Modelling & Weighting

Highly recommended: look at the user guides for a detailed discussion of modelling and weighting in the NLSCY.

This overview is taken from the user guide for Cycle 4, which includes revisions made to the weights of previous cycles. Note that the most recent information is the best information.

Overview
Revisions to weights of Cycles 1, 2 and 3
User Guides


Overview


Weights must be used to derive meaningful estimates of the characteristics measured by the NLSCY, since the survey is a highly complex probability sample. (Weighting is the calculation of how many people in the population a respondent represents.)

The weights are designed for the child, which is the unit of analysis in the NLSCY.

Since the NLSCY sample is based on the Labour Force Survey (LFS) household sample, the derivation of its weights is tied to the weighting procedure used for the LFS. The user guides describe how the household-to-child adjustments are made.

Conceptually, each child’s basic weight is approximately equal to the inverse of his/her probability of selection, with adjustments for factors such as non-response, conditioning bias and demographic estimates.

The NLSCY contains both cross-sectional and longitudinal data. As the target population is not the same in the cross-sectional and longitudinal samples, the number of people each child represents is not the same. The NLSCY consequently has more than one set of weights.

• In Cycles 1, 2 and 3, there are two weight variables: cross-sectional and longitudinal (which pertain to the child population in Cycle 1). In each cycle, the longitudinal weights are recalculated to take into account the sample erosion that occurs between cycles.
• In Cycle 4 an additional longitudinal weight was added, a “funnel” weight, that applies only to Cycle 4 longitudinal respondents who were also respondents in Cycles 1, 2 and 3 (i.e. a respondent in all four cycles).

Weights are calculated for each longitudinal and cross-sectional cohort. Only the cross-sectional weight is used when doing analysis on the cross-sectional sample, and only the longitudinal weight is used when doing analysis on the longitudinal data. The choice of which sample to use depends on the type of analysis.

For any analysis where a significance measure is required, such as correlation analysis, it is recommended that a “sample” weight be used. This weight is obtained by multiplying the survey weight by the sample size and dividing the total by the total estimated population. This produces a mean weight of 1 and a sum of weights equal to the sample size (14.2: Sample Weighting Guidelines for Tabulation, Cycle 4 user guide).

Guidelines for statistical analysis:

• Use survey weights for survey estimates and analyses to be free of bias.
• As the variance estimates calculated by most statistical packages are not adequate for the NLSCY, use the Sampling Variability Tables for simple estimates such as totals, proportions and ratios.
• For other analysis techniques, such as linear regression, logistic regression and analysis of variance, a method exists to rescale the weights in the standard packages – for an example of rescaling, see 14.4 of the Cycle 4 user guide.

Top

Revisions to weights of Cycle 1, 2 and 3


Demographic estimates are used in the final stage of weighting. As updated census information becomes available, the design weights for the earlier cycles are adjusted accordingly.

Revised microdata files for Cycles 1, 2 and 3 have been released with Cycle 4 files. For details on the revisions, including the magnitude of the changes, see Appendix II: Changes from Previous Cycles, Cycle 4 user guide.

The cycle weights will be finalized once the census data are finalized.

Top

User Guides


Each of the user guides include an entire chapter on weighting as well as discussions within other chapters.

(Note: you may need to right-click the links below and save the file to your computer. If this doesn’t work, go directly to the DLI link.)

Cycle 4: Microdata User Guide

• Chapter 12: Weighting and treatment of non response (p. 98) and also
• 9.2.3 Factor analysis using weighted data (p. 53)
• 14.2: Sample weighting guidelines for tabulation (p. 130)
• 17.12 Updating the weights for previous cycles (p. 171)
• Appendix II – 1: Revision of weights for previous cycles (p. 191)

Cycle 3: NLSCY Data Users Guide (2001/2002)

Cycle 2: User Guide

Cycle 1 Release 1: User's Handbook and Microdata Guide

Cycle 1 Release 2: User Guide and Data Dictionary - click on link below

Enquiries: Special Surveys Methods Division, Statistics Canada at 613-951-3321 or 1-800-461-9050; ssd@statcan.ca

nlscy3a.pdf 4293 (Kbytes)

Top