Microdata User Guide National Longitudinal Survey of Children and

Microdata User Guide National Longitudinal Survey of Children and
Microdata User Guide
National Longitudinal Survey of Children and
Youth
Cycle 3
September 1998 to June 1999
Table of Contents
Chapter 1 - Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
General Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter 2 – Survey Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Cycle 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Chapter 3 – Response Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Chapter 4 - Data Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Computer Assisted Interviewing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Household Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
School Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Interview Training, Supervision and Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
11
15
20
22
Chapter 5 - Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Editing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
Data Capture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Minimum Completion Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
Head Office Editing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Consistency Editing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
Naming Convention and Coding Structure for NLSCY Variables . . . . . . . . . . . . . . 32
Acronym Names for Questionnaire Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
Examples of Variables Names . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Coding Structure for NLSCY Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Coding of Open-ended Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Naming Imputation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Derived Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Chapter 6 - Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
The Longitudinal Sample or Cross-sectional? . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Weighting Procedures for the Cross-sectional and Longitudinal Samples . . . . . .
Cross-Sectional Weights for Children Surveyed for the First Time in Cycle 3 . . .
Weighting of Children Sampled in 1994 and 1996 . . . . . . . . . . . . . . . . . . . . . . . . .
43
43
44
45
48
50
Chapter 7 - NLSCY Concepts and Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cross-Sectional and Longitudinal estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
NLSCY Units of Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PMK and Spouse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Family Derived Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Socio-Economic Derived Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
53
55
56
59
61
Chapter 8 - Content and Validation of NLSCY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Validation of Scale Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Factor Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Calculation of Scores and Item Imputation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
Reliability Measures for Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Parent-Reported Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Temperament Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Education (Child) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Behaviour Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Motor and Social Development . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
Relationships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Parenting Scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Analysis of NLSCY Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Results (Cycle 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
Parenting Scales: 12-15 Year Olds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Analysis of NLSCY Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Depression Scale (PMK) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .100
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Family Functioning Scale (Parent) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Activities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
My Parents and Me Scale (BPRCbS07 and BPRCbS08) - Parent . . . . . . . . . . . 105
Child Scales from Self-completed Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . 108
Feelings and Behaviour (self-complete, 10-15) . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
My Parents and Me (self-complete 10-15) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
About me (self-complete 10-15) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
Depression Scale (self-complete 12-15) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Education (Parent) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Socio-demographic Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Labour Force (Parent) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
Work Duration Derived Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Demographic Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
Medical/Biological . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Chapter 9 - Data Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Cross-sectional and longitudinal response rates . . . . . . . . . . . . . . . . . . . . . . . . . .
Component Non-Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Parent Questionnaire Response Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Child Questionnaire Response Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
NLSCY School component . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Response Rates for Math and Reading tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Response Rates for the Teacher’s and Principal’s Questionnaires . . . . . . . . . . .
129
129
135
135
136
136
137
144
Chapter 10 - Guidelines for Tabulation, Analysis and Release . . . . . . . . . . . . . . . . . . . .
Rounding Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Sample Weighting Guidelines for Tabulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Definitions of Types of Estimates: Categorical vs. Quantitative . . . . . . . . . . . . .
Tabulation of Categorical Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Tabulation of Quantitative Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Guidelines for Statistical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C.V. Release Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
150
150
151
152
154
155
155
157
Chapter 11 - Approximate Sampling Variability Tables . . . . . . . . . . . . . . . . . . . . . . . . . .
How to Use the C.V. Tables For Categorical Estimates . . . . . . . . . . . . . . . . . . . .
Examples of using C.V. Tables for Categorical Estimates . . . . . . . . . . . . . . . . . .
How to Use the C.V. Tables to Obtain Confidence Limits . . . . . . . . . . . . . . . . . . .
Example of Using the C.V. Tables to Obtain Confidence Limits . . . . . . . . . . . . .
How to Use the C.V. Tables to Do a T-test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Example of Using C.V. Tables to do a T-Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Coefficients of Variations for Quantitative Estimates . . . . . . . . . . . . . . . . . . . . . .
Release Cut-offs for the NLSCY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
160
164
167
170
171
172
173
173
174
Chapter 1 - Introduction
General Information
Purpose of
this manual
This manual will assist National Longitudinal Survey of Children and
Youth data users.
It’s purpose is to:
<
document data quality and other analytical issues regarding the
NLSCY;
and,
< facilitate the manipulation of the micro data files.
National
Longitudinal
Survey of
Children and
Youth
The National Longitudinal Survey of Children and Youth (NLSCY) is a
long-term study conducted in partnership by Human Resources
Development Canada (HRDC) and Statistics Canada. The primary
objective of the NLSCY is to monitor the development and well being of
Canada’s children as they grow from infancy to adulthood
Survey
Population
The NLSCY is designed to follow a representative sample of Canadian
children, aged newborn to 11 years, into adulthood, with data collection
occurring at two-year intervals.
Collection
Cycle
Each collection cycle used by NLSCY consists of a number of months
sometimes over the period of two calendar years during which
interviews with respondents are completed. Each cycle marks the
beginning of the collection phase when the longitudinal survey
respondents are followed up.
Collection
Cycles
Cycle
1
2
3
NLSCY Data Users Guide
Collection Start Collection End
December 1994
December 1996
October 1998
1
April 1995
April 1997
June 1999
2001/2002
Objectives of
the NLSCY
The objectives of the NLSCY are:
Data Release
Strategy
Cycle 4 data will be released….??
Contact
Person at
Statistics
Canada
All questions about the data set or its use should be directed to:
Contact
Person at
Human
Resources
Development
Canada
The contact person for Human Resources Development Canada is:
Ø To determine the prevalence of various risk and protective factors
for children and youth.
Ø To understand how these factors, as well as life events, influence
children’s development.
Ø To make this information available for developing policies and
programs that will help children and youth.
Ø Collect information on a wide variety of topics – biological, social,
economic.
Ø Collect information about the environment in which the child is
growing up – family, peers, school, community
Ø Information comes from different sources (parent, child, teacher)
and from direct measures (PPVT, math/reading tests, etc.)
Lecily Hunter, Project Manager
NLSCY Special Surveys Division, Statistics Canada
7(C8) Jean Talon Building, Tunney's Pasture, Ottawa, Ontario
K1A 0T6
Telephone:(613) 951-0597
Facsimile:(613) 951-7333
Internet: [email protected]
Toll free #: 1-800-461-9050
Susan McKellar, NLSCY Project Coordinator
Applied Research Branch, Human Resources Development Canada
Place du Portage, Phase II, 165 Hôtel de Ville, Hull, Québec
K1A 0J2
Telephone:(819) 953-8101
Facsimile:(819) 953-8868
Internet: [email protected]
NLSCY Data Users Guide
2
2001/2002
Chapter 2 – Survey Methodology
Definition of
the NLSCY
Population
The NLSCY survey population consists of two sample groups.
They are the:
Ø longitudinal sample,
and
Ø cross-sectional sample.
Longitudinal
Sample
The longitudinal sample consists of different cohorts.
The first cohort consists of the children who were sampled in Cycle 1 at
age 0 -11; these children will be followed until they are 25 years of age.
The second cohort consists of children who were sampled in Cycle 2 at
age 0 - 1; these children will be followed until they are 5 years of age.
The third cohort consists of children who were sampled in Cycle 3 at
age 0 - 1; these children will be followed until they are 7 years of age
(possibly 9 years of age).
The longitudinal sample is also used for cross-sectional purposes to
cover specific age groups.
Crosssectional
Sample
From the Cycle 3 file, we can produce cross-sectional estimates for
ages 0-15 years. A large sample of 5 year olds was included in Cycle
3 to allow for reliable provincial estimates of this age group.
Nonresponse and
missing
information
With each cycle there are respondents for which we are unable to
collect information. Based on interviewer notes from previous cycles
we determine hard-core non-respondents, exclude them from the
sample and do not attempt to trace them.
In most longitudinal surveys, only respondents to the first cohort are
followed and interviewed. However, a number of surveys including the
NLSCY try to re-contact people from the initial cohort, even if they
missed one or more waves of interview.
In Cycle 3 and Cycle 4, an attempt was made to re-interview people
who responded to Cycle 1 but not Cycle 2 or 3.
NLSCY Data Users Guide
3
2001/2002
Cycle 1
In Cycle 1, 22,831 children were interviewed. After sub-sampling,
16,903 respondents to Cycle 1 form the longitudinal cohort that will be
followed until these respondents reach the age of 25.
Close to 5,000 children used in the sample for Cycle 1 were selected
from the National Population Health Survey (NPHS). That sample
(5000 children) was given back to the NPHS. These 5000 children will
remain part of NPHS and will not be followed and interviewed by
NLSCY.
Cycle 2
Due to costs and response burden the sample for Cycle 2 was
reduced.
To try and decrease response burden for families that had more than 2
selected children, a sub-sample from Cycle 1 was taken to keep only
two children per household for the Cycle 2 interviews.
Why use
these
children for
Cycle 2 –
Longitudinal
The siblings selected for Cycle 2 were part of a responding longitudinal
household; this was an inexpensive way of adding children to the
sample. It also allowed us to continue the comparisons of children
within a family versus between families. These children are not
considered longitudinal respondents (even though they live with a
longitudinal family) because they were not eligible in Cycle 1.
The new sample of children from the Labour Force Survey was
included to ensure an unbiased sample in Cycle 2.
NLSCY Data Users Guide
4
2001/2002
Cycle 3
Sample Size
In total 38, 035 children were sampled in Cycle 3. Of those sampled
children 1,089 (3%) were out of scope either because the respondent
had moved permanently outside of Canada or because the household
did not contain a child who was eligible to complete the NLSCY.
The sample size of the 1 and 5 year old children for Cycle 3 was
increased from that used in previous cycles. This was due to the
federal government’s 1997 Speech of the Throne, which outlined the
intent to have measures of the early years and the commitment to
report on the measures. This enabled us to produce provincial
estimates of “readiness to learn” for the 5 year old children.
Cycle 3
–Crosssectional
Sample of
Children
0-11 Months
Children from the age group of 0 to 11 months were taken from the
LFS sample. Unlike Cycle 2, no siblings of the longitudinal cohort
were selected.
2086 households representing 2,123 children were added to the
NLSCY sample in Cycle 3. Seven households were selected for the
LFS sample but were excluded from the NLSCY sample since these
households were already in our sample (37 households had twins).
It should be noted that these children were sampled to ensure that they
would be 0-11 months old at the time of the interview and that
collection work for this component began in October 1998 and ended
in July 1999.
NLSCY Data Users Guide
5
2001/2002
NLSCY Data Users Guide
6
2001/2002
Chapter 3 – Response Rates
Response
Rates at
Child Level
Since the child is the unit of analysis, response rates are presented at
the child level rather than the household level. In many cases there may
be more than one child per household, consequently there may be data
obtained for one child but not for another in the same household.
Computer
Generated
Response
Codes
On the computer each household represents a case and status codes
are automatically given to the case each time an interviewer enters it.
Within each case are components - for example there each child has
it’s own component. Consequently, it’s possible to have one household
with different response codes for each component. Complete
information may be available for one child but not another, in this case a
“partial” or “non-response” code would appear for one child component
while a “fully complete” response code appears for the other. At the
household level this case would have a “partial” response code”.
As the panel ages, a larger proportion of the sample will come from the
early years cohort in which only one child per household is selected.
Thus, the response rates at the household and at the person level
should gradually become very similar.
Table 1: Overall Child Level Response Rates, NLSCY Cycle 3
Number
Sample
Not eligible
%
Longitudinal
Cohort (A)*
%
Other
%
(N,T)
38035
16718
21317
1089
144
945
Eligible
36946
100%
16574
100%
20372
100%
Full
32097
87%
14677
89%
17420
86%
254
1%
103
1%
151
1%
Refusal
2328
6%
1203
7%
1125
6%
Unable to trace
1182
3%
228
1%
954
5%
Other nonresponse
1085
3%
363
2%
722
3%
Partial
* includes longitudinal respondents who did not respond in Cycle 2.
NLSCY Data Users Guide
7
2001/2002
Response
Rates
Cycle 3 NonResponse
As a percentage of all eligible children, a response rate of 88% was
achieved with 87% of cases being fully completed and 1% of cases
being partially completed. In 3% of cases, non-responses occurred
because the respondent no longer resided at the address or phone
number on file and attempts to trace their current location were
unsuccessful. In 6% of cases, households refused to participate and
in 3% of cases other non-responses occurred. Examples of other
non-responses include unable to interview due to unusual
circumstances (i.e. death in the family, illness), the household was
absent during the collection period, and unable to interview due to
language problems.
Responding
Sample by
Age and
Province
In total 31,194 children were retained on the final data file. The
following two tables present the responding sample by province and
age group.
Table 2 - Province and Sample Size
Province
Responding
Sample Size
Newfoundland
1612
PEI
948
Nova Scotia
2019
New Brunswick
1956
Quebec
6298
Ontario
8658
Manitoba
2254
Saskatchewan
2307
Alberta
3125
British Columbia
2817
TOTAL
NLSCY Data Users Guide
31194
8
2001/2002
Table 3 - Age of Child and Responding Sample Size
Age
NLSCY Data Users Guide
Responding
Sample Size
Age
Responding
Sample Size
0
1736
8
1381
1
6391
9
940
2
1589
10
1238
3
2029
11
842
4
1.983
12
1264
5
6958
13
875
6
1536
14
1262
7
1053
15
916
9
2001/2002
NLSCY Data Users Guide
10
2001/2002
Chapter 4 - Data Collection
Computer Assisted Interviewing
ComputerAssisted
Interviewing
Data collection for the household component of the NLSCY relied
heavily on CAI technology.
The use of computer-assisted personal interviewing (CAPI)
technology allows for high quality collection of complex populationspecific content sections. For example, the system facilitates the
collection of the relationships of all household members to each
other (i.e., the relationship grid). This wealth of information will
enable a detailed analysis of family structures, an important concept
for analysis of the child information. This type of collection would be
very difficult to implement in a paper and pencil environment..
The CAI
System
The CAI system has two main parts
1.
2.
Case Management, and
the survey-specific components
Case
Management
The Case Management system controls the case assignment and
data transmission for the survey. For the NLSCY, a case refers to a
household selected for the NLSCY sample. The Case Management
system also automatically records management information for each
contact (or attempted contact) with respondents and provides
reports for the management of the collection process.
Transmission
of Cases
The Case Management system routes the questionnaire
applications and sample file from headquarters to the regional
offices, and from the regional offices to the interviewer laptops. The
returning data take the reverse route. To assure confidentiality, all
data is encrypted for transmission. The data are unencrypted only
once they are on a separate secure computer with no external
access.
NLSCY Data Users Guide
11
2001/2002
Survey Specific
Components
The survey-specific component of CAPI includes an introductory
component with procedures for contact and selection of households.
Once contact has been made and household composition has been
established, the CAPI system generates applicable questionnaire
components dependent on the household composition and the
outcome of the selection procedures.
Some of the specific components generated included a Parent and
General Questionnaire for the Person Most Knowledgeable (PMK)
and spouse/partner and Child's Questionnaire for selected children
in each household.
Household
Roster
The household roster becomes more difficult when a longitudinal
survey interviews more than one longitudinal respondent per
household. Eligibility rules need to be defined to know when to trace
and when to interview. An added complexity was due to the fact that
households of the Cycle 3 sample were divided into two groups:
1.
longitudinal households, that is, those that had already
participated in Cycle 1 and/or 2 of the survey;
2.
new households with children aged 0 to 23 months, 1 years of
age or 5 years of age.
NLSCY Data Users Guide
12
2001/2002
Longitudinal
Households
1st Contact
About 50% of all households contacted had already participated in
either the first and/or second cycle of the survey. Of the selected
children in these Cycle 2 households, a maximum of two were
chosen for whom data was to be collected.
The first contact was established with these households using the
address and telephone number provided during Cycle 2. Next, the
interviewer confirmed that at least one member of the household list
provided in 1996-97 still lives at the address. If none of the
individuals on the list were in the household contacted, the file for the
household was transferred to the trace folder and the interview with
the household was ended. If one of the individuals on the list was a
member of the household contacted, the interview continued
beginning with the confirmation or updating of the contact
information (mailing address and residence, telephone number),
and the updating of the list of household members.
The Final
Phase of 1 st
Contact
During this final phase, if one of the children selected was no longer
part of the household, information as to why (parents' separation,
departure for a foster home, etc.), the date of the child's departure
and the child's new address or other relevant information for tracing
them was obtained. Then, the new members of the household were
added to the list. If at least one of the selected children was no
longer a member of the household, a new household file was
created and transferred to tracing.
The Tracing
File
The Tracing file includes all household members from the first cycle
who were no longer part of the contacted household. The interview
with the contacted household was discontinued if all the selected
children had left, but was continued if at least one of the selected
children was still a member of the household.
NLSCY Data Users Guide
13
2001/2002
Contacting
Nonresponding
Households
In order to achieve the desired response rate, an effort was made to
recontact non-responding households to the first collection in the
second collection period and to recontact non-responding
households to the second collection period in the third collection
period. For example, if in the first collection period, a household
could not be reached because no one was at home, then this case
was sent out again with the February sample and further attempts
were made at that time to contact the household.
Demographic
Information
Collected
For households with eligible children, basic demographic
information was then gathered (age, date of birth, sex, marital
status) and relationships between the members of the household
were completed.
Person Most
Knowledgeable
(PMK)
Some questions about dwelling conditions were asked and this
questionnaire ended with a question designed to select from among
those individuals aged 15 or older the Person Most Knowledgeable
(PMK) about the selected child(ren). This person became the
primary respondent and was labeled as the PMK for this household.
In most cases, the PMK was the mother of the child.
New Crosssectional
Households
The second group of households included 2,087 new households
with children aged 0-11 months, 7,932 new households with children
1 year of age and 6,952 new households with children aged 5 years.
For these households, the initial contact procedures were the same,
except for the fact that no tracing was done for people who had
moved. Households were updated and the interviewer gathered the
demographic data and relationships. After this stage, if there were
no eligible children in the household, the interview ended; otherwise,
it continued in the same way as for the households in the first group
with questions asked about dwelling conditions and the selection of
the PMK.
NLSCY Data Users Guide
14
2001/2002
Household Collection
Household
Collection
Period
There were three collection periods for the household collection,
1.
2.
3.
November and December 1998
February and March 1999
April-May 1999.
The overall sample was split evenly among the three collection
periods and each period lasted approximately six weeks.
The
Household
Collection -
For the household collection, data were collected from a variety of
respondents using different data collection instruments. Except for the
questionnaires asked of 10 to 15 year olds all of the information for
the household collection was collected in a face-to-face or telephone
interview using computer-assisted interviewing (CAI).
Instruments
Completed by
the PMK
After completing the contact and demographic data questionnaire, the
PMK was asked to complete a series of questionnaires. The Parent
Questionnaire for this person and their spouse, if applicable; a Child's
Questionnaire for each child selected in the survey; and a
computerized consent form about contacting the schools attended by
the children.
The Parent
Questionnaire
The first part of this questionnaire was completed by both the PMK
and his/her spouse/partner and was designed to gather socioeconomic and health data about these two individuals. Topic areas
included education, labour force and income. The second part of the
Parent Questionnaire was completed by and for one of the parents
only, usually the PMK. The purpose was to gather information about
the child's family environment, notably the mental health of the PMK
and family functioning.
NLSCY Data Users Guide
15
2001/2002
Child’s
Questionnaire
The Child's Questionnaire was completed for selected children in the
household aged newborn to15 years. Topic areas included health,
birth information, temperament, behaviour, education, activities,
literacy, social relationships, parenting, and legal custody of the
children.
The Informed
Consent
Questionnaire
For each child who attended school in 1997-98, the PMK also
answered a computerized questionnaire in which his/her consent was
requested to: (a) contact the child's teacher and the school principal,
and (b) administer a test of about 45 minutes measuring the child's
mathematics computation and reading comprehension skills. In this
questionnaire, school contact information was also gathered
(principal's name, school address, telephone number).
Cognitive
Measure
Two tests were administered to respondents in order to assess
cognitive measures.
They are:
<
and
<
Math and
Reading
Skills
Indicator
Math and Reading Skills Indicator,
The Peabody Picture Vocabulary Test Revised (PPVT-R).
School children in grade 2 or higher were given a brief mathematics
and vocabulary/reading test of about 12 questions. This placement
test was designed to make it possible to determine the level of the
math computation and reading comprehension tests that would
subsequently be administered in the schools.
For grade 2 children, the interviewer read the questions and recorded
the answers on an answer sheet. For children in grade 3 or above,
the child read the questions and gave the interviewer the answer.
NLSCY Data Users Guide
16
2001/2002
The Peabody
Picture
Vocabulary
Test - Revised
(PPVT-R)
The Peabody Picture Vocabulary Test - Revised (PPVT-R) was
administered by the interviewer to each selected child between 4 and
5 years old, as well as to children aged 6 years and older who were
not yet in grade 2. The oral consent of the PMK was obtained before
the test was administered. The purpose of the test was to assess the
child's level of receptive vocabulary.
After having completed the full NLSCY interview and leaving the
household, the interviewer completed an administrative questionnaire
describing the conditions in which the test was administered. This is
done in order to identify any factors that might have influenced the
child's answers and overall reaction to the test.
Self
Completed
Questionnaire
- 10-15 yrs
The objective of the Self-completed Questionnaire is to collect
information directly from the child on a variety of aspects of his/her life.
These self-completed questionnaires are used to supplement, and in
subsequent analyses, compare with information obtained from the
parent and teacher.
Starting at age 10, with the PMK’s permission, the interviewer
provides a questionnaire to the child and encourages him/her to
complete it in a private setting. Upon completion, the questionnaire is
sealed in an envelope to ensure confidentiality.
The PMK was informed of the confidentiality of the questionnaire
before giving permission for the child to complete it. The PMK is not
allowed to see the completed questionnaire. It was hoped that this
procedure would increase the likelihood that the child would provide
accurate and honest information.
The following table contains the content of the questionnaires completed by those 10-15
years of age:
Table 4
The Self-completed
Questionnaire for
those aged:
NLSCY Data Users Guide
Contains questions on the topics of:
17
2001/2002
10 -11 years
<
friends and family, school, feelings and behaviours,
smoking and drinking and activities.
12-13 years
<
<
friends and family, school, feelings and behaviours,
delinquent behaviour, smoking, drinking, drug use,
health (general, depression and puberty) and about
work and sources of money.
14-15 years
<
<
friends and family, school, feelings and behaviours,
delinquent behaviour, smoking, drinking, drug use,
health (general, depression and puberty) and about
work and sources of money.
work during the school year, summer work, sources of
money and how they spent their money.
<
Interview
Length for
Household
Collection
For the household collection, the interview length for responding NLSCY
households was approximately two hours.
The total amount of time that it took to complete the major
questionnaires that were part of the NLSCY household collection are
presented in the table below. The table gives median interview times
(i.e., the time at which 50% of the cases took more time and 50% took
less). It should be noted that all extreme times (high and low) were
removed before these times were derived, since they often represent a
problem with the time clock/procedure rather than a real interview time.
Table 5 outlines the length of time required to complete the various questionnaires:
Table 5
Type of Questionnaire
Time in Minutes
All questionnaires in the household interview
All Child Questionnaires for the household
All Parent Questionnaires for the household (for the
PMK and spouse/partner)
Total for major components (Child, Parent, General &
PPVT & Informed Consent)
NLSCY Data Users Guide
18
98
31
21
75
2001/2002
Remaining Components1
34
Tables 6 gives the median interview times for a family with PMK a spouse and 1-3
children. The number of selected children (0 to 13) in the household was the factor that had
the strongest impact on interview length.
Table 6:
PMK, spouse
and :
1 child
2 children
3 children
Time in
Minutes
81
134
162
Table 7 gives the median interview times for a family with a PMK (no spouse) and 1-3
children. For households in which the PMK had a spouse/partner and more than two
selected children, the interview length was over two hours.
Table 7 :
PMK, spouse
and :
1 child
2 children
3 children
Time in
Minutes
84
139
171
School Collection
1
This is the difference between the total time and the time required for the major components. This would include
time for the interviewer to introduce the survey, complete the household roster, the relationships, set-up time for the
10 to 11 Questionnaire, the 12-13 Questionnaire and the math and reading skills test, time for the computer to
generate the various questionnaires, etc.
NLSCY Data Users Guide
19
2001/2002
The School
Collection
The school collection took place from April to June 1998. For all
children in the Cycle 3 sample who were attending school, the PMK
was asked to give written permission to allow the collection of
information from the child's teacher and principal. In cases where the
child was in grade 2 or higher the PMK was asked to give permission
to allow the teacher to administer a skills test in math computation and
reading comprehension to the child.
The school collection involved three questionnaires. These
questionnaires were mailed out to teachers and principals, who were
asked to complete the questionnaires and mail them back to
Statistics Canada in the envelopes provided.
Collection
Strategy for
the School
Collection
Questionnaire packages were mailed to principals with instructions on
how the various instruments should be completed. The principals were
then asked to distribute the questionnaires and tests to the teachers.
Approximately one week after the initial mailing a postcard was sent
out to thank all respondents and to remind those who had not yet
responded to do so.
Roughly two weeks later, a second questionnaire package was sent
out to teachers and principals who still had not responded. Finally
three weeks later non-responding teachers and principals were
contacted by telephone and encouraged to participate.
Teacher's
Questionnaire
The goal of the teacher’s questionnaire was to collect information
about the child's academic achievement and behaviour at school, as
well as information on characteristics of the class and the teacher's
instructional practices.
There were three teacher questionnaires which were completed
depending on the circumstances of the child:
•
•
•
NLSCY Data Users Guide
a kindergarten questionnaire,
a teacher questionnaire, for students who had one teacher for
the basic academic subjects;
a different teacher questionnaire for students who had different
teachers for the basic academic subjects.
20
2001/2002
The
Principal's
Questionnaire
The goal of the Principal's Questionnaire was to gather information on
the school environment in order to assess how this may impact child
development. Consequently, the Principal's Questionnaire collected
information on school policies, resources and educational climate,
rather than data about a specific child.
The Math
Computation
and Reading
Comprehensi
on Test
The math portion of the skills test to be administered to the child was a
shortened version of the Mathematics Computation Test of the
standardized Canadian Achievement Tests, Second Edition (CAT/2).
The CAT/2 is a series of tests designed to measure achievement in
basic academic skills. Some of the test's questions on reading
comprehension are taken from the CAT/2 test, and some are new
questions developed for the NLSCY.
NLSCY Data Users Guide
21
2001/2002
Interview Training, Supervision and Control
Interviewers
The NLSCY was conducted by Labour Force Survey interviewers. All
LFS interviewers are under the supervision of a staff of senior
interviewers who are responsible for ensuring that interviewers are
familiar with the concepts and procedures involved in the survey, and
also for periodically monitoring their interviewers and reviewing their
completed documents. Senior interviewers ensure that prompt follow-up
action is taken for refusal and other non-response cases. If necessary,
non-response cases were transferred to the senior and reassigned. The
senior interviewers are, in turn, under the supervision of the LFS
program managers, located in Statistics Canada regional offices.
Training
For the NLSCY a combination of classroom training and self-study
materials were prepared to ensure that interviewers had a proper
understanding of survey concepts.
Self-study
•
involved the interviewers reading the Interviewer's Manual
prepared for the survey and completing home study exercises.
Classroom
•
NLSCY Data Users Guide
a program manager or a senior interviewer presented an
overview of the survey, went through a mock interview with the
participants, gave more specific training on administering the
PPVT-R and presented exercises to help interviewers minimize
non-responses. In total, 14 hours were devoted to these training
activities for each interviewer.
22
2001/2002
NLSCY Data Users Guide
23
2001/2002
Chapter 5 - Data Processing
Editing
Introduction
The main output of the NLSCY is a "clean" master data file. This section
presents a brief summary of some of the processing steps involved in
producing this file.
Computer
Generated
Edits
As discussed earlier, all of the information for the household collection
(except for the 10-11 year old and 12-13 year old self-completed
questionnaires) was collected in a face-to-face or telephone interview
using computer-assisted interviewing (CAI). As such, it was possible to
build various edits and checks into the questionnaire for the various
household CAI components, in order to ensure high quality of the
information collected.
Types of
Computer
Edits
Various types of computer generated edits were used to check data
while the interviewer was completing the interview.
The NLSCY computer generated survey used the following:
<
Review Screens,
<
Range Edits,
<
Flow Patterns Edits,
<
Consistency Edits.
Review
Screens
Review screens were created for important and complex information.
Example:
The selection procedures for the PMK, a critical element of the
survey, were based on the household roster. The household
roster screen showed the demographic information for each
household member and his/her relationship to every other
household member. The collected information was displayed
on the screen for the interviewer to confirm with the respondent
before continuing the interview.
NLSCY Data Users Guide
24
2001/2002
Range Edits
Range edits were used for continuous variables, to confirm or correct
unusual answers during collection.
Example:
For the question regarding the weight of a child at birth, if a
weight entered into the computer was either significantly high or
low, a pop-up message would appear asking the interviewer to
confirm the answer with the respondent.
Flow Pattern
Edits
All flow patterns were automatically built into the CAI system.
Example:
In the Child Care Section, the PMK is asked he/she used
daycare or babysitting in order that he/she (or a partner/spouse)
could work or study. Based on the response given the flow of the
questions could be different. If Child Care was used, the CAI
system continued with a series of questions about the specific
care method(s) used for the child. If not, the CAI system
automatically skipped this series of questions.
General
Consistency
Edits
Some consistency edits were included as part of the CAI system, and
interviewers were able to "slide back" to previous questions to correct
for inconsistencies. Instructions were displayed to interviewers for
handling or correcting problems such as incomplete or incorrect data.
Example:
In the collection of the Labour Force Section, the number of
weeks working, not working, and looking for work should not total
more than 52 weeks. If this was the case, the system generated
a pop-up window which stated the error and instructed the
interviewer to slide back to the appropriate question to confirm
the data and make corrections as required.
NLSCY Data Users Guide
25
2001/2002
Consistency
Edits
Between
Cycles
For this second cycle of the NLSCY edits were also performed to
ensure consistency between cycles for data that was not expected to
change. Data from the previous cycle (feedback variables) were
included in the CAI system for the current cycle. When inconsistencies
were identified, the interviewer was asked by the system to confirm the
Cycle 2 data with the respondent through a series of questions.
Example:
For the Chronic Conditions questions, if a chronic condition such
as asthma was reported in the previous cycle but not indicated
as being present in the current cycle, the system prompted the
interviewer to ask questions to determine if the current data was
in fact correct, or if the condition had changed since the previous
cycle.
Data Capture
Paper and
Pencil
Questionnaire
s
Some questionnaires for the NLSCY were completed on paper and
pencil questionnaires (PAPI). The 10-11, 12-13 and 14-15 year old
Self-Completed Questionnaires, the Teachers’ Questionnaires and
the Principals’ Questionnaire were all completed by PAPI. All of these
documents were completed directly by a survey respondent.
Data Capture
for PAPI
Questionnaire
s
Data capture for these questionnaires were accomplished through
scanning at a centralized area at Statistics Canada’s Head Office.
Questionnaire
Grooming
Prior to scanning, the documents were groomed and verified for
completeness. During this process, any document containing at least
one respondent-completed item was scanned and a file containing
each record was provided to Head Office processing staff for further
processing. As part of the scanning system, some quality checks were
built in to flag unusual entries to warn the operators of potentially
incorrect entries.
The operator visually reviewed the questionnaire responses and
manually entered the correct values. In cases where more than one
response was checked off by the respondent, the operators were
instructed to accept the first response. Errors remaining within the
questionnaires were then edited at a later stage.
NLSCY Data Users Guide
26
2001/2002
Minimum Completion Requirements
Defining
Requirements
One of the first steps in the NLSCY processing was to define the
requirements for a responding household.
.
No
Information
Collected
In some cases, no NLSCY information was collected for a sampled
household. This happened, for example, when an interviewer was
unable to make contact with a selected household for the entire
collection period, in other cases the household refused to participate
in the survey, special circumstances such as an illness or death in a
family or extreme weather conditions sometimes prevented an
interview from taking place.
For cases where no information was collected for a household, the
household was dropped from the NLSCY file and the sampling
weights for responding households were inflated to account for these
"dropped" households
Partial
Information
In other cases, it was possible to carry out some of the interview, but a
complete interview was not obtained for a variety of reasons. Some
respondents were willing to give only a certain amount of time to the
completion of the survey. In some cases an interviewer completed a
portion of the survey with the respondent and made an appointment to
continue at another time but was unable to re-contact the respondent.
Criteria for
Partial
Response
It was necessary to come up with a criteria for deciding what to do
with these "partial" interviews. If the majority of the survey had been
completed, obviously the preference was to keep this case and label
it as a responding household. However, if only very minimal
information was collected the decision was made to drop the
household and treat it as a non-responding household. In order to
make this assessment, the data collected for each selected child in
the household were examined. This was done by looking at certain
key questions across the Child Questionnaire. An assessment was
made as to whether or not there was an adequate amount of
information collected for at least one child in each household. If there
was, the household was maintained in the responding sample.
NLSCY Data Users Guide
27
2001/2002
Missing
Variables
Longitudinal
Child
Records
All missing variables for households were set to not-stated or
imputed. If there was not adequate information for at least one child
then the household was dropped from the responding sample and
treated as a non-response.
In total, 17,618 longitudinal child records were determined to be
complete enough to be kept (codes 000 and 001). These children
came from 12,100 longitudinal households, which is the number of
households maintained in the Cycle 3 NLSCY files.
There were 18,612 child records for the responding longitudinal
households. Out of these, there were 994 longitudinal child records
that were "not acceptable" but were kept because there was at least
one "acceptable" child record for the household.
Missing
Components
Variables on missing components for the household were imputed or
set to not-stated.
The longitudinal file also contains 194 records that were created for
some longitudinal children for whom no data was collected in this
cycle. These are children who are now deceased or who have moved
out of the country, but who will be kept on the longitudinal file for
weighting purposes. For these records, all variables except for the
longitudinal weight (CWTCW01L) have been set to ‘not stated’.
NLSCY Data Users Guide
28
2001/2002
Head Office Editing
Stages of
Editing
For CAI questionnaires for the NLSCY, two stages of editing were
conducted.
<
<
Pre-edit
Consistency Editing
The purpose of the Pre-edit was to carry out some basic formatting and preliminary
editing. Table 9 outlines some of the procedures used.
Table 9
Step
Action
Done to the:
1
< Non-response values from the CAI system were recoded to
standard non-response codes for refusals, don't know and
not-stated.
< Mark All That Apply’ questions were destrung and values
converted to Yes (1) or No (2) responses.
< Databases files were created for each section of the Adult
and Child questionnaires
complete Adult
and Child file
2
< Small data base files were created for each section of each
questionnaire. A record was kept for the section only if the
section was applicable. For example, the section on
temperament was only applicable for children 3 months to 3
years old. Therefore a temperament record was only created
for children in this age group.
Separate DBF
files from Step
1
< Within several sections, different wording was used for
different age groups. For example, in the Activities section,
Question 3 asks "In the past 12 months, outside of school
hours, how often has (the child) taken part in any clubs,
groups or community programs with leadership....”. The
wording for 4 to 5 year-olds (CAACQ3D1) was “such as
Beavers, Sparks or church groups?”. The wording for 6 to 9
year olds (CAACQ3D2) was “such as Brownies, Clubs or
church groups?” Initially these questions were stored as
separate variables. As part of the pre-edit the two variables
were collapsed into one output variable CAACQ3D. The
various wordings are given for these types of questions in the
data dictionary in Appendix 4.
< The flow patterns for each section were processed and valid
skips were assigned ‘not applicable’ codes (6, 96, 996..).
NLSCY Data Users Guide
29
2001/2002
NLSCY Data Users Guide
30
2001/2002
Consistency Editing
The Goal of
Consistency
Editing
After the pre-edit, consistency editing was carried out to verify the
relationship between two or more variables.
Example:
In the Socio-Demographic Section, for children who were not born in
Canada, Question CSDCQ2B asks on what year they first immigrated
to Canada. There was a consistency edit which compared this question
to the year of birth of the child. If the year of immigration was before
year of birth then year of immigration was set to not-stated in the edit.
Editing was also performed to ensure consistency between cycles.
Consistency
Between
Cycles
Example: The responding child’s height in Cycle 3 should not be less
than the height reported in Cycle 2.
Flags were set for inconsistencies between cycles. These variables
appear on the Secondary data file (Appendix 5) and contain ‘Z’ in the
variable name. For PMK and Spouse variables, the data was linked
using a unique person identifier, allowing the comparison to be made if
the PMK was the same in both cycles or if the PMK was the spouse in
the previous cycle and vice versa.
Consistency
Edits for
PAPI
For the questionnaires that were collected using a paper version,
essentially the same steps of editing were carried out. In the pre-edit,
however, there was an additional requirement. In some cases a value
was captured that was not allowable for a particular item. This was
possible due to the fact that the scanning operator was given the ability
to overwrite the edits. These invalid entries were set to "not stated"
values in the pre-edit. Editing for flow patterns was carried out at the
consistency editing stage for the paper questionnaires.
Data File for
10 to 15 year
olds
One data file was produced for the 10-11,12-13 and 14-15
questionnaires. For questions that did not apply to an age group, the
variables were set to ‘not applicable’ codes (6,96,996..).
Data File for
Teacher’s
File
In this cycle there were 3 Teachers’ questionnaires with many of the
same questions. These are to be released in July 2001. Questions that
were not asked from a teacher were set to ‘not applicable’ codes
(6,96,996..).
NLSCY Data Users Guide
31
2001/2002
Naming Convention and Coding Structure for NLSCY Variables
Introduction
The NLSCY microdata file documentation system has employed certain
standards to label variable names and values. The intent is to make
data interpretation more straight-forward for the user.
Naming
Convention
for Variables
A naming convention has been used for each variable on the NLSCY
data file in order to give users specific information about the variable.
All variable names are at most eight characters long so that these
names can easily be used with analytical software packages such as
SAS or SPSS.
C SE C Q nnx or B SE C b Q nnx
Format for
Variable
Names
C refers to the NLSCY Cycle
“A” indicates the first cycle,
"B" the second cycle,
"C" the third etc...
SE - refers to the section of the questionnaire where the question
was asked or the section from which the variable was derived.
C - refers to the collection unit or the unit to which the variable refers.
There are four possibilities 2:
“C” is the variable refers to the child,
“P” the PMK.
“S” the spouse/partner
“H” the household
b - the lower case letter refers to the NLSCY Cycle in which the
variable first appeared on the file.
2
It should be noted that while variables do exist for various units of analyses (i.e., the PMK, the spouse/partner and
the household), it will only be possible to produce "child estimates" from the NLSCY file. The characteristics of the
PMK, spouse/partner and household can be used to describe attributes of the child. For example it will be possible
to estimate the number of children living in a household with low income, or the number of children for whom the
PMK has scored high on the depression scale etc. However it will not be possible to produce estimates of the
number of low income households or depressed PMKs.
NLSCY Data Users Guide
32
2001/2002
_______
Format
for
Variable
Names
Cont’
Example: "b" indicates the variable was new in Cycle 2. In subsequent
cycles, new variables will also be identified using the lowercase letter
representing the cycle. New variables in Cycle 3 will contain a "c", in
Cycle 4 a "d" , etc. Some revisions were made to the content of the
questionnaire between cycles. If the revision resulted in a change to the
meaning or the values of a question, the variable was treated as new
and contains a "c".
________________________________________________________
Q refers to the variable type. There are six possibilities:
refers to the variable for a question that was asked directly on one of the
NLSCY questionnaires
“S” refers to a score calculated for one of the scales used on the
questionnaire
“D” means the variable was derived from other questions that were
asked on the questionnaire
“I” means the variable is a flag created to indicate that an item has been
imputed
“X” means the variable is a flag created to indicate an inconsistency in
reported data between the current and previous cycles
“nnx” refers to the question or variable identification. Generally nn is a
sequential number assigned to the variable; and x is a sequential
alphabetic indicator for a series of variables of a similar type
NLSCY Data Users Guide
33
2001/2002
Acronym Names for Questionnaire Sections
The following table gives the acronym names that were used for each section of the
various NLSCY questionnaires. This acronym is embedded in the variable name for all
variables on the NLSCY data file. The acronym is the second and third characters of the
variable name.
Table 10
Variable
Geographic
Household
Collected or Derived from the:
sample information
dwelling characteristics
Basic demographic variables for each household
member. These variables are included on the
NLSCY data file for the child, the PMK and the
spouse/partner
information of the household roster and
relationship grid
SD
Variables collected as
part of the household
roster.
.
Demographic- derived
to explain the living
arrangements of the
child:
Socio-demographic
HL
Health
CH
RS
Adult Chronic
Conditions
Restriction of Activities :
DP
Depression scale
ED
Education
LF
Labour force
IN
Income
FN
Family functioning scale
MD
Medical/biological
.
GE
HH
MM
DM
NLSCY Data Users Guide
child on the Child's Questionnaire and for the PMK
and spouse/partner on the Adult Questionnaire.
PMK and Spouse on the Adult questionnaire, and
for the Child on the Child questionnaire
PMK and Spouse in the Health section of the
Adult questionnaire
PMK and Spouse in the Health section of the
Adult questionnaire
Parent Questionnaire (this scale was
administered to the PMK)
children 4 to 13 years old on the Child's
Questionnaire and about the PMK and
spouse/partner on the Adult Questionnaire
PMK and spouse/partner on the Adult
Questionnaire.
household income and personal income of the
PMK, collected on the Adult Questionnaire.
Adult Questionnaire (this scale was administered
to the PMK or spouse/partner to measure how
family members relate to each other.)
Child's Questionnaire (0 to 3 years of age)
34
2001/2002
TM
LT
AA
BE
MS
RL
PR
CR
PP
Temperament
Literacy
Activities
Behaviour
Motor and social
development
Social relationship
Parenting style
Child care
PPVT test:
PA
PPVT assessment:
FF
SC
AM
FB
PM
Friends and Family
School
About Me
Feelings and Behaviour
My Parents and Me
PU
Puberty
DR
Smoking, drinking and
drugs
AT
Activities
HT
WK
DA
Health
Work and Sources of
Money
Dating
EP
Principal's Education
ET
Teacher's Education
RE
MA
Reading test
Math computation test
NLSCY Data Users Guide
Child's Questionnaire (3 months to 3 years old)
Child's Questionnaire (0 to 6 years)
Child's Questionnaire (0 to 13 years)
Child's Questionnaire (0 to 13 years)
Child's Questionnaire (0 to 3 years)
Child's Questionnaire (4 to 9 years)
Child's Questionnaire (0 to 13 years)
Child's Questionnaire (0 to 13 years)
4 to 6 years old (if child in grade 1 or less included
those over 6 years of age)
interviewer to describe the conditions under which
the PPVT was administered to the child.
10 to 13 Self-complete Questionnaires: Section A
10 to 13 Self-complete Questionnaires: Section B
10 to 13 Self-complete Questionnaires: Section C
10 to 13 Self-complete Questionnaires: Section D
10-11 questionnaire Section E, 12-13
questionnaire, Section G
10 to 13 Self-complete Questionnaires:
Section F, 12-13 year Health questions in Section
H
10-11 questionnaire, Section G; 12-13
questionnaire Section F
10-11 questionnaire, Section H, 12-13
questionnaire, Section E
12-13 Self-complete Questionnaire: Section H
12-13 Self-complete Questionnaire: Section I
12-13 Self-complete Questionnaire: Taken from
questions in the Family and Friends and the Health
Sections
Child's Principal about the school and the
resources available to the staff
Child's Teacher about the child and the classroom
environment
children in grade 2 and over
children in grade 2 and over.
35
2001/2002
Examples of Variables Names
In order to illustrate the naming convention used for variables included on the NLSCY data
file the following examples are given.
Table 11
Variable
Name
Refers to:
CLFSQ2
Q2 in the Labour Force Section for the spouse/partner
C
a Cycle 3 variable
LF
the Labour Force Section
S
the spouse/partner
Q
an item asked directly on the questionnaire
2
the ID of the item.1
Variable
Name
Refers to:
CPRCS0
3
a positive interaction score on the parenting scale for a 2 to
15 year-old child
C
PR
a Cycle 2 variable
the Parenting Section
C
the child.
S
a score
3
ID of the variable
1
There is a possibility that this name will not correspond to the questionnaire in the present cycle,
given that we keep the same names of variables in the data dictionary. This usually happens
when a number or section changes from one cycle to another. For example, cmdcbq31
corresponds to question 3 in the section on Working After Birth, whereas in Cycle 2 it
corresponded to question 31 in the Medical and Biological Information section.
NLSCY Data Users Guide
36
2001/2002
Variable
Name:
Refers to:
CHLCbZ3
a flag that indicates an inconsistency in the child’s height
between the current and previous cycles.
C
a Cycle 2 variable
HL
the Health Section
C
to the child.
b
a new variable in Cycle 2.
Z
a longitudinal flag
3
ID of the variable
Coding Structure for NLSCY Variables
Introduction
Some standards have been developed for the coding structure of
NLSCY variables in order to explain certain situations in a consistent
fashion across all variables. The following describes these various
situations and the code used to describe the situation.
Refusal
During a CAI interview, the respondent may choose to refuse to provide
an answer for a particular item. The CAI system has a specific function
key that the interviewer presses to indicate a refusal. This information
is recorded for the specific item refused and transmitted back to Head
Office.
On the NLSCY data file an item which was refused is indicated by a
code "8".
For a variable that is one digit long the code will be "8", for a 2 digit
variable "98" for a three digit variable "998" etc.
NLSCY Data Users Guide
37
2001/2002
Don’t Know
The respondent may not know the answer to a particular item. Again the
CAI system has a specific function key to describe this situation.
On the NLSCY data file, the code used to indicate that the respondent
did not know the answer to an item is "7". For a variable that is one digit
long the code will be "7", for a two-digit variable "97" for a three-digit
variable "997" etc.
Not
Applicable
In some cases a question was not applicable to the survey respondent.
A code "6", "96" "996" ... has been used on the data file to indicate that
a question or derived variable is not applicable.
< In some cases a single question or series of questions was not
applicable. For example, the question on number of hours per week
the child is cared for in a daycare centre (CCRCQ1G1) is only
applicable for children for whom this type of care is used
(CCRCQ1G=1). Otherwise there will be a code 996 for this question
<
In other cases an entire section of the questionnaire was not
applicable or even an entire questionnaire. For example, the Motor
and Social Development Section was applicable only to children 0
to 3 years old. For all children outside of this age group (i.e., 4 years
and older) the motor and social development variables have been
set to not-applicable ("6", "96", "996" etc.).
For cases where the PMK did not have a spouse or common-law
partner residing in the household, all "spouse" variables (e.g., the
Labour Force Section and the Education Section for the spouse) have
been set to not applicable.
NLSCY Data Users Guide
38
2001/2002
Not-Stated
In some cases, as part of Head Office processing the answer to an item
has been set to not-stated. The not-stated code indicates that the
answer to the question is unknown. Not-stated codes were assigned
for three main reasons.
1. As part of the CAI interview, the interviewer was permitted to enter a
refusal or don't know code, as described above. When this
happened the CAI system was often programmed to skip out of this
particular section of the questionnaire. In the case of refusal, it was
assumed that the line of questioning was sensitive and it was likely
that the respondent would not answer any more questions on this
particular topic area. In the case of a don't know it was assumed that
the respondent was not well enough informed to answer further
questions. As part of the NLSCY processing system, it was decided
that all of these subsequent questions should be assigned a notstated code. A not-stated code means that the question was not
asked to the respondent. In some cases it is not even known if the
question was applicable to the respondent.
2. In some cases a specific questionnaire was not started or it was
started but ended prematurely. For example, there may have been
some kind of an interruption, or the respondent decided that she/he
wished to terminate the interview. If there was enough information
collected to establish this household as a responding household,
then all remaining items on the questionnaire (and on questionnaires
that had not yet been started) were set to not-stated. The one
exception was that if it was known that a certain section or a certain
questionnaire was not applicable, then these questions were set to
not applicable.
3. The third situation in which not-stated codes were used was as a
result of consistency edits. When the relationship between groups of
variables was checked for consistency, if there was an error, often
one or more of the variables was set to not-stated.
For derived variables if one or more of the input variables to the derived
variable had a refusal, don't know or not-stated code, then the derived
variable was set to not-stated.
NLSCY Data Users Guide
39
2001/2002
Coding of Open-ended Questions
Open-ended
Format
A few data items on the NLSCY questionnaire were recorded by
interviewers in an open-ended format. For example, in the Labour
Force Section, a PMK who had worked in the previous 12 months was
asked a series of open-ended questions about the current or most
recent job:
< What kind of business, service or industry is/was this?
< What kind of work are/were you doing?
< At this work, what are/were your most important duties or activities?
How they
are
recorded
The interviewer recorded in words the answer provided by the PMK. At
Head Office, these written descriptions were coded into industry and
occupation codes to describe the nature of the work of the PMK.
Similar information was collected for the spouse/partner and codes
assigned to describe the nature of the work.
How they
are coded
The coding systems used were the 1980 Standard Occupational
Classification codes (SOC) and the 1980 Standard Industrial
Classification codes (SIC). Grouped versions of these codes are
available on the data file (CLFPD07 and CLFPD08 for the PMK, and
CLFSD07 and CLFSD08 for the spouse/partner).
NLSCY Data Users Guide
40
2001/2002
Naming Imputation
Missing
Variables
Imputation
For various reasons there are certain variables that may be missing
for responding households on the NLSCY file. This is usually referred
to as item non-response. Earlier in the chapter the various codes that
have been used to describe the reason for the item non-response
("refusal", "don't know", "not stated") are described.
For some variables on the NLSCY file, however, rather than using a
special non-response code, imputation has been carried out.
Imputation is the process whereby missing or inconsistent items are
"filled in" with plausible values. For the NLSCY, imputation was
carried out for household income and PMK income.
Imputation flags have been included on the NLSCY file so that users
will have information on the extent of imputation and what specific
items have been imputed on what records.
All imputation flags on the NLSCY data file have an "I" as the fifth
character of the variable name. For example, the name of the
imputation flag for household income (CINHQ03) is CINHI03.
Derived Variables
Combining
Items
A number of data items on the data file have been derived by
combining items on the questionnaire in order to facilitate data analysis.
For example, in the Labour Force section, one of the question is on the
Number of Weeks Worked but in the Adult Education section, the
question is Whether They Are Presently Going To School. The
combination of these two questions forms a variable that is based on
the Actual Situation Of Work And Study.
Longitudinal Longitudinal derived variables were created to indicate changes
between data reported in the current and previous cycles for family
derived
structure and PMK and Spouse changes.
variables
NLSCY Data Users Guide
41
2001/2002
Derived
Variable
Name
All derived variables on the NLSCY data file have a "D" as the fifth
character of the variable name. The name of the variable for the
primary care arrangement is CLFPD51.
NLSCY Data Users Guide
42
2001/2002
Chapter 6 - Weighting
Estimation
Estimation
The principle behind estimation in a probability sample, such as the
NLSCY is that each respondent in the sample "represents," several
other persons in the general population. For example, generally
speaking, each child in the NLSCY sample represents about 300
children in the population.
The
“Weighting
Phase”
The weighting phase is a step which calculates how many people each
respondent represents. As the target population is not the same for the
cross-sectional sample and the longitudinal sample, the number of
persons each child represents is not the same. Consequently, two
series of weights must be calculated:
<
one for the cross-sectional sample,
<
one for the longitudinal sample.
These weights appear on the NLSCY data files (CWTCW01C) for
cross sectional weight, CWTCW01L for longitudinal weight), and must
be used to derive meaningful estimates of the characteristics measured
by the survey.
For example, to estimate the number of children living in single-parent
families in 1996 we would select the records in the cross-sectional
sample of Cycle 2 with that characteristic and sum the weights found on
those records.
NLSCY Data Users Guide
43
2001/2002
The Longitudinal Sample or Cross-sectional?
Choice of
Sample
Dependant
on Analysis
The choice of which sample to use depends on the type of analysis to
be done. The longitudinal sample pertains to the child population at the
time this sample was selected (i.e., 1994-95). The sum of the
longitudinal weights is equal to the available demographic estimates for
January 1995.
Longitudinal Only the longitudinal children, i.e., those selected in 1994, are given a
longitudinal weight other than 0. For each cycle, the longitudinal weight
Weight
of the panel is recalculated to take into account the further erosion (nonresponse) that occurs between the two cycles of the survey, i.e., about
two years. It is this one that is usually better suited to longitudinal
analysis based on a comparison of the data for more than one year, as
it allows for the life courses of the children to be quantified over time.
Crosssectional
Weight
Flows
The cross-sectional sample makes it possible to do estimates based
on data from a single cycle. A separate cross-sectional weight is
calculated for each cycle. For Cycle 1, the longitudinal sample and the
cross-sectional sample have the same target population. As the target
populations are identical, only one series of weights was needed for
this cycle.
Flows may be calculated using cross-sectional estimates produced for
two cycles. However, the flows thus measured are net flows. They are
calculated based on a snapshot taken for each reference period. As a
result, they mask all transitions that cancel each other out.
Here is an example to illustrate this phenomenon:
A researcher wishes to know whether the number of young people
who smoke increased between 1994 and 1996. He can therefore
calculate the number of smokers in 1994 using the Cycle 1 sample,
and a second estimate for 1996 using the cross-sectional sample
for Cycle 2. By comparing these two estimates, he can determine
whether the number of smokers increased or decreased. However,
this comparison conceals the fact that a number of young people
quit smoking in the interim. From this analysis, it would therefore
not be possible to verify whether a program designed to reduce the
number of young people who smoke is effective. Again using our
example, the cross-sectional sample would make it possible to
quantify each transition, and therefore to calculate the gross flows.
NLSCY Data Users Guide
44
2001/2002
Weighting Procedures for the Cross-sectional and Longitudinal Samples
NLSCY
Weighting
Strategy
The NLSCY weighting strategy is based on a series of cascaded
adjustments applied to a basic (or initial) weight. Conceptually, the
basic weight of each child is approximately equal to the inverse of the
child's probability of selection. In the case of the selected households
of the LFS in 1996, the basic weight was the sub-weight calculated
by this survey. For the longitudinal children, that is, those sampled in
1994, the basic weight was determined using the weight calculated
for Cycle 1. The final weight, cross-sectional or longitudinal, was
obtained by multiplying the basic weight by many adjustments.
This section explains the various corrections made to the basic
weight and the procedures used to weight the cross-sectional and
longitudinal samples
Weighting of
Longitudinal
Sample
We will discuss the longitudinal weighting process first, as it is the
simpler of the two. Furthermore, this weight is used later to
determine cross-sectional weight.
Two steps are involved in obtaining the longitudinal weight for
children selected in Cycles 1 and 2. These adjustment factors are
applied to the basic weight in order to obtain the final longitudinal
weight. As concerns the sample of children selected in Cycle 1, the
basic weight is the final weight, before post-stratification, obtained in
Cycle 1. With respect to the sample of children selected in Cycle 2,
the basic weight is the final weight, before post-stratification,
obtained in Cycle 2. For further information on these weights, please
consult the documentation from previous cycles.
Step One Adjustment
Factor
In step one, an adjustment factor is calculated that accounts for the
erosion (non-response) observed since the sample was selected.
As regards the sample of children selected in Cycle 1, this factor
corrects the erosion that affected this cohort in Cycles 2 and 3. With
respect to the cohort selected in Cycle 2, the factor inflates the basic
weight in order to mitigate the non-response for this cohort observed
in Cycle 3. These factors are determined by means of models.
NLSCY Data Users Guide
45
2001/2002
Homogeneous
Response
Group - HRG
Regardless of whether they responded in Cycle 3, a considerable
amount of information on these children was gathered during
previous cycles. The non-response correction strategy makes use of
this information. It is based on the homogeneous response group
(HRG) method, which involves an attempt to consolidate those
individuals with the same propensity to respond. These groups are
formed using the characteristics for each child reported in Cycle 1. A
correction factor is then derived for each HRG, as follows:
Sum of adjusted weights in the HRG
Sum of adjusted weights of respondents in the HRG
Two HRG Sets
Two distinct HRG sets were constructed: one for the sample of
children selected in Cycle 1, and another for the sample of children
selected in Cycle 2. Both sets are required, as these samples do not
necessarily react to the same non-response mechanism. As there is
every reason to believe that this mechanism changes in accordance
with the number of times an individual is surveyed, the non-response
adjustment model must take this fact into account. Lastly, the
constraints represented by adjustment-factor range and minimum
HRG size are imposed during HRG formation in order to obtain
reasonable, reliable correction factors.
Poststratification
The purpose of the second adjustment factor is to ensure consistency
between survey estimates and demographic estimates produced by
Statistics Canada. This method is known as post-stratification. For
the sample of children selected in Cycle 1, the target population is the
set of children aged 0 to 11 in early 1995. As a result, the poststratification adjustment for this sample ensures consistency between
the sum of the weights and demographic estimates from January
1995 for each combination of province, age and sex. As regards the
sample of children selected in Cycle 2, the adjustment is made using
demographic estimates from January 1997.
NLSCY Data Users Guide
46
2001/2002
Weighting of
the Crosssectional
Sample
As explained earlier, the cross-sectional sample is comprised of
children selected in 1994 and children selected in 1996. In the
following paragraphs, we present the correction factors which, when
applied to the basic weights, make it possible to calculate the
weights of the cross-sectional sample. These correction factors differ
according to whether the child was selected in 1994 or in 1996.
First of all, cross-sectional weights were calculated separately for the
children selected in 1994 and those selected in 1996. Thereafter,
each of these two components represents its respective target
population. However, these target populations are not entirely
separate. It is therefore necessary to apply other correction factors
to take this overlap into account. The purpose of the last step (poststratification) is to ensure consistency between survey estimates and
demographic estimates produced by Statistics Canada.
NLSCY Data Users Guide
47
2001/2002
Cross-Sectional Weights for Children Surveyed for the First Time in Cycle 3
Children
Selected from
the Labour
Force Survey
(LFS) Sample
The weighting strategy applied to these children is similar to that used
in Cycles 1 and 2.
1st Correction
For Number
of Rotation
Groups
The LFS sample is made up of six “rotation groups”, each of which is
a representative sub-sample of the LFS target population. In the
NLSCY, we used 13 rotation groups. Consequently, the first
adjustment is 6/13. Further to this adjustment, the adjusted weight is
obtained by multiplying the LFS weight by 6/13.
2nd Correction
For
Household
NonResponse
In surveys such as the NLSCY, some households do not provide
responses1 for a variety of reasons: refusal, special circumstances,
language problems, temporary absence. This non-response is usually
compensated for by proportionally correcting the sub-weights of the
responding households. The correction is made by multiplying the
sub-weight of the responding households by the following factor:
Sum of adjusted weights of households sampled within a stratum of the NLSCY
Sum of adjusted weights of responding households within a stratum of the NLSCY
In this equation, the adjusted weight is the weight obtained after
Correction 1. A different correction was made in each of the strata
specially defined for non-response by the LSF. The strata are defined
using the following information: province, economic region, census
metropolitan area, type of sector (urban, rural), apartment frame,
whether special region or not. Each of the strata has at least 30
children and a response rate of at least 70%. 2 Strata that are too
small or have a response rate of less than 30% are grouped together
until these restrictions are met.
1
Following the survey, it is possible that information is gathered only for one child in a household, although two
children are in the sample. According to the NLSCY release strategy, both these children are considered
respondents, as we have considerable information about their parent(s). For this reason, it is not necessary to apply
a correction factor for the non-response of the children.
2
These restrictions are designed to ensure that the adjustment factor is relatively stable and not too large.
NLSCY Data Users Guide
48
2001/2002
3rd Correction
For
Households
With More
Than One
Economic
Family
Sometimes a household includes more than one economic family.
When this occurs, the child selection procedure requires the selection
of one of these families at random. This correction is the inverse of
the selection probability of the family in the household in question.
This correction affected only two households.
4th Correction
For
Households
With More
Than Two
Eligible
Children
For Cycle 3, a maximum of 2 children were to be interviewed in the
new households. If the economic family has more than 2 eligible
children, 2 children are chosen at random. This correction takes this
selection process into account in economic families, and affects only
5 households, as very few households have more than two children
under 1 year of age.
This is the last correction to be calculated for these children before
weight integration.
NLSCY Data Users Guide
49
2001/2002
Weighting of Children Sampled in 1994 and 1996
Weighting of
Children
Selected
from the
Birth
Register in
1998
Two correction factors are required for these children. The first inflates
the survey weight in order to account for the non-response observed
during data collection in Cycle 3. For this adjustment, the
homogeneous response group method is used once more. However,
as there is little information on non-responding households from this
sample, HRGs in this particular case correspond to the strata used to
select the birth-register sample.
The second adjustment calculated for these children accounts for the
fact that we interviewed twins. The basic weight was modified for
couples of twins, as households with twins have a higher probability of
selection than those with just one eligible child.
Weighting of
Children
Sampled in
1994 and
1996
It is not necessary to apply all the corrections described in the previous
section to these children, as this was done in Cycles 1 and 2. The
basic weight we use is therefore the weight obtained in previous
cycles after the adjustment for non-response and before poststratification. Only two corrections were necessary for these children.
1st Correction
The first correction inflates the basic weight in order to account for nonresponse. The adjustment used in this step is identical to that
calculated in determining longitudinal weight.
NLSCY Data Users Guide
50
2001/2002
2nd
Correction
The second adjustment attempts to minimize the impact of rare
interprovincial migrations. Some children selected in 1994 or 1996
had moved or changed province since the first interview. This can
sometimes distort weights for the new province of residence. For
example, the weight of a child selected in Ontario is far greater than
that for a child selected in Prince Edward Island. When a child
selected in Ontario moves to Prince Edward Island, this will have an
enormous impact on the estimates for Prince Edward Island if he/she
retains his/her original weight. This type of migration is very rare
among the target population. In this context, it is not reasonable to
assume that the sampled child who has moved from Ontario to Prince
Edward Island represents a large number of children in the target
population who have followed the same life course. Rather, such a
case should be considered uncharacteristic. The weight of these
children has therefore been corrected downward
Weight
Integration
Using the three weight calculation methods presented in the previous
sections, it is possible to produce estimates for their respective target
population. In some cases, however, these target populations are not
mutually exclusive. It is therefore necessary to derive a correction
factor that takes this overlap into account. In addition, one final factor
is needed to ensure that these weights produce estimates consistent
with the demographic estimates produced from other sources.
Correction
For Overlap
Of Target
Populations
We are dealing with three types of households: those selected in
Cycle 1, those selected in Cycle 2, and those selected in Cycle 3.
However, the target populations for these three samples overlap in the
cases of children selected in 1994, who are now 5 years of age, and
children aged 5 selected from the birth register in 1998. This overlap
must be taken into account in order to ensure that our sample does not
systematically overestimate the characteristics of the population.
NLSCY Data Users Guide
51
2001/2002
Multiplier
Factor
To take the relative contribution of each sample into account, we
calculated a multiplier factor for each province. If an optimal
combination of these samples is to be obtained, this factor must take
into account the accuracy of each sample’s estimates. For example,
an estimate from a highly accurate sample is considered more
important than that from a sample of low accuracy. Accordingly, the
former would have a high adjustment factor, and the latter, a low
adjustment factor.
An example will illustrate this approach. Let us suppose that 30
longitudinal children aged 5 years were sampled in New Brunswick in
1994 and 10 children from the same age group were selected from the
birth register in 1998. Moreover, suppose the design for the birth
register sample is twice as effective 1 as that for the 1994 sample. In
this case, the correction factor for the longitudinal children would be:
30
2
30 +10 = 0.6, while the correction factor for the birth register
sample would be 0.4.
2
(
)
Note that the sum of the two adjustment factors is 1.
Correction
For PostStratification
Post-stratification was carried out on the weights thus far to ensure that
the national and provincial estimates agreed with the January 1997
demographic estimates of the population of children aged 0 to 13. For
Cycle 3, post-stratification was done by province, age group and sex.
This correction factor was derived for each post- stratification, as
follows.
Demographic estimate
Sum of weights in the post-strata
This correction ends the weighting process of the cross-sectional
sample for the second cycle of the NLSCY.
1
In this context, a sample is more effective if its sampling variance is smaller than that of another sample of equal size
selected using a different sampling design.
NLSCY Data Users Guide
52
2001/2002
Chapter 7 - NLSCY Concepts and Definitions
Introduction
There are many variables and concepts which are critical to the analysis
of the NLSCY data. In this section there is a brief discussion regarding
the types of possible analyses with the NLSCY data. This is followed by
a description of key variables which have been derived to explain the
living arrangements of the child and the socio-economic conditions
under which the child lives.
The content areas for each section of the various questionnaires used
for the first cycle of the NLSCY are presented in the next section.
Cross-sectional and Longitudinal Estimates
NLSCY
Design
The NLSCY design and sample has been constructed so that it will be
possible to produce both cross-sectional and longitudinal estimates. At
present, it is possible to obtain cross-sectional estimates with Cycle 1,
Cycle 2 and more recently with Cycle 3 data. It is also possible to obtain
longitudinal information from the longitudinal file.
The allocation of the Cycle 1, 2 and 3 sample was such that it is
possible to produce estimates at the national level for the specific age
cohorts and at the provincial level for aggregated age groups. This is
true for cross-sectional data as well as longitudinal data.
NLSCY Data Users Guide
53
2001/2002
There are two longitudinal cohorts, those who were in the sample
Longitudinal beginning with Cycle 1 (aged 0-11 years at cycle 1) and those who
were in the sample beginning in Cycle 2 (0-1 year olds). The Cycle 1
Cohorts
longitudinal sample is comprised of all children sampled for Cycle 1 of
the survey in responding households (excluding those from the
integrated sample (NPHS) and the 3rd and 4th child of each family).
The plan is to follow these children over time, and revisit them every two
years. Analyses of these children will permit researchers the opportunity
to perform in-depth studies of the long-term impact of risk factors (such
as divorce or the onset of a health condition) and protective factors
(such as positive interactions with parents or academic success at
school) on these children as they move into adulthood. If a child moves
out of the household where he or she was sampled at Cycle 1, that child
will be traced to wherever he or she resides during future cycles of the
survey. From a longitudinal perspective, the child, not the household, is
the statistical unit of analysis.
Attrition
It should be noted that some children who were participants in Cycle 1
of the NLSCY did not participate in the second cycle or may not
participate in subsequent cycles due to a variety of reasons. This is
usually referred to as attrition. The number of these children is being
carefully monitored and we are making every effort to keep these
numbers at a minimum. The Cycle 1 sample and its allocation were
designed with this in mind and as long as future response rates are not
lower than expected the sample will still permit longitudinal research by
age cohort at the national level.
Augmenting
the Sample
In the second and subsequent cycles, it is intended that the NLSCY will
add children belonging to age groups no longer covered in the
longitudinal sample. For example, for Cycle 3 a panel of children 0 and
1 year of age was added to the Cycle 3 sample. This augmented
sample will allow for ongoing cross-sectional analyses to supplement
the primary longitudinal research. As such, at each cycle it will be
possible to get a snapshot of Canadian children of all ages. At the
present time, it is not planned to follow this augmented component of
the sample longitudinally, or if so it will be done on a limited scope.
NLSCY Data Users Guide
54
2001/2002
Children
Who
Immigrate to
Canada
It should be noted that the children who immigrate to Canada at any
point in time after the Cycle 1 sample was selected and who are in the
age cohorts covered in the Cycle 1 sample, will not be included in either
cross-sectional or longitudinal estimates at this time. The number of
children excluded by this criterion is small. Estimates of the number of
children immigrating to Canada will be monitored and a decision may
be made in the future to introduce a new sample into the NLSCY to
cover these children.
NLSCY Units of Analyses
Unit of
Analysis Child
The unit of analysis for the NLSCY is intended to be the child and
eventually the young adult. For each cycle of the NLSCY, extensive
information will be gathered on the child’s family, parent(s), and
neighborhood.
Defining
It is true that families or households are relatively straightforward units of
Longitudinal analysis with cross-sectional data but the situation becomes
Households problematic with longitudinal data. Households change composition
frequently, due to divorce of parents or children leaving the parental
nest. Attempts have been made in other studies to define “longitudinal
households”, but the implementation of this concept has never been
straightforward. No single definition has been found to be appropriate
for most analytic tasks, and many definitions exclude the portion of the
population that has undergone the change. Unfortunately, this is often a
significant as well as interesting population to study. It has been
suggested that a superior alternative is to use the individual as the unit
of analysis and present family and household variables as a
characteristic of the individual.1
Thus the files which have been constructed for all NLSCY data consist
of child records. In order to understand the family situation, estimates
such as of the number of children in single parent families, or the
number of children living in low-income households, can be produced.
1
For a more detailed examination of units of analysis in longitudinal studies, see G.D. Duncan and M.S. Hill,
“Conceptions of Longitudinal Households: Fertile or Futile?” Journal of Economic and Social Measurement (1985)
13: 361-375.
NLSCY Data Users Guide
55
2001/2002
PMK and Spouse
Person Most
Knowledgeabl
e
In each NLSCY household for Cycle 3, for each selected child, a
question was asked about who in the household was the person most
knowledgeable about this child. This person was labeled as the PMK.
The intention was that the PMK would provide the information for all
selected children in the household and then give socio-demographic
information about himself/herself and him/her spouse/partner. In some
rare cases it might have been appropriate to label two different
people in a household as PMKs. For example, in the case of a step
family, it may have been appropriate to label the mother as the PMK
for one child and the father for another. However, in order to simplify
the interview procedures, only one PMK was selected per household.
Relationship
of PMK to
NLSCY
Children
The following is the breakdown of the relationship of the PMK to the
NLSCY children for Cycle 3.
For 93.0 % of responding children, the PMK was the mother (92.1 %
the biological mother and 0.9 % the step, adoptive or foster mother)
For 6.4 % of the children the PMK was the father for 0.6 % of children
the PMK was not a parent.1
Cases Where
the PMK was
not the Parent
For the majority of cases of the PMK not being a parent, the child had
a parent living in the household, but the parent was not selected as the
PMK. For the most part this situation occurred when a child had a very
young mother living with her own parents, i.e., the child's
grandparents, and the grandmother was selected as the PMK
Spouse/Partne
r as PMK
If the PMK had a partner residing in the household at the time of the
interview, then this person was labeled as the spouse. Spouses
included both married and common-law partners. Detailed socioeconomic information was collected about the spouse/partner in order
to describe the family situation of the child
See the table below.
1
These numbers for the PMK and spouse/partner are based on unweighted data.
NLSCY Data Users Guide
56
2001/2002
The following is the breakdown of the relationship of the spouse/partner to the children.
For families where the PMK (’s)...
The % was...
did not have a spouse/partner residing in the household
14.8
Spouse/partner was the father
78.9
was the biological father
74.8
was the step, adoptive or foster father)
4.1
spouse/partner was the mother (biological, step, adoptive or foster
5.4
spouse/partner was not a parent
0.9
Change in
PMK
Between
Cycles
For several reasons, the PMK and his/her spouse could be different
people than those designated in the first and second cycles. For this
reason, a variable flagging the change in individual on the longitudinal
file was created (see CDMPcD27 for the PMK change and
CDMScD28 for the change in spouse). This new variable indicates
whether there was any change in the PMK from one cycle to the other. It
is highly recommended that this variable be used when doing
longitudinal analyses involving the characteristics of the parents.
Here is a breakdown of the consistency of the relationship between the NLSCY children
and the PMK and his/her spouse:
For families where the PMK was ...
The % was
the same person in both cycles
91.6
the spouse of the PMK in Cycle 2
7.1
a new individual
1.2
For families where the PMK (’s)...
The % was
had no spouse living in the household for either of the two survey cycles
10.7
spouse was the same person for both cycles of the survey
74.4
spouse of the PMK for Cycle 3 had been the PMK for Cycle 2
6.4
had no spouse for Cycle 2, but did have a spouse for Cycle 3
4.6
had no spouse for Cycle 2, but did have a spouse for Cycle 3
3.2
was the same person for both survey cycles, but had a different spouse
0.7
NLSCY Data Users Guide
57
2001/2002
NLSCY Data Users Guide
58
2001/2002
Family Derived Variables
Relationship
Grid
Using NLSCY data, a child's family may be described in several
different ways. Many of the family variables used to describe the
NLSCY children were derived from what is known as the relationship
grid. As part of the household roster some basic demographic
information was collected for all members of the child's household.
As part of this questionnaire, the relationship of everyone in the
household to everyone else was asked. Using this information it was
possible to create an extensive set of variables to describe the
child's family situation.
The following are some of the family derived variables for the children
that exist on this second micro data file for the NLSCY. The names of
the derived variable are given in brackets.
Single Parent
Families
There are two ways to describe the parental situation of children
using NLSCY data.
Using the relationship grid, a child's single-parent status was derived.
There were 84.4% of children living with two parents, 15.4% with one
parent and 0.2% without a parent1 (CDMCD04).
A child's parental status can also be defined in terms of the PMK.
There were 85.2% of the NLSCY children living in a household where
the PMK had a spouse/partner; and for 14.8% of children the PMK
did not have a spouse/partner (CDMPD06A).
The two ways of describing the child's family are very similar. The
only reason for the small differences is a result of the few cases
where the child lived with a parent, but the parent was not selected to
be the PMK.
1
These estimates for family derived variables are based on unweighted data.
NLSCY Data Users Guide
59
2001/2002
Step, Blended
and Intact
Families
Children living with two parents are classified as being members of
intact, step and/or blended families based on the relationship of
these children to the parents.1
Intact Family
An intact family consists of a married or common-law couple in which
all children are the natural and/or adopted offspring of both members
of the couple.
For the NLSCY children, 75.9% were a member of an intact family
(CDMCD16). For the NLSCY children, 4.3% were step children
themselves (CDMCD03) and 8.5% lived in a step family
(CDMCD15).
Step Family
A step family consists of a married or common-law couple residing in
the same household, with at least one step child living with them who
is the biological or adopted child of one parent but not the other. It
should be noted that a child who is the biological child of both parents
is said to belong to a step family if at least one of these parents has a
step child residing in the household.
For the NLSCY children, 4.3% were step children themselves
(CDMCD03) and 8.5% lived in a step family (CDMCD15).
1
Foster children and children living with only one parent are not included in step, blended or intact families. In
the derivation of blended, intact and step families, if a child was the adoptive child of one parent and the
biological child of the other parent, then this child was treated like a step child, and thus the family labelled as a
step family. In other Statistics Canada publications children of this type are treated as if they were biological
children of both parents.
NLSCY Data Users Guide
60
2001/2002
Blended
Families
Blended families combine children who have different relationships
with their parents. A blended family consists of a married or
common-law couple living with at least two children, one of whom
does not share the same natural and/or adoptive parents as the other
child(ren). The following are examples of blended families:
<
a couple with biological children of the female partner as well
as biological children of the male partner (i.e., hers and his)
<
a couple with biological children of the female partner as well
as children out of the new union (i.e., hers and theirs).
The blended family is a sub-set of the step family. For the NLSCY
children, 6.2% were members of a blended family (CDMCD14).
Economic
Family
For the NLSCY, an economic family is defined as all family members
related by blood, marriage, common-law relationship or adoption.
Foster children are considered to be part of the economic family. For
example, if a woman lives in a household with her spouse and two
children as well as her sister and her sister's child then all of these
individuals would be part of one economic family. If a boarder also
resided in the household with her child then this would constitute a
second economic family.
Siblings
For the NLSCY data, siblings include full, half, step, adopted and
foster siblings. Only siblings residing in the household have been
included in the calculation of the sibling derived variables included on
the micro data file. In the case of common-law relationships, if both
members have brought their own children into the relationship then
these children are considered siblings. It should be noted that the
classification of siblings was age independent. If an NLSCY child had
an adult sibling (for example, 21 years of age) living in the household
then this sibling was included in the calculation of the sibling derived
variables. The sibling derived variables include total siblings, as well
as number of older siblings, younger siblings and siblings of exactly
the same date of birth (i.e., twins) (CDMCD08, 09, 10 and 11).
Socio-Economic Derived Variables
NLSCY Data Users Guide
61
2001/2002
Derived
Variables
There were two derived variables produced from Cycle 1 data to
assist analysts in understanding and explaining the socio-economic
situation of the child's family: socio-economic status, and income
ratio.
In the second and third cycle of the survey, two distinct measures of
socio-economic status were calculated: one longitudinal, and one
cross-sectional. The derivations of cross-sectional SES and of
longitudinal SES differ only with respect to the standardization of the
components. The derivation of the non-standardized components of
SES (i.e., parents' education level, parents' occupational prestige
and household income) was the same for both SES measures.
SocioEconomic
Status
Sociologists often use the term "socio-economic status" (SES) to
refer to the relative position of a family or individual in an hierarchical
social structure, based on their access to, or control over, wealth,
prestige and power. In studies of children's academic and socialemotional development, SES is often operationally defined through
measures describing the occupational prestige, educational levels,
and economic positions of children's parents.
The measure of SES is calculated for each household assigned to
each selected child in that household.1 It was derived from five
sources: the level of education of the PMK, the level of education of
the spouse/partner, the prestige of the PMK's occupation, the
prestige of the occupation of the spouse/partner, and household
income. The method of constructing each component of SES, and
the construction of the overall cross-sectional and longitudinal SES
measure are described below.
1
This particular definition of SES was proposed by Dr. Douglas Willms, Atlantic Centre for
Policy Research in Education. University of New Brunswick.
NLSCY Data Users Guide
62
2001/2002
Education Years of
School
The education variable used in the construction of SES was years of
schooling. Two such variables were derived independently; one for
the PMK and one for the spouse/partner (CEDPD04 for the PMK and
CEDSD04 for the Spouse/partner). For the PMK the years of
schooling variable was derived based on items CEDPQ01 (years of
elementary and high school) and CEDPQ04 (highest level of
education attained beyond high school). To create a somewhat
continuous interval-level education variable, these two items were
recoded to form years of schooling in the following manner:1
CEDPD04
00
03
06
07
08
09
10
11
12
13
16
18
20
Condition
CEDPQ01=1 (no schooling)
CEDPQ01=2 (1 to 5 years)
CEDPQ01=3 (6 years)
CEDPQ01=4 (7 years)
CEDPQ01=5 (8 years)
CEDPQ01=6 (9 years)
CEDPQ01=7 (10 years)
CEDPQ01=8 (11 years)
CEDPQ01=9 (12 years)
CEDPQ01=10(13 years)
CEDPQ04=6 (BA/BSC)
CEDPQ04=7 (Masters)
CEDPQ04=8 or 9 (MD/PHD)
An extra year was then added to CEDPD04 if the PMK had a
diploma from a trade school or community college (i.e., if
CESPDQ04= 4 or 5 then CEDPD04 = CEDPQ04+1).
The same procedure was used to set up a years of schooling
variable for the spouse/partner (CEDSD04).2
1
In cases where the PMK had not graduated from high school but had completed a post-secondary
degree or certificate, then the post-secondary degree or certificate took precedence. For example, if the PMK had
completed only grade 10, but had masters, then AEDPD04 was set to 18.
2
It was decided that years of schooling was an interesting derived variable itself and therefore this
variable has been included on the NLSCY master file for the PMK and spouse/partner (CEDPD04 and CEDSD04).
NLSCY Data Users Guide
63
2001/2002
Occupational
Prestige
Occupational status is an important indicator of SES. The occupation
variable used in the derivation of SES was a modified version of a
scale developed by Pineo, Porter and McRoberts (1977). The
classification system groups occupations described in Statistics
Canada's 1980 Standard Occupational Classification into 16
somewhat homogeneous categories, ordered from 1 to 16, where
code 1 represents the highest level of occupation and code 16 the
lowest. The 16-category scale provides a ranking of occupations
according to their social standing or prestige. For the NLSCY, for
both the PMK and the spouse/partner, a detailed description was
taken of the job considered to be his or her main job during the
previous 12 months. The information was used to code occupations
into the 1980 classification, and in turn into the 16 prestige
categories. For the purposes of deriving both SES, the order of the
Pineo-Porter-McRoberts scale was reversed. The final scale used in
the derivation of both SES had the following values:
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
96
99
NLSCY Data Users Guide
Farm labourer
Unskilled manual
Unskilled Clerical/sales/service
Semi-skilled manual
Semi-skilled clerical/sales
Farmer
Skilled crafts and trade
Skilled clerical/sales/service
Foreman/forewoman
Supervisor
Middle manager
Technician
Semi-professional
High-level management
Employed professional
Self-employed professional
Not-applicable - this was assigned for the
Spouse/partner for cases where the PMK did not have
a spouse/partner
Not stated
64
2001/2002
This ordinal scale can be used to rank individuals into the
various occupation groups but one cannot assume that the
intervals between ranks are equal. For example, in this scale a
middle manager (code 11) is ranked higher than a supervisor
(code 10), which in turn ranked higher than a foreman (code
09). However, this does not imply that the difference in
occupation between the middle manager and a supervisor is
equivalent to the difference between a supervisor and a
foreman. By assuming that the underlying latent construct has
a particular distribution, one can assign intervals to the various
categories. Mosteller and Tukey (1977) propose a logit
transformation to re-express ordinal data on an interval scale.
To do this, the percentage of individuals in each occupation
group is considered a piece of the logistic distribution. The
code assigned to each occupation is the centre of its piece in
the logistic distribution. This transformation was employed to
scale the 16 occupations.
For each occupation group x, the following values were
computed:
p
=
the percentage of individuals with an
occupation less than occupation x
(based on the Pineo-Porter-McRoberts
category)
pp
=
the percentage of individuals with an
occupation less than or equal to
occupation x (based on the Pineo-PorterMcRoberts category)
phi(p) =
p*ln(p) + (1-p)*ln (1-p)
phi(pp) =
pp*ln(pp) + (1-pp)*ln(1-pp)
The recoded (logit) value for occupation x was assigned to be:
PINEOLOG=phi(pp) - phi(p)
pp-p
PINEOLOG (for both the PMK and spouse/partner) was then used in
the derivation of both SES.
NLSCY Data Users Guide
65
2001/2002
Household
Income
The last variable used in the derivation of SES was household
income. More detail regarding the collection of household income
and data quality issues can be found in Section 9.17. To derive SES,
income was coded in $1,000s of dollars, and a few outliers with
incomes greater than $150,000 were recoded to $150,000.
Final
Derivation of
Crosssectional and
Longitudinal
SES
Thus the five variables that were used to derive both SES were:
Final
Derivation of
Crosssectional SES
Each of the five variables was standardized to have a mean of zero
and a standard deviation of one.
- CEDPD04 (years of schooling for the PMK),
- CEDSD04 (years of schooling for the spouse/partner),
- PINEOLOG-PMK (the pineo occupation code for the PMK
transformed to the logit distribution),
- PINEOLOG-SP (the pineo occupation code for the
spouse/partner transformed to the logit distribution) and
HHINC (household income in thousands of dollars)
NLSCY Data Users Guide
66
2001/2002
Consideration
of Missing
Data for the
Derivation of
CrossSectional SES
In the case of cross-sectional SES, the components were
standardized using the means and standard deviations of the
variables for all households as observed in Cycle 3. Thus, new
standards were established based on the data for Cycle 3 families
with selected children aged 0 to 15. Given the change in age of the
selected children between Cycle 3 and Cycle 2 (0-13 years) it is
expected that our sample allowing for the production of Cycle 3
standards consists of slightly older families. This characteristic
difference is of some importance, as older families are generally
expected to present more favourable socio-economic characteristics
than younger families. From one cycle to the next, this difference
might not be felt, but over the long term (or over several cycles),
differences will likely be noticeable. The income variable which is
utilized to derive SES is expressed in current dollars. Thus, the costof-living increase and the subsequent adjustment of salary and
income level will also, over the long term, have a significant impact on
the value of the means and standard deviations used to standardize
the components of cross-sectional SES. The variable for crosssectional SES is labeled CINHD08.
Final
Derivation of
Longitudinal
SES
The final derivation of longitudinal SES is based on the standards
calculated for the first cycle of the survey. The same raw values of the
components helpful in deriving cross-sectional SES are used, but the
standardization differs in this way. Thus, unlike cross-sectional SES,
the standardization is not expected to produce a mean of zero and a
standard deviation of one for each of the variables. By definition, the
use of longitudinal SES is relevant only for analyses based on
longitudinal children.
Initial
Standards of
the First Cycle
The initial standards of the first cycle which were used to derive
longitudinal SES were created based on the characteristics specific
to households having children aged 0 to 11. These same families, in
the third cycle of the survey, have children aged 4 to 15. The value of
longitudinal SES therefore allows us to calculate the net progression
of each child in relation to the initial characteristics of his/her
household.
NLSCY Data Users Guide
67
2001/2002
Values of SES
A child living in a household where the income has improved
appreciably (all things being equal), will see the value of his/her
longitudinal SES improve as well. However, in the same
circumstances, the value of cross-sectional SES may decline. This
would be the case, notably, if all children were living in households
that experienced on average an improvement in socio-economic
status.
Two SES
Values
It is therefore essential to be familiar with the rules used to derive the
two SES values in order to use the variables properly in the analyses.
The differences observed from one cycle to the other for the
standards of both SES are not yet very pronounced. Therefore, the
use of one measure rather than another, in the short term, should not
produce significant differences in research results. But over the long
term, the proper use of both measures should become more
important. Normally, it is recommended that cross-sectional SES be
used to accurately measure the relative position of a child in relation
to other children in a given cycle, whereas the use of longitudinal SES
provides a better indication of the progression of an individual's
situation from one cycle to the other.
NLSCY Data Users Guide
68
2001/2002
Consideration
of Missing
Data
Missing values (i.e., not-stated values) were ignored in the
standardization. In the standardization of the spouse/partner
variables (CEDSD04 and PINEOLOG-SP), if the PMK did not have
a spouse/partner these records were ignored. The SES composite
was then calculated by taking the (unweighted) average of the five
standardized variables. If one of the five variables had missing data
due to non-responses (refusal, don't know, etc.) then the average was
taken over the remaining non-missing items. If there was no
spouse/partner in the household (i.e., the PMK had no
spouse/partner) then the average was taken over the three
applicable variables (CEDPD04, PINEOLOG-PMK, and HHINC).1
For two-parent families (i.e., for cases where there was a PMK and a
spouse/partner), if two or more out of the five input variables were
missing, then SES was set to "not-stated." For single-parent families
(i.e., there was no spouse/partner), if one or more out of the three
input variables were missing, then SES was set to "not-stated."
Examples of
SES
The values for SES range from -2.000 to +1.750. The distribution of
SES scores is as follows for children on the file.
SES SCORE RANGE
% CHILDREN WITH SCORE IN RANGE
Cross-Sectional
Longitudinal
1.5 or over
1.0 to less than 1.5
1.92
5.44
2.17
5.13
0.5 to less than 1
10.53
11.03
0 to less than 0.5
21.28
24.77
-0.5 to less than 0
30.06
30.11
-1.0 to less than -0.5
17.74
15.2
-1.5 to less than -1.0
7.68
4.4
Less than -1.5
3.35
1.11
Not-stated
2
6.08
Note: These numbers are based on unweighted data.
1
With this procedure, the SES score for single-parent families will tend to be lower because household
income, on average, will be lower. However, the SES score will properly reflect the level of education and the
occupational prestige of the single parent. Nevertheless, for most regression analyses where SES is used as a
control variable, it would be useful to include a dummy variable denoting whether the family was a single- or twoparent family.
NLSCY Data Users Guide
69
2001/2002
Types of
Families
SES SCORE
Crosssectional
1.5
0.5
0.0
-0.5
-1.0
-1.5
-2.0
In order to give a flavour for the types of families associated with
various SES scores the following examples are given for illustration
purposes. It should be noted that the SES scores given in these
examples are approximate and do not correspond to actual records
on the NLSCY file. Many more examples are possible for each score
involving both one and two parent
families.
EXAMPLE
A Family in Which...
!both the PMK and spouse have a university degree (BA/BSC)
!they are both employed professionals
!the household income is $80,000
!the PMK has a university degree (BA/BSC) and the spouse has
grade 13
!the PMK is employed as a semi-professional and the spouse is
employed in a semi-skilled clerical position
!household income is approximately $65,000
!the PMK has grade 13 and the spouse grade 12
!the spouse is employed in a semi-skilled manual position and the
PMK has a semi-skilled clerical position, is not in the labour force
!household income is approximately $55,000
!the PMK and spouse have both completed grade 12
!the PMK is employed in a semi-skilled manual position and the
spouse in an unskilled manual position
!household income is approximately $30,000
!neither the PMK nor the spouse have completed high school
!the PMK is employed in an unskilled manual position and the spouse
is employed in an unskilled manual position
!household income is approximately $25,000
!neither the PMK nor the spouse have completed high school
!neither the PMK nor the spouse are in the labour force
!household income is approximately $15,000
!there is no spouse
!the PMK has not completed high school
!the PMK is not in the labour force
!the household income is less than $10,000
NLSCY Data Users Guide
70
2001/2002
Chapter 8 - Content and Validation of NLSCY
Introduction
The NLSCY was designed to follow an ecological or holistic
approach to measuring child development. The survey captures the
diversity and dynamics of the factors affecting children. To ensure
that all relevant topic areas affecting child development were
adequately addressed by the survey, a multidisciplinary consultation
was carried out at the inception of the survey. The selection of
specific subject areas, priorities and survey questions was very
much a group effort with input and advice from:
- the NLSCY expert advisory group which consists of researchers in
the area of child development and the social sciences;
- federal departments;
- representatives from the provinces and territories responsible for
child development programs.
Factors
Affecting Child
Growth
It was recommended that the NLSCY cover a broad range of
characteristics and factors affecting child growth and development.
Extensive information was gathered about the child, as well as the
child's parent(s), characteristics of the family and the
neighbourhood. This section provides an outline of the content for
each section of the questionnaire included in the NLSCY data.
NLSCY
Processing
System
As part of the NLSCY processing system, there were some basic
quality checks performed for each section of the questionnaire. Any
items for which there was a high level of non-response or which were
frequently involved in edit failures were looked at in detail. Where
appropriate, comparisons were made to external data sources and
analyses were carried out to investigate possible reasons for
differences from these other sources. Any concerns about potential
data quality problems for any items in a particular section of the
questionnaire are discussed in this section of the documentation.
NLSCY Data Users Guide
71
2001/2002
General
Validation
Procedures
Before the section-by-section discussion of content and validation
results, the general validation procedures used for the "scale" data
are presented. PLEASE NOTE THAT MOST SCALES WERE
DEVELOPED AND VALIDATED IN CYCLE 1. IN SUBSEQUENT
CYCLES, THE SAME FACTOR STRUCTURE WHICH EMERGED
FROM THE CYCLE 1 ANALYSIS WAS IMPOSED. THIS
ENSURES THAT THE SCALES ARE CONSISTENT ACROSS
TIME TO ALLOW FOR LONGITUDINAL ANALYSIS AND CROSS
SECTIONAL COMPARISONS. IN THE SECTIONS, DESCRIBING
THE VALIDATION OF THE SCALES, WHERE THE ANALYSIS
WAS DONE USING CYCLE 1 DATA, THE VARIABLES CITES
WILL BEGIN WITH AN “A” (I.E. ABECB10). LIKEWISE WHEN THE
SCALE IS NEW IN CYCLE 3, THE SCALE ANALYSIS WILL CITE
CYCLE 3 VARIABLES
Validation of Scale Data
Scale Definition
For some of the concepts deemed important to measure in the
NLSCY it was decided that the concept would most appropriately be
measured through the use of a scale. A scale is simply a group of
questions or items that measure a certain concept when the
answers to the items are put together.
For example, on the child’s questionnaire it was determined that it
was important to have an assessment of certain parenting
behaviours. The Parenting Scale that was employed was one that
was proposed by Dr. M. Boyle at Chedoke-McMaster Hospital,
based on work by Dr. Ken Dodge (Vanderbilt University) which was
an adaptation of Strayhorn and Weidman’s Parent Practices Scale.
The scale is intended to measure three different constructs or
factors related to parenting; positive interaction, hostile/ineffective
parenting and consistent parenting.
NLSCY Data Users Guide
72
2001/2002
Scales and
Calculations
For each factor measured by a scale, a score is calculated. The
score for a particular factor can be used to give an ordering of
individuals. For example, for the Parenting Scale, for children with
higher scores for the “positive interaction” factor, the PMK reported
having more positive encounters with the child (e.g., laughed with
them more, praised them more etc.). The score for a particular
factor is usually based on a series of items, since one single item
usually cannot measure the factor or construct with adequate
precision.
During the development of the NLSCY, when consideration was
being made of what specific scales should be used to measure a
particular concept, scales were as much as possible selected that
had been used in other studies where the psychometric properties
of the measures produced by the scale were available with
complete references.
Evaluation of
Scale Data
In many instances, the wording of certain questions was modified
and in some cases new questions were added. Sometimes the
scale that was used had not previously been used for children in
Canada, or had only been used for very small samples. Given these
concerns and further concerns regarding interviewing conditions, it
was felt that the factor structures of the scales used in the NLSCY
could be different from the ones given in the literature. Therefore the
project team felt the need to carry out an extensive evaluation of the
scale data to ensure that the psychometric properties found in other
studies also held true for the NLSCY experience.
There were three major steps in the analyses of the scale data. First
a new factor analysis was performed on all scales to determine the
constructs or factors inherent in each scale. Then scale scores were
calculated based on this factor structure. Finally reliability measures
were produced. The general procedures followed for each of these
steps are described in detail in the following pages.
NLSCY Data Users Guide
73
2001/2002
Factor Analyses
Factor Analysis
for Scales
The factor structure of each scale was determined based on data
from the first cycle. The factor structure imposed on the scales
already used in the first cycle and repeatedly utilized in the second
cycle of the survey was the result of analyses of data from the first
cycle.
The following is a summary of the procedures used in the factor
analysis for each scale.
1/ The sample of respondents for each scale (and age group, if the
scale used different questions for different groups), was randomly
divided into two half-samples. This was done to find out whether
different samples would yield the same results.
2/ Principal component analysis was carried out separately on each
half-sample to find out how many factors should be extracted in the
factor analysis performed subsequently. In principle, the same
number of factors as was found in the literature was expected. In
practice, however, some scales showed a different number of
factors because in some cases factors combined while in others
new factors emerged.
3/ Factor analysis was done on each half-sample and the factor
structure and loading of each factor were compared across the halfsamples.
4/ In the factor analysis, the items for each child in the appropriate
age group were used, multiplied by the child's normalized weight.
An individual's statistical weight is normalized by dividing his/her
weight (AWTCW01) by the average weight for all individuals. Thus,
the sum of the normalized weights is equal to the sample size.
NLSCY Data Users Guide
74
2001/2002
5/ Once the factor structures were analysed and the items included
in each factor were determined, scores were calculated. To produce
the scores, 1 was subtracted from each item so that the lowest
possible score would be 0. A score of 0 indicates that the child has
no problems for all factors in the behaviour scale except for the
Prosocial factor, where a score of 0 indicates the absence of
prosocial behaviour. Some items were imputed. The imputed values
were computed by a procedure (the SAS PRINQUAL procedure)
that determines which of the possible values for an item is the most
plausible for an individual in view of his/her response profile, the
response profiles of others in the sample, and the number of factors
included in the analysis.
6/ The score for each factor on the scale was derived by totaling the
values of the items that made up that factor (including imputed
values). The score was set to "missing" if too many of the values of
an items included in the factor were unreported. A value may be
missing if the parent refused to answer or did not know the answer
to the item.
Distance
Between
Answer
Categories
Factor analysis requires that the data have the property of interval or
ratio data, that is the distance between each answer category of the
question should be the same. For example, in scales where the
answer choices are: Never, Sometimes, Often, and Always, one
must assume that the distance between Never and Sometimes is
the same as that between Sometimes and Often in the respondent's
perception. It was felt that this was not necessarily true in the case
for the scales used in the NLSCY.
Data
Transformation
Using Optimal
Scaling
Therefore before performing the factor analysis for each of the
NLSCY scales, the data were transformed using optimal scaling.
The method used was one proposed by Young and several
associates (Young, 1981) which is a variant of Fisher's optimal
scaling technique. The method is presented as a means of
transforming data which are fundamentally nominal or ordinal in
nature to interval or ratio level data so that statistical techniques
which are appropriately applied only to interval and ratio data may
be utilized.
NLSCY Data Users Guide
75
2001/2002
Factor Analysis
Using
Weighted Data
Initially the factor analysis for each scale to be included in the
NLSCY data was carried out using unweighted data. At that point in
time the final weights had not yet been calculated. Once the weights
were available, work started on repeating the factor analyses using
the weighted data. (See Section 7 for a description of the weighting
procedures.) With the weights, the same factor structure was not
always observed. When there was a discrepancy, results emerging
from the weighted analysis were used.
Calculation of Scores and Item Imputation
Calculation of
Scores for
Each Factor
The results of the factor analyses were used to determine what items
"loaded" into each factor (i.e., were a part of each factor). The next
step was to calculate a score for each factor. This was done by
summing the values for each individual item that made up the factor.
In some cases some rescaling of values was done before the final
score was calculated. The following example illustrates how factor
scores were computed.
NLSCY Data Users Guide
76
2001/2002
Example of
Factor Score
Computation
One of the constructs that emerged in the factor analysis for the
Parenting Scale on the Child's Questionnaire was the ineffective
parenting factor. In the factor analysis on cycle 1 data seven items
were found to load into this factor.
APRCQ04
How often do you get annoyed with your child
for saying or doing something he/she is not
supposed to?
APRCQ08
Of all the times you talk to your child about his/her
behaviour, what proportion is praise?
APRCQ09
Of all the times you talk to your child about his/her
behaviour, what proportion is disapproval?
APRCQ13
How often do you get angry when you punish your
child?
APRCQ14
How often do you think the kind of punishment
you give your child depends on your mood?
APRCQ15
How often do you feel you have problems
managing your child in general?
APRCQ18
How often do you have to discipline your child
repeatedly for the same thing?
The answer categories for these items were of two types:
1 - never
2 - about once a week or less
3 - a few times a week
4 - one or two times a day
5 - many times each day
1 – never
2 - less than half the time
3 - about half the time
4 - more than half the time
5 - all the time
NLSCY Data Users Guide
77
2001/2002
Example of
Factor Score
Computation:
Continued
Negative
Loading
NLSCY Data Users Guide
In the calculation of the score for this hostile/ineffective parenting
factor, the categories were rescaled to 0 to 4 (i.e., the category
"never" was scored as 0, the category "about once a week or
less/less than half the time" was scored as 1, ... and the category
"many times each day/all the time" was scored as 4). In order to
compute the score these values were summed across the seven
items involved in the factor resulting in a hostile/ ineffective
parenting score in the range 0 to 28. A score of 0 represents the
absence of a problem and a score of 28 is the highest possible
score with respect to problems. For most of the scores calculated
for the NLSCY, a score of 0 represents the absence of a problem.
However there are exceptions to this which are noted in the
documentation for each particular scale.
Note that the second item that loaded into the hostile/ineffective
parenting factor, APRCQ08 (Of all the times you talk to your child
about his/her behaviour, what proportion is praise?) is in the
opposite direction compared to the other items. In fact the item
loaded "negatively" into the factor. Therefore when computing the
score the values for this item were reversed - all the time was
scored as 0, more than half the time as 1, ... and never as 4.
In the documentation for each scale any item that was reversed for
the scoring algorithm due to a negative loading is indicated.
78
2001/2002
NonResponse
Code
The score for the hostile/ineffective parenting factor is labeled as
APRCS04 on the record layout for the micro data file. An "S" in the
5th position of the variable name indicates a score.
When the score was being calculated for each factor there was a
possibility that one or more of the items making up the score had a
non-response code (don't know, refusal or not-stated). If the
number of items with a non-response code was above a certain
threshold, the factor score was set to not-stated. Generally this
threshold value was set at 10% of the items. If less than 10% of the
items had a missing value then the items with non-response codes
were imputed before the score was computed. The procedure
used to impute these missing items is a routine available in SAS
in the procedure called PRINQUAL. This procedure indicates,
among valid item values, the one that seems the most plausible for
a given record. It considers the response profile of the record with
the missing item, the response profile of other responding records
in the sample as well as the number of factors considered in the
analyses.
Imputation
Flags
A flag was created for many of the items for which values have
been imputed to indicate the records for which imputation has
taken place. Where these exist, the flags have been included on
the micro data file. The flag on the file which corresponds to an
item has the same name as the item itself except that the Q
(question indicator) in the variable name is replaced by I. For
example some imputation was carried out for APRCQ04 (How
often do you get annoyed with your child for saying or doing
something he/she is not supposed to?). The imputation flag for this
item is labeled APRCI04.
Raw Items
It should be noted that in addition to the scores, the raw items for
each scale are included on the micro data file. This will allow
researchers to consider alternate factor structures if desired. For
the raw items the original values (in the 1 to 5 range for the
parenting scale) have been retained before any rescaling or
reversal of values took place.
NLSCY Data Users Guide
79
2001/2002
Reliability Measures for Scales
Introduction
Reliability refers to the accuracy, dependability, consistency or
repeatability of score results. In more technical terms, reliability
refers to the degree to which the scores are free of measurement
errors. There are many ways to measure
reliability.
Cronbach’s
Alpha
One of the most commonly used reliability coefficients is
Cronbach's alpha (Cronbach, 1951). Alpha is a measure of the
internal consistency of the items within the factor. It is based on the
average covariance of items within the factor. It is assumed that
items within a factor are positively correlated with each other
because they are attempting to measure, to a certain extent, a
common entity or construct.
Interpretation
s of
Cronbach’s
Alpha
Cronbach's a has several interpretations. It can be viewed as the
correlation between this scale or factor and all other possible
scales containing the same number of items, which could be
constructed from a hypothetical universe of items that measure the
characteristic of interest. In the hostile/ineffective parenting factor,
for example, the seven questions actually used for inclusion on the
scale can be viewed as a sample from the universe of many
possible items. Parents could also have been asked: "How often
do you raise your voice when you discipline your child?" or "How
often do you threaten punishment more often than you use it?"
Cronbach's a tells how much correlation can be expected between
the scale which was used and all other possible seven-item scales
measuring the same thing.
NLSCY Data Users Guide
80
2001/2002
Another
Interpretation
of Cronbach’s
Alpha
Another interpretation of Cronbach's a is the squared correlation
between the score an individual obtains on a particular factor (the
observed score) and the score he/she would have obtained if
questioned on all possible items in the universe (the true score).
Since a can be interpreted as a correlation coefficient, it ranges
from 0 to 1.
It has been shown that in general, a is a lower bound to the
reliability of a scale of n items (Novick and Lewis, 1967). In other
words in most situations, a provides a conservative estimate of a
score's reliability.
Satisfactory
Level of
Reliability
What is a satisfactory level of reliability? It is difficult to specify a
single level that should apply in all situations. Some researchers
believe that reliabilities should not be below 0.8 for widely used
scales. At that level, correlations are affected very little by random
measurement error. At the same time, it is often very costly in
terms of time and money to obtain a higher reliability coefficient. It
should be noted that for some of the factors for which scores were
computed for the NLSCY, the reliabilities are below this level. The
Cronbach a is given in the documentation for each score that has
been calculated. Researchers can determine for themselves
whether or not the score has adequate reliability for their specific
purposes.
Finally, it should be mentioned that for the NLSCY the Cronbach a
for each factor score was computed using SAS. Typically, the a
coefficients calculated using SAS are lower than those calculated
using SPSS.
Parent-Reported Scales
Temperament Scale
NLSCY Data Users Guide
81
2001/2002
Introduction
The Infant
Characteristic
s
Questionnaire
Temperament scales are used to measure the temperament of
young children (up to and including the age of three) based on the
parents' answers to questions about the degree of difficulty their
child presents for them. This measure is founded on the
assumption that a child's temperament is not solely dependent on
biological factors, but is also influenced by the parents' perception
of the difficulty of the child.
The temperament scale used in the NLSCY for children 3 to 5
months old was developed by Dr. John Bates of the University of
Indiana. This well-established scale, originally known as the Infant
Characteristics Questionnaire (ICQ), has been used in large-scale
studies and is considered by specialists to be the best available
measure for use in household surveys.
The ICQ has been adapted for use in other surveys covering
different age groups: 6 to 11 months, 12 to 23 months and twoyear-olds. A revised version of the scale, devised by Dr. Jo-Anne
Finegan at Toronto's Hospital for Sick Children, is used for threeyear-olds.
Questions
Measuring
Aspects of
Temperament
: Children 3 to
5 Months
Aspects of
Temperament
Children 1 to
3 Years
NLSCY Data Users Guide
For children aged 3 to 5 months, the scale is made up of
questions ATMCQ01 to ATMCQ12, ATMCQ14 to ATMCQ20,
ATMCQ23 and ATMCQ33 is intended to measure the extent to
which the child is fussy, unadaptable, unpredictable and dull. For
children 6 to 11 months old, the foregoing list was expanded to
include ATMCQ13 and ATMCQ24 to ATMCQ27. The expanded
list of questions measures the same four aspects of temperament
as for children 3 to 5 months old.
For children between 1 and 3 years-old, questions ATMCQ1 to
ATMCQ15 and ATMCQ17 to ATMCQ33 should theoretically
measure the degree to which the child is difficult, irregular,
unadaptable, affectively negative and persistent/unstoppable.
82
2001/2002
Meanings of
Ratings for
Specified
Behaviours
The respondent, in most cases a parent, is required to answer
each question in the scale by assigning a rating between 1 and 7.
For all questions except ATMCQ14, a 1 means that the child has a
favourable response or usually exhibits the specified behaviour,
while a 7 indicates that the child reacts negatively or seldom
displays the behaviour in question. If the child is in the middle, a 4
is assigned. In question ATMCQ14, the meanings of the ratings
are reversed.
Education (Child)
Introduction
The objective of this section was to obtain some basic information
about the child's educational experiences. The amount and type of
information collected varied depending upon the age of the child,
with more information being collected for the older children who
have had greater school experience.
Basic information was collected for all age groups, such as: the
child's grade level, type of school and language of instruction,
whether the child looks forward to school, behaviour problems at
school, absenteeism, parental hopes for the child's educational
outcomes, number of school changes and residential moves.
For children in grade 1 or higher, additional questions were asked
concerning other aspects such as skipping and repeating grades,
achievement, special education, parents' perception of school
climate and importance of good grades to parents.
NLSCY Data Users Guide
83
2001/2002
The
Teacher’s
and
Principal’s
Questionnaire
s
The Teacher's and Principal’s Questionnaires provide additional
information about the child and his/her school achievement and
behaviour.
Child’s Grade
The child's grade was also collected on the Teacher's
Questionnaire. There was not always consistency across the data
collection units on what the correct grade was. In the edit, priority
was placed on what the teacher said in the case of discrepancies.
At the data collection stage, six different questions were asked to
determine the child's grade. This was done due to differences in
grade classification among provinces. At the processing stage,
these six questions were collapsed into one variable. On the
record layout, an indication is given as to what the code means for
each province. For example, if the grade code (CEDCD01) is 10,
this refers to secondary 1 for Québec and grade 7 for all other
provinces. A similar procedure was carried out for grade skipped
(CEDCD02) and grade repeated (CEDCD03).
On the micro data file the variables on language of instruction
(BEDCQ12A) and type of school (BEDCQ08) were set to notstated because of confidentiality concerns.
Education
Section
In the Education Section, there was one question (BEDCQ13)
which asked the number of days the child had missed since the
beginning of the school year. The answer to this question obviously
depends on the collection date which has not been included on the
micro data file because of confidentiality concerns. Therefore this
variable has been suppressed and a derived variable was created
(BEDCD04) to indicate the percent of days missed since the
beginning of the school year.
Behaviour Scale
NLSCY Data Users Guide
84
2001/2002
Objective of
the Behaviour
Scale
The objective of the behaviour scale is to assess aspects of the
behaviour of children two years of age and older.
Initially, an attempt was made to measure the following behaviours
for children aged 2 and 3:
! hyperactivity,
! emotional disorder,
! anxiety,
! physical aggression,
! inattention,
! prosocial behaviour,
! separation anxiety and
! opposition.
Similar
Behaviours
For children between 4 and 11 years of age, an attempt was made
to measure similar behaviours; separation anxiety and opposition
were omitted, and indirect aggression and some aspects of
conduct disorder were added.
Factor
Analysis for
the Behaviour
Scale
The following indicates the items that were included on the
questionnaire to measure these various constructs of behaviour.
As discussed in Section 9.1, a complete factor analysis was
carried out for the behaviour scale to assess the psychometric
properties of this scale for the NLSCY population. As part of this
analysis, the items that loaded into each construct or factor were
compared to the expected result described below. The results of
this analysis are presented later on in this section.
Theoretical
Constructs
Below are the theoretical constructs used for the factor analysis.
The actual scales that emerged from the analysis vary from these
constructs.
NLSCY Data Users Guide
85
2001/2002
Two- and three-year-olds:
! Conduct disorder
Items include BBECQ6G from the Ontario Child Health Study
(OCHS).
! Hyperactivity
Items include BBECQ6B, Q6I, Q6N, Q6P, Q6S and Q6W from the
OCHS and ABECQ6HH from the Montreal Longitudinal Survey.
! Emotional disorder
Items include BBECQ6F, Q6K, Q6Q, Q6V, Q6CC, Q6MM and
Q6RR from the OCHS.
! Anxiety
Items include several of the OCHS emotional disorder questions
(BBECQ6F, Q6Q, Q6V and Q6CC).
! Physical aggression
Items include BBECQ6X from the Montreal Longitudinal Survey
and BBECQ6G from the OCHS.
! Inattention
Items include BBECQ6P from the OCHS and ABECQ6EE, Q6KK
and Q6QQ from the Montreal Longitudinal Survey.
! Prosocial behavior
Items include BBECQ6D, Q6U, Q6BB, Q6SS and Q6UU from the
Montreal Longitudinal Survey; the last four items are from a scale
developed by K. Weir and G. Duveen.
! Separation anxiety
Items include BBEC6DD1, 6LL1, 6PP1 and Q6TT1 from
Achenbach's Child Behavior Checklist (CBCL).
! Opposition
Items include BBECQ6E1, Q6J1, Q6R1 and Q6T1 also drawn
from Achenbach's CBCL.
NLSCY Data Users Guide
86
2001/2002
Children aged 4 to 11:
! Conduct disorder
Items include BBECQ6C, Q6E, Q6G, Q6L, Q6O (this item is
coded "not applicable" for children not in school), Q6T, Q6AA,
Q6DD, Q6FF, Q6JJ and Q6PP from the Ontario Child Health
Study (OCHS).
! Hyperactivity
Items include BBECQ6B, Q6I, Q6N, Q6P, Q6S and Q6W from the
OCHS and Q6HH from the Montreal Longitudinal Survey.
! Emotional disorder
Items include BBECQ6F, Q6K, Q6Q, Q6V, Q6CC, Q6MM and
Q6RR from the OCHS.
! Anxiety
Items include BBECQ6Y and Q6II from the Montreal Longitudinal
Survey along with several of the OCHS emotional disorder items
(BBECQ6F, Q6Q, Q6V and Q6CC).
! Indirect aggression
Items include BBECQ6J, Q6R, Q6Z, Q6LL and Q6TT from
Lagerspetz, Bjorngvist and Peltonen of Finland.
! Physical aggression
Items include BBECQ6X from the Montreal Longitudinal Survey
and BBECQ6G, Q6AA and Q6NN from the OCHS.
! Inattention
Items include BBECQ6P from the OCHS and BBECQ6EE, Q6KK
and Q6QQ from the Montreal Longitudinal Survey.
! Prosocial behaviour
Items include BBECQ6A, Q6H, Q6M, Q6GG and Q6OO from the
OCHS and ABECQ6D, Q6U, Q6BB, Q6SS and Q6UU from the
Montreal Longitudinal Survey; the last four items are from a scale
devised by K. Weir and G. Duveen.
NLSCY Data Users Guide
87
2001/2002
Results: Twoand threeyear-olds
There were 3,909 two- and three-year-olds in the sample. The
group was split into two sub-samples of 1,932 and 1,977
individuals, and the analysis for this age group was performed
separately for each sub-sample. The non-response rate for most
items was about 2.2%. Some individuals were excluded from the
analysis that produced the factors. The exclusion criteria were as
follows: individuals with eight or more items coded "missing,"
individuals with one or more refusals, individuals with two or more
missing items under hyperactivity and emotional disorder, and
individuals with one or more missing items for the other theoretical
factors. After the criteria were applied, there were 1,742 and
1,773 individuals left in the sub-samples to be analysed. Data
were imputed for only 12 items. The number of imputations ranged
between 1 and 8 for those 12 items. A total of 34 values were
imputed.
Factor
Analysis
The factor analysis derived five factors for this age group:
hyperactivity-inattention (ABECS01), prosocial behaviour
(ABECS02), emotional disorder-anxiety (ABECS03), physical
aggression-opposition (ABECS04) and separation anxiety
(ABECS05). The items making up each factor are listed in the
table below.
BEHAVIOUR SCALE FOR 2- AND 3-YEAR-OLDS
FACTOR
SCORE
ITEMS
Hyperactivity – inattention
ABECS0
1
ABECS0
2
ABECS0
3
ABECS0
4
ABECS0
5
ABECQ6B, 6I, 6N, 6P, 6S, 6HH,
6QQ
ABECQ6D, 6U, 6BB, 6SS, 6UU
Prosocial behaviour
Emotional disorder – anxiety
Physical aggression – opposition
Separation anxiety
NLSCY Data Users Guide
88
ABEQC6F, 6K, 6Q, 6V, 6MM,
6RR
ABECQ6G, 6W, 6X, 6E1, 6R1,
6T1, 6Z1, 6NN
ABECQ6CC, 6DD1, 6PP1, 6LL1,
6TT1
2001/2002
Cronbach's alpha (raw value) was computed with SAS using
normalized weighted data (in general, Cronbach's alphas
computed by SAS are lower than those produced by SPSS). For
hyperactivity-inattention (ABECS01), Cronbach's alpha was 0.798.
The item that had the greatest effect on this factor was ABECQ6P,
as removing it lowers Cronbach's alpha to 0.762. The table below
shows the Cronbach's alpha for each factor, first including all items,
then excluding the item having the greatest effect.
Cronbach’s
Alpha for 2and 3- Year
Olds
CRONBACH'S ALPHA FOR THE BEHAVIOUR SCALE
FOR 2- AND 3-YEAR-OLDS
FACTOR
CRONBACH'
S ALPHA
(RAW)
Hyperactivity-inattention
(ABECS01)
Prosocial behaviour
(ABECS02)
Emotional disorder-anxiety
(ABECS03)
Physical aggressionopposition (ABECS04)
Separation anxiety
(ABECS05)
Children
aged 4 to 11:
0.798
ITEM THAT LOWERS
CRONBACH'S
ALPHA THE MOST IF
IT IS EXCLUDED
ABECQ6P
CRONBACH'
S ALPHA IF
THE ITEM IS
EXCLUDED
0.761
0.847
ABECQ6SS
0.795
0.593
ABECQ6MM
0.539
0.754
ABECQ6Z1
0.717
0.561
ABECQ6DD1
0.431
There were 14,226 children in the 4 to 11 age group. Two subsamples of 7,073 and 7,153 were created for analysis. The item
non-response rate was approximately 2.1% for most of the 47
items involved in the analysis. Individuals were excluded from the
analysis on the basis of the following criteria: individuals with eight
or more items coded "missing," individuals with one or more
refusals; individuals with two or more missing items under prosocial
behaviour, conduct disorder, hyperactivity, anxiety and emotional
disorder; and individuals with one or more missing items for the
other factors. After the criteria were applied, 6,620 and 6,683
individuals remained in the sub-samples to be analysed. Data were
imputed for 26 items. The number of imputations ranged between 1
and 159 for those 26 items. A total of 363 values were imputed.
NLSCY Data Users Guide
89
2001/2002
Six Factors
for 4- to 11Year Olds
Six factors were identified for this age group: hyperactivityinattention (ABECS06), prosocial behaviour (ABECS07),
emotional disorder-anxiety (ABECS08), physical aggressionconduct disorder (ABECS09), indirect aggression (ABECS10) and
a new factor, property offence (ABECS11). The items making up
each factor are listed in the table below.
BEHAVIOUR SCALE FOR 4- TO 11-YEAR-OLDS
FACTOR
SCORE
ITEMS
Hyperactivity – inattention
ABECS0
6
ABECS0
7
ABECQ6B, 6I, 6N, 6P, 6S, 6W,
6HH, 6QQ
ABECQ6A, 6D, 6H, 6M, 6U, 6BB,
6GG,
6OO, 6SS, 6UU
ABECQ6F, 6K, 6Q, 6V, 6CC, 6II,
6MM, 6RR
ABECQ6G, 6X, 6AA, 6FF, 6JJ,
6NN
ABECQ6J, 6R, 6Z, 6LL, 6TT
Prosocial behaviour
Emotional disorder – anxiety
Physical aggression – conduct
disorder
Indirect aggression
Property offence
NLSCY Data Users Guide
ABECS0
8
ABECS0
9
ABECS1
0
ABECS1
1
90
ABECQ6C, 6E, 6L, 6T, 6DD,
6PP
2001/2002
Cronbach’s
Alpha for 4to 11- Year
Olds
Cronbach's alphas for these factors are given in the table below.
Normalized weighted data were used in the computations.
CRONBACH'S ALPHA FOR THE BEHAVIOUR SCALE
FOR 4-TO 11-YEAR-OLDS
FACTOR
CRONBACH'
S ALPHA
(RAW)
Hyperactivity-inattention
(ABECS06)
Prosocial behaviour
(ABECS07)
Emotional disorder –
anxiety (ABECS08)
Physical aggression –
conduct disorder
(ABECS09)
Indirect aggression
(ABECS10)
Property offence
(ABECS11)
0.838
ITEM THAT LOWERS
CRONBACH'S
ALPHA THE MOST IF
IT IS EXCLUDED
ABECQ6I
CRONBACH'
S ALPHA IF
THE ITEM IS
EXCLUDED
0.810
0.816
ABECQ6BB
0.789
0.794
ABECQ6II
0.756
0.770
ABECQ6AA
0.716
0.781
ABECQ6LL
0.733
0.637
ABECQ6C
0.553
The scores for these factors could not be computed in 338, 647, 324, 358, 814 and 310
cases respectively because of unreported values.
Motor and Social Development
Objective for
Motor and
Social
Development
Section
The Motor and Social Development Section of the Child's
Questionnaire was completed for children in the 0 to 3 age group.
The objective was to measure motor, social and cognitive
development of young children. A scale was used to assess these
concepts (BMSCQ01 to BMSCQ48).
NLSCY Data Users Guide
91
2001/2002
Overview of
the Motor
and Social
Development
(MSD) Scale
The Motor and Social Development (MSD) Scale was developed
by Dr. Gail Poe of the U.S. National Center for Health Statistics.
The MSD scale consists of a set of 15 questions that measure
dimensions of the motor, social and cognitive development of
young children from birth through 3 years; the questions vary by age
of the child. Each item asks whether or not a child is able to
perform a specific task. The scale has been used in collections of
the National Longitudinal Survey of Youth in the United States and
in recent versions of the National Child Development Survey in
England.
Standardized
Scores
A score was calculated for each child by summing the number of
"yes" answers to each item in the scale (BMSCS01). Although
there were different sets of questions depending on the age in
months of the child, differences were observed when comparing
score within these age bands. For example, there was a specific
set of questions for children 4 to 6 months old. It was found that
children who were 6 months old had scores that were on average
higher than those 4 months olds. Therefore a decision was made to
produce standardized scores. Each child was assigned a standard
score, such that the mean MSD score was 100 and the standard
deviation was 15 for all age groupings of Cycle 2 and Cycle 3. This
standardization had been done by 1 month age groups. Therefore
children who are 0 months old had in Cycle 1 an average MSD
score of 100, children who are 1 month old had an average MSD
score of 100, ..., and children 47 months old had an average MSD
score of 100. Using a standardized score (BMSCS02) makes it
possible to compare scores of children across the 0 to 3 age
group, not controlling for age.
Standardized
Scores from
Previous
Cycles
In the previous cycle the name was based on the cycle 1 child.
However, since the number of children is not very big we decided to
create new names in cycle 3 based on the combination of the
scores from cycle 1, 2 and 3. The standardized scores from cycle 1
and 2 will be recalculated based on these names.
NLSCY Data Users Guide
92
2001/2002
Relationships
Objective
The Relationships Section of the Child's Questionnaire was
completed for all children 4 years of age and older. The objective
was to provide information about the child's relationships with
others. Positive relationships with other children and adults may
help to counteract other factors which place a child at risk.
The section collects information about how the child gets along with
parents, brothers and/or sisters, teachers, friends, and classmates,
with some variation by age of the child. Parents' knowledge of the
names of the friends of 8- to 13-year-olds is also investigated,
along with their perception of these other children's behaviours, and
whether their own child is shy or outgoing.
Questions
from the
Ontario Child
Health Study
The questions on number of days spent doing things with friends,
number of friends, and getting along with friends, parents, teachers
and siblings (BRLCQ01, Q02, Q06-Q09) are based on those in the
Ontario Child Health Study.
Parenting Scale
Objective
The objective of this scale is to measure certain parenting
practices. Specifically, two scales were used. The first was
designed to measure the positive interaction, hostility/
ineffectiveness and consistency of the parenting of the child. The
second scale was designed to measure parental practices that
may or may not provoke aversion.
The questions from the Child's Questionnaire used to measure
these aspects of parenting are identified in the following
paragraphs. As mentioned in Section 9.1, complete factor analyses
were done on the parenting scales to evaluate the psychometric
properties of these scales for the NLSCY population. The make-up
of each factor obtained during these analyses was compared to
that which had been indicated in the literature. The results of these
analyses are presented later in this section.
NLSCY Data Users Guide
93
2001/2002
Questions
Questions
for the 0-11
Age Groups
Questions BPRC-Q1 to BPRC-Q18 on positive interaction, hostility
or ineffectiveness and on coherence were provided by Dr. M. Boyle
of the Chedoke-McMaster Hospital, based on the work of Dr. Ken
Dodge (Venderbilt University) and an adaptation of the Parent
Practices Scale of Strayhorn and Weidman. (For children ages 0 to
23 months, only questions APRCQ1 to APRCQ7 were asked.)
Questions
for the 2-11
Years Age
Groups
Questions BPRC-Q19 to BPRC-Q25 which measure parental
practices which may or may not cause aversion were provided by
Dr. M. Boyle.
Analysis of NLSCY Data
Factor
Structure
The factor structure of each scale was determined based on data
from the first cycle. The factor structure imposed on the scales
already used in the first cycle and repeatedly used in the second
cycle of the survey was the result of analyses done based on data
from the first cycle.
To conduct the analysis on the parenting scales for the NLSCY
data, a factor analysis was conducted on the scale for the 0 to 23
months age group and the two scales for the 2 to 11 age group
separately. New factor structures emerged which are described in
the Results Section below.
NLSCY Data Users Guide
94
2001/2002
New Factor
Structures
Once the factor structures were analysed and the items included in
each factor were determined, scores were calculated. To produce
the scores, 1 was subtracted from each item so that the lowest
possible score value would be 0. For each of the four factors, a
score of 0 indicates:
- the absence of positive interaction for the positive interaction
factor;
- the absence of hostile/ineffective interaction for the
hostile/ineffective factor;
- the absence of consistent parenting for the consistency factor;
- the absence of punitive interaction or aversion producing
practices for the hostility/ineffective parenting factor.
Results (Cycle 1)
Children
aged 0 to 23
months
There were 4,696 children in the sample for the age group 0 to 23
months. The group was split into two sub-samples of 2,311 and
2,385 individuals, and the analysis for this age group was
performed separately for each sub-sample. The non-response rate
for the seven items ranged from 1.9 to 2.5%. Some individuals
were excluded from the analysis that produced the factors. The
exclusion criterion was as follows: individuals with one or more
missing items. After the criterion was applied, there were 2,245
and 2,307 individuals left in the sub-samples to be analysed. No
imputation was done. The factor analysis derived two factors for
this age group: positive interaction (APRCS01), and
hostile/ineffective (APRCS02). The items making up each factor
are listed in the table below.
PARENTING SCALE FOR CHILDREN AGED 0 TO 23 MONTHS
FACTOR
SCORE
ITEMS
Positive interaction
Hostile/ineffective
APRCS01
APRCS02
APRCQ1, 2, 3, 6, 7
APRC4, 5
NLSCY Data Users Guide
95
2001/2002
Cronbach’s
Alpha for
Children
aged 0 to 23
Months
Cronbach's alpha (raw value) was computed with SAS using
normalized weighted data. (In general, Cronbach's alphas
computed by SAS are lower than those produced by SPSS.) For
the positive interaction factor (APRCS01), Cronbach's alpha was
0.727. The item that had the greatest effect on this factor was
APRCQ7. Removing it lowers Cronbach's alpha to 0.656. For the
hostile/ineffective factor (APRCS02), Cronbach's alpha was 0.394.
(It should be noted that there were only two items for this factor, and
the alpha can only be derived if one of the 2 items is removed.)
After identifying the two factors, the next step was to calculate
scores for each.
Missing
Values
Scores could be calculated for only 132 individuals for the positive
interaction factor, and for only 124 individuals for the
hostile/ineffective factor because of missing values for the items for
these factors.
Children
aged 2 to 11
There were 18,135 children in the sample for the age group 2 to 11.
The group was split into two sub-samples of 9,090 and 9,045
individuals, and the analysis for this age group was performed
separately for each sub-sample. The non-response rate for each of
the eighteen items ranged from 2.1 to 2.7%. Some individuals were
excluded from the analysis that produced the factors. The exclusion
criteria were as follows: individuals with two or more items coded
"missing" under positive interaction and hostility, and individuals
with a single missing item under consistency. After the criteria were
applied, there were 8,815 and 8,772 individuals left in the subsamples to be analysed. Data were imputed for 12 items. The
number of imputations ranged between 1 and 16. A total of 91
values were imputed. The factor analysis derived three factors for
this age group: positive interaction (APRCS03), and hostility
(APRCS04), and consistency (APRCS05). The items making up
each factor are listed in the table below.
PARENTING SCALE FOR CHILDREN AGED 2 TO 11
FACTOR
SCORE
ITEMS
Positive interaction
Ineffective
APRCS03
APRCS04
APRC Q1, 2, 3, 6, 7
APRC Q4, 8*, 9, 13, 14, 15, 18
NLSCY Data Users Guide
96
2001/2002
Consistency
APRCS05
* Item inverted when computing the score.
APRC Q10, 11, 12*, 16*, 17*
Cronbach's alphas for these factors are given in the table below. Normalized weighted
data were used for the computations.
CRONBACH'S ALPHA FOR THE PARENTING SCALE
FOR 2- AND 3-YEAR-OLDS
Factor
Cronbach's
alpha (raw)
0.808
Item that lowers
Cronbach's alpha the
most if it is excluded
APRCQ2
Cronbach's
alpha if the item
is excluded
0.749
Positive interaction
(APRCS03)
Ineffective (APRCS04)
0.706
APRCQ13
0.654
Consistency
(APRCS07)
0.660
APRCQ12
0.569
The scores for these factors could not be computed in 408, 482 and 534 cases
respectively because of unreported values.
Parenting
scale for
children
aged 2 to 11
There were 18,135 children in the sample for the age group 2 to 11.
The group was split into two sub-samples of 9,090 and 9,045
individuals, and the analysis for this age group was performed
separately for each sub-sample. The non-response rate for the
seven items analysed was about 2.5%. The exclusion criterion was
as follows: individuals with one or more items coded "missing"
were excluded. After this criterion was applied, there were 8,848
and 8,801 individuals left in the sub-samples to be analysed. No
unreported values were imputed.
NLSCY Data Users Guide
97
2001/2002
Aversion
Factor
Derived for
Children
aged 2 to 11
A factor was derived for this age group: aversion (APRCS06). The
items making up this factor are APRCQ21, 22, 23 and 24. Items 21
and 23 were inverted when computing the scores. The factor
weights of variables APRCQ19, 20 and 25 were insufficient to be
included.
Cronbach's alpha for this factor was 0.569. The item that had the
greatest effect on this factor was APRCQ22. Removing it lowers
Cronbach's alpha to 0.377. (Normalized weighted data were used
in the computations.)
The score for this factor could not be computed in 478 cases
because of unreported values.
Parenting Scales: 12-15 Year Olds
Conflict
Tactics Scale
The conflict tactics score was created for children aged 12-15. The
following items were used in the factor analysis: CPRCBb30a,
CPRCBb30b, CPRCBb30c, CPRCBb30d, CPRCBb30e,
CPRCBb30f, CPRCBb30g, CPRCBb30h, CPRCBb30i, and
CPRCBb30j.
NLSCY Data Users Guide
98
2001/2002
Analysis of NLSCY Data
Factor
Structure
The factor structure of this scale was determined based on data
from cycle3. To conduct the analysis on this scale for the NLSCY
data, a factor analysis was conducted splitting the data into two
separate files. Once the factor structure was analysed and the
items included in the factor were determined, the final score were
calculated. To produce the scores, 1 was subtracted from each
item so that the lowest possible score value would be 0. For the
factors, a high score indicates the presence
Results
(Cycle 3)
There were 4,296 children in the sample for the age group 12 to 15
years. The group was split into two sub-samples of 2,140 and
2,156 individuals, and the analysis for this scale was performed
separately for each sub-sample. The non-response rate for the ten
items ranged from 1.9 to 2.5%. In total 310 cases who had one or
more missing values and were excluded from the analysis. These
cases were given a missing value for the overall score since no
imputation was completed. The factor analysis revealed one strong
factor -- conflict tactics--(CCRCS09). Items I and J were not
included in the factor as they reduced the Alpha Cronbach score.
The final score included items items A,B,C,D,E,F,G, and H . Items
A and H were reversed in the calculation of the score. All values
were recoded from 1-5 to 0-4. The final score ranges from 0-28
with a high score indicating a higher degree of parent-child
disagreements. The Alpha Cronbach value for the score is 0.75.
Parent-Child
Cohesion
Scale
The parent child cohesion score was created for children aged 1215. The following items were used in the factor analysis:
CPRCBb31a, CPRCBb31b, CPRCBb31c, CPRCBb31d,
CPRCBb31e, CPRCBb31f, CPRCBb31g, and CPRCBb31h.
NLSCY Data Users Guide
99
2001/2002
Depression Scale (PMK)
Introduction
The depression scale was administered to the PMK as part of the
Parent Questionnaire. Questions for this scale (BDPPQ12A to
BDPPQ12L) are a shorter version of the depression rating scale
(CES-D), comprising 20 questions, developed by L. S. Radloff of
the Epidemiology Study Center of the National Institute of Mental
Health in the United States. This rating scale is used to measure
the frequency of symptoms in the public at large. The occurrence
and severity of symptoms associated with depression during the
previous week are measured. The rating scale was reduced to 12
questions by Dr. M. Boyle of the Chedoke-McMaster Hospital of
McMaster University.
Symptoms of
Depression
This rating scale is aimed at gathering information about the mental
health of respondents, with particular emphasis on symptoms of
depression. Several members of the NLSCY advisory group of
experts pointed out that the best way of proceeding was to
measure one particular aspect of the PMK's mental health instead
of trying to measure overall mental health. It was proposed that this
section focus on depression for the following reasons: depression
is a prevalent condition; it has been demonstrated that depression
in a parent affects the children; present research on this subject is
generally based on demonstration groups and not on population
samples; and it is felt that introducing policies in this area could
make a difference.
Questions
for the
Depression
Rating Scale
The depression rating scale includes twelve questions, each of
which contains four response categories. In order for the lowest
score value to be 0, the value for each question was reduced by 1
in calculating the score. As well, the answer categories were
reversed for questions having a negative loading (BDPPQ12F,
Q12H, and Q12J). The total score (BDPPS01) may therefore vary
between 0 and 36, a high score indicating the presence of
depression symptoms.
NLSCY Data Users Guide
100
2001/2002
Results
Unweighted
Data
The factor structure of each scale was determined based on data
from the first cycle. The factor structure imposed on the scales
already used in the first cycle and repeatedly used in the second
cycle of the survey was the result of analyses done based on data
from the first cycle.
In analysing this scale, unweighted data 1 were used. The sample
size was 13,439 PMKs. However, once the observations
containing mostly missing values were eliminated, the analysis
dealt with only 13,140 PMKs. The non-response rate for the various
questions in the rating scale was roughly 2.0%, whereas for the
total score, a non-response rate of 2.2% was obtained. There was
no imputation for the variables in this rating scale.
Single-Factor
Analysis
In spite of the possibility of extracting more than one factor from the
depression rating scale, a single-factor analysis was used since
the interest was in developing a global depression index. Following
the analysis, the 12 variables of the scale were all kept as
components of this factor since all 12 loading values met the
established threshold. The Cronbach alpha coefficient (calculated
using SAS software) was 0.82. The variable ADPPQ12D showed
the highest correlation (0.68) with the total score (once the variable
was removed), whereas the variable showing the lowest correlation
was ADPPQ12L with a correlation of 0.33. The Cronbach alpha
coefficient calculated by omitting one variable was between 0.79
and 0.82 for the 12 variables.
1
Weighted data could not be used since the weights developed for the NLSCY are for children only, and
not for parents.
NLSCY Data Users Guide
101
2001/2002
Family Functioning Scale (Parent)
Introduction
Questions related to family functioning, i.e., BFNHQ01A to
BFNHQ01L, were developed by researchers at the ChedokeMcMaster Hospital of McMaster University and have been used
widely both in Canada and abroad. This scale is used to measure
various aspects of family functioning, (e.g. problem solving,
communications, roles, affective involvement, affective
responsiveness and behaviour control).
Global
Assessment
of Family
Functioning
Question BFNHQ01M, drawn from the Follow-up to the Ontario
Child Health Study, was added to the original scale to determine
whether alcohol consumption had an effect on global family
dynamics. However, it was not used in the analysis of the scale.
This scale is aimed at providing a global assessment of family
functioning and an indication of the quality of the relationships
between parents or partners. For this reason and because of the
small number of questions, no attempt was made to measure the
various aspects of family functioning.
Effect of
Family
Relations on
Children
Other surveys have shown that the relationship between family
members has a considerable effect on children. The results of the
Ontario Child Health Study have shown, for example, that there is
an important link between family dysfunction and certain mental
conditions in children.
Administerin
g the Family
Functioning
Scale
The family functioning scale was administered to either the PMK or
the spouse/partner as part of the Parent Questionnaire. The unit of
analysis for the scale is the family. The scale includes twelve
questions, each of which contains four response categories. In
order for the lowest score value to be 0, the value of the categories
was reduced by 1 in calculating the score. The order of the
categories was reversed for questions having a negative loading
(BFNHQ01A, Q01C, Q01E, Q01G, Q01I, and Q01K). The total
score (BFNHS01) may therefore vary between 0 and 36, a high
score indicating family dysfunction.
NLSCY Data Users Guide
102
2001/2002
Results
The factor structure of each scale was determined based on data
from the first cycle. The factor structure imposed on the scales
already used in the first cycle and repeatedly used in the second
cycle of the survey was the result of analyses done based on data
from the first cycle.
Nonresponse
Rate
In analysing this scale, unweighted data 1 were used. The sample
size for the scale was 13,439 families. However, once the
observations containing missing values were eliminated, the
analysis dealt only with 13,190 families. The non-response rate for
the different variables was between 1.3 and 1.4%, whereas for the
total score, a non-response rate of 1.9% was obtained. There was
no imputation for the variables in this scale.
Cronbach’s
Alpha for
Family
Functioning
Scale
Following single-factor analysis, all 12 variables of the scale were
kept since the loading values were well above the established
threshold. The Cronbach alpha coefficient (calculated using SAS
software) was 0.88. The variable AFNHQ01L showed the highest
correlation (0.66) with the total score (once the variable was
removed), whereas the variable showing the lowest correlation was
AFNHQ01A with a correlation of 0.51. The Cronbach alpha
coefficient calculated by omitting one variable was stable at about
0.87 for the 12 variables.
Distribution
of Values for
the Family
Functioning
Scale
When the values for the factor score for the family functioning scale
are examined for the NLSCY children, the distribution that is
observed is not a continuous one. In fact, the most common score
is 12. This is a result of the fact that there are 12 items in the scale
and four possible rescaled values (0 to 3). Many respondents had a
rescaled score of 1 for every item in the scale and thus an overall
score of 12. This means that the respondent answered "agree" to
all of the items in the scale which were positive and "disagree" to
all of the negative items, as opposed to the more extreme answers
of "strongly agree" or "strongly disagree." Basically this artifact in
the scale score is due to the fact than many respondents were
consistent in their answering pattern across items.
1
Weighted data could not be used since the weights developed for the NLSCY are for children only, and
not for families.
NLSCY Data Users Guide
103
2001/2002
Activities
Activities
Scale-10/13
Years
(BACCS6)
The object of the activities scale is to measure the child’s
participation in home responsibilities. In Cycle 2, the factor scores
were derived based on the factorial structure identified in Cycle 1.
Below is a description of the items that were included on the
questionnaire to measure activities, the analysis used to construct
the scale and the results of these analyses, all from Cycle 1.
Questionnair
e Items
In Cycle 1, questions ACCSQ6A- ACCSQ6F were tested and
questions ACCSQ6A- ACCSQ6E were used to construct the
scale. Only Children aged 10 and 11 years answered these
questions. This set of questions about responsibilities is from the
Home Observation for Measurement of the Environment-Short
Form questionnaire in the National Longitudinal Survey of Youth,
Ohio State University.
Analysis of
the NLSCY
Data
To construct the Activities Scale for the NLSCY, a factor analysis
was conducted to test the theoretical construct. In the factor
analysis the items were multiplied by the child’s normalized weight.
An individual’s statistical weight is normalized by dividing his/her
weight (AWTCW01) by the average weight of all individuals.
Consequently, the sum of the normalized weights is equal to the
sample size.
Once the factor structures were analysed and the items included in
the factor was determined, the score was calculated. No imputation
was done on the values. If any values were missing the final score
was set to missing. A value may be missing if the child refused to
answer or did not know the answer to the question.
To produce the score, 1 was subtracted from each item so that the
lowest score would be 0. The final score was derived by totaling
the values of all items with non-missing values. The score ranges
from 0 to 15. A score of 0 indicates the respondent does not
participate in home responsibilities.
NLSCY Data Users Guide
104
2001/2002
Results
In the sample there were 3,434 children aged 10 or 11 years. They
were divided into two sub samples of size 1,705 and 1,729 and an
analysis was done on each sample. The non-response rates for
the 5 items was 1.3%. Individuals with missing values were
excluded from the analysis conducted for the purpose of
constructing the factor. After these exclusions. The sub-samples
contained 1,680 and 1,709 individuals respectively, for analysis
purposes. No imputation took place. As a result of factor analysis,
one factor was identified: the activities factor (AACCS6). Items
AACCQ6A-AACCQ6E loaded into the factor.
Cronbach’s
Alpha for
Activities
Cronbach’s alpha coefficients (raw values) were calculated with
SAS, using the normalized weighted data. Please note that, in
general, Cronbach’s alphas calculated with SAS are lower than
those produced by the SPSS software package. The Cronbach
alpha for the activities score was 0.778. The item that affects the
factor the most is AACCQ6B. If it were removed from the analysis,
the Cronbach’s alpha would drop to 0.705. The final activities
score could not be calculated for 45 (1.3%) individuals, due to
missing values for the items comprising this factor.
My Parents and Me Scale (BPRCbS07 and BPRCbS08) - Parent
Objective
The objective of the My Parents and Me scale is to measure the
parent’s perception of his/her relationship with his/her child. This
was asked only for children 12 or 13 years of age. Below is a
description of the items that were included in the My Parents and
Me section of the parent report questionnaire to measure family
relations, the analysis used to construct the scale and the results of
these analyses.
Questionnair
e Items
Questions BPRCQ29A to BPRCQ29R were taken from the
Western Australia Child Health Survey. The scale was developed
by Lempers et al. (1989) based on work of Schaefer (1965) and
Roberts et al. (1984) and measures parental nurturance, rejection
and monitoring.
NLSCY Data Users Guide
105
2001/2002
Analysis of
the NLSCY
Data
To construct the My Parents and Me Scale for the NLSCY, a factor
analysis was conducted to test the theoretical construct. In the
factor analysis the items were multiplied by the child’s normalized
weight. An individual’s statistical weight is normalized by dividing
his/her weight (BWTCW01C) by the average weight of all
individuals. Consequently, the sum of the normalized weights is
equal to the sample size.
Imputation
for Missing
Values
Once the factor structures were analysed and the items included in
each the factor were determined, the scores was calculated.
Imputation was done for missing values. The imputed values were
imputed using the SAS PRINQUAL procedure that determines
which of the possible values for an item is the most plausible for an
individual in view of his/her response profile, the response profiles
of others in the sample, and the number of factors included in the
analysis.
Missing
Values
If too many values were missing the final score was set to missing.
To produce the final scores, 1 was subtracted from each item so
that the lowest score would be 0. The final score was derived by
totaling the values of all items with non-missing values. A score of
0 indicates the following for the two factors that were found to exist
in the My Parents and Me scale:
-a low degree of parental nurturance for the parental nurturance
score;.
-a low degree of parental rejection for the parental rejection score;
and
Results
In the sample there were 2,258 children aged 12 or 13 years. They
were divided into two sub samples and analysis was done on each
sub-sample. Individuals with missing values were excluded from
the analysis conducted for the purpose of constructing the factor.
After these exclusions the sub-samples contained 1,076 and 1.146
individuals respectively. As a result of the factor analyses, two
factors were identified: the parental nurturance factor and the
parental rejection factor. The items that comprised each factor are
described in the following table.
NLSCY Data Users Guide
106
2001/2002
MY PARENTS AND ME SCALE FOR CHILDREN AGED 12 AND 13 YEARS OLD
(PARENT REPORT).
FACTOR
SCORE
ITEMS
Parental
Nurturance
BPRCbS07
BPRCQ29A, BPRCQ29H, BPRCQ29I, BPRCQ29L,
BPRCQ29N, BPRCQ29R
Parental
Rejection
BPRCbS08
BPRCQ29C, BPRCQ29G, BPRCQ29J, BPRCQ29K,
BPRCQ29M, BPRCQ29P, BPRCQ29Q
Cronbach’s
Alpha For 12
/ 13 Year
Olds
Cronbach’s alpha coefficients (raw values) were calculated with
SAS, using the normalized weighted data. Please note that, in
general, Cronbach’s alphas calculated with SAS are lower than
those produced by the SPSS software package. Cronbach’s
alphas for these factors are given in the table below.
CRONBACH’S ALPHA VALUES FOR MY PARENTS AND ME SCALE: 12/13 YEAR
OLDS (PARENT REPORT)
Factor
Cronbach’s
alpha
Parental Nurturance
(BPRCbS07)
Parental Rejection
(BPRCbS08)
NLSCY Data Users Guide
0.780
Items that lowered
cronbach’s alpha the
most if excluded
BPRCQ29N
Cronbach’s
alpha if the item
is excluded
0.729
0.747
BPRCQ29M
0.710
107
2001/2002
Child Scales from Self-completed Questionnaire
Friends and
Family (selfcomplete, 1013)
Friends and Family was one of the sections on the questionnaire
completed by children in the 10 to 13 age group. The objective was
to determine how well the child felt he/she was getting along with
others.
Information
Collection
The section collected information on numbers of close friends, time
spent with friends, presence of someone the child can confide in,
and the quality of relationships with others, such as parents, peers
and teachers. This information is important in identifying the extent
and quality of the child's social support network. To allow for
comparison, the section includes questions which are also included
on the Child's Questionnaire completed by the PMK.
Peer
Relations
Sub-scale
There was one group of questions in this section which were part of
a scale. Items BFFCQ01, BFFCQ02, BFFCQ03 and BFFCQ04
are intended to measure how well the child gets along with peers. It
is part of the Peer Relations Sub-scale from the Marsh SelfDescription Questionnaire, developed by H.W. Marsh.
Friends
Scale
(BFFCS01)
The object of the friends scale is to measure how well the child
feels he/she gets along with his/her peers. In order to understand
how the factorial structure was determined in Cycle 1, below is a
description of the items that were included on the questionnaire in
Cycle 1 to measure peer relations, the analysis used to construct
the scale and the results of these analyses.
Questionnair
e Items
In Cycle 1, questions AA1CQ01 to AA1CQ04 were used to
construct the scale. This set of questions on getting along with
peers is the Peer relations Subcale from the Marsh SelfDescription Questionnaire.
Analysis of
the NLSCY
Data
To construct the Friends Scale for the NLSCY, a factor analysis
was conducted to test the theoretical construct. In the factor
analysis the items were multiplied by the child’s normalized weight.
NLSCY Data Users Guide
108
2001/2002
Score
Calculation
Once the factor structures were analysed and the items included in
the factor was determined, the score was calculated. No imputation
was done on the values. If any values were missing the final score
was set to missing. A value may be missing if the child refused to
answer or did not know the answer to the question.
To produce the score, 1 was subtracted from each item so that the
lowest score would be 0. The final score was derived by totaling the
values of all items with non-missing values. The score ranges from
0 to 16. A score of 0 indicates the respondent does not have a lot
of friends and does not have positive relations with other children.
Results
In the sample in Cycle 1, there were 3,434 children aged 10 or 11
years. They were divided into two sub samples of size 1,705 and
1,729 and analysis was done on each sample. The non-response
rates for the 4 items ranged from 10.9% to 11.5%. Individuals with
missing values were excluded from the analysis conducted for the
purpose of constructing the factor. After these exclusions, the subsamples contained 1,508 and 1,529 individuals respectively, for
analysis purposes. No imputation took place. As a result of factor
analysis, one factor was identified: the friends factor (AA1CS01).
All items - AA1CQ01 to AA1CQ04 - loaded into the factor.
Cronbach’s alpha coefficients (raw values) were calculated with
SAS, using the normalized weighted data. Please note that, in
general, Cronbach’s alphas calculated with SAS are lower than
those produced by the SPSS software package. The Cronbach
alpha for the friends score was 0.779. The item that affects the
factor the most is AA1CQ04. If it were removed from the analysis,
the Cronbach’s alpha would drop to 0.689. The final friends score
could not be calculated for 397 (11.6%) individuals, due to missing
values for the items comprising this factor.
NLSCY Data Users Guide
109
2001/2002
Feelings and Behaviour (self-complete, 10-15)
Feelings and
Behaviour
This section was part of the self-complete questionnaire given to
children in the 10 to 15 age group. The objective of this section was
to determine the child's perception of his/her general behaviour and
the child's engagement in risk-taking behaviours.
Behaviour
Checklist
This section replicates the behaviour checklist included on the
Child's Questionnaire completed by the PMK for those aged 10-11
and the one on the Teacher's Questionnaire. It is intended to
provide indicators of the following behaviours: conduct disorder,
hyperactivity, inattention, physical aggression, indirect aggression,
emotional disorder, anxiety and prosocial behaviours. In Cycle 2,
the factor scores were derived based on the factorial structure
identified in Cycle 1.
Analysis of
the NLSCY
Data
The following indicates the constructs or factors that the behaviour
scale was intending to measure, the items that were included in the
factor and the sources for the items.
NLSCY Data Users Guide
110
2001/2002
! Conduct disorder:
Items include AD1CQ01C, E, G, L, O, T, AA, DD, FF, JJ, and PP
from the Ontario Child Health Study (OCHS).
! Hyperactivity
Items include AD1CQ01B, I, N, P, S and W from the Ontario Child
Health Study and AD1CQ1HH from the Montreal Longitudinal
Survey.
! Emotional disorder
Items include AD1CQ01F, K, Q, V, CC, MM, and RR from the
Ontario Child Health Study.
! Anxiety
Items include AD1CQ01Y and AD1CQ1II from the Montreal
Longitudinal Survey and several of the OCHS emotional disorder
items - AD1CQ01F, Q, V and CC.
! Indirect aggression
Items include AD1CQ01J, R, Z, LL and TT from Lagerspetz,
Bjorngvist and Peltonen of Finland.
! Physical aggression
Items include AD1CQ01X from the Montreal Longitudinal Survey
and AD1CQ01G, AA and NN from the Ontario Child Health Study.
! Inattention
Items include AD1CQ01P from the Ontario Child Health Study and
AD1CQ1EE, KK, QQ from the Montreal Longitudinal Survey.
! Prosocial behaviour
Items include AD1CQ01A, H, M GG and OO from the Ontario Child
Health Study and AD1CQ01D, U, BB, SS, and UU from the
Montreal Longitudinal Survey.
Constructing
the
Behaviour
Scale
In Cycle 1, to construct the Behaviour Scale for the NLSCY, a factor
analysis was conducted to test the theoretical construct. In order to
be consistent with the behaviour scale created from the parent
questionnaire, the factor structure which emerged from the 4-11
behaviour scale was imposed on the 10/11 behaviour scale.
NLSCY Data Users Guide
111
2001/2002
Normalized
Weight
In the factor analysis the items were multiplied by the child’s
normalized weight. An individual’s statistical weight is normalized
by dividing his/her weight (AWTCW01) by the average weight of all
individuals. Consequently, the sum of the normalized weights is
equal to the sample size.
Score
Calculation
Once the factor structures were analysed and the items included in
each the factor were determined, the scores was calculated. Some
items were imputed. The imputed values were imputed using the
SAS PRINQUAL procedure that determines which of the possible
values for an item is the most plausible for an individual in view of
his/her response profile, the response profiles of others in the
sample, and the number of factors included in the analysis.
Producing
Final Scores
To produce the final scores, 1 was subtracted from each item so
that the lowest score would be 0. The score for each factor on the
scale was computed at by totaling the values of the items that made
up the factor (including imputed values). The score was set to
‘missing’ if too many of the values of any items included in the
factor were unreported. A value may be missing if the child refused
to answer the item. A score of 0 indicates that the child has no
problems for any of the factors in the behaviour scale with the
exception of the prosocial factor, where a score of 0 indicates the
absence of prosocial behaviour.
Results
In the sample there were 3,434 children aged 10 or 11 years. They
were divided into two sub samples of size 1,705 and 1,729 and
analysis was done on each sample. The non-response rates for the
8 items ranged from 13.6% to 16.7%. Individuals with missing
values were excluded from the analysis conducted for the purpose
of constructing the factor. After these exclusions, the sub-samples
contained 1,352 and 1,398 individuals respectively, for analysis
purposes. As a result of imposed factor analysis, five factors were
identified: hyperactivity-inattention, prosocial behaviour, emotionaldisorder-anxiety, physical aggression-conduct disorder, and
indirect aggression. The items that comprised each factor are
described in the following table.
NLSCY Data Users Guide
112
2001/2002
BEHAVIOUR SCALE FOR 10 AND 11 YEARS OLD.
FACTOR
Indirect aggression
SCORE
AD1CS01
Emotional disorder
AD1CS02
Conduct disorder and
physical aggression
Hyperactivity/inattention
AD1CS03
Prosocial behaviour
AD1CS05
Cronbach’s
Alpha for
Behaviour
Scale
AD1CS04
ITEMS
AD1CQ01J, AD1CQ01R, AD1CQ10Z,
AD1CQ10LL, and AD1CQ01TT
AD1CQ1F, AD1CQ1K, AD1CQ1Q,
AD1CQ1V, AD1CQ1CC, AD1CQ1II,
AD1CQ1MM, and AD1CQ1RR
AD1CQ1G, AD1CQ1X ,AD1CQ1AA,
AD1CQ1FF, AD1CQ1JJ, and AD1CQ1NN
AD1CQ1B, AD1CQ1I, AD1CQ1N, AD1CQ1P,
AD1CQ1S, AD1CQ1W, AD1CQ1HH and
AD1CQ1QQ
AD1CQ1A , AD1CQ1D, AD1CQ1H,
AD1CQ1M, AD1CQ1U, AD1CQ1BB,
AD1CQ1GG, AD1CQ1OO, AD1CQ1SS, and
AD1CQ1UU
Cronbach’s alpha coefficients (raw values) were calculated with
SAS, using the normalized weighted data. Please note that, in
general, Cronbach’s alphas calculated with SAS are lower than
those produced by the SPSS software package. Cronbach’s
alphas for these factors are given in the table below.
NLSCY Data Users Guide
113
2001/2002
CRONBACH’S ALPHA VALUES FOR BEHAVIOUR SCALE: 10/11 YEAR OLDS
Factor
Cronbach’s
alpha
0.728
Items that lowered
cronbach’s alpha the
most if excluded
AD1CQ1LL
Cronbach’s
alpha if the item
is excluded
0.657
Indirect aggression
(AD1CS01)
Emotional disorder
(AD1CS02)
Conduct disorder and
physical aggression
(AD1CS03)
Hyperactivity/inattention
(AD1CS04)
Prosocial behaviour
(AD1CS05)
0.760
AD1CQ1II
0.717
0.738
AD1CQ1AA
0.678
0.751
AD1CQ1QQ
0.717
0.766
AD1CQ1SS
0.741
The scores for these factors could not be computed in, 566 (16.5%), 597 (17.4%), 585
(17%), 621 (18.1%) and 587 (17.1%) cases respectively because of unreported values.
My Parents and Me (self-complete 10-15)
Objective - My
Parents and
Me
This section was part of the self-complete questionnaire given to
children in the 10 to 15 age group. The objective was to
complement the Parenting Section on the Child's Questionnaire
completed by the PMK by gathering information directly from the
child regarding his/her perception of his/her relationship with
parents. For the self-completed questionnaire, it was also
considered important to obtain a measure of parental supervision
(i.e., monitoring), as this has been shown to be linked to child
outcomes - there is a correlation between a lack of supervision and
negative outcomes, such as juvenile delinquency and other risktaking behaviours.
NLSCY Data Users Guide
114
2001/2002
Scale Used for
My Parents
and Me
The scale that was used was also used in the Western Australia
Child Health Survey. It was developed by Lempers et al (1989)
based on work of Schaefer (1965) and Roberts et al (1984) and
measures parental nurturance, rejection and monitoring. This
information will complement the constructs measured in the parentcompleted Child's Questionnaire (positive child-parent interaction,
hostile/ineffective child-parent interaction, and consistent childparent interaction, aversive and non-aversive parent management
techniques.)
My Parents
and Me Scale
(CPMCcS1,
CPMCbS2B,
CPMCcS3)
The objective of the My Parents and Me scale is to measure the
child’s perception of his/her relationship with his/her parents and
parental supervision. Below is a description of the items that were
included on the 10-15 year old questionnaires to measure family
relations, the analysis used to construct the scale and the results of
these analyses.
Questionnaire
Items
Questions CPMCcQ1A to CPMCcQ1Q were taken from the
Western Australia Child Health Survey. In addition to these
questions, questions CPMCcQ1R to CPMCcQ1T were also used.
The scale was developed by Lempers et al. (1989) based on work
of Schaefer (1965) and Roberts et al. (1984) and measures
parental nurturance, rejection and monitoring.
Analysis of the
NLSCY Data
To construct the My Parents and Me Scale for the NLSCY, a factor
analysis was conducted to test the theoretical construct. In the
factor analysis the items were multiplied by the child’s normalized
weight. An individual’s statistical weight is normalized by dividing
his/her weight (CWTCW01C) by the average weight of all
individuals. Consequently, the sum of the normalized weights is
equal to the sample size.
NLSCY Data Users Guide
115
2001/2002
Score
Calculation
Once the factor structures were analysed and the items included in
each factor were determined, the scores were calculated.
Imputation was done for missing values. The imputed values were
imputed using the SAS PRINQUAL procedure that determines
which of the possible values for an item is the most plausible for an
individual in view of his/her response profile, the response profiles
of others in the sample, and the number of factors included in the
analysis.
Missing
Values
If too many values were missing the final score was set to missing.
To produce the final scores, 1 was subtracted from each item so
that the lowest score would be 0. The final score was derived by
totaling the values of all items with non-missing values. A score of
0 indicates the following for the three factors that were found to
exist in the My Parents and Me scale:
-a low degree of parental nurturance for the parental nurturance
score;
-a low degree of parental rejection for the parental rejection score;
-a low degree of parental monitoring for the parental monitoring
score.
Results (Cycle
3)
In the sample of 10-15 year olds there were 5,539 children. The
sample was divided into two sub-samples and an analysis was
done on each sample. Individuals with missing values were
excluded from the analysis conducted for the purpose of
constructing the factor. After these exclusions, the two subsamples contained 2509 and 2584 individuals respectively.
Three Factors
Identified for
10-15 Year
Olds
As a result of the factor analyses, three factors were identified for
the 10-15 year olds: the parental nurturance factor, the parental
rejection factor and the parental monitoring factor. The items that
comprised each factor are described in the following table.
MY PARENTS AND ME SCALE FOR CHILDREN AGED 10 TO 15 YEARS OLD.
NLSCY Data Users Guide
116
2001/2002
FACTOR
Parental
Nurturance
Parental
Rejection
Parental
Monitoring
Cronbach’s
Alpha for My
Parents and
Me
SCORE
CPMCcS1
CPMCbS2B
CPMCcS3
ITEMS
CPMCcQ1A, CPMCcQ1D, CPMCcQ1K, CPMCcQ1M,
CPMCcQ1Q CPMCcQ1H CPMCcQ1I
CPMCcQ1C, CPMCcQ1G, CPMCcQ1J, CPMCcQ1L,
CPMCcQ1O, CPMCcQ1P, CPMCcQ1R
CPMCcQ1B, CPMCcQ1F, CPMCcQ1N, CPMCcQ1E,
CPMCcQ1T
Cronbach’s alpha coefficients (raw values) were calculated with
SAS, using the normalized weighted data. Please note that, in
general, Cronbach’s alphas calculated with SAS are lower than
those produced by the SPSS software package. Cronbach’s alphas
for these factors are given in the table below.
CRONBACH’S ALPHA VALUES FOR MY PARENTS AND ME SCALE: 10-15 YEAR
OLDS
Factor
Parental Nurturance
(CPMCcS1)
Parental Rejection
(CPMCbS2B)
Parental Monitoring
(CPMCcS3)
Cronbach’
s alpha
0.88
Items that lowered cronbach’s
alpha the most if excluded
CPMCQ1M
Cronbach’s alpha if
the item is excluded
0.855
0.73
CPMCQ1O
CPMCcQ1R
CPMCcQ1T
0.504
0.680
0.459
0.57
About me (self-complete 10-15)
Objective About Me
Scales
(BAMCS01,
BAMCS02)
The objective of the About me scale is to measure the child’s overall
self-esteem and perception of physical appearance. Specifically,
two scales were used: one was designed to measure overall selfesteem and the other was designed to measure perceptions of
physical appearance.
Factor Scores
In Cycle 2, the factor scores were derived based on the factorial
structure identified in Cycle 1. Below is a description of the items
that were included on the questionnaire to measure these scales,
the analysis used to construct the scale and the results of these
analyses, all from Cycle 1.
NLSCY Data Users Guide
117
2001/2002
Questionnaire
Items
In Cycle 1, questions AA1CQ01A to AA1CQ01D on overall self
esteem were taken from the General-Self Scale of the Marsh Self
Description Questionnaire developed by H.W Marsh. Questions
AA1CQ01E to AA1CQ01H on perceptions of physical appearance
were taken from the Physical Appearance Scale of the Marsh Self
Description Questionnaire developed by H.W Marsh
Analysis of the
NLSCY Data
To construct the About me Scale for the NLSCY, a factor analysis
was conducted to test the theoretical construct. In the factor analysis
the items were multiplied by the child’s normalized weight. An
individual’s statistical weight is normalized by dividing his/her weight
(AWTCW01) by the average weight of all individuals. Consequently,
the sum of the normalized weights is equal to the sample size.
Missing Values
Once the factor structures were analysed and the items included in
each the factor were determined, the scores were calculated. No
imputation was done for missing values. If any values were missing,
the final score was set to missing. To produce the final scores, 1
was subtracted from each item so that the lowest score would be 0.
The final score was derived by totaling the values of all items with
non-missing values. A score of 0 indicates the following for the two
factors that were found to exist for in the About me scales:
-a lack of general self esteem for the general self scale; and
-a negative perception of physical appearance for the physical
appearance score.
Results
In the sample there were 3,434 children aged 10 or 11 years. They
were divided into two sub-samples of sizes 1,705 and 1,729 and
analysis was done on each sample. The non-response rates for the
8 items ranged from 14% to 15.8%. Individuals with missing values
were excluded from the analysis conducted for the purpose of
constructing the factor. After these exclusions, the sub-samples
contained 1,371 and 1,413 individuals respectively, for analysis
purposes. As a result of factor analysis, two factors were identified:
the general self factor and the physical appearance factor. The
items that comprised each factor are described in the following
table.
NLSCY Data Users Guide
118
2001/2002
GENERAL SELF SCALE FOR CHILDREN AGED 10 AND 11 YEARS OLD.
FACTOR
General Self
Physical
Appearance
Cronbach’s
Alpha for
About Me
Scales
SCORE
AC1CS02
AC1CS01
ITEMS
AC1CQ01A, AC1CQ01B AC1CQ01C AC1CQ01D
AC1CQ01E, AC1CQ01F AC1CQ01G AC1CQ01H
Cronbach’s alpha coefficients (raw values) were calculated with
SAS, using the normalized weighted data. Please note that, in
general, Cronbach’s alphas calculated with SAS are lower than those
produced by the SPSS software package. For the general self score
the Cronbach alpha was 0.728. The item that affects the factor the
most is AC1CQ01C. If it were removed from the analysis, the
Cronbach’s alpha would drop to 0.629. For the physical appearance
score the Cronbach alpha was 0.874. The item that affects the factor
the most is AC1CQ01E. If it were removed from the analysis, the
Cronbach’s alpha would drop to 0.811. Once the factors were
determined, the next step was to calculate the scores for each of the
two factors. For the general self factor, scores could not be
calculated for 555 individuals (16.2%), due to missing values for the
items comprising this factor. For the physical appearance factor,
scores could not be calculated for 589 individuals (17.2%), due to
missing values for the items comprising this factor.
NLSCY Data Users Guide
119
2001/2002
Depression Scale (self-complete 12-15)
Depression
Scale
(BHTCbS1B)
Score
Calculation
In order to be consistent with the depression scale created from the
parent questionnaire, the factor structure which emerged from the
parental scale for PMK depression was imposed on the 12/13
depression scale.
In order to produce the score, 1 was subtracted from each item so
that the lowest score would be 0. The final score was derived by
totaling the values of all items with non-missing values. As well, the
answer categories were reversed for questions having a negative
loading (BHTCb11F, 11H, and 11J). The total score (BHTCbS1B)
may therefore vary between 0 and 36, a high score indicating the
presence of depression symptoms.
Education (Parent)
ObjectiveEducation
(Parent)
The Education Section was completed for both the PMK and
spouse/partner. The objective was to gather information on the years
of school completed, educational attainment, and current attendance
at an educational institution.
Research (for example, the Ontario Child Health Study and the
National Longitudinal Survey of Youth in the United States) has
indicated a link between maternal educational attainment, the home
environment and child development. The questions on full-time and
part-time school attendance provide an indicator of the main
activities of the PMK and the spouse/partner.
Values for
CEDPD02 and
CEDSD02
The variables (CEDPD02 for the PMK and CEDSD02 for the
spouse/partner) have the following values.
! less than secondary
! secondary school graduation
! beyond high school
! college or university degree (including trade).
The other education variable included is current school status and
whether attendance is full-time or part-time.
NLSCY Data Users Guide
120
2001/2002
Socio-demographic Characteristics
Objective Sociodemographic
Characteristics
The objective of the Socio-demographic Section was to gather
information on immigration, ethnic background and the language
profile of household members. This will allow analysis for various
components of the Canadian population and will permit
identification of visible minorities. As well, there were questions on
religious affiliation and frequency of attendance at religious
services. Religion, particularly frequency of attendance, is
acknowledged as having a positive influence on a child's
development.
Suppression of
Variables
It was necessary to suppress many of the variables in this section
on the micro data file due to confidentiality concerns. The questions
on country of birth, ethnicity and religion have all been suppressed
while frequency of attendance at religious services has been
included.
Questions on
Mother Tongue
and Language
of Conversation
The questions on mother tongue and language of conversation are
included on the micro data file but only with aggregated answer
categories:
Aggregated
Variables for
Language
The aggregated variables for language of conversation are labeled
CSDPD05B, CSDSD05B, and CSDCD05B, for the PMK,
Spouse/partner and Child on the micro data file. The mother tongue
variables are CSDPD06B, CSDSD06B and CSDCD06B.
! English only
! French only
! English and French only
! at least one "other" language indicated.
For the immigrant population, a derived variable was created to
indicate the number of years since first immigrating to Canada. It
was possible to put a grouped version of this derived variable on
the micro data file (CSDPD02B, CSDSD02B, CSDCD02B).
NLSCY Data Users Guide
121
2001/2002
Remote Access
Since there are many variables in this section which have been
suppressed for the micro data file, researchers who are particularly
interested in conducting analyses on socio-demographic variables
are encouraged to consider making use of the remote access
service described in Section 13.3.
Labour Force (Parent)
Employment
Stability
Employment stability impacts the home environment, both in terms of
income and stress levels. Research, conducted for the Ontario Child
Health Study, indicates that parental unemployment can adversely
impact child mental health.
Objective Labour Force
The Labour Force Section was completed for both the PMK and the
spouse/partner. The main objective of the section was to determine
employment stability as an indicator of the continuity of employment
income. Questions included, periods of absence from work, reason
for the most recent absence, hours worked, and work arrangements
(e.g. shifts) during the previous year. Information was collected on the
main job and on all jobs for a one-year period.
Respondents
and
Employment
Respondents were asked to identify what they considered to be their
main job over the previous year (if they had more than one job). A
complete description was recorded for this main job and industry and
occupation coding was carried out (using 1980 Standard Industrial
Classification codes and 1980 Standard Occupational Classification
codes).
Wages and
Salaries
Data on wages and salaries for this main job were collected. Wage
rate data provides an additional source of information on income.
This data will be useful in analysing choices which parents,
particularly mothers, face in deciding to stay at home or to return to
the labour force.
NLSCY Data Users Guide
122
2001/2002
NLSCY Data Users Guide
123
2001/2002
Work Duration Derived Variables
Work Duration
Derived
Variables
Jobs Held
During the
Previous Year
With the data collected in the Labour Force Section it was possible
to create a series of derived variables to describe the stability of
work for the PMK and spouse/partner over the previous year.
As mentioned above, a series of questions were asked about all jobs
the PMK and spouse/partner held during the previous year. As well,
in order to address absences within a job the following question was
asked as the initial lead-in question to a job:
Did you have that job one year ago, without a break in employment
since then?
There is, moreover, a derived variable (CL FPD33) for indicating the
number of weeks worked by the PMK in a job or company the
previous year.
Response
Burden
Current
Collection
Tool
In the first cycle of the survey, an employment vector of 53 weeks was
established based on information about each job held, to a maximum
of six jobs. To reduce the respondent's response burden, this
collection method was abandoned in favour of a more general
section. A good many variables derived from Cycle 1 were
reproduced, but it should be noted that while considerable effort was
made to keep the same definitions, the collection tool was changed
substantially.
With the current collection tool, it is still possible to gather labour
force data for the previous year, but in a more general way. A series
of questions was used to determine the number of weeks worked in
the 12 previous months, the number of weeks the individual was
absent from work, the number of weeks the individual was without
work but seeking employment, and so on. Moreover, the tool focuses
on the current main job or, if applicable, the most recent job. A
detailed description of this job was obtained (employer, type of
company, nature of the work, main duties, status, hours worked,
salary).
NLSCY Data Users Guide
124
2001/2002
Other Derived
Variables
This release includes other derived variables which describe the
employment picture over the reference year, such as number of
weeks worked part-time, number of weeks worked full time, etc.
Demographic Variables
Demographic
Variables
The demographic variables discussed in this section refer to
variables collected on the household roster. As part of the household
roster, some basic demographic information (e.g., age, gender,
marital status) was collected for all members of the child's household.
The relationship grid was also completed as part of this
questionnaire i.e., the relationship of everyone in the household to
everyone else. Using this information it was possible to create an
extensive set of variables to describe the child's family situation. Most
of these derived variables are critical to the analyses of NLSCY data
and are described in Section 8 (NLSCY Concepts and Definitions).
Edits on the
Relationship
Grid Data
If was necessary to perform an extensive series of edits on the data
that were collected as part of the relationship grid. There were some
edits that were carried out as part of the CAI system during collection.
However in the data that were received at Head Office there were still
inconsistencies.
Examples of
Editing
The following are some examples of the types of editing that was
carried out:
! in all relationships reported, a person could not have more than two
parents; and
! the difference in age between a husband and wife had to be less
than 29 years.
In total there were over 30 relationship edits performed. Some of the
edits were what is known as "soft" edits and some were "hard." The
first example was a hard edit and the second a soft edit. For all edit
failures, the records for the entire household were reviewed manually
for obvious mistakes. A correction had to be made for the hard edit
failures. For the soft edit failures a correction was made if it was
deemed appropriate to do so.
NLSCY Data Users Guide
125
2001/2002
Sources of
Error
The major source of error for relationship data had to do with step
children. There were several cases where a female parent was living
with a biological child and a spouse or common-law partner. The
relationship of the male partner to the child was coded as
"unrelated." For questionnaires completed in French this relationship
was often coded as "in-law." In the edit, the relationship code was
changed to step child for these cases. As a result of the relationship
edits the number of children in step families increased by close to
40%.
Medical/Biological
Medical /
Biological
Children
Under 2
Birth Weight
The Medical /Biological Section was completed for children in the 0
to 3 age group. The major objective was to collect information on
factors such as gestational age and birth weight. These factors have
been shown to have a direct impact on a child's growth and
development. For example, in the long term, underweight babies face
higher risks of poor health as well as longer-lasting developmental
difficulties.
For each child under two, the nature of the delivery, general health of
the child at birth and the use of specialized services following the
birth were collected in this section. The NLSCY also investigated the
biological mother's pregnancy and delivery history, topics such as the
mother's breast-feeding experiences and prenatal lifestyle.
Since birth weight is such an important variable, caution was taken in
editing this variable. The records for children with very low birth
weights (< 1.5 kilograms) were examined to verify that the response
was legitimate. Other variables considered in the edit were the length
of the baby at birth, the number of days early of the delivery, the
conditions of the delivery (e.g., multiple birth and special medical
care) and the health of the child at birth. If there was nothing to
corroborate the low birth weight it was set to "not-stated."
NLSCY Data Users Guide
126
2001/2002
Gestational
Age of the
Child
There were a couple of derived variables created for this section that
bear note. Two variables were derived to indicate the gestational
age of the child. CMDCD06 gives the gestational age in days and
CMDCD07 indicates if the child was born prematurely (gestational
age 258 days or less), in the normal range (gestational age 259 to
293 days) or late (gestational age 294 days or later).
A variable was derived (CMDCD08) to indicate if the child was of
normal birth weight (2500 grams), moderately low birth weight (1500
to 2499 grams) or very low birth weight (< 1500 grams).
NLSCY Data Users Guide
127
2001/2002
NLSCY Data Users Guide
128
2001/2002
Chapter 9 - Data Quality
Types of
Errors
The estimates derived from this survey are based on a sample of
children. Somewhat different values might have been obtained if a
complete census had been taken using the same questionnaires,
interviewers, supervisors, processing methods, etc. The difference
between the estimates obtained from the sample and the results from
a complete count taken under similar conditions is called the sampling
error of the estimates.
NonSampling
Errors
Interviewers might misunderstand the instructions, respondents might
make errors while answering the questions, the answers might be
incorrectly entered on the questionnaire, and errors might be
introduced while processing and tabulating the data. These are all
examples of non-sampling errors.
Defining the
Term
Respondent
In certain circumstances, it is not possible to gather all the data about a
child. The definition of the term respondent used in Cycle 1 was again
used for Cycle 2. According to this definition, a child is a respondent if
there is enough information about at least one child in his household.
Cross-sectional and longitudinal response rates
CrossSectional
Response
Rate
The cross-sectional response rate (or collection rate), at the household
level, is shown in the following table. This rate does not provide an
indicator of the quality of cross-sectional estimates, as such an
indicator would account for the non-response rate in previous cycles.
Instead, the rates shown below reflect the efficiency of the datacollection process in Cycle 3.
NLSCY Data Users Guide
129
2001/2002
NLSCY - Cross-Sectional Response Rate by Province
Province
Households
Contacted
1,781
1,030
2,235
Respondent
Households
1,612
948
2,018
Response Rate
New Brunswick
Quebec
2,181
6,963
1,954
6,294
90%
90%
Ontario
10,501
8,651
82%
Manitoba
2,528
2,250
89%
Saskatchewan
2,619
2,306
88%
Alberta
3,583
3,117
87%
British Columbia
TOTAL
3,315
36,736
2,813
31,963
85%
87%
Newfoundland
Prince Edward Island
Nova Scotia
91%
92%
90%
The cross-sectional sample included longitudinal households sampled in Cycles 1 and 2,
as well as households contacted for the first time in Cycle 3 (newborn children selected
from the LFS and the birth register). Since a good number of households were
contacted for the first time in Cycle 3, the overall response rate for Cycle 3 is lower than
that for Cycle 2.
Cross-Sectional
Response Rate
by Sample
Source
The table below gives the response rate for households contacted
for the first time in Cycle 3 as well as for respondent households
contacted in at least one previous cycle.
NLSCY - Cross-Sectional Response Rate by Sample Source
Longitudinal Households Selected in
Cycle 1
Longitudinal Households Selected in
Cycle 2
Newborn Children Selected from the
LFS
1-Year-Old Children Selected from
Birth Register
5-Year-Old Children Selected from
Birth Register
Total
NLSCY Data Users Guide
Households
Contacted
16,563
Respondent
Households
14,777
Response
Rate
90%
3,947
1,999
3,640
1,736
92%
86%
7,542
6,390
85%
6,685
5,420
81%
36,736
31,963
87%
130
2001/2002
As well, the reason for household non-response will be different depending on whether the
household is longitudinal. In fact, longitudinal households are usually more apt to take part
in the survey (having already done so in the past). However, some households may have
moved between the second and third collection cycles. As a result, it is sometimes
necessary to track down the longitudinal children before proceeding with collection. This
operation is not always successful. Longitudinal children who move may thus lead to some
erosion of our longitudinal sample.
New Households
Added to Cycle 2
Non-Respondents by
Reason for Not
Responding
The following tables show the distribution of nonresponding, longitudinal and new households, by reason for
not responding.
NLSCY – New Households Added to Cycle 2
Non-Respondents by Reason for Not Responding
Refusal
No one at home
Language barrier
Special circumstances (sickness, weather conditions, etc.)
Partial response (rejected for lack of information)
Not tracked down
Other/reason unknown
Total
NonResponding
Households
1,051
129
34
190
149
861
266
2,680
%
45%
5%
1%
7%
6%
32%
42%
100%
NLSCY – Longitudinal Households Not Responding
to Cycle 2, by Reason for Not Responding
Refusal
Not tracked down
No one at home
Language barrier
Special circumstances (sickness, weather conditions,
etc.)
Partial response (rejected for lack of information)
Other or reason unknown
Total
NLSCY Data Users Guide
131
NonResponding
Households
1,393
244
35
3
100
66%
12%
2%
0.1%
5%
234
84
2,093
11%
4%
100%
%
2001/2002
Longitudinal
Response Rate
Given the survey method applied to the first two collection cycles,
it was unfortunately impossible to obtain an exact longitudinal
response rate taking into consideration all the components of
erosion. Ideally, this rate would be the simple ratio of the number
of longitudinal children responding to the second cycle to the
number of children contacted for the first cycle. However, the
number of children present in non-responding households during
the first cycle is unknown. The number of children present in
households not responding to the LFS is also unknown. It is
therefore impossible to compute an exact rate since the exact
denominator of this rate is unknown.
In keeping with the custom for longitudinal surveys, we decided to
publish the response rate among respondents for Cycle 1. In the
table below, which gives these rates by province, the percentage
reported is the ratio between the number of respondents for the
cycle in question and the number of respondents in Cycle 1.
NLSCY - Longitudinal Response Rate by Province – Children Selected in Cycle 1
Province
Newfoundland
Prince Edward Island
Nova Scotia
New Brunswick
Quebec
Ontario
Manitoba
Saskatchewan
Alberta
British Columbia
Canada
NLSCY Data Users Guide
No. of Respondents No. of Respondents No. of Respondents
in Cycle 1
in Cycle 2
in Cycle 3
950
467
1,191
1,070
3,182
4,342
1,232
1,413
1,599
1,457
16,903
892 (94%)
443 (95%)
1,068 (90%)
958 (90%)
2,994 (93)%
3,899 (90%)
1,162 (94%)
1,305 (92%)
1,465 (92%)
1,333 (92%)
15,468 (92%)
132
846 (90%)
434 (92%)
1,085 (91%)
958 (90%)
2,845 (90%)
3,762 (87%)
1,114 (90%)
1,257 (89%)
1,420 (89%)
1,284 (88%)
15,005 (89%)
2001/2002
Non-Response
Bias
Non-response is a type of error that can result in bias in survey
estimates. Biased estimates can occur when the characteristics of
non-respondents differ significantly from those of survey
respondents. Bias resulting from non-response during the first
contact was dealt with in the manual for the first cycle. As few
households were added for the second cycle, and since similar
results would be obtained, this study is not taken up for the second
cycle.
A considerable amount of information is available to evaluate this
potential bias. As a result, we attempted to model the “nonresponse to Cycle 2” event using variables obtained during the first
collection cycle. In this context, the non-response event may have
two causes: (a) the decision made by the respondent not to
cooperate; (b) our inability to contact the respondent. This second
cause may be the result of a move or of a temporary absence when
attempts at contact were made. The model must therefore include
two distinct phenomena: mobility and cooperation.
Separate models have been developed for each region in the
country in order to take into consideration the characteristics of each
one. Note that the decision to cooperate or not in a survey is made
by an adult. As a result, the explanatory variables for these models
are in fact characteristics of adults.
Regional
Models
Without entering into the details of each regional model, here are
some of the conclusions that were drawn:
·
- People with a lower income show lower response rates than
people with a higher income.
·
- People with a lower level of education show lower response rates
than people with a higher level of education.
- People living in a large city show lower response rates than people
living in smaller cities.
- The presence of a spouse in the household is associated with
better response rates.
NLSCY Data Users Guide
133
2001/2002
Weighting
Process
In order to minimize the risk associated with this potential bias, the
models were used for the weighting process (see Section 7). This
technique helps correct sampling weights in order to account for the
potential bias resulting from non-response. However, it does not
guarantee that there is no bias induced by non-response. There
remains a latent risk, and we must remain watchful. That is why there
is considerable effort to minimize and study non-response, during
both collection and processing.
Other Sources
of Bias
All children covered by the NLSCY have been selected among
households having already taken part in the Labour Force Survey.
This method of selection leads to three problems which might
produce bias in our estimates.
First Problem
The first problem stems from the fact that only respondents to the
LFS have been considered for the NLSCY sample. It could be that
some of the LFS non-respondents had children in the appropriate
age group. These households were not included in the NLSCY
sample, which could be a source of bias.
Second
Problem
The second problem is due to the fact that only households having
children when the LFS was conducted were included in the NLSCY
sample. It could be that some households were not included in the
sample because the dwelling was vacant or their members were outof-scope for the NLSCY at the time of the LFS. Some of these
households may have had children (0 to 13) living in them a few
months later when the NLSCY interview took place. Since these
households were not eligible to be selected, some bias may have
been introduced.
NLSCY Data Users Guide
134
2001/2002
Third Problem
The third and last problem complements the second. In some cases,
the sampled address, where a child was living at the time of
selection, was no longer occupied by a family having in-scope
children at the time of collection. In a way, this is a frame
undercoverage issue linked to the time lag between the LFS
interview and the NLSCY interview. This situation might occur when
the selected occupants have moved before collection takes place.
As a result, it is possible that the NLSCY sample undercovers the
population of highly mobile children.
Component Non-Response
Component
Non-Response
As discussed in Section 5, there were several respondents or
components to the NLSCY interview. The PMK provided detailed
information about each selected child. In the Parent Questionnaire
and the general questionnaire, the PMK provided information about
himself or herself and his or her spouse/partner. The PPVT-R test
was administered to children in the 4 to 5 age group. Children in the
10 to 15 age group completed a questionnaire on their own. For
school-aged children the teacher completed a questionnaire about
the child, and if the child was in grade 2 or above, a Math Test was
administered. There was a potential for non-response for each of
these components.
Responding
Household
It should be noted, however, that when a household was deemed to
be a responding household, then all required components were
created for that household, even if there were no data provided for a
particular component. For example, if there was a 10 year-old in a
responding household who did not complete the 10 to 11
Questionnaire, then this component still exists for the child, with all
variables set to not-stated. Likewise if a parent completed a Child
Questionnaire for one child in the household but refused to do so for
a second child, then there is a record for this second child (with notstated values for all variables).
Parent Questionnaire Response Rates
NLSCY Data Users Guide
135
2001/2002
Response
Rates
The PMK and his or her spouse/partner answered this
questionnaire. Again, we determined the valid response rate
obtained in order to assess the completeness of the data. Out of
the 24,692 PMKs and their spouse/partners:
! there were answers to all relevant questions in 74% of the cases;
! a valid answer was obtained for more than 90% of questions
submitted to 95% of the adults;
! less than 50% valid answers were gathered for 1.5% of the adults.
Child Questionnaire Response Rates
Response
Rates
In order to assess completeness of the child data, we determined
the rate of answered questions among those that were relevant to
the child. In the sample of respondents consisting of
20,102 children:
- there were answers to all relevant questions in 63% of the cases;
- a valid answer was obtained for more than 90% of questions
submitted to 98% of the children;
- less than 50% valid answers were gathered for less than 1% of the
children.
NLSCY School component
NLSCY Data Users Guide
136
2001/2002
School
component
The School component of the NLSCY collects information every two
years on a nationally representative group of children, within their
school environment. This information is used jointly with the
information collected earlier about the same children in the
Household component of the survey. By collecting both the child’s
household information and school performance, the NLSCY can
obtain a more complete picture of the child’s development.
Every child surveyed in the NLSCY, from grades two to ten, are given
mathematics and reading tests. In order for a test to be
administered, the consent from parents and the school board are
required. These tests were constructed by selecting a subset of
questions from the Canadian Test Centre’s Canadian Achievement
Tests, second edition (CAT/2).
The mathematics test is a shorten version of the CAT/2
mathematical operations test. This test measures the student's
ability to do addition, subtraction, multiplication and division
operations on whole numbers, decimals, fractions, negatives and
exponents. Problem solving involving percentages and the order of
operations are also measured.
School
component
The reading comprehension test is developed in part from the
CAT/2. Since the CAT/2 contains only English passages, French
passages were developed in co-operation with educators at
Université de Sherbrooke. The test is designed to measure basic
reading skills. The test's objectives cover information recall, analysis
of passages, identifying the main idea, interpretation of various
types of writing and critical evaluation. Each test consists of two
original English passages and two original French passages in
order to make the test as linguistically equivalent as possible.
Response Rates for Math and Reading tests
NLSCY Data Users Guide
137
2001/2002
Response
Rates
Among the 9,542 children eligible for the math and reading tests for
Cycle 3:
<
86% of parents (representing 8,206 children) consented to
having the school board contacted to administer the math
and reading tests.
<
The school boards of 97% of the 8,206 children consented to
administer the tests. That meant that from the total number of
9,542, consent was obtained for 7,920 or 83% of all eligible
children.
<
65% of the tests that were administered were returned.
<
Due to an operational error, approximately 2% of the tests
were not sent to the schools.
To summarize, out of the total of 9,542 eligible children we received
5,153 or 54% completed math and reading tests for Cycle 3. This
rate is much lower than the 74% total that was obtained from Cycle
2.
NLSCY Data Users Guide
138
2001/2002
Testing
strategy
In cycle 3, the evolution of the NLSCY testing strategy continued:
2) The reading comprehension test introduced in Cycle 2 of the
survey, has seen few modifications, while the mathematics test
has evolved considerably from its inception in Cycle 1. These
tests are administered in the school to children from grades 2 to
10. During the household interview, the parents’ consent is
requested before the tests can be administered to the child at
school.
3) Since Cycle 2, a skills indicator of mathematics and reading
abilities is administered at the home to pre-assess their
abilities. The indicator consists of 10 to 13 questions with
multiple choice answers: the questions were taken from the
second edition of the Canadian Achievement tests (CAT/2). The
CAT/2 is a series of tests to measure basic skills in a variety of
subjects taught in schools.
4) In Cycle 3 of the NLSCY, a separate version of the mathematics
and reading comprehension tests was administered for each
academic grade level, except for grades 9 and 10 which got the
same level tests, for a total of eight in all. Thus, students in Grade
2 completed the level 2 test, students in Grade 3, the level 3 test,
and so on to level 9 for students in Grades 9 and 10. In some
instances, students were given a higher level test. Fifty per cent
of the children who scored a perfect score on the home
administered skills indicator test were given a higher level
school test than their actual level. This approach was used to
offset the potentially serious problem of “ceiling effect”
encountered during Cycle 1 with the mathematics test, especially
in Grades 3 and 5.
The mathematics and reading comprehension tests were
administered from the same booklet by the child's teacher, in class,
using a multiple-choice questionnaire.
NLSCY Data Users Guide
139
2001/2002
Mathematics
test
This test was a shortened version of the CAT/2 mathematical
operations test. The CAT/2 mathematical operations test measures
the student's ability to do addition, subtraction, multiplication and
division operations on whole numbers, decimals, fractions,
negatives and exponents. Problem solving involving percentages
and the order of operations are also measured. The short version of
the test developed for the purposes of the NLSCY now consists of
20 questions at each level, except for level 9-10, which consists of
15 questions. The tests were expanded in Cycle 3 to include
overlapping items between each level. An extra five items were
added to each level selected from the test of the next level.
Each child who took the mathematics test was given a raw (gross)
score, a scaled score referred to as the classical scaled score and
an IRT scaled score. The raw (gross) score is obtained simply by
adding the number of correct answers. The Classically derived
scale score and the IRT scaled score are described as follows.
The IRT derived
scaled score
The approach of the item response theory (IRT) was used
successfully in Cycle 2 to derive scores for the reading
comprehension tests. Unlike the approach of the classical theory,
the IRT makes it possible to scale the scores without preset
population standards. Using common test items linking grades,
standards are estimated from the entire population of children
taking the test for this cycle. Scores are derived ranking each child
within a level then the scores are scaled to reflect the progression
of scores throughout all the levels. In order to ensure comparability
from year to year, each sample from each cycle must represent
equivalent populations.
Among the single dimension models, the two-parameter and the
three-parameter logistical model were chosen for math and the
reading tests respectively. The two-parameter model takes into
consideration both the difficulty and the discrimination of the item
while the three parameter model also considers the pseudoguessing component. In this way, the IRT takes into consideration
the pattern of responses. Two children with the same raw (gross)
score will not have the same scaled score unless they answered
exactly the same way. For example, a child who only answered the
5 easiest questions correctly would have a lower scaled score than
the one who only answered the 5 hardest questions correctly. The
scaled scores of the Cycle 3 mathematics test ranged from 100 to
600.
NLSCY Data Users Guide
140
2001/2002
The classically
derived scaled
score
The scaled score is derived from standards (norms) established by
the Canadian Test Centre (CTC). The CTC developed these
standards from a sample of Canadian children from all 10 provinces
(however, the test has been developed in English only and so in
Quebec, the sample represents only the English schools), which is
referred to as the normative sample. The children from the
normative sample received the complete test. The scaled scores
are units of a single scale with equidistant intervals that covers all of
the grade levels. The scale was developed using a Thurstone
procedure derived from the classical testing theory.
The fact that a short test was used for children in the NLSCY sample
meant that it was not possible to directly associate the CTC scaled
scores with the gross scores obtained in the survey. For this reason,
the CTC normative sample was used to calculate the percentile rank
for each gross score but using only the 15 of the 20 questions of the
short NLSCY test. Only 15 of the 20 items were normalized for the
appropriate grade level while the remaining 5 items were taken from
the test of the next grade level. The normative score was then
interpolated by inserting the percentile rank obtained with the 15
questions of the short test between the percentiles of the complete
test. For example, using level 6, we find in the short test a percentile
rank of 2.2% for a raw (gross) score of 1. On the complete test, the
percentile ranks of 2.0% and 3.7% correspond to raw (gross)
scores of 5 and 6 and to scaled scores of 332 and 348 respectively.
After linear interpolation, we obtain a scaled score of 334 for the
gross score of 1 on the short version of the test.
The table below shows the relation between the raw (gross) scores
and the scaled scores by grade for the NLSC mathematics test. The
scaled scores for this test range from 200 to 999 for Cycle 3
NLSCY Data Users Guide
141
2001/2002
Relation between raw (gross) scores and scaled scores by grade for the Cycle 3
mathematics test
Gross
scores
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Scaled scores
Grade 2
Grade 3
Grade 4
Grade 5
Grade 6
Grade 7
Grade 8
Grades 9-10
204
215
232
249
259
268
277
286
294
302
310
320
331
345
368
402
238
264
281
300
315
326
338
348
358
368
379
391
405
420
442
479
264
281
312
336
357
376
392
405
417
428
439
451
466
487
515
550
293
321
349
372
390
407
421
434
445
458
471
484
498
516
539
569
314
334
364
388
408
428
446
460
476
490
505
520
537
562
585
624
330
349
389
427
457
484
506
526
546
563
580
602
620
644
665
701
362
384
417
448
477
499
519
537
561
583
603
625
649
673
712
794
406
427
464
504
533
558
582
603
627
652
677
701
727
754
789
871
With the expansion of the tests in Cycles 2 and 3, the ceiling effect
measured in Cycle 1 has been greatly reduced. Although the skill
indicator was still used to elevate children to the next level test, the
process is problematic and prone to human error. During the
implementation of the tests in Cycle 3, children who received a
perfect score on the skills indicator were divided into two equal size
groups; one group to receive their regular level tests, the other to
receive next level tests. Unfortunately, half of the children targeted
through this process, namely those who were to receive their regular
level tests, were not sent tests. Although these missing tests would
not affect the classically derived scaled scores, they had to be
adjusted for during the IRT calculation of scaled scores.
NLSCY Data Users Guide
142
2001/2002
Reading
comprehension
test
The reading comprehension test, like the mathematics test, was
also developed in part from the CAT/2. However, since the CAT/2
contain only English passages, French passages had to be chosen
from another source by educators at Université de Sherbrooke.
The test is designed to measure basic reading skills. The test's
objectives cover information recall, analysis of passages,
identification of the main idea, interpretation of various types of
writing and critical evaluation. For each grade level, the test
developed for the NLSCY consists of four reading passages
totalling 20 questions. Each test consists of two original English
passages and two original French passages in order to make the
test as fair as possible. In addition, between two consecutive
grades, there are always two common passages with more or less
10 questions.
Similarly to the mathematics test, each child who took the reading
test was also given a raw (gross) score and a scaled score. Since
the CTC did not have standards for this test, the approach of the
item response theory (IRT) described for the mathematics tests
was the only option for this test. Unlike the approach of the
mathematics test, the three-parameter logistical model was
chosen for the reading test. This model takes into consideration the
difficulty and the discrimination of the item and a pseudo-guessing
parameter that seems more prevalent for this test.
A number of corrections were made to the English and French
passages in Cycle 3 in order to improve the translation. Other
items which had a negative biserial correlation with the ability
being measured were also changed to improve their fit into the
logistic model. The Bilog-MG software was used to calculate the
scaled scores for both the reading and mathematics test.
NLSCY Data Users Guide
143
2001/2002
Response Rates for Teachers and principals questionnaire
Response
Rates
The response rates for the Teacher’s Questionnaire and the
Principal’s Questionnaire were computed by province, type of
school and the child’s age. In Cycle 3, the parents of 16,558 children
reported that their children were attending school. The children
ranged in age from 4 to 15 and in level from Kindergarten to Grade
10. The parents of 94% of the 16,558 children gave us written
permission to collect data from their children’s teacher and
principal. The school boards of 97% of the children whose parents
had given permission (that is, 91% of all children attending school)
agreed to take part in the survey. After obtaining parental and
school board consent, we sent a questionnaire to each child’s
teacher. The response rate for the latter questionnaire was 67%.
Taking into account the questionnaires that were never mailed to
teachers because we were unable to obtain parental or school
board permission, we collected questionnaires from the teachers of
61% of all children attending school.
Response Rates by province (Teacher’s Questionnaire)
Province
Newfoundland
Prince Edward Island
Nova Scotia
New Brunswick
Québec
Ontario
Manitoba
Saskatchewan
Alberta
British Columbia
Total
Children
Going to
School
824
335
992
1001
3256
4778
1066
1155
1621
1526
16554*
Consent of Parent
Consent of School Board
Freq.
Cum %
Freq.
803
315
965
953
2978
4536
994
1070
1546
1417
15577
97.5
94.0
97.3
95.2
91.5
94.9
93.2
92.6
95.4
92.9
94.1
803
314
962
951
2882
4494
930
1015
1507
1261
15119
Cond
%
100.0
99.7
99.7
99.8
96.8
99.1
93.6
94.9
97.5
89.0
97.1
Questionnaires Returned
Cum %
Freq.
97.5
93.7
97.0
95.0
88.5
94.1
87.2
87.9
93.0
82.6
91.3
519
226
726
708
1700
2842
684
778
1049
854
10086
Cond
%
64.6
72.0
75.5
74.4
59.0
63.2
73.5
76.7
69.6
67.7
66.7
Cum
%
63.0
67.5
73.2
70.7
52.2
59.5
64.2
67.4
64.7
56.0
60.9
* The province of residence is missing for 4 children.
NLSCY Data Users Guide
144
2001/2002
Response
Rates by
province
The response rate for the Teacher’s Questionnaire appeared to be
lower in Quebec, British Columbia and Ontario. For Quebec and
Ontario, this may be due to a lower rate of return by teachers. In
British Columbia’s case, while the questionnaire return rate was
average, the low response rate may have been caused by a lower
school board consent rate.
Response Rates by school type (Teacher’s Questionnaire)
Type of School
Children Going to
School
Public
Catholic
Private
Other/Missing
Total
12825
2337
660
736
16558
Response
Rates by
school type
Consent of Parent
Consent of School Board
Freq.
Cum %
Freq.
12098
2226
591
666
15581
94.3
95.3
89.5
90.5
94.1
11774
2201
517
630
15122
Questionnaires Returned
Cond % Cum %
97.3
98.9
87.5
94.6
97.1
91.8
94.2
78.3
85.6
91.3
Freq.
7925
1401
357
406
10089
Cond
%
67.3
63.7
69.1
64.4
66.7
Cum
%
61.8
59.9
54.1
55.2
60.9
The response rate for the Teacher’s Questionnaire appeared to be
nearly equal for public school students and Catholic school students,
but it appeared to be lower for children attending private schools.
This could be due to the lower percentage of consent given by
private school boards.
Response Rates by age (Teacher’s Questionnaire)
Age
4
5
6
7
8
9
10
11
12
13
14
15
Total
Children Going
to School
Consent of Parent
617*
5520
1219
1161
1147
1063
981
965
1035
985
1017
848*
16558
Freq.
576
5244
1175
1120
1094
1024
911
898
963
894
923
759
15581
Cum %
93.4
95.0
96.4
96.5
95.4
96.3
92.9
93.1
93.0
90.8
90.8
89.5
94.1
Consent of School Board
Freq.
515
5046
1153
1090
1070
1005
893
880
944
875
905
746
15122
Cond %
89.4
96.2
98.1
97.3
97.8
98.1
98.0
98.0
98.0
97.9
98.0
98.3
97.1
Cum %
83.5
91.4
94.6
93.9
93.3
94.5
91.0
91.2
91.2
88.8
89.0
88.0
91.3
Questionnaires Returned
Freq.
314
3235
785
726
716
671
606
607
668
612
635
514
10089
Cond %
61.0
64.1
68.1
66.6
66.9
66.8
67.9
69.0
70.8
69.9
70.2
68.9
66.7
Cum %
50.9
58.6
64.4
62.5
62.4
63.1
61.8
62.9
64.5
62.1
62.4
60.6
60.9
* There were 21 children that are 3 years old and 4 children that are 16 years old. The 3 year olds were
regrouped with the 4 year olds and the 16 year olds were regrouped with the 15 year olds.
NLSCY Data Users Guide
145
2001/2002
Response
Rates by age
The response rate for the Teacher’s Questionnaire appeared to be
lower for four-year-olds. The percentage of questionnaires returned
and the school board consent rate were also lower for four-year-old
students.
Principal’s questionnaire
Principal’s
questionnaire
As in the case of the Teacher’s Questionnaire, when we had the
permission of the parents and the school boards, we mailed a
questionnaire to the school principal. In cases where more than one
child was attending the same school, the principal received only
one questionnaire. The response rate for the Principal’s
Questionnaire was 68%. This means that we collected Principal’s
Questionnaires for 69% of the children whose parents and school
boards gave permission (or 63% of all children attending school).
Response Rates by province (Principal’s Questionnaire)
Province
Newfoundland
Prince Edward Island
Nova Scotia
New Brunswick
Québec
Ontario
Manitoba
Saskatchewan
Alberta
British Columbia
Total
Principal of the Schools That Have
Received a Questionnaire
Questionnaires Returned
Frequency
153
52
250
216
782
1499
277
296
515
455
4495
225
63
332
279
1355
2245
370
389
727
659
6644*
%
68.0
82.5
75.3
77.4
57.7
66.8
74.9
76.1
70.8
69.0
67.7
* The province of the school is missing for 3 principals.
NLSCY Data Users Guide
146
2001/2002
Principal’s Questionnaire
Province
Children
Going to
School
Newfoundland
Prince Edward Island
Nova Scotia
New Brunswick
Québec
Ontario
Manitoba
Saskatchewan
Alberta
British Columbia
Total
824
335
992
1001
3256
4778
1066
1155
1621
1526
16554*
Consent of
Parent
Freq.
803
315
965
953
2978
4536
994
1070
1546
1417
15577
Consent of School Board
Cum %
97.5
94.0
97.3
95.2
91.5
94.9
93.2
92.6
95.4
92.9
94.1
Freq.
803
314
962
951
2882
4494
930
1015
1507
1261
15119
Cond % Cum %
100.0
97.5
99.7
93.7
99.7
97.0
99.8
95.0
96.8
88.5
99.1
94.1
93.6
87.2
94.9
87.9
97.5
93.0
89.0
82.6
97.1
91.3
Children whose Principal
have returned their
questionnaire
Freq.
520
253
749
745
1698
2991
720
798
1070
878
10422
Cond %
64.8
80.6
77.9
78.3
58.9
66.6
77.4
78.6
71.0
69.6
68.9
Cum %
63.1
75.5
75.5
74.4
52.1
62.6
67.5
69.1
66.0
57.5
63.0
* The province of residence is missing for 4 children.
Response
Rates by
province
The response rate for the Principal’s Questionnaire (and hence the
percentage of children whose principals returned their
questionnaire) varied widely from province to province and
appeared to be lower in Quebec.
Response Rates by school type ( Principal’s Questionnaire)
Type of School
Principal of the Schools
That Have Received a
Questionnaire
Public
Catholic
Private
Other/Missing
Total
4948
1069
316
314
6647
Type of School Children
Going to
School
Public
Catholic
Private
Other/Missing
Total
12825
2337
660
736
16558
NLSCY Data Users Guide
Questionnaires Returned
Frequency
3376
735
199
188
4498
%
68.2
68.8
63.0
60.0
67.7
Consent of Parent
Consent of School Board
Frequency Cum %
12098
94.3
2226
95.3
591
89.5
666
90.5
15581
94.1
Frequency
11774
2201
517
630
15122
147
Children whose Principal have
returned their questionnaire
Cond % Cum % Frequency Cond % Cum %
97.3
91.8
8176
69.4
63.8
98.9
94.2
1510
68.6
64.6
87.5
78.3
327
63.2
49.5
94.6
85.6
412
65.4
56.0
97.1
91.3
10425
68.9
63.0
2001/2002
Response
Rates by
school type
As in the case of the Teacher’s Questionnaire, the response rate for
the Principal’s Questionnaire appeared to be about equal for public
school students and Catholic school students, but it appeared to be
lower for children attending private schools.
Response Rates by age (Principal’s Questionnaire)
Age
Children
Going to
School
4
5
6
7
8
9
10
11
12
13
14
15
Total
617*
5520
1219
1161
1147
1063
981
965
1035
985
1017
848*
16558
Consent of Parent
Freq.
576
5244
1175
1120
1094
1024
911
898
963
894
923
759
15581
Cum %
93.4
95.0
96.4
96.5
95.4
96.3
92.9
93.1
93.0
90.8
90.8
89.5
94.1
Consent of School Board
Freq.
515
5046
1153
1090
1070
1005
893
880
944
875
905
746
15122
Cond %
89.4
96.2
98.1
97.3
97.8
98.1
98.0
98.0
98.0
97.9
98.0
98.3
97.1
Cum %
83.5
91.4
94.6
93.9
93.3
94.5
91.0
91.2
91.2
88.8
89.0
88.0
91.3
Children whose Principal have
returned their questionnaire
Freq.
354
3316
840
765
763
733
629
646
685
587
605
502
10425
Cond %
68.7
65.7
72.9
70.2
71.3
72.9
70.4
73.4
72.6
67.1
66.9
67.3
68.9
Cum %
57.4
60.1
68.9
65.9
66.5
69.0
64.1
66.9
66.2
59.6
59.5
59.2
63.0
* There were 21 children that are 3 years old and 4 children that are 16 years old. The 3 year olds were regrouped with
the 4 year olds and the 16 year olds were regrouped with the 15 year olds.
Response Rates
by age
As in the case of the Teacher’s Questionnaire, the response rate for
the Principal’s Questionnaire appeared to be slightly lower for fouryear-olds. The school board consent rate was also lower for fouryear-old students.
NLSCY Data Users Guide
148
2001/2002
NLSCY Data Users Guide
149
2001/2002
Chapter 10 - Guidelines for Tabulation, Analysis and Release
Introduction
This section of the documentation outlines the guidelines to be
adhered to by users tabulating, analyzing, publishing or otherwise
releasing any data derived from the survey microdata file. With the
aid of these guidelines, users of microdata should be able to
produce the same figures as those produced by Statistics Canada
and, at the same time, will be able to develop currently unpublished
figures in a manner consistent with these established guidelines.
Rounding Guidelines
Statistics
Canada
Guidelines
In order that estimates for publication or other release derived from
the NLSCY microdata file correspond to those produced by
Statistics Canada, users are urged to adhere to the following
guidelines regarding the rounding of such estimates:
a) Estimates in the main body of a statistical table are to be rounded
to the nearest hundred units using the normal rounding technique. In
normal rounding, if the first or only digit to be dropped is 0 to 4, the
last digit to be retained is not changed. If the first or only digit to be
dropped is 5 to 9, the last digit to be retained is raised by one. For
example, in normal rounding to the nearest 100, if the last two digits
are between 00 and 49, they are changed to 00 and the preceding
digit (the hundreds digit) is left unchanged. If the last digits are
between 50 and 99 they are changed to 00 and the preceding digit
is incremented by 1.
b) Marginal sub-totals and totals in statistical tables are to be derived
from their corresponding unrounded components and then are to be
rounded themselves to the nearest 100 units using normal rounding.
c) Averages, proportions, rates and percentages are to be
computed from unrounded components (i.e., numerators and/or
denominators) and then are to be rounded themselves to one
decimal using normal rounding.
NLSCY Data Users Guide
150
2001/2002
d) Sums and differences of aggregates (or ratios) are to be derived
from their corresponding unrounded components and then are to be
rounded themselves to the nearest 100 units (or the nearest one
decimal) using normal rounding.
e) In instances where, due to technical or other limitations, a rounding
technique other than normal rounding is used resulting in estimates
to be published or otherwise released which differ from
corresponding estimates published by Statistics Canada, users are
urged to note the reason for such differences in the publication or
release document(s).
f) Under no circumstances are unrounded estimates to be published
or otherwise released by users. Unrounded estimates imply greater
precision than actually exists.
Sample Weighting Guidelines for Tabulation
Sample Design
The sample design used for the NLSCY was not self-weighting.
When producing simple estimates, including the production of
ordinary statistical tables, users must apply the proper
demographic load. If proper weights are not used, the estimates
derived from the microdata file cannot be considered to be
representative of the survey population, and will not correspond
to those produced by Statistics Canada. In effect, the weight
assigned to each child reflects the number of children
represented by a particular respondent.
For any analysis dealing with correlation analysis or any other
statistics where a significance measure is required, it is
recommended that a “sample” weight be used. This weight is
obtained by multiplying the demographic load by the sample size
and dividing this total by the total estimated population. This
produces a mean weight of 1 and a sum of weights equal to the
sample size.
NLSCY Data Users Guide
151
2001/2002
Benefit of Using
an Adjusted
Weight
The benefit of this adjusted weight is that an over estimation of
the significance (which is very sensitive to sample size is
avoided while maintaining the same distributions as those
obtained when using the population weight. The disadvantage is
that the numerator is not weighted up to the target population and
the Coefficient of Variance Tables described in section 12 and
presented in Appendix 3 are no longer useful as a measure of
data quality.
Software
Differences
Users should also note that some software packages may not
allow the generation of estimates that exactly match those
available from Statistics Canada, because of their treatment of
the weight field.
Definitions of Types of Estimates: Categorical vs. Quantitative
Unit of Analysis
The NLSCY file has been set up so that the child is the unit of
analysis. The weight that can be found on each record
(CWTCW01C for the cross-sectional sample and CWTCW01L
for the longitudinal sample) is a “child” weight. Estimates of
parents or families cannot be made from the NLSCY microdata
file.
Categorical
Estimates
Categorical estimates are estimates of the number, or
percentage of the surveyed population possessing certain
characteristics or falling into some defined category. An
estimate of the number of persons possessing a certain
characteristic may also be referred to as an estimate of an
aggregate.
NLSCY Data Users Guide
152
2001/2002
Examples of
Categorical
Questions
Q: Was (the child) born before, after or on the due date?
R: Before
After
On due date
Q: Compared to other babies in general, would you say the (the
child's) health at birth was:
R: Excellent
Very good
Good
Fair
Poor
Quantitative
Estimates
Quantitative estimates are estimates of totals or of means,
medians and other measures of central tendency of quantities
based upon some or all of the members of the surveyed
population.
^
They also specifically involve estimates of the form
X
^
^
where
Y
^
X
is an estimate of the surveyed population total quantity and Y is
an estimate of the number of people in the surveyed population
contributing to that total quantity.
Example of a
Quantitative
Estimate
NLSCY Data Users Guide
An example of a quantitative estimate is the average number of
days of care received by babies who required special medical
care following birth. The numerator is an estimate of the total
number of days for which babies required special care. The
denominator is the number of babies who required special care
at birth.
153
2001/2002
Example of a
Quantitative
Question
Q: For how many days, in total, was this care received?
R: Days
Q: What was the child's weight at birth in pounds and ounces?
R: Pounds Ounces
Tabulation of Categorical Estimates
Estimates of the
Number of
Children
Estimates of the number of children with a certain characteristic
can be obtained from the microdata file by summing the final
weights of all records possessing the characteristic(s) of interest.
These estimates may be cross-sectional or longitudinal.
Proportions and
Ratios
^
Proportions and ratios of the form
X
^
are obtained by:
Y
(a) summing the final weights of records having the characteristic
^
of interest for the numerator ( X );
(b) summing the final weights of records having the characteristic
of interest for the denominator (
^
Y ), then;
(c) dividing the numerator estimate by the denominator estimate.
NLSCY Data Users Guide
154
2001/2002
Tabulation of Quantitative Estimates
Estimates of
Quantities
Estimates of quantities can be obtained from the microdata file
by multiplying the value of the variable of interest by the final
weight for each record, then summing this quantity over all
records of interest.
Example of a
Quantitative
Estimate
For example, to obtain an estimate of the total number of days of
special care received by infants who were born prematurely:
- multiply the number of days for which special care was received
by the final weight;
- then sum this value over all records for which the child was born
prematurely.
To obtain a weighted average of the form
^
X
^
^
, the numerator ( X ) is calculated as for a
Y
quantitative estimate and the denominator (Y), is calculated as
for a categorical estimate. For example, to estimate the average
number of days spent in special care by premature babies:
(a) estimate the total number of days as described above;
(b) estimate the number of children in this category by summing
the final weights of all records for babies which were premature;
then
(c) divide estimate (a) by estimate (b).
Guidelines for Statistical Analysis
Sample Design
NLSCY Data Users Guide
The NLSCY is based upon a complex sample design, with
stratification, multiple stages of selection, and unequal
probabilities of selection of respondents. Using data from such
complex surveys presents problems to analysts because the
survey design and the selection probabilities affect the
estimation and variance calculation procedures that should be
used. In order for survey estimates and analyses to be free from
bias, the survey weights must be used.
155
2001/2002
Variance
Estimates
While many analysis procedures found in statistical packages
allow weights to be used, the meaning or definition of the weight
in these procedures differ from that which is appropriate in a
sample survey framework, with the result that while in many
cases the estimates produced by the packages are correct, the
variance estimates that are calculated are not adequate.
Variances for simple estimates such as totals, proportions and
ratios (for qualitative variables) are provided in the
accompanying Sampling Variability Tables.
Rescaling the
Weights
For other analysis techniques (for example linear regression,
logistic regression and analysis of variance), a method exists
which can make the variances calculated by the standard
packages more meaningful, by incorporating the unequal
probabilities of selection. The method rescales the weights so
that there is an average weight of 1.
Example of
Rescaling the
Weights
For example, suppose that analysis of all male children is
required. The steps to rescale the weights are as follows:
-Select all respondents from the file with SEX = male (variable
CMMCQ02).
-Calculate the AVERAGE weight for these records by summing
the original person weights (BWTCW01C ) from the microdata
file for these records and then dividing by the number of records
with SEX = male.
-For each of these records, calculate a RESCALED weight
equal to the original person weight divided by the AVERAGE
weight.
-Perform the analysis for these respondents using the
RESCALED weight.
However, because the stratification and clustering of the
sample's design are still not taken into account, the variances
calculated in this way are likely to be under-estimated.
NLSCY Data Users Guide
156
2001/2002
Calculation of
Variance
Estimates
The calculation of truly meaningful variance estimates requires
detailed knowledge of the design of the survey. Such detail
cannot be given in this microdata file because of confidentiality.
Variances that take the complete sample design into account
can be calculated for many statistics by Statistics Canada on a
cost-recovery basis.
C.V. Release Guidelines
Release
Guidelines
Before releasing and/or publishing any estimate from the
NLSCY, users should first determine the quality level of the
estimate. The quality levels are acceptable, marginal and
unacceptable. As discussed in Chapter 10, sampling and nonsampling errors both influence data quality. For the purposes of
this document, however, estimate quality is based solely on the
sampling error illustrated by the coefficient of variation, as shown
in the table below.
First, the number of children who contribute to the calculation of
the estimate should be determined. If this number is less than
30, the weighted estimate should be considered to be of
unacceptable quality.
For weighted estimates based on sample sizes of 30 or more,
users should determine the coefficient of variation of the estimate
and follow the guidelines below. These quality level guidelines
should be applied to weighted rounded estimates.
All estimates can be considered releasable. However, those of
marginal or unacceptable quality level must be accompanied by
a warning to caution subsequent users.
NLSCY Data Users Guide
157
2001/2002
QUALITY LEVEL GUIDELINES
Quality Level of
Estimate
1. Acceptable
2. Marginal
3. Unacceptable
Guidelines
Estimates have:
a sample size of 30 or more, and low coefficients of
variation in the range 0.0% to 16.5%.
No warning is required.
Estimates have:
a sample size of 30 or more, and high coefficients of
variation in the range 16.6% to 33.3%.
Estimates should be flagged with the letter M (or some
similar identifier). They should be accompanied by a
warning to caution subsequent users about the high
levels of error, associated with the estimates.
Estimates have:
a sample size of less than 30, or very high coefficients
of variation in excess of 33.3%.
Statistics Canada recommends not to release
estimates of unacceptable quality. However, if the
user chooses to do so then estimates should be
flagged with the letter U (or some similar identifier)
and the following warning should accompany the
estimates:
“The user is advised that…(specify the data)…do not
meet Statistics Canada’s quality standards for this
statistical program. Conclusions based on these data
will be unreliable, and most likely invalid. These data
and any consequent findings should not be published.
If the user chooses to publish these data or findings,
then this disclaimer must be published with the data.”
NLSCY Data Users Guide
158
2001/2002
NLSCY Data Users Guide
159
2001/2002
Chapter 11 - Approximate Sampling Variability Tables
Introduction
In order to supply coefficients of variation which would be
applicable to a wide variety of categorical estimates produced
from this microdata file and which could be readily accessed by
the user, a set of Approximate Sampling Variability Tables has
been produced. These “look-up” tables, which can be found in
Appendix 3, allow the user to obtain an approximate coefficient
of variation based on the size of the estimate calculated from
the survey data.
Coefficients of
Variation
The coefficients of variation (c.v.) are derived using the variance
formula for simple random sampling and incorporate a factor
which reflects the multi-stage, clustered nature of the sample
design. This factor, known as the design effect, was
determined by first calculating design effects for a wide range of
characteristics and then choosing from among these a
conservative value to be used in the look-up tables which would
then apply to the entire set of characteristics.
Sample
Requirements
For the NLSCY, the sample was constructed taking account the
following requirements.
! A sufficient sample was required in each of the 10 provinces
to allow for the production of reliable estimates for all
longitudinal children who were 0 to 11 years of age in Cycle 1.
! It was also necessary to have a large enough sample to
produce estimates for Cycle 1 at the Canada level by seven key
age groupings or cohorts: 0 to 11 months, 1 year, 2 to 3 years,
4 to 5 years, 6 to 7 years, 8 to 9 years, and 10 to 11 years.
! In each province, a sufficient sample size was required for
Cycle 2 to produce reliable estimates for all children who were 0
to 11 years of age in Cycle 1.
NLSCY Data Users Guide
160
2001/2002
Design Effect,
Sample Size,
Population
The tables that follow show the design effects, sample sizes and
population counts by province and age groupings used to
produce the Approximate Sampling Variability Tables.
First, the tables for the cross-sectional samples:
CROSS-SECTIONAL SAMPLE
Province
Design
Effect
Sample Size
Population
Newfoundland
Prince Edward Island
Nova Scotia
New Brunswick
Québec
Ontario
Manitoba
Saskatchewan
Alberta
British Columbia
Atlantic provinces
Prairies
Total
2.1
2.2
2.7
2.5
4.4
4.3
3.8
2.9
3.1
3.7
2.6
3.7
4.1
1,001
545
1,293
1,664
3,757
5,195
1,484
1,589
1,827
1,670
4,503
4,900
20,025
100,089
26,932
167,311
133,481
1,275,660
2,107,791
213,543
203,197
568,358
686,174
427,813
985,098
5,482,536
NLSCY Data Users Guide
161
2001/2002
CROSS-SECTIONAL SAMPLE
Age Group
0 to 23 years
2 to 3 years
4 to 5 years
6 to 7 years
8 to 9 years
10 to 11 years
12 to 13 years
0 to 3 years
4 to 11 years
4 to 7 years
8 to 11 years
Total (0 to 13
years)
Design
Effect
2.1
2.4
2.7
2.9
2.5
2.4
2.8
2.7
3.4
4.2
3.5
4.1
Sample Size
4,154
3,866
2,928
2,418
2,161
2,240
2,258
8,020
9,747
5,346
4,401
20,025
Population
740,151
766,998
804,057
812,201
773,433
792,572
793,124
1,507,149
3,182,263
1,616,258
1,566,005
5,482,536
Design effects for the longitudinal sample are as follows:
CYCLE-1 LONGITUDINAL SAMPLE
Province
Design
Effect
Sample Size
Population
Newfoundland
2.0
892
89,533
Prince Edward
Island
2.0
443
23,161
Nova Scotia
New Brunswick
Québec
Ontario
Manitoba
Saskatchewan
Alberta
British
Columbia
Atlantic
provinces
Prairies
Total
2.9
2.3
4.9
4.2
3.4
2.8
3.2
3.6
1,068
958
2,944
3,899
1,161
1,305
1,465
1,333
144,722
115,913
1,099,033
1,777,525
183,268
176,449
489,604
574,160
2.7
3,361
373,351
3.6
5.3
3,931
15,468
849,321
4,673,390
NLSCY Data Users Guide
162
2001/2002
CYCLE-1 LONGITUDINAL SAMPLE
Age Group
Design
Effect
2.7
Sample Size
3,654
752,598
4 to 5 years
6 to 7 years
8 to 9 years
10 to 11 years
12 to 13 years
3.2
3.3
3.0
3.1
3.2
2,697
2,429
2,169
2,249
2,270
791,754
800,064
763,632
783,049
782,293
2 to 5 years
6 to 13 years
3.3
3.8
6,351
9,117
1,544,352
3,129,038
6 to 9 years
10 to 13 years
3.9
4.1
4,598
4,519
1,563,696
1,565,342
Total (2 to 13
years)
5.3
15,468
4,673,390
2 to 3 years
Approximate
Sampling
Variability Tables
Population
All coefficients of variation in the Approximate Sampling Variability
Tables are approximate and, therefore, unofficial. The use of
actual variance estimates would likely result in estimates with lower
variances; for example, estimates listed as “unacceptable” in the
Approximate Sampling Variability Tables could move up to the
“marginal” category.
Remember: If the number of observations on which an estimate is
based is less than 30, the weighted estimate should be classified
as “unacceptable” regardless of the value of the coefficient of
variation for this estimate. This is because the formulas used for
estimating the variance do not hold true for small sample sizes.
NLSCY Data Users Guide
163
2001/2002
How to Use the C.V. Tables For Categorical Estimates
Introduction
The following rules should enable the user to determine the
approximate coefficients of variation from the Sampling Variability
Tables for estimates of the number, proportion or percentage of
the surveyed population possessing a certain characteristic and
for ratios and differences between such estimates.
Rule 1:
Estimates of
Numbers
Possessing a
Characteristic
(Aggregates)
The coefficient of variation depends only on the size of the
estimate itself. On the Sampling Variability Table for the
appropriate geographic area or age group, locate the estimated
number in the left-most column of the table (headed “Numerator of
Percentage”) and follow the asterisks (if any) across to the first
figure encountered. This figure is the approximate coefficient of
variation.
Rule 2:
Estimates of
Proportions or
Percentages
Possessing a
Characteristic
The coefficient of variation of an estimated proportion or
percentage depends on both the size of the proportion or
percentage and the size of the total upon which the proportion or
percentage is based. Estimated proportions or percentages are
relatively more reliable than the corresponding estimates of the
numerator of the proportion or percentage, when the proportion or
percentage is based upon a sub-group of the population. For
example, the proportion of female babies who were of low birth
weight is more reliable than the estimated number of “female
babies who were of low birth weight”. Note that in the tables the
c.v.’s decline in value reading from left to right.
When the proportion or percentage is based upon the total
population of the geographic area or age group covered by the
table, the c.v. of the proportion or percentage is the same as the
c.v. of the numerator of the proportion or percentage. In this case,
Rule 1 can be used.
When the proportion or percentage is based upon a subset of the
total population, reference should be made to the proportion or
percentage (across the top of the table) and to the numerator of the
proportion or percentage (down the left side of the table). The
intersection of the appropriate row and column gives the
coefficient of variation.
NLSCY Data Users Guide
164
2001/2002
Rule 3:
Estimates of
Differences
Between
Aggregates or
Percentages
The standard error of a difference between two estimates is
approximately equal to the square root of the sum of squares of
each standard error considered separately.
That is, the standard error of a difference
where
is estimate 1,
is:
is estimate 2, and alpha 1 and alpha 2
are the coefficients of variation of
coefficient of variation of
respectively. The
is given by
This formula is accurate for the difference between separate and
uncorrelated characteristics, but is only approximate otherwise.
Rule 4:
Estimates of
Ratios
Where the numerator is not a subset of the denominator (for
example, the ratio of the number of low birth-weight female babies
to that of low-birth weight male babies), the standard deviation of
the ratio of the estimates is approximately equal to the square root
of the sum of squares of each coefficient of variation considered
separately multiplied by the ratio itself.
The standard error of ratio
where
and
of low-birth
is therefore:
are the coefficients of variation of
weight female babies) and
babies) respectively.
The coefficient of variation of
(the number of low birth-weight male
is given by
The formula will tend to overstate the error, if
positively correlated and
understate the error if
NLSCY Data Users Guide
165
and
(the number
.
and
are
are negatively correlated.
2001/2002
Rule 5:
Estimates of
Differences of
Ratios
In this case, Rules 3 and 4 are combined. The c.v.’s for the two
ratios are first determined using Rule 4, and then the c.v. of their
difference is found using Rule 3.
Warning Note on
Confidence
Intervals
Release guidelines applying to estimates also apply to confidence
intervals. For example, if the estimate is “marginal”, then the
confidence interval is marginal and should be accompanied by a
warning note to caution subsequent users about high levels of
error.
NLSCY Data Users Guide
166
2001/2002
Examples of using C.V. Tables for Categorical Estimates
Introduction
The following are examples using actual NLSCY data to illustrate
how to apply the foregoing rules.
Example 1:
Estimates of
Numbers
Possessing a
Characteristic
(Aggregates)
Using NLSCY data, 84,085 babies were estimated to be of low
birth weight (i.e., less than 2,500 grams). How does the user
determine the coefficient of variation of this estimate?
(1) Refer to the c.v. table for children in 0 to 3 age group. It should
be noted that, because the question on birth weight applied only to
children in this age group, this table should be used to determine
the c.v. for this estimate.
(2) The estimated aggregate (84,085) does not appear in the
left-hand column (the “Numerator of Percentage” column), so it is
necessary to use the figure closest to it, namely 85,000.
(3) The coefficient of variation for an estimated aggregate is found
by referring to the first non-asterisk entry on that row, namely, 7.3%.
(4) The approximate coefficient of variation of the number of low
birth-weight babies is estimated to be 7.3%. The finding that there
were 84,085 babies that were of low birth weight is “acceptable”
and no warning message is required to produce this estimate
since the c.v. for the estimate is in the 0.0% to 16.5% range.
NLSCY Data Users Guide
167
2001/2002
Example 2:
Estimates of
Proportions or
Percentages
Possessing a
Characteristic
Using NLSCY data, it is estimated that 70. 8% (59,567/84,085) of
low birth-weight babies were born prematurely (gestational age
258 days or less). How does the user determine the coefficient of
variation of this estimate?
(1) Refer to the c.v. table for children in 0 to 3 age group. It should
be noted that, because the questions on birth weight and delivery
time applied only to children in this age group, this table should be
used to determine the c.v. for this estimate.
(2) Because the estimate is a percentage which is based on a
subset of the total population (i.e., low birth-weight babies who
were born prematurely), it is necessary to use both the percentage
(70.8%) and the numerator portion of the percentage (59,567) in
determining the coefficient of variation.
(3) The numerator, 59,567, does not appear in the left-hand column
(the “Numerator of Percentage” column) so it is necessary to use
the figure closest to it, namely 60,000. Similarly, the percentage
estimate does not appear as any of the column headings, so it is
necessary to use the figure closest to it, 70.0%.
(4) The figure at the intersection of the row and column used,
namely 5.0% is the coefficient of variation to be used.
(5) The approximate coefficient of variation of the percentage of
low birth-weight babies who were premature is estimated to be
5.0%. Since the c.v. for the estimate falls in the 0.0% to 16.5%
range, this estimate is “acceptable”, and the finding that 70.8% of
low birth-weight babies were born prematurely requires no warning
note.
Example 3:
Estimates of
Differences
Between
Aggregates or
Percentages
NLSCY Data Users Guide
Using NLSCY data, it is estimated that 6.1% (45,690/753,203) of
female babies were born prematurely, while 4.9%
(38,395/791,149) of male babies were born prematurely. How
does the user determine the coefficient of variation of the
difference between these two estimates?
(1) Using the c.v. table for the 0 to 3 age group in the same manner
as described in example 2 gives the c.v. of the estimate for female
babies as 10.3%, and the c.v. of the estimate for male babies as
10.9%.
168
2001/2002
Example 4:
Estimates of
Ratios
Suppose now a user wants to compare the number of low birthweight female babies to the number of low birth-weight male
babies. The user is interested in comparing these estimates in the
form of a ratio. How does the user determine the coefficient of
variation of this estimate?
(1) First of all, this estimate is a ratio estimate, where the
numerator of the estimate = (
) is the number of low birth-weight
female babies and denominator = ( ) of the estimate is the
number of low birth-weight male babies.
(2) Refer to the table for the 0 to 3 age group. The questions on
birth weight were applicable only to children in the 0 to 3 age
group.
(3) The numerator of this ratio estimate is 45,690. The figure
closest to it is 45,000. The coefficient of variation for this estimate
is found by referring to the first non-asterisk entry on that row,
namely, 10.3%.
(4) The denominator of this ratio estimate is 38,395. The figure
closest to it is 40,000. The coefficient of variation for this estimate
is found by referring to the first non-asterisk entry on that row,
namely, 10.9%.
(5) The approximate coefficient of variation of the ratio estimate is
therefore given by Rule 4, which is
are the coefficients of variation of
, where
and
and
, respectively.
That is:
The ratio of low birth-weight female babies versus low birth-weight
male babies is 45,690/38,395, or 1.19:1. Since the c.v. for the
estimate falls in the 0.0% to 16.5% range (15.0%), this estimate is
“acceptable”, and the finding that 70.8% of low birth-weight babies
were born prematurely requires no warning note.
NLSCY Data Users Guide
169
2001/2002
How to Use the C.V. Tables to Obtain Confidence Limits
Introduction
Although coefficients of variation are widely used, a more intuitively
meaningful measure of sampling error is the confidence interval of
an estimate. A confidence interval constitutes a statement on the
level of confidence that the true value for the population lies within a
specified range of values. For example a 95% confidence interval
can be described as follows:
If sampling of the population is repeated indefinitely, each sample
leading to a new confidence interval for an estimate, then in 95% of
the samples the interval will cover the true population value.
Using the standard error of an estimate, confidence intervals for
estimates may be obtained under the assumption that under
repeated sampling of the population, the various estimates
obtained for a population characteristic are normally distributed
about the true population value. Under this assumption, the
chances are about 68 out of 100 that the difference between a
sample estimate and the true population value would be less than
one standard error, about 95 out of 100 that the difference would
be less than two standard errors, and about 99 out 100 that the
differences would be less than three standard errors. These
different degrees of confidence are referred to as the confidence
levels.
Confidence intervals for an estimate are generally expressed as
two numbers, one below the estimate and one above the estimate,
as where k is determined depending upon the level of confidence
desired and the sampling error of the estimate.
Confidence intervals for an estimate can be calculated directly
from the Approximate Sampling Variability Tables by first
determining from the appropriate table the coefficient of variation
of the estimate and then using the following formula to convert to a
confidence interval CI:
where
NLSCY Data Users Guide
is the determined coefficient of variation
170
and
2001/2002
t = 1 if a 68% confidence interval is desired
t = 1.6 if a 90% confidence interval is desired
t = 2 if a 95% confidence interval is desired
t = 3 if a 99% confidence interval is desired.
Note Regarding
Release
Guidelines
Release guidelines applying to estimates also apply to confidence
intervals. For example, if the estimate is “marginal”, then the
confidence interval is marginal and should be accompanied by a
warning note to caution subsequent users about high levels of
error.
Example of Using the C.V. Tables to Obtain Confidence Limits
Example
A 95% confidence interval for the estimated proportion of babies
who were of low birth weight would be calculated as follows.
Estimate of X = 5.5%
t=2
alpha estimate of X = 7.3% (.073 expressed as a proportion)
is the coefficient of variation of this estimate as determined by the
tables
CIx = {0,055 - (2)(0,055)(0,073), 0,055 + (2)(0,055)(0,073)}
CIx = {0,055 - 0,008, 0,055 + 0,008}
CIx = {0,047, 0,063}
With 95% confidence it can be said that between 4.7% and 6.3%
of babies who were 0 to 3 years old at the time of the survey were
of low birth weight.
NLSCY Data Users Guide
171
2001/2002
How to Use the C.V. Tables to Do a T-test
Hypothesis
Testing
Standard errors may also be used to perform hypothesis testing, a
procedure for distinguishing between population parameters using
sample estimates. The sample estimates can be numbers,
averages, percentages, ratios, etc. Tests may be performed at
various levels of significance, where a level of significance is the
probability of concluding that the characteristics are different when,
in fact, they are identical.
Let
and
be sample estimates for two characteristics
of interest. Let the standard error on the difference
be
If
is between -2 and 2, then no conclusion about the
difference between the characteristics is justified at the 5% level of
significance. If however, this ratio is smaller than -2 or larger than
+2, the observed difference is significant at the 0.05 level. That is
to say that the characteristics are significantly different.
NLSCY Data Users Guide
172
2001/2002
Example of Using C.V. Tables to do a T-Test
Example
Let us suppose we wish to test, at 5% level of significance, the
hypothesis that there is no difference between the proportion of low
birth-weight female babies and that of low birth-weight male
babies. From example 3 (Section 12.1.1), the standard error of
the difference between these two estimates was found to be =
.008.
Hence,
Since t = 1.5 is between -2 and 2, no conclusion at the 0.05 level of
significance can be made regarding the difference in proportions
of low birth-weight male or female babies.
Coefficients of Variations for Quantitative Estimates
Quantitative
Estimates
For quantitative estimates, special tables would have to be
produced to determine their sampling error. Since most of the
variables for the NLSCY are categorical in nature, this has not
been done.
As a general rule, however, the coefficient of variation of a
quantitative total will be larger than the coefficient of variation of the
corresponding category estimate. If the corresponding category
estimate is not releasable, the quantitative estimate will not be
either. For example, the coefficient of variation of the total number
of days of special medical care received for low birth-weight
babies would be greater than the coefficient of variation of the
corresponding proportion of babies who were of low birth weight.
Hence if the coefficient of variation of the proportion is not
releasable, then the coefficient of variation of the corresponding
quantitative estimate will also not be releasable.
NLSCY Data Users Guide
173
2001/2002
Pseudo
Replication
Coefficients of variation of such estimates can be derived as
required for a specific estimate using a technique known as
pseudo replication. This involves dividing the records on the
microdata files into subgroups (or replicates) and determining the
variation in the estimate from replicate to replicate. Users wishing
to derive coefficients of variation for quantitative estimates may
contact Statistics Canada for advice on the allocation of records to
appropriate replicates and the formulae to be used in these
calculations.
Release Cut-offs for the NLSCY
Cut-off Numbers
In the tables that follow, cut-off numbers are given for NLSCY
estimates in order for them to be of “acceptable”, “marginal” or
“unacceptable” quality. Users are encouraged to use these cutoffs when publishing data from the NLSCY. First a table is given to
show the cut-offs at the provincial, regional and Canada level.
Then a table is given to show the cut-offs for the various age
cohorts. An interpretation of what is meant by the various cut-off
levels can be found in Section 11.4.
For example, an estimate for Nova Scotia of 5,000 would fall into
the “marginal” range. This would mean that the estimate should be
flagged and a note of caution would be attached for subsequent
users about the high level of error associated with the estimate.
NLSCY Data Users Guide
174
2001/2002
GEOGRAPHICAL RELEASE CUT-OFFS
CROSS-SECTIONAL SAMPLE
Province
Acceptable Estimates at or
above
Marginal Estimates
between
Unacceptable
Estimates at or
below
Newfoundland
Prince Edward
Island
Nova Scotia
New Brunswick
Québec
Ontario
Manitoba
Saskatchewan
Alberta
7,500
3,500
2,000 to 7,500
1,000 to 3,500
2,000
1,000
12,000
7,000
52,500
62,000
18,500
13,000
33,500
3,000 to 12,000
2,000 to 7,000
13,500 to 52,500
15,500 to 62,000
5,000 to 18,500
3,500 to 13,000
8,500 to 33,500
3,000
2,000
13,500
15,500
5,000
3,500
8,500
51,500
13,500 to 51,500
13,500
9,000
2,500 to 9,000
2,500
26,000
6,500 to 26,000
6,500
41,000
10,000 to 41,000
10,000
British
Columbia
Atlantic
provinces
Prairie
provinces
Total
NLSCY Data Users Guide
175
2001/2002
RELEASE CUT-OFFS BY AGE GROUP
CROSS-SECTIONAL SAMPLE
Age Group
Acceptable Estimates at
or above
15,500
20,000
35,500
42,000
37,500
37,000
40,500
Marginal Estimates
between
4,000 to 15,500
5,000 to 20,000
9,000 to 35,500
11,000 to 42,000
9,500 to 37,500
9,500 to 37,000
10,500 to 0,500
Unacceptable
- Estimates at
or below
4,000
5,000
9,000
11,000
9,500
9,500
10,500
0 - 3 years
4 - 11 years
18,500
41,000
4,500 to 18,500
10,000 to 41,000
4,500
10,000
4 - 7 years
8 - 11 years
43,000
43,000
11,000 to 43,000
11,000 to 43,000
11,000
11,000
TOTAL
41,000
10,000 to 41,000
10,000
0 - 23 months
2 - 3 years
4 - 5 years
6 - 7 years
8 - 9 years
10 - 11 years
12 - 13 years
NLSCY Data Users Guide
176
2001/2002
GEOGRAPHICAL RELEASE CUT-OFFS
LONGITUDINAL SAMPLE
Province
Acceptable Estimates at
or above
7,000
3,500
13,000
9,500
63,500
67,500
18,000
Marginal Estimates
between
2,000 to 7,000
1,000 to 3,500
3,500 to 13,000
2,500 to 9,500
16,500 to 63,500
17,000 to 67,500
4,500 to 18,000
Unacceptable Estimates at or
below
2,000
1,000
3,500
2,500
16,500
17,000
4,500
Saskatchewan
Alberta
British Columbia
13,000
36,500
52,000
3,500 to 13,000
9,500 to 36,500
13,500 to 52,000
3,500
9,500
13,500
Atlantic provinces
Prairie provinces
10,500
27,500
2,500 to 10,500
7,000 to 27,500
2,500
7,000
Total
58,000
14,500 to 58,000
14,500
Newfoundland
Prince Edward Island
Nova Scotia
New Brunswick
Québec
Ontario
Manitoba
NLSCY Data Users Guide
177
2001/2002
RELEASE CUT-OFFS BY AGE GROUP
LONGITUDINAL SAMPLE
Age Group
0 - 23 months
2 - 3 years
4 - 5 years
6 - 7 years
8 - 9 years
10 - 11 years
0 - 3 years
4 - 11 years
4 - 7 years
8 - 11 years
NLSCY Data Users Guide
Acceptable Estimates at
or above
Marginal Estimates
between
Unacceptable
Estimates at
or below
19,500
33,000
38,000
37,000
36,500
38,500
29,000
47,000
47,000
50,500
58,000
5,000 to 19,500
8,500 to 33,000
9,500 to 38,000
9,500 to 37,000
9,500 to 36,500
10,000 to 38,500
7,000 to 29,000
11,500 to 47,000
12,000 to 47,000
12,500 to 50,000
14,500 to 58,000
5,000
8,500
9,500
9,500
9,500
10,000
7,000
11,500
12,000
12,500
14,500
178
2001/2002
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement