Assessment Systems Corporation DFIT8 Statistical Software User's Manual

DFIT8 for Windows
User’s Manual
Differential Functioning of Items and
Tests
T. C. Oshima, Steve Kushubar,
John C. Scott, and Nambury S. Raju
DFIT8 for Windows – User’s Manual
System Requirements
Microsoft Windows 2000/XP/Vista
7 Mb free disk space
2 Mb free RAM
For Further Information
Assessment Systems Corporation
2233 University Avenue, Suite 200
St. Paul, Minnesota 55114, U.S.A.
Voice: 651/647-9220
Fax: 651/647-0412
E-Mail: [email protected]
World Wide Web page: www.assess.com
User’s Manual for DFIT8: Differential Functioning of Items and Tests
First Edition, Windows Version (DFIT8.04)
DFIT8 for Windows – User’s Manual
Your DFIT8 License
Unless you have purchased multiple licenses for DFIT8, your license is a single-user license.
You may install DFIT8 on up to two computers (e.g., a desktop and a laptop) so long as there is
no possibility that the two copies will be used simultaneously.
Technical Assistance
If you need technical assistance using DFIT8, please visit the Support section of our Web site,
www.assess.com. If the answer to your question is not posted, please email us at
[email protected]. Technical assistance for DFIT8 is provided for one year from the date you
purchase your license. Please provide us with the invoice number for your license purchase when
you request technical assistance.
Citation
Oshima, T. C., Kushubar, S., Scott, J.C. & Raju N.S. (2009). DFIT8 for Window User’s
Manual: Differential functioning of items and tests. St. Paul MN: Assessment Systems
Corporation.
DFIT8 for Windows – User’s Manual
Contents
1 Introduction .............................................................................................................1
Item Response Theory Models ................................................................................................................. 1
Differential Functioning of Items and Tests ............................................................................................. 3
The DFIT Framework for DIF and DTF ................................................................................................... 4
Significance Tests in DFIT ....................................................................................................................... 6
References ................................................................................................................................................. 7
2 Formatting Input Data for Analysis ........................................................................8
Dichotomous IRT Calibration................................................................................................................... 8
Polytomous IRT Calibration ..................................................................................................................... 8
Linking ...................................................................................................................................................... 9
References ............................................................................................................................................... 10
3 Definition File .......................................................................................................11
Running DFIT ......................................................................................................................................... 11
Dichotomous Run Definition File ........................................................................................................... 11
Polytomous Graded Response Model Definition File ............................................................................ 14
4 Output File ............................................................................................................17
Viewing Output....................................................................................................................................... 17
Content of the Output File ...................................................................................................................... 17
Appendix A. Examples ............................................................................................18
1. A dichotomous example (Example 1)................................................................................................. 18
2. A polytomous example (Example 2) .................................................................................................. 25
Appendix B. Troubleshooting ..................................................................................33
DFIT8 for Windows – User’s Manual
1
Introduction
Nambury S. Raju (1937 – 2005) developed a framework for assessing the differential functioning
of items and tests (DFIT) based on item response theory (IRT). In DFIT, differential item
functioning (DIF) and differential test functioning (DTF) are assessed by using item and score
parameter estimates obtained from an IRT calibration. The framework can be applied to test data
that are either dichotomous (Raju, van der Linden, & Fleer, 1995) or polytomous (Flowers,
Oshima, & Raju, 1999). Although both unidimensional and multidimensional DFIT models
have been developed (Oshima, Raju, & Flowers, 1997), this program implements only the
unidimensional models.
Item Response Theory Models
For dichotomously scored items, the three-parameter logistic model describes the probability of
an individual answering item i correctly as a function of ability (θ) as follows.
Pi (θ s ) = ci + (1 − ci )
exp  Dai (θ s − bi ) 
1 + exp  Dai (θ s − bi ) 
.
(1)
The item characteristics are represented by the ai, bi, and ci parameters. The location parameter,
bi reflects the difficulty of the item. The discrimination parameter, ai, relates to the steepness of
the curve. The ci parameter reflects the lower asymptote, or the probability that a person with
extremely low θ would correctly answer the item. D is a scaling constant typically set at 1.702.
Polytomous IRT models require the estimation of multiple item response functions (IRF)
representing the different response categories. Although various polytomous models have been
developed in the IRT literature, the current version of DFIT8 uses Samejima’s (1969) graded
Page 1
DFIT8 for Windows – User’s Manual
response model (GRM) which is designed for ordered response categories. Theoretically,
however, DFIT can be applied to different types of polytomous models.
According to GRM, for an item with m response categories, there will be m-1 boundary response
functions (BRF). A BRF represents the probability of person s responding above response
category k on item i,
Pik* (θ s ) =
exp  Dai (θ s − bik ) 
1 + exp  Dai (θ s − bik ) 
,
(2)
where bik is a location parameter that designates the boundary between response categories k and
k+1, and ai is the item discrimination parameter.
The probability of responding in a particular response category can be computed from the
difference between adjacent BRFs. This function is referred to as the category response function
(CRF):
Pik (θ ) = Pi*( k −1) (θ ) − Pik* (θ ) .
(3)
Because the first and last response categories lack an adjacent boundary, Samejima (1969)
defined Pi*0 (θ ) = 1, and Pim* (θ ) = 0. There will be as many CRFs for an item as there are response
categories.
The expected score of individual s on item i, ESsi(θs), can be defined as a weighted average of the
category values, where the weights reflect the probability of the individual selecting each
category (i.e, the CRFs),
m
ES i (θ s ) = ∑ Pik (θ s ) X ik ,
(4)
k =1
where Xik is the value assigned to category k on item i. For a dichotomously scored item, the
expected score is equal to the probability of answering an item correctly (Equation 1).
Page 2
DFIT8 for Windows – User’s Manual
The total score on a test can be defined as the sum of the scores on the individual items. This
total test score can also be modeled as a function of θ and the resulting curve is called the test
response function (TRF). The TRF is defined as sum of expected score functions across n items,
n
T (θ s ) = ∑ ES i (θ s ) .
(5)
i =1
Differential Functioning of Items and Tests
The DFIT analysis starts with two sets of item parameter estimates (one from a reference group
and another from a focal group) which are placed on the scale of the focal group along with the
focal group’s θ estimates. According to IRT, the item response functions (IRFs; also called item
characteristic curves or item characteristic functions) are invariant over subgroups of examinees.
Therefore, DIF is conceptualized by measuring the differences between the two IRFs .
Figure 1 graphically depicts the differences between a reference group IRF and a focal group
IRF in the dichotomous model, by plotting probability (Y axis) against θ (X axis). The curve
plotted is the IRF. According to IRT, the two curves should be invariant after the item parameter
estimates from each group are placed on the common scale. Thus, any gap between the two
IRFs indicates DIF. The gap can be measured in various ways. For example Raju (1988)
developed a DIF index based on the area of the gap. In the DFIT framework, on the other hand,
the gap is measured by the average squared distance of the probability difference at θ levels
based on the focal group. This approach, compared to the area approach, offers substantial
advantages as it can be easily expanded to different models (unidimensional or multidimensional
models) in various scoring formats (dichotomous or polytomous data). The DFIT approach also
makes several applications possible that are useful in DIF research (e.g., DTF, differential bundle
functioning (Oshima, Raju, Flower, & Slinde, 1998)).
Page 3
DFIT8 for Windows – User’s Manual
Figure 1. Two IRFs/ICFs
The DFIT Framework for DIF and DTF
The DFIT framework offers two types of DIF, non-compensatory DIF (NCDIF) and
compensatory DIF (CDIF). NCDIF is defined as
2
NCDIFi = EF  di (θ )  ,


(6)
where
d i (θ ) = PiF (θ ) − PiR (θ )
(7)
for the dichotomous model (see Figure 1), and
d i (θ ) = ES iF (θ ) − ES iR (θ )
(8)
for the polytomous model. EF denotes the expectation taken over the θ distribution from the focal
group.
Page 4
DFIT8 for Windows – User’s Manual
DTF is similarly defined but at the test level. As shown in Figure 2, the two curves compared
are the TRFs and the distance is defined as,
D(θ ) = TF (θ ) − TR (θ ) .
(9)
DTF is then defined as the expected value of the squared difference between focal and reference
groups, where the expectation is taken across the θ distribution from the focal group,
2
DTF = EF  D (θ )  .


(10)
Figure 2. Two TRFs/TCFs
Despite the mathematical similarity between NCDIF and DTF, the relationship between the two
is not straightforward as NCDIF does not take the item covariances into account. CDIF, on the
other hand, has a straightforward relationship with DTF,
n
DTF = ∑ CDIFi .
(11)
i =1
CDIF is defined as
CDIFi = E F (d i D ) = Cov(d i , D ) + µ di µ D ,
(12)
where Cov stands for covariance.
Page 5
DFIT8 for Windows – User’s Manual
NCDIF and CDIF both relate to DIF, but address different issues. Like most item-level DIF
indices, NCDIF assumes that all items other than the studied item are DIF free, and can be used
like any other DIF indices. CDIF, on the other hand, takes into account the pattern of DIF across
items, and can be used to investigate the effect of removing the DIF items on DTF.
Significance Tests in DFIT
DFIT8 employs a recently developed significance test for NCDIF and DTF. The test is called the
item parameter replication (IPR) method (Oshima, Raju & Nanda, 2006; Raju, Fortmann, Kim,
Morris, Nering, & Oshima, in press) and provides a means of deriving cutoff values that are
tailored to a particular data set. The IPR method begins with estimates of item parameters for the
focal group and the sampling variances and covariances of these estimates. Based on these initial
estimates, a large number of replications (typically 1,000 pairs) of item parameters are simulated.
Then the cutoff value for each alpha level is determined by the empirical sampling distribution of
NCDIF/DTF obtained under the null hypothesis that focal and reference groups have identical
parameters. The IPR method produces cutoff values for each item.
The significance of CDIF is not tested directly. Instead, items with large CDIF are removed one
by one until DTF reaches non-significance by a chi-square test. Those removed CDIF items are
then considered significant.
Page 6
DFIT8 for Windows – User’s Manual
References
More detailed information on DFIT can be found in the following references. Researchers new
to DFIT might find the instructional module developed for the National Council on Measurement
in Education (Oshima & Morris, 2008) useful.
Flowers, C. P., Oshima, T. C., & Raju, N. S. (1999). A description and demonstration of the
polytomous-DFIT framework. Applied Psychological Measurement, 23, 309-326.
Oshima, T. C., & Morris, S. B. (2008). An NCME Instructional Module on Raju’s Differential
Functioning of Items and Tests (DFIT). Educational Measurement: Issues and Practice, 27, 4350.
Oshima, T. C., Raju, N. S., & Flowers, C. P. (1997). Development and demonstration of
multidimensional IRT-based internal measures of differential functioning of items and tests.
Journal of Educational Measurement, 34, 253-272.
Oshima, T. C., Raju, N. S., Flowers, C. P., & Slinde, J. (1998). Differential bundle functioning
(DBF) using the DFIT framework: Procedures for identifying possible sources of differential
functioning. Applied Measurement in Education, 11, 353-369.
Oshima, T. C., Raju, N. S., & Nanda (2006). A new method for assessing the statistical
significance in the differential functioning of items and tests (DFIT) framework. Journal of
Educational Measurement, 43, 1-17.
Raju, N. S. (1988). The area between two item characteristic curves. Psychometrika, 53, 495-502.
Raju, N. S., K. A. Fortmann, Kim, W., Morris, S. B., Nering, M., & Oshima, T. C. (in press).
The item parameter replication method for detecting differential functioning in the DFIT
framework. Applied Psychological Measurement.
Raju, N. S., van der Linden, W. J., & Fleer, P. F. (1995). An IRT-based internal measure of test
bias with applications for differential item functioning. Applied Psychological Measurement, 19,
353-368.
Samejima, F. (1969). Estimation of latent ability using a response pattern of graded scores.
Psychometrika Monograph, No. 17.
Page 7
DFIT8 for Windows – User’s Manual
2
Formatting Input Data for Analysis
Dichotomous IRT Calibration
Any dichotomous IRT calibration program can be used to obtain item parameter estimates, their
variance and covariance for each group (focal or reference). BILOG-MG3 (Zimowski, Muraki,
Mislevy, & Bock, 2002), for example, provides an estimated covariance file (*.cov) which
contains item parameter estimates, error variances and covariances. The *.cov file needs to be
rearranged to be readily read into DFIT8. The newly created file should contain the following
information in the specific order for each item for the focal group: b, a, c, V(b), V(a), V(c),
Cov(b,a), Cov(b,c) and Cov(a,c), where V is the variance and Cov is the covariance. BILOGMG3 also provides the ability file (*.sco) which contains the θ estimates. DFIT depends on the
accurate estimation of item parameters, and accurate parameter estimation typically requires
large samples of examinees. Therefore, DFIT is not recommended for application with small
samples. For dichotomous IRT models, the required sample size for each group will depend on
the number of parameters estimated: Conservative sample sizes (N) are N > 200 for the oneparameter model, N > 500 for the two-parameter model, and N > 1,000 for the three-parameter
model (Crocker & Algina, 1986).
Polytomous IRT Calibration
Any polytomous IRT calibration program can be used to obtain item parameter and variance
estimates. The covariance information may not be readily available, however. For example,
PARSCALE (Muraki & Bock, 2003) does not provide the covariance information. A Fortran
program (Polycov; Morris 2007) is available to calculate the item variances and covariances
from Parscale output and to create the input data file for DFIT8. It is not recommended to
assume that the covariances are equal to 0. The input data file to DFIT8 has to provide focal
group information in a specific order. Using a 5-category item as an example, it will be a, b1, b2,
Page 8
DFIT8 for Windows – User’s Manual
b3, b4, V(a), V(b1), V(b2), V(b3), V(b4), Cov(a,b1), Cov(a,b2), Cov(a,b3), Cov(a,b4), Cov(b1,b2),
Cov(b1,b3), Cov(b1,b4), Cov(b2,b3), Cov(b2,b4), and Cov(b3,b4) in that order for each item for the
focal group.
As with the dichotomous case, large sample sizes are required for polytomous DFIT. A sample
size recommendation for accurate estimation for the polytomous models should be followed (e.g.,
N > 500 is recommended by Reise & Yu, 1990).
Linking
In the DFIT framework, the item parameter estimates from the reference group are placed on the
scale of the focal group. Linking coefficients (a multiplicative coefficient and an additive
coefficient) can be obtained by using linking programs such as IPLINK (Lee & Oshima, 1996) or
PIE for dichotomous models, IRTEQ (Han, 2007) for either dichotomous or polytomous models,
or POLYEQUATE for polytomous models. Iterative linking, or two-stage linking is
recommended. Iterative linking is a process by which only items that are considered to be DIFfree are used as linking items. The steps for conducting two-stage linking are as follows:
1. Obtain linking coefficients using all items (the first stage),
2. Run DFIT to identify large DIF items,
3. Remove those large DIF items from the test,
4. Obtain revised linking coefficients (the second stage), and finally
5. Run DFIT for all items again using those revised linking coefficients.
Page 9
DFIT8 for Windows – User’s Manual
References
Crocker, L. & Algina, J. (1986). Introduction to classical and modern test theory. New York:
Holt, Rinehart & Winston.
Han, K. T. (2007). IRTEQ: Windows application that implements IRT scaling and equating
[computer program]. Amherst, MA: University of Massachusetts Amherst, Center for
Educational Assessment. Available for download at http://www.umass.edu/remp/software/irteq/
Lee, K., & Oshima, T. C. (1996). IPLINK: Multidimensional and unidimensional item parameter
linking in item response theory. Applied Psychological Measurement, 20, 230. Available for
download at http://education.gsu.edu/eps/4493.html
Morris, B. S. (2008). Polycov [Computer program]. Chicago: Illinois Institute of Technology.
Available for download at http://www.iit.edu/~morris/polycov/polycov.html
Muraki, E., & Bock, R. D. (2003). PARSCALE 4: IRT Item Analysis and Test Scoring for
Rating-Scale Data. Scientific Software: Chicago, IL.
Reise, S. P. & Yu, J. (1990). Parameter recovery in the graded response model using
MULTILOG. Journal of Educational Measurement, 27, 133-144.
Zimowski, M. F., Muraki, E., Mislevy, R. J., & Bock, R. D. (2002). BILOG-MG3 [Computer
software]. St.Paul, MN: Assessment Systems Corporation.
Page 10
DFIT8 for Windows – User’s Manual
3
Definition File
Running DFIT
In order to perform a DFIT analysis, the user must create a DFIT run definition file (*.dft). The
run definition file provides all the information necessary to read the necessary data files, perform
the analysis and generate the output file.
Definition files are created by using the File
New menu option.
The user can create the definition file by using the Normal View or Raw View (choices under the
View pull-down menu). In the Normal View, the Windows frames assist the user to fill out the
necessary information. In the Raw View, the user can type the syntax directly in text mode.
After the definition file is saved, the user can run the program by clicking Run in the pull-down
menu.
Dichotomous Run Definition File
Here is a sample dichotomous run definition file:
Each line in the dichotomous run definition file is described below:
Page 11
DFIT8 for Windows – User’s Manual
Line 1
Any title up to 80 characters long.
Line 2
The first number represents the number of items (I5 format – i.e., an integer field five characters
wide). The maximum number of Items allowed is 100.
The second number represents the IRT model. This number should be 1, 2, or 3 (I5 format).
The third number represents the number of examinees or subjects in the focal group. The
maximum number of examinees allowed is 5,000 (I5 format).
The fourth number is the number of replications for generating (using the IPR method) itemlevel and test/scale-level cut-off scores for assessing the significance of NCDIF and DTF indices
(I5 format). The maximum is 99999.
The fifth number is the number for generating random item parameters in the Monte Carlo
procedure and it MUST be a negative integer (I5 format). If the Windows interface is used to
make the present file, the number for generating random item parameters is created by the
program.
The sixth number is D (F8.4 format. i.e., a 8-character field with a decimal point and four
significant digits after the decimal point), the IRT constant (see Equations 1 and 2). It must be
set 1.0 if the IRT model is logistic or 1.7 if the IRT model is normal ogive.
The seventh and eighth numbers are the multiplicative and additive constants (F8.4 format for
both). These are used for placing the reference group item parameters on the same metric as the
focal group item parameters. In the case of the 1-paramter logistic or Rasch Model, the
multiplicative constant must be set to 1.0. If the focal and reference group parameters are
already on a common metric, the multiplicative and additive constants must be 1.0 and 0.0
respectively.
Page 12
DFIT8 for Windows – User’s Manual
Line 3
A ‘1’ tells the program to include the item in the DIF/DTF analysis. A ‘0’ excludes the item
from the analysis. For example, if there are 30 items in a test and you want to exclude Item 5
from the DIF/DTF analysis, then Line 3 should be as follows:
111101111111111111111111111111
If you want to include all 30 items in the DIF/DTF analysis, Line 3 should read as follows:
111111111111111111111111111111
Line 4
The format for reading the item parameters and the variance-covariance matrix as a vector [b, a,
c, V(b), V(a), V(c), Cov(b,a), Cov(b,c) and Cov(a,c) in that order] for each item in the focal
group. The specific format could vary from one application to the next. In the example above,
9F8.4 means nine fields of 8 characters each, including a decimal point and four significant
digits after the decimal point. Formats must be enclosed in parentheses.
Line 5
The format for reading item parameters as a vector (b, a, and c in that order) for each item in the
reference group. Information about the variances and covariances is not needed for the reference
group. The specific item format could vary from one application to the next.
Line 6
The format for reading the θ values for the focal group. The specific θ format could vary from
one application to the next. In the example above. 12x means skip the first 12 columns and
begin reading in column 13.
Line 7
The file path to the focal group item data file. If no path is specified (i.e., just a file name, as in
the above example) , the program will look for the files in the same folder in which the program
exists
Page 13
DFIT8 for Windows – User’s Manual
Line 8
The file path to the reference group item data file.
Line 9
The file path to the θ values file.
Line 10
The file path to the output file.
Polytomous Graded Response Model Definition File
Here is a sample polytomous run definition file:
Line 1
Any title up to 80 characters long
Line 2
The first number represents the number of items (I5 format). The maximum number of items
allowed is 100.
The second number represents the number of examinees or subjects in the focal group. The
maximum number of examinees allowed is 5,000 (I5 format).
Page 14
DFIT8 for Windows – User’s Manual
The third number is the number of replications for generating (via Monte Carlo procedures)
item-level and test/scale-level cut-off scores for assessing the significance of NCDIF and DTF
indices (I5 format).
The fourth number is the number for generating random item parameters in the Monte Carlo
procedure and it MUST be a negative integer (I5 format). If the Windows interface is used to
make the present file, the number for generating random item parameters is created by the
program.
The fifth number is D (F8.4 format). It must be set to 1.0 if the IRT model is logistic or 1.7 if the
IRT model is normal ogive.
The sixth and seventh numbers are the multiplicative and additive constants (F8.4 format for
both). These are used for placing the reference group item parameters on the same scale as the
focal group item parameters. In the case of one-paramter logistic or Rasch model, the
multiplicative constant must be set to 1.0. If the focal and reference group parameters are
already on a common metric, the multiplicative and additive constants must be 1.0 and 0.0,
respectively.
Line 3
A ‘1’ tells the program to include the item in the DIF/DTF analysis. A ‘0’ excludes the item
from the analysis. For example, if there 25 items in a test and you want to exclude Item 5 from
the DIF/DTF analysis, then Line 3 should be as follows:
1111011111111111111111111
If all 25 items are to be included in the DIF/DTF analysis, Line 3 should read as follows:
1111111111111111111111111
Line 4
Each number represents the number of categories for a given item. The maximum number of
categories allowed is 9. If you have a scale with 25 items with 5 categories per item, Line 4
should read:
5555555555555555555555555
Page 15
DFIT8 for Windows – User’s Manual
Line 5
The format for reading item parameters and the variance-covariance matrix as a vector [For a 5category item, it will be a, b1, b2, b3, b4, V(a), V(b1), V(b2), V(b3), V(b4), Cov(a,b1), Cov(a,b2),
Cov(a,b3), Cov(a,b4), Cov(b1,b2), Cov(b1,b3), Cov(b1,b4), Cov(b2,b3), Cov(b2,b4), and Cov(b3,b4)
in that order] for each item in the focal group. The specific format could vary from one
application to the next.
Line 6
The format for reading item parameters and their variances as a vector [for an item with 5
categories: a, b1, b2, b3, b4, V(a), V(b1), V(b2), V(b3) and V(b4) in that order] for each item in the
Reference Group. The specific format could vary from one application to the next.
Line 7
The format for reading the θ values for the focal group. The specific θ format could vary from
one application to the next.
Line 8
The file path to the focal group item data file.
Line 9
The file path to the reference group item data file.
Line 10
The file path to the θ values file.
Line 11
The file path to the output file.
Line 12 and beyond: Each row indicates the initial score values associated with the categories for
a given item. For example, if you have 25 items in a scale, you need 25 lines (Lines 12-36).
Page 16
DFIT8 for Windows – User’s Manual
4
Output File
Viewing Output
The output file is created after a successful run of the definition file and presented on the screen.
It can be also viewed later using File
Open in the program, or using any other programs that
display text data.
Content of the Output File
Items 1 – 5 listed below are input information to verify that the input files are read into DFIT8
correctly. Items 6 – 10 present the results from the DFIT analysis. See Appendix A for examples.
1. Title
2. Input variables
3. Focal group item parameters
4. Reference group (unequated) item parameters
5. Reference group (equated) item parameters
6. Monte Carlo generated item-level and test-level cut-offs (from the IPR method). Cutoff values
for alpha = .001, .005, .01, 05, .10, and .50 are provided. DTF cutoff values are listed at the
bottom.
7. Main output where CDIF, NCDIF, and NCDIF significance (ns indicates not significant) are
reported. DIF category is for future use and it is blank.
8. Summary statistics for true scores
9. DTF related statistics, including the chi-square test
10. CDIF items to be removed to achieve non-significant DTF.
Page 17
DFIT8 for Windows – User’s Manual
Appendix A.
Examples
1. A dichotomous example (Example 1)
This is a 40-item simulated dichotomous test with 1,000 examinees in each group (reference and
focal). DIF was embedded in Items 5, 10, 15, and 20. Items were calibrated separately for each
group using BILOG-MG3 (2-parameter model). The item parameter estimates and their variance
covariance estimates are listed in EX1R.PAR and EX1F.PAR for the reference group and focal
group, respectively. In each file, there are five columns (b, a, V(b), V(a), and Cov(b,a) in that
order). The linking coefficients were calculated using IPLINK (1.0240 and −0.0180 for the
multiplicative and additive coefficients, respectively, after the second-stage linking). In the
score file (EX1F.THT), the third column contains the θ estimates. For the IPR method, 1,000
replications were used to conduct the significance test.
Definition File in the Normal View
Page 18
DFIT8 for Windows – User’s Manual
Definition File in the Raw View
Page 19
DFIT8 for Windows – User’s Manual
EX1F.PAR
EX1R.PAR
EX1F.THT
Page 20
DFIT8 for Windows – User’s Manual
EX1.TXT (Selected output)
1. Title
2. Input variables
3. Focal group item parameters
4. Reference group (unequated) item parameters
Page 21
DFIT8 for Windows – User’s Manual
5. Reference group (equated) item parameters
6. Monte Carlo generated item-level and test-level cut-offs (the IPR method)
Page 22
DFIT8 for Windows – User’s Manual
7. Main output where CDIF, NCDIF, and NCDIF significance (ns indicates not-significant) are
reported. DIF category is for future use and it is left blank.
8. Summary statistics for true scores
Page 23
DFIT8 for Windows – User’s Manual
9. DTF related statistics including the chi-square test
10. CDIF items to be removed to achieve non-significant DTF.
Results indicate that Items 5, 10, 15, and 20 show NCDIF at the .001 level. Item 13 shows
NCDIF at the .05 level. The DTF of .51 is significant at the .001 level. After removing two
significant CDIF items (Items 15 and 5), DTF is no longer significant.
Page 24
DFIT8 for Windows – User’s Manual
2. A polytomous example (Example 2)
This is a 20-item simulated polytomous test (5-category responses scored 0, 1, 2, 3, 4) with 1,000
examinees in each group (reference and focal). DIF was embedded in Items 3, 8, 13, and 18.
Items were calibrated separately for each group using PARSCALE (Muraki & Bock, 2003). The
item parameter estimates and their variance covariance estimates are listed in EX2R.PAR and
EX2F.PAR for the reference group and focal group, respectively. Those *.PAR files were
created by POLYCOV which read the PARSCALE parameter files, added covariances, and
arranged the information as required for DFIT input. In each file, there are 20 variables [a, b1, b2,
b3, b4, V(a), V(b1), V(b2), V(b3), V(b4), Cov(a,b1), Cov(a,b2), Cov(a,b3), Cov(a,b4), Cov(b1,b2),
Cov(b1,b3), Cov(b1,b4), Cov(b2,b3), Cov(b2,b4), and Cov(b3,b4) in that order]. The linking
coefficients were calculated using EQUATE (1.0335 and 0.0492 for the multiplicative and
additive coefficients, respectively, after the second-stage linking). In the score file (EX2F.SCO),
the θ can be found at (/58x,F10.4). For the IPR method, 1000 replications were used to conduct
the significance test.
Definition File in the Normal View
Page 25
DFIT8 for Windows – User’s Manual
Page 26
DFIT8 for Windows – User’s Manual
Definition File in the Raw View
Page 27
DFIT8 for Windows – User’s Manual
EX2F.PAR
EX2R.PAR
EX2F.SCO
Page 28
DFIT8 for Windows – User’s Manual
EX2.TXT (Selected output)
1. Title
2. Input variables
3. Focal group item parameters
Page 29
DFIT8 for Windows – User’s Manual
4. Reference group (unequated) item parameters
5. Reference group (equated) item parameters
Page 30
DFIT8 for Windows – User’s Manual
6. Monte Carlo generated item-level and test-level cut-offs (the IPR method)
7. Main output where CDIF, NCDIF, and NCDIF significance (ns indicates not-significant) are
reported. DIF category is for future use and it is left blank.
Page 31
DFIT8 for Windows – User’s Manual
8. Summary statistics for true scores
9. DTF related statistics including the chi-square test
10. CDF items to be removed to achieve non-significant DTF.
Results indicate that Items 3, 8, 13, and 15 show NCDIF at the .001 level. Item 14 shows
NCDIF at the .01 level. The DTF of .92 is significant at the .001 level. After removing two
significant CDIF items (Items 3 and 8), DTF is no longer significant.
Page 32
DFIT8 for Windows – User’s Manual
Appendix B.
Troubleshooting
Most problems that occur when running DFIT are a result of incorrect or poorly formatted data.
If you open the file in Normal View, when you start the run DFIT will perform validation of the
definition file.
The most likely error is an end of file error. The figure below shows a typical end of file error.
This can be caused by several different errors in the data files:
1. Insufficient data in the data file
2. Incorrect format specification in the run definition file
3. Incorrect number of items or examinees specified on line 2 of the run definition file
Page 33
">