Science&quot; data are
rescored from 0,1,2 to 0,1,3 with a null category at 2. the categories now mean &quot;disagree, neutral, agree-ish,
agree&quot;. We can imagine that no child in this sample selected the half-smile of agree-ish.
The category frequencies of categories 0,1,2,3 are 378, 620, 0, 852
The three Rasch-Andrich threshold parameters are -.89, +infinity, -infinity.
The +infinity is because the second parameter is of the order log(620/0). The –infinity is because the third
parameter is of the order log(0/852).
Mark Wilson's 1991 insight was that the leap from the 2nd to the 4th category is of the order log(620/852). This is
all that is needed for immediate item and person estimation. But it is not satisfactory for anchoring rating scales.
In practice however, a large value substitutes satisfactorily for infinity. So, a large value such as 40 logits is used
for anchoring purposes. Thus the approximated parameters become -.89, 40.89, -40.00 for SAFILE=. With these
anchored threshold values, the expected category frequencies become: 378.8, 619.4, .0, 851.8. None of these
are more than 1 score point away from their observed values, and each represents a discrepancy of .2% or less
of its category count.
Extreme incidental null categories (unobserved top or bottom categories) are essentially out of range of the
sample and so the sample provides no direct information about their estimates. To estimate those estimates
requires us to make an assertion about the form of the rating scale structure. The Rasch &quot;Poisson&quot; scale is a
good example. All its infinitude of thresholds are estimable because they are asserted to have a specific form. But
see Example 12 for a different apporach to this situation.
Our recommendation is that structural zeroes be rescored out of the data. If categories may be observed next
time, then it is better to include a dummy data record in your data file which includes an observation of the
missing category and reasonable values for all the other item responses that accord with that missing category.
This one data record will have minimal impact on the rest of the analysis.
322.
One observation per respondent
Question: I'm trying to analyze a dataset where there are four test forms, and on each test form there is only one
4-point polytomous item. That is, each student took one and only one test question. Can this type of dataset be
calibrated using Winsteps?
Reply: If there is only one response per person, there is not enough information to construct measures, but only
enough to order the people by the raw score of that one response. But .....
If the people taking each of the 4 forms are supposed to be randomly equivalent, then we can equate the forms,
and discover how a &quot;3&quot; on one form relates to a &quot;3&quot; on another form. To do this:
Enter the 4 forms as 4 items in Winsteps.
For each &quot;item&quot; enter the column of responses.
Anchor the rows at 0.
Set ISGROUPS=0
Run the analysis.
The measure corresponding to each score on each item is given in Table 3.2, &quot;Score at Cat&quot;, and shown in Table
312
2.2. Use the measures in the &quot;At Cat.&quot; column to correspond to the polytomous observations in summary analyses.
Example: The responses to the 4 forms, A, B, C,D, were:
A
B
C
D
1
2
3
4
3
4
2
4
2
3
2
3
4
1 1 3
3 1 4 1
2 1
Note that the order of the persons within form doesn't matter, and the number of respondents per form doesn't
matter. Here is the Winsteps control file:
Title = &quot;Measurement with 4 forms&quot;
NI=4
Item1=1
Name1=1
; there aren't any row names.
Codes=1234
ISGROUPS=0
; allow each form its own rating (or partial credit) scale
Item=Form ; rename to remind ourselves
Person=Row ; Rows are anchored at zero, and so are all equivalent.
Pafile=*
1-7 0
; anchor all rows at &quot;0&quot;. 7 is the largest number of students who took any form.
*
CONVERGE=L
; only logit change is used for convergence
LCONV=0.005
; logit change too small to appear on any report.
&amp;end
A ; the 4 items are the 4 forms
B
C
D
END LABELS
1234 ; responses per form entered as columns with students in any order.
3424
2323
4132
.111
.34.
.1..
Resulting Table 2.2:
TABLE 2.2 Measurement with 4 forms
ZOU767ws.txt Jan 22 6:42 2003
INPUT: 7 ROWS, 4 FORMS MEASURED: 7 ROWS, 4 FORMS, 16 CATS
WINSTEPS 3.38
-------------------------------------------------------------------------------EXPECTED SCORE: MEAN (&quot;:&quot; INDICATES HALF-POINT THRESHOLD)
-3
-2
-1
0
1
2
3
|---------+---------+---------+---------+---------+---------|
1
1
:
2
:
3
:
4
4
|
|
|
|
1
1
:
2
:
3
:
4
4
1
1
:
2
:
3
:
4
4
|
|
1
1
:
2
: 3
:
4
4
|---------+---------+---------+---------+---------+---------|
-3
-2
-1
0
1
2
3
7
M
NUM
2
1
3
4
NUM
FORM
B
A
C
D
FORM
ROWS
Table 3.2:
SUMMARY OF CATEGORY STRUCTURE. Model=&quot;R&quot;
FOR GROUPING &quot;0&quot; FORM NUMBER: 1
A
FORM ITEM DIFFICULTY MEASURE OF .00 ADDED TO MEASURES
+-----------------------------------------------------------------|CATEGORY
OBSERVED|OBSVD SAMPLE|INFIT OUTFIT||STRUCTURE|CATEGORY|
|LABEL SCORE COUNT %|AVRGE EXPECT| MNSQ MNSQ|| MEASURE | MEASURE|
|-------------------+------------+------------++---------+--------+
| 1
1
1 14|
.00
.00| 1.00 1.00|| NONE
|( -1.59)| 1
313
| 2
2
1 14|
.00* .00| 1.00 1.00||
.00 |
-.42 | 2
| 3
3
1 14|
.00* .00| 1.00 1.00||
.00 |
.42 | 3
| 4
4
1 14|
.00* .00| 1.00 1.00||
.00 |( 1.59)| 4
|-------------------+------------+------------++---------+--------+
|MISSING
3 43|
.00
|
||
|
|
+-----------------------------------------------------------------AVERAGE MEASURE is mean of measures in category.
+-------------------------------------------------------------------+
|CATEGORY
STRUCTURE
| SCORE-TO-MEASURE
| 50% CUM.| COHERENCE|
| LABEL
MEASURE S.E. | AT CAT. ----ZONE----|PROBABLTY| M-&gt;C C-&gt;M|
|------------------------+---------------------+---------+----------|
|
1
NONE
|( -1.59) -INF
-1.01|
|
0%
0%|
|
2
.00
1.15 |
-.42 -1.01
.00|
-.61 | 50% 100%|
|
3
.00
1.00 |
.42
.00
1.01|
.00 | 50% 100%|
|
4
.00
1.15 |( 1.59) 1.01 +INF |
.61 |
0%
0%|
+-------------------------------------------------------------------+
1
2
3
4
Form B:
+-------------------------------------------------------------------+
|CATEGORY
STRUCTURE
| SCORE-TO-MEASURE
| 50% CUM.| COHERENCE|
| LABEL
MEASURE S.E. | AT CAT. ----ZONE----|PROBABLTY| M-&gt;C C-&gt;M|
|------------------------+---------------------+---------+----------|
|
1
NONE
|( -1.09) -INF
-.63|
|
0%
0%|
|
2
1.10
.76 |
-.11
-.63
.28|
-.14 | 14% 100%|
|
3
-.69
.76 |
.70
.28
1.34|
.14 |
0%
0%|
|
4
.70
1.08 |( 2.02) 1.34 +INF |
1.02 |
0%
0%|
+-------------------------------------------------------------------+
323.
1
2
3
4
Order of elements in Control file
Element
Condition for Occurrence
&amp;INST (optional, for backwards compatibility only)
TITLE= title of analysis
recommended
ITEM1= starting column of items
Required
NI= number of items
Required
ISGROUPS= grouping information
optional, with GRPFRM=N (the standard)
MODELS= model information
optional, with MODFRM=N (the standard)
RESCORE= rescore information
optional, with RESFRM=N (the standard)
KEY1= key information
optional, if KEYFRM= omitted (the standard)
KEY2= ..
optional, if KEYFRM= omitted (the standard)
KEYn= ..
optional, if KEYFRM= omitted (the standard)
(n=1 to 99, number of largest key)
other control variables
optional
; comments
optional
&amp;END
Required
ISGROUPS= (in data file format)
required if GRPFRM=Y
MODELS= (in data file format)
required if MODFRM=Y
RESCORE= (in data file format)
required if RESFRM=Y
KEY1= (in data file format)
required if KEYFRM=1 or more
KEY2= ..
required if KEYFRM=2 or more, and so on up to
KEYn= (in data file format)
required if KEYFRM=n
Item Names (must be NI= names) required if INUMB=N (the standard)
END NAMES
required if INUMB=N (the standard)
data records
required if DATA=&quot; &quot; (the standard)
324.
Partial Credit model
The &quot;Partial Credit&quot; Rasch model was devised for multiple-choice questions in which credit is given for almostcorrect distractors. But there is no reason to believe that the almost-correctness of distractors to different
questions is the same. Consequently, each item is modeled to have its own response structure.
This model was extended to any questionnaire using ordered polytomies in which the response structure is
314
modeled to be unique to each item.
Winsteps estimates response structures by item groupings, using ISGROUPS=. From this perspective, the
Andrich Rating Scale Model includes all items in one grouping. The Masters Partial Credit Model allocates each
item to its own grouping.
The conventional representation of the Partial Credit model is
log ( Pnij / Pni(j-1) ) = Bn - Dij
Winsteps parameterizes Dij as Di + Fij where sum(Fij) = 0. And Di is the average (Dij).
log ( Pnij / Pni(j-1) ) = Bn - Di -Fij
Algebraically these two representations are identical.
Thus every item has a mean difficulty, Di. This simplifies communication, because the results of a Partial Credit
analysis now have the same form as any other polytomous analysis supported by Winsteps.
325.
Plausible values
Plausible values are estimates intended to represent the distribution of measures that could produce the observed
scores. They were developed for large-scale educational assessments from which group-level measures are to
be obtained, but with data too thin to support individual-level measurement.
Winsteps is designed for individual measurement. When this is possible, then group-level reporting can be done,
e.g., with PSUBTOT=. The Winsteps estimate approximates the mean of the plausible-value distribution. For
Rasch software that produces plausible values, see www.winsteps.com/rasch.htm.
326.
Plotting with EXCEL
First try the &quot;Plots&quot; pull-down menu.
Plotting: This is conveniently and flexibly done with EXCEL:
(A) Check that your copy of EXCEL works.
(B) Download the free Excel chart-labeler add-in from www.appspro.com/Utilities/ChartLabeler.htm
(C) Run XYChartLabeler.exe
The Excel add-in &quot;XY Chart Labels&quot; is added to the Excel Tools pull-down menu.
(D) To plot
Write PFILE= or IFILE= from two analyses, or copy from Output Tables.
Copy and paste each of these into EXCEL:
Use &quot;Data&quot; &quot;Text to Columns&quot; to put values in columns
Put the columns to be plotted next to each other:
x-axis values to left of y-axis values.
Highlight numbers to be cross-plotted.
To make a plot:
Click on &quot;Chart Wizard&quot;
Click on &quot;XY Scatter&quot;
&quot;Next&quot;
&quot;Next&quot;
Fill in &quot;Chart Title&quot; and &quot;Value&quot; names
&quot;Next&quot;
315
Click &quot;A new sheet&quot;
&quot;Finish&quot;
On the plot:
Click &quot;Series 1&quot;
&quot;Clear&quot;
Right click a data point
Click &quot;Format data&quot;
Click &quot;Marker none&quot;
(the data points will soon disappear!! - Don't worry!)
Click &quot;OK&quot;
Right click a gridline
&quot;Clear&quot;
Add point labels:
Click on &quot;Chart&quot; tab
Click &quot;Tools&quot;
Click &quot;XY Chart labels&quot;
Click &quot;XY Add&quot;
Click &quot;Centered&quot;
Click &quot;Select label _&quot;
Click &quot;Sheet 1&quot; tab
Highlight point labels
Click red marker
Click &quot;OK&quot;
Point-labeled XY plot appears.
Use the plot to communicate:
Click on plot
Use handles to make the plot square
If drawing toolbar is not active:
Right click a toolbar
Click on Drawing
Click on &quot;line&quot; tool
Draw in a useful line.
327.
Poisson counts
The Winsteps program can analyze Poisson count data, with a little work.
Poisson counts are a rating (or partial credit) scale with pre-set structure. The structure measures are loge(n),
n=1 upwards.
You can define a structure anchor file in this way:
XWIDE=2
STKEEP=YES
CODES=0001020304050607080901011121314.......979899
SAFILEe=*
0 0
1 0
; the value corresponding to log(1) - the pivot point for the item measure
2 .693
; the value corresponding to loge(2)
3 1.099
; the value corresponding to log(3)
.....
99 4.595
; the value corresponding to log(99)
*
Arrange that the observations have an upper limit much less than 99.
You may find that you need to multiply all structure measures by a constant to adjust the &quot;natural&quot; form of the
Poisson counts to the actual discrimination of the empirical Poisson process. (The Facets program does this
316
automatically, if so instructed.)
You need to adjust the constant so that the average mean-square is about 1.0. (See RMT 14:4 about using
mean-squares to adjust logit user-scaling.)
But my experience with running Poisson counts in the Facets program (which supports them directly) is that most
&quot;Poisson count&quot; data do not match the Poisson process well, and are more usefully parameterized as a rating (or
partial credit) scale. There is nearly always some other aspect of the situation that perturbs the pure Poisson
process.
328.
Point-measure correlation
PTMEA, the point-measure correlation is reported instead of PTBIS when PTBIS=N or PTBIS=RPM is specified.
PTMEA or RPM is the point-measure correlation, rpm. It is computed in the same way as the point bi-serial,
except that Rasch measures replace total scores. Since the point-biserial loses its meaning in the presence of
missing data, specify PTBIS=N when data are missing or CUTLO= or CUTHI= are specified.
The formula for this product-moment correlation coefficient is:
rpbis = {sum {(x-x bar)(y-y bar``)}} over
{sort {{sum {(x-x bar``)} sup 2}`` {sum {(y-y bar``)} sup 2}} }
where x = observation for this item (or person), y = measure for this item (or person). The range is -1 to +1. Its
maximum values approximate the point-biserial correlation, reported in Rasch Measurement Transactions 5:4 to
be:
329.
Polytomous mean-square fit statistics
Response String
INFIT OUTFIT
RPM (PTMEA)
Easy....................Hard MnSq MnSq Corr. Diagnosis
I. modelled:
33333132210000001011
31332332321220000000
33333331122300000000
33333331110010200001
.98 .99 .78 Stochastically
.98 1.04
.81 monotonic in form,
1.06
.97 .87 strictly monotonic
1.03
1.00
.81 in meaning
II. overfitting (muted):
33222222221111111100
33333222221111100000
32222222221111111110
32323232121212101010
.18
.31
.21
.52
.22
.35
.26
.54
.92
.97
.89
.82
Guttman pattern
high discrimination
low discrimination
tight progression
317
III. limited categories:
33333333332222222222
22222222221111111111
33333322222222211111
.24 .24 .87 high (low) categories
.24 .34 .87 central categories
.16 .20 .93 only 3 categories
IV. informative-noisy:
32222222201111111130
33233332212333000000
33133330232300101000
33333333330000000000
.94 1.22
.55
1.25
1.09
1.49
1.40
1.37
1.20
noisy outliers
.77 erratic transitions
.72 noisy progression
.87 extreme categories
V. non-informative:
22222222222222222222
12121212121212121212
01230123012301230123
03030303030303030303
03202002101113311002
.85 1.21
.00
1.50
1.96
3.62
4.61
5.14
6.07
2.99
3.59
one category
-.09 central flip-flop
-.19 rotate categories
-.09 extreme flip-flop
-.01 random responses
VI. contradictory:
11111122233222111111
11111111112222222222
22222222223333333333
00111111112222222233
00000000003333333333
1.75
2.56
2.11
4.00
8.30
.00 folded pattern †
-.87 central reversal
-.87 high reversal
-.92 Guttman reversal
-.87 extreme reversal
2.02
3.20
4.13
5.58
9.79
from Smith R.M. (1996) Rasch Measurement Transactions 10:3 p. 516
The z-score standardized statistics report, as unit normal deviates, how likely it is to observe the reported meansquare values, when the data fit the model. The term z-score is used of a t-test result when either the t-test value
has effectively infinite degrees of freedom (i.e., approximates a unit normal value) or the Student's t-distribution
value has been adjusted to a unit normal value.
† &quot;folded data&quot; can often be rescued by imposing a theory of &quot;not reached&quot; and &quot;already passed&quot; on to the
observations. For instance, in archaeological analysis, the absence of bronze implements can mean a &quot;stone
age&quot; or an &quot;iron age&quot; society. A useful recoding would be &quot;1&quot; = &quot;stone age&quot;, &quot;2&quot; = &quot;early bronze&quot;, &quot;3&quot; = &quot;bronze&quot;,
&quot;2=&gt;4&quot; = &quot;late bronze&quot;, &quot;1=&gt;5&quot; = &quot;iron age&quot;. This can be done iteratively to obtain the most self-consistent set of
4's and 5's. (Folding is discussed in Clive Coombes' &quot;A Theory of Data&quot;.)
330.
Quality-control misfit selection criteria
Rasch measurement does not make any presumptions about the underlying distribution of the parameters.
Maximum likelihood estimation expects &quot;errors&quot; in the observations to be more or less normally distributed around
their expected values. Since all observations are integral values, this expectation can be met only asymptotically
as the number of persons and items becomes infinite. The information-weighted fit statistic, &quot;infit&quot;, and the outliersensitive fit statistic, &quot;outfit&quot;, are described in BTD and RSA. Possible values, and hence interpretation, of these
statistics is influenced by the observed distribution the person and item statistics. This is particularly true of their t
standardized values which are designed to follow standard normal (0,1) distributions. The local significance of
these statistics is best interpreted in terms of their means and sample standard deviations reported in Table 3.
Start investigating the misfit causing the most extreme values of these statistics, and stop your investigation when
the observed responses become coherent with your intentions.
The fit statistics reported will not exactly match those printed in BTD or RSA, or those produced by another
program. This is because the reported values of these statistics are the result of a continuing process of
development in statistical theory and practice. Neither &quot;correct&quot; fit statistics nor &quot;correct&quot; values exist, but see the
Appendices for guidance.
Report measure in Tables 6 and 10 if any of:
318
Statistic Less than
Greater than
t standardized INFIT
-(FITP or FITI)
t standardized OUTFIT -(FITP or FITI)
mean-square INFIT
1-(FITP or FITI)/10
mean-square OUTFIT 1-(FITP or FITI)/10
FITP or FITI
FITP or FITI
1+(FITP or FITI)/10
1 + (FITP or FITI)/10
or point-biserial correlation negative
To include every person, specify FITP=0. For every item, FITI=0.
For Table 7, the diagnosis of misfitting persons, persons with a t standardized fit greater than FITP= are reported.
Selection is based on the OUTFIT statistic, unless you set OUTFIT=N in which case the INFIT statistic is used.
For Table 11, the diagnosis of misfitting items, items with a t standardized fit greater than FITI= are reported.
Selection is based on the OUTFIT statistic, unless you set OUTFIT=N in which case the INFIT statistic is used.
331.
Rank order data
Rankings and partial rankings, with or without ties, can be conveniently analyzed using ISGROUPS=0.
Each row is an element to be ranked.
Each column is a set of rankings. In the item label, place any interesting demographics about the person doing
the ranking.
Note: if every ranking set includes every element, and ties are not allowed, then elements can be columns and
ranking sets as rows. ISGROUPS=0 is not required.
In general, we allow each ranking (column) to define its own &quot;ranking scale&quot;. This is equivalent to the Partial
Credit model.
Measures for the elements are obtained. Measures for the ranking sets are meaningless and can be ignored.
Fit statistics, DIF and DPF analysis, and contrast analysis of residuals are all highly informative.
332.
Rectangular copying
To copy a rectangle of numbers:
1. Select the lines of text that include the rectangle of numbers.
2. Copy the lines to the clipboard
3. Paste the lines into a word-processing document or an Excel spreadsheet cell.
4.Set the font of the lines to Courier.
5A. In Word, select rectangles with Alt+Mouse (see below)
5B. In TextPad, select rectangles with Alt+Mouse
5C. In WordPerfect, select &quot;Edit &gt; Select &gt; Rectangle&quot;
5D. In Excel, use &quot;Data &gt; Text to Columns&quot; to select the column of numbers into a column.
You could also display the column of numbers on your computer screen and do a graphical copy. PrintScreen
saves the screen to the clipboard, then paste into Paint and do a rectangle selection of what you want. Paste the
selection into your document as a Figure.
Rectangular copy-and-paste with Microsoft Word
In Word, ctrl-A the whole document.
Select a &quot;Courier&quot; font. Now everything lines up neatly in columns.
319
Click the left mouse button to unlhighlight everything.
Move the mouse pointer to the top-left corner of the rectangle you want to copy.
Press and hold down the Alt-Key. Left-click and hold down the mouse. Release the Alt-Key
Drag down to the right-hand corner of what you want. The rectangle should high-light.
Release all keys and mouse-buttons.
Ctrl+C to copy the high-lighted section.
Move the mouse pointer to where you want the rectangle to go.
Ctrl+V to paste.
Add or delete blank spaces to line things up neatly
or use the free text editor TextPad
333.
Reliability and separation of measures
The Winsteps &quot;person reliability&quot; is equivalent to the traditional &quot;test&quot; reliability. Low values indicate a narrow
range of person measures, or a small number of items. To increase person reliability, test persons with more
extreme abilities (high and low), lengthen the test. Improving the test targeting may help slightly.
The Winsteps &quot;item reliability&quot; has no traditional equivalent. Low values indicate a narrow range of item
measures, or a small sample. To increase &quot;item reliability&quot;, test more people. In general, low item reliability means
that your sample size is too small for stable item estimates based on the current data. If you have anchored
values, then it is the item reliability of the source from which the anchor values emanate which is crucial, not the
current sample.
The traditional &quot;test reliability&quot;, as defined by Charles Spearman in 1904 is the &quot;true person variance / observed
person variance&quot; for this sample on these test items. So it is really a &quot;person sample reliability&quot; rather than a &quot;test
reliability&quot;, where reliability = reproducibility of person ordering. The &quot;true person variance&quot; cannot be known, but it
can be approximated. KR-20 approximates it by summarizing item point-biserials. Cronbach Alpha approximates
it with an analysis of variance. Winsteps approximates it using the measure standard errors.
The &quot;model&quot; person reliability (including measures for extreme scores) is an upper bound to this value, when
persons are ordered by measures.
The &quot;real&quot; person reliability (including measures for extreme scores) is a lower bound to this value, when persons
are ordered by measures
KR-20 value is an estimate of the value when persons are ordered by raw scores. CRONBACH ALPHA (KR-20)
KID RAW SCORE RELIABILITY is the conventional &quot;test&quot; reliability index. It reports an approximate test reliability
based on the raw scores of this sample. It is only reported for complete data. An apparent paradox is that extreme
scores have perfect precision, but extreme measures have perfect imprecision.
Winsteps computes upper and lower boundary values for the True Reliability. The lower boundary is the Real
Reliability. The upper boundary is the Model Reliability. The unknowable True Reliability lies somewhere
between these two. As contradictory sources of noise are remove from the data, the True Reliability approaches
the Model Reliability
Conventionally, only a Person (&quot;Test&quot;) Reliability is reported. The relationship between raw-score-based reliability
(i.e., KR-20, Cronbach Alpha) and measure-based reliability is complex, see www.rasch.org/rmt/rmt113l.htm - in
general, Cronbach Alpha overestimates reliability, Rasch underestimates it. So, when it is likely that the Rasch
reliability will be compared with conventional KR-20 or Cronbach Alpha reliabilities (which are always computed
assuming the data match their assumptions), then include extreme persons and report the higher Rasch
reliability, the &quot;Model&quot; reliability, computed on the assumption that all unexpectedness in the data is in accord with
Rasch model predictions. The big differences between Score and Measure reliabilities occur when
(a) there are extreme scores. These increase score reliability, but decrease measure reliability.
(b) missing data. Missing data always decreases measure reliability. If the missing data are imputed at their
expected values (in order to make conventional reliability formulas computable), they increase score reliability.
Winsteps attempts to adjust the raw-score reliability for this inflation in the raw-score reliability, but can only do the
adjustment in an approximate way.
Winsteps also reports an item reliability, &quot;true item variance / observed item variance&quot;. When this value is low, it
320
indicates that the sample size may be too small for stable comparisons between items.
Anchored values are treated as though they are the &quot;true values&quot; of the MLE estimates. Their local standard
errors are estimated using the current data in the same way as unanchored MLE standard error estimates. It is
the measures (anchored or unanchored) and local standard errors that are used in the reliability computations. If
you wish to compute reliabilities using different standard error estimates (e.g., the ones when the anchor values
were generated), then please perform a separate reliability computation (using Excel).
You can easily check the Winsteps reliability estimate computation yourself.
Read the Winsteps PFILE= into an Excel spreadsheet.
Compute the STDEVP standard deviation of the person measures. Square it. This is the &quot;Observed variance&quot;.
&quot;Model&quot; Reliability: Take the standard ERROR column. Square each entry. Sum the squared entries. Divide that
sum by the count of entries. This is the &quot;Model Error variance&quot; estimate. Then,
Model Reliability = True Variance / Observed Variance = (Observed Variance - Model Error Variance) / Observed
Variance.
&quot;Real&quot; Reliability: Take the standard ERROR column. Square each entry, SE^2. In another column, put
SE^2*Maximum [1.0, INFIT mean-square). Divide that sum by the count of entries. This is the &quot;Real Error
variance&quot; estimate. Then,
Real Reliability = True Variance / Observed Variance = (Observed Variance - Real Error Variance) / Observed
Variance.
Separation and Reliability
The crucial elements in the computation of reliability are the &quot;True&quot; variance and the Error variance. These are
squared distances and so difficulty to conceptualize directly. It is easier to think of their square-roots, the &quot;True&quot;
standard deviation (TSD) and the root-mean-square standard error (RMSE).
SEPARATION is the ratio of the PERSON (or ITEM) ADJ.S.D., the &quot;true&quot; standard deviation, to RMSE, the error
standard deviation. It provides a ratio measure of separation in RMSE units, which is easier to interpret than the
2
reliability correlation. This is analogous to the Fisher Discriminant Ratio. SEPARATION is the signal-to-noise
ratio, the ratio of &quot;true&quot; variance to error variance.
RELIABILITY is a separation reliability. The PERSON (or ITEM) reliability is equivalent to KR-20, Cronbach
Alpha, and the Generalizability Coefficient. See much more at Reliability. The relationship between separation
SEPARATION and RELIABILITY is
2
2
RELIABILITY = SEPARATION /(1+SEPARATION )
0.5
or SEPARATION = (RELIABILITY/(1-RELIABILITY)) .
Error
RMSE
1 0
1 1
1 2
1 3
1 4
True
S.D.
0 1
1 2
4 5
9 10
16 17
True
Obs
Signal- Separation=True SD/Error RMSE
Var Var.
to-noise
Reliability = True Var/Obs Var
0 0 0
1 1 0.5
2 2 0.8
3 3 0.9
4 4 0.94
Fisher R.A. (1936) The Use of Multiple Measurements in Taxonomic Problems. Annals of Eugenics, 7: 179-188.
http://www.library.adelaide.edu.au/digitised/fisher/138.pdf
334.
Right-click functions
Mouse Right-Click on the Winsteps screen
Right-click anywhere on the Winsteps screen, and a menu will be displayed.
During estimation: It is the File menu.
After estimation: It is the Table menu.
321
Mouse Right-Click on the Task Bar
Right-click on the task bar, to obtain the window-control menu.
Mouse Right-Click on File Dialog Box:
Right-click a file name, and get the Send To menu.
Add functionality to the Send To menu by copying short-cuts into c:\windows\sendto (or the equivalent sendto
folder in your version of Windows) - a useful program to add is WordPad or your own text editor. To do this:
Start
Find
Files or Folders
Named: WordPad in C:
when Wordpad.exe appears, Right-click on it.
Send To: Desktop: Create shortcut
Exit from Find
On Desktop:
My Computer
C:
Windows (if this does not appear: then View: Folder Option: View: Show all files)
Send To
Drag WordPad shortcut from Desktop into SendTo folder.
WordPad is now in the Send To menu.
335.
Rules for assigning values to control variables (key-words)
Do not worry about these unless WINSTEPS does not respond to your control file the way you expected. If
possible, compare your control file with what is shown in Table 0 of your report output file in order to isolate the
problem.
1. Values are assigned to control variables by typing the name of the control variable (or enough of it to
disambiguate it), an equal sign, &quot;=&quot;, and the value, e.g.
TABLES=11011011100
or
322
TAB=1101101110
this is enough of TABLES to make clear what you mean.
2. You must use one line for each assignment, but continuation lines are permitted.
To continue a line, put a + at the end of the line. Then put a + at the start of the text in the next line. The two lines
will be joined together so that the + signs are squeezed out, e.g.,
TITLE = &quot;Analysis of medical+
+ research data&quot;
is interpreted as
TITLE = &quot;Analysis of medical research data&quot;
Continuation lines &quot;+&quot; are helpful to make control files fit on your screen.
CODES = 01020304+
+05060708
is interpreted as
CODES = 0102030405060708
To comment out a continuation line:
; CODES = 01020304+
+05060708
or
; CODES = 01020304+
; +05060708
3. The control variables may be listed in any order.
4. Character strings must be enclosed in 'single quotes' or &quot;double quotes&quot; when they contain blanks, e.g.,
TITLE=&quot;Rasch Analysis of Test Responses&quot;
or
TITLE='Rasch Analysis of Test Responses'
Quotes are not required for single words containing no blanks, e.g. PFILE=kctpf.txt
5. The control variables may be in upper or lower case or mixed,
e.g., Pfile = Person.txt
6. Blanks before or after control variables, and before or after equal signs are ignored, e.g.
TITLE=&quot;Test Report&quot;
and
TITLE = &quot;Test Report&quot;
are equally correct.
7. Commas at the end of lines are ignored, so equally correct are:
NAME1 = 33,
and
NAME1 = 33
8. Control variables can be made into comments, and so be ignored, by entering a semi-colon in column 1, e.g.
; FITP=3 is ignored
9. When all control variables (required or optional) have been assigned values,
type &amp;END (in upper or lower case) on the next line, e.g.,
Title =&quot;A 30 Item test&quot;
NI = 30
; this is a comment: person names in columns 1-20.
ITEM1= 21
&amp;END
323
336.
Shortcut Keys
Some very frequent operations are quicker to perform using shortcut keystokes than by menu selection. Here are
the shortcut keys implemented in Winsteps:
Alt+ hold down the Alt key and press the letter key.
Ctrl+ hold down the Ctrl key and press the letter key.
Alt+A
Alt+E
Alt+H
Alt+R
Alt+S
Alt+X
start Another copy of Winsteps
Edit the control file
display Help file
Restart Winsteps with this control file
Specification entry
eXit this Winsteps, and restart with this control file
Ctrl+F
Ctrl+O
Ctrl+Q
Ctrl+S
Ctrl+P
Finish iterating
Open a control or output file
Quit Winsteps
Save on-screen activity log
Print on-screen activity log
Esc Escape from action
337.
Specifying how data are to be recoded
You will need to choose how this is done.
First, use CODES= to specify the response codes in your data file.
If there is only one type of recoding to be done, use
NEWSCORE=
If this one type of rescoring only applies to some of the items, also use
RESCORE=
If the rescoring is more complex, use
IREFER= and IVALUE=
If the items are multiple-choice, use
KEYn=
If missing values are not to be ignored, i.e., treated as not-administered, you will need
MISSCORE=
If alphabetical codes are used to express two-digit numbers in one column, use
ALPHANUM=
338.
Standard errors: model and real
A standard error quantifies the precision of a measure or an estimate. It is the standard deviation of an imagined
error distribution representing the possible distribution of observed values around their &quot;true&quot; theoretical value.
This precision is based on information within the data. The quality-control fit statistics report on accuracy, i.e., how
closely the measures or estimates correspond to a reference standard outside the data, in this case, the Rasch
model.
Model &quot;Ideal&quot; Standard Error
The highest possible precision for any measure is that obtained when every other measure is known, and the
data fit the Rasch model. For well-constructed tests with clean data (as confirmed by the fit statistics), the model
324
standard error is usefully close to, but slightly smaller than, the actual standard error. The &quot;model&quot; standard error
is the &quot;best case&quot; error. It is the asymptotic value for JMLE. For dichotomous data this is, summed over items
i=1,L for person n, or over person n=1,N for item i:
S.E . =
∑ (P (1 − P ))
ni
ni
For polytomies (rating scales, partial credit, etc.), with categories j=0,m:
2
m

m

S.E. = ∑ ∑ jPnij − ∑ jPnij  


n or i j =0 
j =0
 

Misfit-Inflated &quot;Real&quot; Standard Error
Wright and Panchapakesan (1969) discovered an important result for tests in which each examinee takes more
than a handful of items, and each item is taken by more than a handful of examinees: the imprecision introduced
into the target measure by using estimated measures for the non-target items and examinees is negligibly small.
Consequently, in almost all data sets except those based on very short tests, it is only misfit of the data to the
model that increases the standard errors noticeably above their model &quot;ideal&quot; errors. Misfit to the model is
quantified by fit statistics. But, according to the model, these fit statistics also have a stochastic component, i.e.,
some amount of misfit is expected in the data. Discovering &quot;perfect&quot; data immediately raises suspicions!
Consequently, to consider that every departure of a fit statistic from its ideal value indicates failure of the data to
fit the model is to take a pessimistic position. What it is useful, however, is to estimate &quot;real&quot; standard errors by
enlarging the model &quot;ideal&quot; standard errors by the model misfit encountered in the data.
Recent work by Jack Stenner shows that the most useful misfit inflation formula is
Real S.E. of an estimated measure = Model S.E. * Maximum [1.0, sqrt(INFIT mean-square)]
In practice, this &quot;Real&quot; S.E. sets an upper bound on measure imprecision. It is the &quot;worst case&quot; error. The actual
S.E. lies between the &quot;model&quot; and &quot;real&quot; values. But since we generally try to minimize or eliminate the most
aberrant features of a measurement system, we will probably begin by focusing attention on the &quot;Real&quot; S.E. as we
establish that measurement system. Once we become convinced that the departures in the data from the model
are primarily due to modelled stochasticity, then we may base our decision-making on the usually only slightly
smaller &quot;Model&quot; S.E. values.
What about Infit mean-squares less than 1.0? These indicate overfit of the data to the Rasch model, but do not
reduce the standard errors. Instead they flag data that is lacking in randomness, i.e., is too deterministic.
Guttman data are like this. Their effect is to push the measures further apart. With perfect Guttman data, the
mean-squares are zero, and the measures are infinitely far apart. It would seem that inflating the S.E.s would
adjust for this measure expansion, but Jack Stenner's work indicates that this is not so. In practice, some items
overfit and some underfit the model, so that the overall impact of low infit on the measurement system is diluted.
339.
Starting Winsteps from the DOS prompt
WINSTEPS can also be invoked from the DOS prompt in a DOS window.
At the prompt enter
C:&gt;WINSTEPS(Enter)
Winsteps proceeds with its standard operations.
You can enter control and output files directly on the prompt line.
C:&gt;WINSTEPS SF.txt SF.OUT(Enter)
Winsteps starts analysis immediately. You will not be prompted for &quot;Extra Specifications&quot;
You can also enter extra specifications here:
C:&gt;WINSTEPS SF.txt SF.OUT chart=yes distractors=no(Enter)
Leave no spaces within specifications, or place them in quotes, e.g.,
C:&gt;WINSTEPS SF.txt SF.OUT &quot;chart = yes&quot; &quot;distractors = no&quot;(Enter)
To perform the previous analysis again, with a temporary report output file:
C:&gt;WINSTEPS @(Enter)
@ is replaced by the top control file on the Files= menu. If no output file is specified, then a temporary one is
325
used.
For Batch file operation, see Batch=
340.
Subtest scoring
A test or protocol may consist of a series of subtests. Code each item in its item label with what subtest it belongs
to.
1. Analysis of subtests.
Items and Persons: Use ISELECT= in your Winsteps control file to select the relevant subtest. This performs an
independent analysis of the items and persons for the subtest.
2. Reporting of subtests.
Items: In an overall analysis, the items of individual subtests can be reported after applying ISELECT= from the
Specification pull-down box.
Persons: a measure for each person on each subtest can be obtained by specifying the subtest character in
DPF= and producing Table 31 or the DPF plot.
341.
Transposing the data matrix
To transpose the rows and columns of the data matrix, select Transpose on the Output Files menu.
then
326
Original data: data codes are those in the data file.
Scored data: data after applying NEWSCORE=, IVALUE=, KEY=, etc.
Recounted data: data after applying STKEEP=No, etc.
Permanent file: file name for the transposed file. Enter in the white box, or use the &quot;Browse&quot; function.
The transposed file is created and displayed.
Temporary file: use a temporary file: this can be &quot;saved as&quot; a permanent file.
The transposed file is created and displayed.
Launch Winsteps: launch Winsteps with the permanent transposed file as its control file.
Display original: show the original, untransposed, control file.
Done: transposing actions completed
Cancel: exit from this routine
Help: show Help file.
producing, for Exam1.txt:
; Transposed from: C:\WINSTEPS\examples\exam1.txt
&amp;INST
TITLE = &quot;TRANSPOSED: KNOX CUBE TEST&quot;
ITEM = KID
PERSON = TAP
NI = 35
; ACTIVE PERSONS BECOME COLUMNS
;NN = 18
; ACTIVE ITEMS BECOME ROWS
ITEM1 = 1
XWIDE = 1
NAME1 = 37
NAMLEN = 16
; ends at 52
CODES = 01
STKEEP = Y
; Add here from original control file: C:\WINSTEPS\examples\exam1.txt
&amp;END
Richard M 1
...... (more item labels)
Helen
F 35
END NAMES
11111111111111111111111111111111111 1-4
1
...... (more data records)
00000000000000000000000000000000000 4-1-3-4-2-1-4 18
342.
Unobserved and dropped categories
If you have data in which a category is not observed, then you must make an assertion about the unobserved
category. There are several options:
For intermediate categories: either
(a) this category will never be observed (this is called a &quot;structural zero&quot;). Generally, these categories are
327
collapsed or recoded out of the rating scale hierarchy. This happens automatically with STKEEP=No.
or (b) this category didn't happen to be observed this time (an &quot;incidental&quot; or &quot;sampling&quot; zero). These categories
can be maintained in the rating scale hierarchy (using STKEEP=Yes), but are estimated to be observed with a
probability of zero.
For extreme categories:
(a) if this category will never be observed, the rating scale is analyzed as a shorter scale. This is the Winsteps
standard.
(b) if this category may be observed, then introduce a dummy record into the data set which includes the
unobserved extreme category, and also extreme categories for all other items except the easiest (or hardest)
item. This forces the rare category into the category hierarchy.
(c) If an extreme (top or bottom) category is only observed for persons with extreme scores, then that category
will be dropped from the rating (or partial credit) scales. This can lead to apparently paradoxical or incomplete
results. This is particularly noticeable with ISGROUPS=0.
In order to account for unobserved extreme categories, a dummy data record needs to be introduced. If there is a
dropped bottom category, then append to the data file a person data record which has bottom categories for all
items except the easiest, or if the easiest item is in question, except for the second easiest.
If there is a dropped top category, then append to the data file a person data record which has top categories for
all items except the most difficult, or if the most difficultt item is in question, except for the second most difficult.
This extra person record will have very little impact on the relative measures of the non-extreme persons, but will
make all categories of all items active in the measurement process.
If it is required to produce person statistics omitting the dummy record, then use PSELECT= to omit it, and
regenerate Table 3.
343.
User-friendly rescaling
Transforming logits into other units using UIMEAN=, UPMEAN=, USCALE=. These can be more meaningful for
particular applications, see Chapter 8 of BTD. Anchor values are treated according to UANCHOR=
Example 1: CHIPs are a useful transformation, in which 1 logit = 4.55 CHIPs. In this user-scaling system,
standard errors tend to be about 1 CHIP in size. The recommended control variable settings are:
USCALE = 4.55
UIMEAN = 50
UDECIM = 1
MRANGE = 50
The probability structure of this relationship is:
Probability of Success
Difference between Person Ability Measure and Item Difficulty Measure in CHIPs
.10 -10
.25 -5
.50 0
.75 5
.90 10
Example 2: WITs are one tenth the size of CHIPs, enabling the elimination of decimals from your output tables.
USCALE = 45.5
UIMEAN = 500
UDECIM = 0
MRANGE = 500
Example 3: You want the lowest reportable person measure to be 0 and the highest to be 100. Looking at Table
20, you see the extreme values are -4.53 and +5.72. You have not used USCALE= and UMEAN=.
328
USCALE= (wanted range) / (current range)
USCALE= (100 - 0)
/ (5.72 - -4.53) = 100 / 10.25 = 9.76
UMEAN= ( wanted low ) - ( current low * USCALE= ) = 0 - ( -4.53 * 9.76 ) = 44.20
Required values are:
USCALE = 9.76
UIMEAN = 44.20
UDECIM = 0
to show no decimal places in report
Example 4: You want the lowest reportable person measure to be 0 and the highest to be 100. Looking at Table
20, you see the extreme values are -4.53 and +5.72. The current values in the output are USCALE=1 and
UIMEAN=0.
USCALE= (previous USCALE=) * (wanted range)
/ (current range) = 1* (100 - 0)/ (5.72 - -4.53) = 1 * 100 /
10.25 = 9.76
UMEAN= (wanted low) - (current low - previous UMEAN=)*(wanted range)/(current range) = 0 - (-4.53 0)*100/10.25 = 44.20
UDECIM = 0
to show no decimal places in report
Double checking, when previous UMEAN=0, USCALE=1:
low value = (current low)*(USCALE=) + (UMEAN=) = (-4.53 * 9.76) + 44.20 = -0.01
high value = (current high)*(USCALE=) + (UMEAN=) = (5.72 * 9.76) + 44.20 = 100.02
Example 5: You want the lowest reportable person measure to be 100 and the highest to be 900. Looking at
Table 20, you see the extreme values are -4.53 and +5.72. Looking at the second page of output, you see the
current values are USCALE=1 and UMEAN=0.
USCALE= (previous USCALE=) * (wanted range: 900 - 100) / (reported range: 5.72 - -4.53) = 1 * 800 / 10.25
= 78.05
UMEAN= (wanted low) - (reported low - previous UMEAN=)*(wanted range)/(reported range) = 100 - (-4.53 0)*800/10.25 = 453.56
UDECIM = 0
to show no decimal places in report
Example 6: You want norm-referenced user-scaling, such that the person mean is 0.0, and the person sample
standard deviation is 1.0.
In a standard analysis, set:
UDECIM=4
USCALE=1
UMEAN=0
Look at Table 18
+-----------------------------|ENTRY
RAW
|NUMBER SCORE COUNT MEASURE
|-----------------------------| MEAN
6.7
14.0 -.3728
| S.D.
2.4
.0 2.2202
+------------------------------
Set (either in a new analysis, or using the &quot;Specification&quot; pull-down menu
USCALE = 1/person S.D. = 1/2.2202 = 0.4504
UMEAN = - person mean/person S.D. = - (-.3728)/2.2202 = 0.1679
Look at Table 18
+-----------------------------|ENTRY
RAW
|NUMBER SCORE COUNT MEASURE
|-----------------------------| MEAN
6.7
14.0
.0000
| S.D.
2.4
.0 1.0000
+------------------------------
329
Example 7: You want to give your pass-fail point the value 400 and 100 to the lowest reported measure.
Inspecting your output you see that the pass-fail point is at 1.5 logits, and that -3.3 logits is the lowest reported
measure.
Then 400-100 new units = 1.5 - (-3.3) logits, so
USCALE = 300 / 4.8 = 62.5
UMEAN = 400 - (1.5) * 62.5 = 306.25
Then: 1.5 logits = 306.25 + 1.5*62.5 = 400
-3.3 logits = 306.25 - 3.3*62.5 = 100
344.
Using a word processor or text editor
If WordPad does not work on your computer or you don't like it, then change word processor
a) Input files: all lines in your control and data files follow DOS text conventions. This means that files created with
a Word Processor, such as &quot;Word Perfect&quot;, must be saved as &quot;DOS-text with line breaks&quot; or &quot;ASCII&quot; files.
1. Lines must not contain tabs or word processor codes.
2. Lines cannot overflow onto the next line, except for data records which are processed using the FORMAT= or
MFORMS= control variables.
3. Lines must end with DOS or ASCII Carriage Return and Line Feed codes.
Be particularly careful to instruct your Word Processor to allow more characters within each line than are present
in the longest line in your control or data files. Then your Word Processor will not break long data or control lines
into two or more text lines with &quot;Soft Return&quot; codes. These cause WINSTEPS to malfunction. Space for a large
number of characters per line is obtained by specifying a very wide paper size and/or a very small type size to
your Word Processor.
When using &quot;Word Perfect&quot; to edit control or data files, select the smallest available type size (often 20 cpi or 5
pt). Define and use a very wide (50 inch) page style. It does not matter whether your printer can actually print it.
Always save control and data files as &quot;DOS-text with line breaks&quot; or ASCII files.
With WordStar, use &quot;Non-Document&quot; mode to avoid these difficulties.
b) Output files: when importing WINSTEPS output into a document file, the following options have proved useful:
Base Font - 17 cpi (or more) or 8 point (or smaller) or 132 characters per line (or more)
Left Justify
Page numbering
Margins: top = 1&quot;, bottom = 0.5&quot;, left = 1&quot;, right = 0&quot;
345.
Weighting items and persons
There are circumstances in which certain items are to be given more influence in constructing the measures than
others. For instance, certain items may be considered critical to the demonstration of competence. Unweighted
data is preferable for calibrating the Rasch items. This is because each observation is modeled to contribute one
unit of independent statistical information. The effect of weighting is to distort the distribution of independent
statistical information in the data.
Step 1. Analyze the data without weighting. Investigate misfit, construct validity etc.
Step 2. Weight the items. Compare the item calibrations with weighted and unweighted data to identify where
there are discrepancies.
Though WINSTEPS supports several methods, IWEIGHT= is simplest.
Another approach is to replicate the data for particular items. This can be done with FORMAT= without
changing the data file.
Items can also be rescored from say, 0-1 to 0-2, but this makes variable maps difficult to interpret.
The weights applied to items are persons are used in computing the measure estimates, standard errors and fit
statistics. When using significance tests with weighting, normalize the weights so that the total amount of
330
independent statistical information in the data is not over- or under-inflated, i.e., when using PWEIGHT= with an
observed sample size of N, multiply all PWEIGHT= values by N / (sum of all weights).
The standard is weights = 1.
When an item or person is weighted as 2, then the estimation acts as though that item or person appears twice in
the data file.
When an item or person is weighted as 0, then that person does not influence the estimates, standard errors or fit
statistics of other persons and items, but does have measure, standard error and fit statistics computed on all
observations for itself. This is useful for evaluating pilot or off-dimensional items, or measuring idiosyncratic
persons.
Weight Selection: On the output tables menu, these are the options for persons and/or items. When IWEIGHT=
or PWEIGHT= are used in estimation, reports can be adjusted to reflect those weights or not. Weights of zero are
useful for pilot items, variant items or persons with unusual characteristics. These can be reported exclusively or
excluded from reports.
(1) all items or persons are reported, with their weights (the standard). Tables 23 and 24 are computed as though
all weights are 1.
(2) items or persons with a weight of 0 are excluded from the reporting. Tables 23 and 24 are computed as
though all weights are 1, but zero weights are omitted.
(3) only items or persons with a weight of 0 are reported. Tables 23 and 24 are computed only from items or
persons with a weight of 0.
(4) all items or persons are reported as though they have a weight of 1.
346.
Winsteps: history and steps
What is the origin of Winsteps and to what does &quot;steps&quot; refer? Winsteps is an outcome of this process of
development:
In 1983, Benjamin D. &quot;Ben&quot; Wright of the University of Chicago and &quot;Mike&quot; Linacre released the first Rasch
analysis program for personal computers. It was also the first to allow missing data.
1983: Microscale (on PCs). &quot;Rasch scaling by microcomputer&quot; - since MSDOS was limited to 8-character program
names, the actual execution name was &quot;MSCALE&quot;.
1987: Mscale (dichotomies and Andrich rating scales) + Msteps (for partial credit &quot;steps&quot;). Ben implemented the
Microscale algorithm on a Unix minicomputer, but kept the PC execution name, &quot;Mscale&quot;.
331
1989: Bigscale (back to PCs). Again under MS-DOS but with much larger datasets. Mike takes over development
again.
1991: Bigsteps (the functionality of Msteps was included). Ben interpreted this to mean &quot;big steps forward in
social science measurement&quot;
1998: Winsteps (when made Windows native). Ben interpreted this to mean &quot;winning steps forward in social
science measurement&quot;
When talking about Rasch measurement, Ben used &quot;step&quot; to mean:
(a) the category number counting up from 0 at the bottom. The bottom step, for dichotomies or polytomies, was
the lowest category, always numbered 0. Ben would talk about going up and down the steps as one moved up
and down the latent variable.
(b) the location of the transition from one category to the next higher category on the latent variable. Now called
the Rasch-Andrich threshold for polytomies and the item difficulty for dichotomies.
(c) the process of moving from one category to the next as one's amount of the latent variable changes. A low
negative threshold below a category indicates that the category is easy to step into as one moves up the latent
variable. A high positive threshold below a category indicates a category that is hard to step into. So &quot;disordered&quot;
thresholds around a category (high below, low above) indicate a category that is &quot;hard to step into and easy to
step out of&quot; as one moves up the latent variable, i.e., a narrow category. The extreme of this is an infinitelynarrow, i.e., unobserved, category. It is infinitely hard to step into and infinitely easy to step out of.
347.
Wordpad or text editor
The default text editor used by Winsteps is WordPad, a Microsoft Accessory. If WordPad is not available,
NotePad is used.
How Winsteps decides what Text Editor to use:
(a) On its first installation, or if you delete the file Winsteps.ini in your Winsteps directory, Winsteps looks on your
C: drive for Wordpad.exe.
If there is more than one version of Wordpad.exe on your computer, Winsteps may try to use the
wrong one. The Output Tables may not display, and you may receive messages like:
&quot;The application has failed to start because MFCANS32.dll was not found. Reinstalling the application
may fix the problem&quot;.
Use Windows &quot;Find&quot; or &quot;Search&quot; to look for Wordpad.exe. If there is more than one found, the correct
one is probably in
&quot;Accessories&quot;. Double-click the WordPad entries until one launches. Go to (e) below and check that the
correct one is referenced in Winsteps.ini
(b) If WordPad is found automatically, then its path is placed in Winsteps.ini as the Editor.
(c) If WordPad is not found, then the Editor associated with .txt files is used. If there is none, NotePad is used.
(d) If file Winsteps.Ini exists, then Winsteps uses the Text Editor or Word Processor assigned by Editor= in
Winsteps.Ini
(e) You can examine Winsteps.ini from the Edit pull-down menu, &quot;Edit initial settings&quot;, or using a word
processor or text editor to look at file Winsteps.ini in the Winsteps directory.
Using your own text editor:
You can make WINSTEPS use your own text editor or word processor by changinf the Editor in Edit initial settings
or editing Winsteps.ini in your Winsteps folder. Put the full pathname of your editor in place of WordPad, e.g., for
332
Word 6.0, it could be: &quot;C:\MSOFFICE\WINWORD\WINWORD.EXE&quot;
You can find the full path to your word processor by doing a &quot;Find&quot; or &quot;Search&quot; for it from the &quot;Start&quot; menu.
A useful replacement for Wordpad is the shareware program TextPad. This has many more features.
348.
Data vanishes
Winsteps runs, but not all your data are reported. Please check:
Your control and data files look correct. Use the Edit menu.
ITEM1= points to the first column of reponses
CODES= has all valid data codes.
CUTLO= and CUTHI= are not specified.
ISELECT=, PSELECT=, IDFILE=, PDFILE=, IDELETE=, PDELETE= are as you want.
349.
Display too big
If your screen resolution is low, you may not see all the Winsteps window on your screen.
1. Start Menu =&gt; Settings =&gt; Control Panel =&gt; Display =&gt; Settings =&gt; Screen Resolution =&gt; More
2. For dialog boxes, right-click on the screen. This will bring up this dialog box:
Click on &quot;Left&quot; to move the over-sized dialog box left, etc. Click &quot;Home&quot; to position it in the top left-hand corner of
your screen.
350.
File misread problems
Some file formatting problems are reported with error messages. Others simply mislead Winsteps.
General rules:
(a)
All control and data files must be in &quot;DOS TEXT&quot; or ASCII format.
(b)
To check if your control specifications have been correctly interpreted, look at Table 0, at the end of your
&quot;output&quot; file.
(c)
To check if your data have been correctly interpreted, produce an RFILE= file and look at the response
strings, particularly for column alignment.
Common Problems and their Diagnoses
1) Message:
VARIABLE UNKNOWN OR VALUE INVALID:
Diagnosis: A line in your control file is not of the format &quot;variable = value&quot;. This may be because your control file
is in not in DOS Text or ASCII format. Return to your word processor or WordPad and save your control file as a
DOS Text file. Then rerun.
2A) Message:
PROBLEM: BLANK LINE NEEDED AT END OF: filename
Diagnosis: The last line of file &quot;filename&quot; does not end with a line-feed (LF) code (so does not meet strict
Windows specifications). This can be fixed by adding a blank line to the end of that file. You can use WordPad to
do this by editing the file from the Winsteps .
333
2B) Message:
Use last entry in File menu to add final blank line to:
filename
Then press enter key, or enter a different file name:
Diagnosis: The last line of file &quot;filename&quot; does not end with a line-feed (LF) code (so does not meet strict
Windows specifications). This can be fixed by adding a blank line to the end of that file. You can use WordPad to
do this by access the file from the Winsteps File menu. It will probably be the last file listed there. You could also
copy the file, add a blank last line, and then type in the file name in Winsteps.
3) Blank characters disappear, causing misalignment of data or control instructions.
Diagnosis:
(a) These &quot;blank&quot; characters could be ASCII &quot;13&quot; (CR) codes. Some word processors treat them as blanks, some
as end-of-line codes, and some ignore them, making them &quot;disappear&quot;. Winsteps ignores them. WordPad
reports them as squares. Replace them with blanks.
(b) These &quot;blank&quot; characters could be ASCII &quot;9&quot; (TAB) codes. Word processors expand these into several
blanks.
Winsteps treats them as one blank. For consistency, globally replace Tab characters with blanks.
4) Data or control lines disappear.
Diagnosis:
(a) Word processors automatically wrap long lines across several lines. Winsteps does not. Make sure that each
line ends with an &quot;end of line&quot; code.
(b) Some word processors use ASCII &quot;13&quot; (CR) as an end-of-line code. Winsteps ignores these codes. Use
ASCII &quot;10&quot; (LF) as the end-of-line code. Press the Alt key, while typing 10 on the numeric keypad to generate this
code.
5) Data or control lines are split across several lines.
Diagnosis: You may have edited your control or data files with a word processor that wrapped one line across
several lines. When you saved this file in DOS text format, each line was then saved as several lines. Re-edit
the file, reduce the point size, maximize the page size (e.g., landscape) until each line only occupies one line.
Then save the file again in DOS text or ASCII format.
6) Message:
BEYOND CAPACITY
Diagnosis: Your data file contains more person records than can be analyzed in one run. Some records have
been bypassed.
Data sets that exceed program capacity to analyze in one run offer opportunities, as well as challenges. There
are several strategies.
I. Analyze a sample of the data. Use this to produce anchor values. Then, using the anchor values, run all the
data one section at a time.
II. Analyze a sample of the data. Analyze another sample. Compare results to identify instability and compute
reasonable anchor values. Remember that small random changes in item calibrations have negligible effect on
person measures.
To select a sample of your data, use the FORMAT= statement. See the example on pseudo-random person
selection on page .
There are versions of Winsteps that support more persons. Contact www.winsteps.com for details.
7) File read errors
Reading Control Variables ..
Input in process..
Opening: c:\input-file.txt
PROBLEM: Access denied (5): for file: c:\input-file.txt in mode I
334
Your data file may be &quot;read only&quot; and missing an end-of-line indicator on the last data record. Copying and paste
your data file as a data file with ordinary file access privileges. Then reference this file in your DATA=
specification.
351.
If Winsteps does not work
1) Repeat the installation process. It will not delete any of your data files.
2) Check that WINSTEPS runs on an example dataset, e.g., exam1.txt
3) If the program will not run, or produces implausible results:
a) It may be a Windows-related problem, see www.winsteps.com/problems.htm
b) There may not be enough disk space for work files, see &quot;Not enough disk space&quot;.
c) There may not be sufficient RAM memory to execute. See &quot;Not enough memory&quot;.
d) It may be a WordPad problem, see Changing your Word Processor setting.
4) If you still have problems, use the comment form at www.winsteps.com
352.
Initialization fails
You've downloaded and run WinstepsInstall.exe or MinistepInstall.exe or a similar installation routine. It ran, and
then automatically started Winsteps or Ministep, but now ...
It &quot;hangs&quot; with the message: &quot;Constructing Winsteps.ini ...&quot;
So then you kill Winsteps or Ministep from the &quot;Ctrl+Alt+Delete&quot; program box
Is this the scenario? If so, here's what to do next:
You need to create a text file called &quot;Winsteps.ini&quot; with NotePad in the same directory that holds Winsteps.exe or
Ministep.exe.
If you cannot do this manually, then this is the reason why Winsteps or Ministep is failing.
The format of the text file is:
Editor=&quot;C:\Program Files\Accessories\WORDPAD.EXE&quot;
Excel=&quot;C:\Program Files\Microsoft Office\Office\EXCEL.EXE&quot;
Filter=All Files (*.*)
Temporary directory=&quot;C:\TEMP\&quot;
Reportprompt=Yes
Welcome=Yes
Replace C:\Program Files\Accessories\WORDPAD.EXE with the path to WordPad on your machine.
Replace or remove the line Excel=
If you do not have &quot;Wordpad.exe&quot; on your computer, then put in Editor=&quot;NOTEPAD.EXE&quot;
Then start Winsteps or Ministep again. (Not the Install procedure).
Pull-down the Winsteps or Ministep &quot;edit&quot; menu immediately, look at &quot;initial settings&quot;.
They should correspond to Winsteps.ini.
Now proceed normally with Starting Winsteps
Other installation snags are solved at www.winsteps.com/problems.htm
353.
Not enough disk space
You need about twice as much work file space on your disk as the size of your data file.
Double-click &quot;My Computer&quot; and right-click on the disk drive. The &quot;properties&quot; shows the amount of available disk
335
space.
Temporary work files are placed in the TEMP directory (see WINSTEPS.INI). The temporary table output files are
placed temporarily in the current directory (reported on your screen when WINSTEPS starts). Delete unwanted
files to give yourself more disk space, or log onto a different disk drive, with more available space, before
executing WINSTEPS. Files with names &quot;Z......ws.txt&quot;, &quot;Z......ws.xls&quot;, &quot;(numbers)ws.txt&quot; and &quot;.TMP&quot; and files in the
TEMP directory are work files. These can be deleted.
354.
Not enough memory
WINSTEPS can use all available memory. If memory is exhausted, then WINSTEPS fails or proceeds very slowly.
Terminate other programs. Due to &quot;memory leakage&quot;, Windows gradually loses access to memory not properly
released when programs terminate. Reboot your computer to free up this memory.
355.
Plotting problems
The Plots menu calls Microsoft Excel to perform many plotting functions. This enables you to use all the
functionality of Excel to customize the plots to your needs.
Many different versions of Excel have been published by Microsoft, and they have some incompatibilities. You
may find that plotting fails with an error code, e.g., &quot;1004&quot;.
The Winsteps-to-Excel interface is in a module in your Winsteps folder named winexcel.exe. Please replace this
module with a module that matches your version of Excel. These are available from
www.winsteps.com/problems.htm
356.
Tab-delimited file problems
Question:
I'm having trouble reading files from SPSS or EXCEL. I save the file in EXCEL or SPSS as a text file and have
tried saving in all formats (tab-delimited, comma, etc.). It always saves the data with tabs separating the column
values. When I read the file in Winsteps,
the tabs are translated into columns, thus producing every other row of empty columns. How do I get around this?
Answer:
There are numerous ways round this difficulty.
(a) Use Winsteps to format your SPSS .sav file into a Winsteps file (use the SPSS pull-down menu in Winsteps).
(b) Use the Winsteps &quot;setup&quot; routine to format your EXCEL data.
Start Winsteps &quot;setup&quot; (Setup pull-down menu in Winsteps)
Copy the cells you want from EXCEL, paste them into the Data area of the Setup screen.
(c) Get EXCEL to write out a file without Tabs: &quot;Save as&quot; a &quot;.prn&quot; file. &quot;Formatted text (space delimited)&quot;
Before you do this, set all data columns widths to 1.
(d) Tell Winsteps the data are Tab separated: use DELIMITER= .... but this is tricky to get to work correctly.
(e) Edit your data file with WordPad. Replace all tabs with nothing. To &quot;find&quot; a Tab, highlight a Tab, copy it
(ctrl+C), then paste it into the &quot;find&quot; box: ctrl+v.
(f) Edit your data file with WordPad. Replace all tabs with a space. To &quot;find&quot; a Tab, highlight a Tab, copy it
(ctrl+C), then paste it into the &quot;find&quot; box: ctrl+v. Press the space bar in the &quot;replace&quot; box. Use XWIDE=2
(g) Use FORMAT= to pick up every other column. This is also tricky.
357.
Winsteps problems and error codes
For Windows-related and other problems, see www.winsteps.com/problems.htm
336
Winsteps expects its input files to be in MS-DOS text file format, i.e., readable by NotePad. If the file appears not
to be, this is displayed:
File may not have CR+LF line breaks
The file is not in standard &quot;text&quot; file format. This can be reported for files created on a Mac.
Solution: Open the file with WordPad. Check that the file looks correct. Then &quot;Save as type: Text Document - MSDOS Format&quot;
VARIABLE=VALUE EXPECTED : {\rtf1\....
Your control file is in RTF, not TXT format. Double-click on the file, and &quot;Save as&quot; &quot;text with line breaks&quot; or or
&quot;MS-DOS format&quot;
NUMERIC VALUE EXPECTED: &yuml;&thorn;T I T L E =
Your control file is in Unicode, not ANSI coding. Double-click on the file, and &quot;Save as&quot; with &quot;Encoding&quot; &quot;ANSI&quot; or
&quot;MS-DOS format&quot;
NUMERIC VALUE EXPECTED: &thorn;&yuml; T I T L E =
Your control file is in Unicode BIgendiam, not ANSI coding. Double-click on the file, and &quot;Save as&quot; with
&quot;Encoding&quot; &quot;ANSI&quot; or &quot;MS-DOS format&quot;
NUMERIC VALUE EXPECTED: &iuml;&raquo;&iquest;TITLE =
Your control file is in UTF-8, not ANSI coding. Double-click on the file, and &quot;Save as&quot; with &quot;Encoding&quot; &quot;ANSI&quot; or
&quot;MS-DOS format&quot;
UNRECOGNIZED VARIABLE (CHECK SPELLING): Facets = 3
This can occur when a Facets control file is submitted to Winsteps.
Error codes are reported by Winsteps for which no automatic action is possible. See Winsteps Help.
358.
Winsteps SPSS error codes
If you encounter one of these when running Winsteps and don't understand the reason, please contact
www.winsteps.com. See Winsteps Help
337
Control Variable Index
&amp;END end of control variables (required)...............................................................................................................180
&amp;INST start of control variables (ignored) ..............................................................................................................180
@Fieldname = name for location in person or item label.......................................................................................179
ALPHANUM= alphabetic numbering ........................................................................................................................69
ASCII= output only ASCII characters = Yes .............................................................................................................69
ASYMPTOTE= item upper and lower asymptotes = No ..........................................................................................70
BATCH= Running in batch mode = No.....................................................................................................................71
BYITEM= display graphs for items = Yes.................................................................................................................74
CATREF= reference category for Table 2 = (item measure) ...................................................................................74
CFILE= scored category label file ............................................................................................................................75
CHART= output graphical plots in Tables 10 &amp; 13-15 = Yes ...................................................................................76
CLFILE= codes label file...........................................................................................................................................77
CODES= valid data codes = 01................................................................................................................................78
CONVERGE= select convergence criteria = Either .................................................................................................80
CSV= comma-separated values in output files = No................................................................................................81
CURVES= curves for Tables 2 &amp; 21 = 111, all.........................................................................................................82
CUTHI= cut off responses with high probability of success = 0, no .........................................................................82
CUTLO= cut off responses with low probability of success = 0, no .........................................................................83
DATA= name of data file = (data at end of control file) ............................................................................................83
DELIMITER= data field delimiters = &quot; &quot;, fixed fields .................................................................................................83
DIF= columns within person label for Table 30 = $S1W1 ........................................................................................85
DISCRIMINATION= report item discrimination = No................................................................................................86
DISFILE= category/distractor/option count file .........................................................................................................87
DISTRT= output option counts in Tables 10 &amp; 13-15 = Yes ....................................................................................87
DPF= columns within item label for Table 31 = $S1W1 ...........................................................................................88
EDFILE= edit data file...............................................................................................................................................88
END LABELS end of control variables .....................................................................................................................89
END NAMES end of control variables ......................................................................................................................89
EQFILE= code equivalences ....................................................................................................................................89
EXTRSC= extreme score correction for extreme measures = 0.3...........................................................................89
FITHIGH= higher bar in charts = 0, none .................................................................................................................90
FITI= item misfit criterion = 2.0 .................................................................................................................................91
FITLOW= lower bar in charts = 0, none ...................................................................................................................91
FITP= person misfit criterion = 2.0 ...........................................................................................................................91
FORMAT= reformat data ..........................................................................................................................................91
FORMFD= the form feed character = ^, MS-DOS standard ....................................................................................97
FRANGE= half-range of fit statistics on plots = 0, auto-size ....................................................................................97
G0ZONE= One-to-zero zone = 50 % .....................................................................................................................102
G1ZONE = Zero-to-one zone = 50 %.....................................................................................................................102
GRFILE= probability curve coordinate output file.....................................................................................................98
GROUPS= assigns items to groups = &quot; &quot;, all in one grouping..................................................................................98
GRPFROM= location of GROUPS= = No, before &amp;END.......................................................................................101
GUFILE= Guttmanized response file......................................................................................................................102
HEADER= display or suppress subtable headings = Yes......................................................................................102
HIADJ= correction for top rating scale categories = 0.25.......................................................................................103
HLINES= heading lines in output files = Yes..........................................................................................................103
IAFILE= item anchor file .........................................................................................................................................103
IANCHQU= anchor items interactively = No ..........................................................................................................105
ICORFILE= item residual correlation file ................................................................................................................105
IDELETE= item one-line item deletion ...................................................................................................................106
IDELQU= delete items interactively = No ...............................................................................................................106
IDFILE= item deletion file .......................................................................................................................................107
IDROPEXTREME= drop items with extreme scores = No .....................................................................................109
IFILE= item output file.............................................................................................................................................109
ILFILE= item label file = (after &amp;END) ....................................................................................................................111
IMAP= item label on item maps Tables 1 &amp; 12 = (whole label)..............................................................................111
INUMB= label items by sequence numbers = No ..................................................................................................111
338
IPMATRIX= response-level matrix (from Output Files menu only) .......................................................................112
IREFER= identifying items for recoding (with IVALUE=)........................................................................................114
ISELECT= item selection criterion = *, all items.....................................................................................................115
ISFILE= item structure output file ...........................................................................................................................116
ISGROUPS= assigns items to groups = &quot; &quot;, all in one grouping ..............................................................................98
ISORT= column within item label for alphabetical sort in Table 15 = 1..................................................................118
ISUBTOTAL= columns within item label for subtotals in Table 27.........................................................................119
ITEM= title for item labels = ITEM ..........................................................................................................................120
ITEM1= column number of first response = (required)...........................................................................................120
ITLEN= maximum length of item label = 30 ...........................................................................................................121
IVALUEx= recoding of data (with IREFER=) ..........................................................................................................121
IWEIGHT= item (variable) weighting ......................................................................................................................122
KEYFORM= skeleton for Excel ..............................................................................................................................248
KEYFROM= location of KEYn= = 0, before &amp;END.................................................................................................125
KEYn= scoring key .................................................................................................................................................124
KEYSCR= reassign scoring keys = 123.. ...............................................................................................................125
LCONV= logit change at convergence = .001 logits ..............................................................................................126
LINLEN= length of printed lines in Tables 7 &amp; 10-16 &amp; 22 = 80.............................................................................127
LOCAL= locally restandardize fit statistics = No ....................................................................................................127
LOGFILE= accumulates control files ......................................................................................................................127
LOWADJ= correction for bottom rating scale categories = 0.25 ............................................................................128
MAKEKEY= construct MCQ key = No....................................................................................................................128
MATRIX= correlation ooutput file layout = No ........................................................................................................128
MAXPAGE= the maximum number of lines per page = 0, no limit ........................................................................129
MFORMS= reformat input data and multiple data forms........................................................................................129
MISSCORE= scoring of missing data codes = -1, ignore ......................................................................................134
MJMLE= maximum number of JMLE (UCON) iterations = 0, no limit....................................................................135
MNSQ= show mean-square instead of t-standardized fit statistics = Yes ............................................................135
MODELS= assigns model types to items = R, dichotomy, rating scale or partial credit ........................................136
MODFROM= location of MODELS= = N, before &amp;END........................................................................................137
MPROX= maximum number of PROX iterations = 10 ...........................................................................................138
MRANGE= half-range of measures on plots = 0, auto-size ..................................................................................138
MUCON= maximum number of JMLE (UCON) iterations = 0, no limit ..................................................................135
NAME1= first column of person label = 1 ...............................................................................................................139
NAMLEN= length of person label = (calculated) ....................................................................................................139
NAMLMP= name length on map for Tables 1 &amp; 12 &amp; 16 = (calculated) ................................................................140
NEWSCORE= recoding values (with RESCORE=) ...............................................................................................140
NI= number of items (required)..............................................................................................................................141
NORMAL= normal distribution for standardizing fit = No, chi-square ....................................................................142
OSORT= category/option/distractor sort = S, score value order............................................................................142
OUTFIT= sort misfits on greater of infit or outfit = Yes...........................................................................................142
PAFILE= person anchor file....................................................................................................................................143
PAIRED= correction for paired comparison data = No...........................................................................................144
PANCHQU= anchor persons interactively = No .....................................................................................................144
PCORFIL= person residual correlation file.............................................................................................................144
PDELETE= person one-line item deletion..............................................................................................................145
PDELQU= delete persons interactively = No .........................................................................................................145
PDFILE= person deletion file..................................................................................................................................146
PDROPEXTREME= drop persons with extreme scores = No ...............................................................................147
PERSON= title for person labels = PERSON.........................................................................................................147
PFILE= person output file .......................................................................................................................................147
PMAP= person label on person map: Tables 1 &amp; 16 = (whole name) ..................................................................149
PRCOMP= residual type for Tables 23-24 = S, standardized...............................................................................149
PSELECT= person selection criterion = *, all persons ...........................................................................................149
PSORT= column within person label for alphabetical sort in Table 19 = 1 ............................................................150
PSUBTOTAL= columns within person label for subtotals in Table 28 ..................................................................151
PTBIS= point-biserial (instead of point-measure) correlation coefficients = No.....................................................152
PVALUE= proportion correct or average rating = No .............................................................................................153
PWEIGHT= person (case) weighting .....................................................................................................................153
339
QUOTED= quote marks around labels = Yes ........................................................................................................154
RCONV= score residual at convergence = 0.1 ......................................................................................................155
REALSE= inflate S.E. of measures for misfit = No.................................................................................................155
RESCORE= response recoding (with NEWSCORE= or KEYn=) ..........................................................................155
RESFROM= location of RESCORE= No, before &amp;END ........................................................................................157
RFILE= scored response file ..................................................................................................................................157
SAFILE= item structure anchor file.........................................................................................................................158
SAITEM= item numbers in SFILE= and SAFILE= (with one ISGROUPS=) = No..................................................163
SANCHQ= anchor structures interactively = No ....................................................................................................163
SCOREFILE= person score file..............................................................................................................................164
SDELQU= delete structure interactively = No ........................................................................................................165
SDFILE= item structure deletion file .......................................................................................................................165
SEPARATOR= data field delimiters = &quot; &quot;, fixed fields ..............................................................................................83
SFILE= structure output file ....................................................................................................................................166
SIFILE= simulated data file.....................................................................................................................................167
SPFILE= supplementary control file .......................................................................................................................168
STBIAS= correct for JMLE statistical estimation bias = No ...................................................................................168
STEPT3= include structure summary in Table 3 (instead of Table 21) = Yes .......................................................170
STKEEP= keep non-observed intermediate categories in structure = Yes............................................................169
T1I#= number of items summarized by &quot;#&quot; symbol in Table 1 = (auto-size).........................................................170
T1P#= number of persons summarized by &quot;#&quot; symbol in Table 1 = (auto-size) ....................................................170
TABLES= output tables ..........................................................................................................................................170
TARGET= estimate using information-weighting = No...........................................................................................171
TFILE= list of Tables to be output ..........................................................................................................................171
TITLE= title for output listing = (control file name)..................................................................................................172
TOTALSCORE= show total scores with extreme observations = No ....................................................................173
UANCHOR= anchor values supplied in user-scaled units = Yes ...........................................................................173
UASCALE= the anchor scale value of 1 logit = 1..................................................................................................174
UCOUNT = number of unexpected responses: Tables 6 &amp; 10 = 50 ......................................................................174
UDECIMALS= number of decimal places reported = 2..........................................................................................174
UIMEAN= the mean or center of the item difficulties = 0 .......................................................................................174
UMEAN= the mean or center of the item difficulties = 0 ........................................................................................174
UPMEAN = the mean or center of the person abilities = (computed) ....................................................................175
USCALE= the user-scaled value of 1 logit = 1 .......................................................................................................175
W300= Output files in Winsteps 3.00 format = No .................................................................................................176
WHEXACT= Wilson-Hilferty exact normalization = Yes.........................................................................................176
XFILE= analyzed response file...............................................................................................................................177
XMLE= consistent - almost unbiased - estimation = No.........................................................................................178
XWIDE= columns per response = 1 .......................................................................................................................179
340
">
Hi, I'm the chat assistant. I've read the WINSTEPS/MINISTEP User's Guide and am ready to answer your questions about the programs' features, Rasch analysis, data input, output interpretation, and statistical methods.
Show related prompts
Assistant typing
Assistant is out of keyboard. Please, try again later.
Key features

Rasch measurement analysis
Rating scale support
Dichotomous data analysis
Item and person anchoring
Differential item functioning (DIF)
Principal components analysis
Frequently asked questions

WINSTEPS is used for Rasch measurement analysis, which is a statistical approach for analyzing categorical data such as test responses or survey ratings.
WINSTEPS can analyze rating scale data, dichotomous data, partial credit items, and paired comparisons.
WINSTEPS produces tables such as distribution maps, fit plots, item statistics, option frequencies, and differential item functioning (DIF) analysis.
What is WINSTEPS used for?

What kind of data can WINSTEPS analyze?

What output tables does WINSTEPS produce?

Related

Winsteps MINISTEP Rasch-Model User Guide | AI Chat & PDF

Tradency Mirror Trader 用户指南 | AI Chat & PDF 下载

Cito TiaPlus Users Manual | AI Chat & PDF Download

ITEMAN Item Analysis Program User manual: AI Chat & PDF

Casio fx-82AU PLUS Calculator User Manual | AI Chat

Rasch Model Application Guide | AI Chat & PDF Download

RAND Panel Program: AI Chat & PDF Access

openXdata mForms 2.4 User Manual: AI Chat & PDF Access

Burn-P3 User Manual: AI Chat & PDF Download

Burn-P3 Software User Manual: AI Chat & PDF Access

RAND Longitudinal Scalogram Analysis Program User's Guide | AI Chat & PDF Download

IEA ICCS 2009 International Database User Guide + AI Chat & PDF Download
