Appendix III: Sampling weights to account for the Labour Force Survey complex survey design
Archived Content
Information identified as archived is provided for reference, research or recordkeeping purposes. It is not subject to the Government of Canada Web Standards and has not been altered or updated since it was archived. Please "contact us" to request a format other than those available.
The Labour Force Survey (LFS) sample is drawn from an area frame and is based on a stratified, multi-stage design that uses probability sampling. First, Canada's population in provinces and regions is partitioned into strata. Instead of selecting the dwellings in the strata directly, a sample of small well-defined areas called clusters is selected in each stratum in the first stage of sampling. All dwellings within selected clusters are listed, and a sample of dwellings is chosen from each list in the second stage of sampling. A handful of three-stage strata are created in Quebec, Ontario, Alberta, and British Columbia to handle isolated urban centres.
In addition to accounting for the different probabilities of selection into the sample as well as accounting for non-response and coverage issues, one should also control for clustering and stratification of the survey design in order to obtain the correct standard error. While stratification typically increases the precision of the parameter estimates, the clustering of the sample will usually reduce it. Two weights are of particular relevance to the variance estimates from descriptive and multivariate analysis from the LFS Tabulations (TABS) files: 1) sub-weights (SUBWT); and 2) final weights (FINALWT). The final weights are used to produce the group means estimates in the paper so that the figures are generated on the basis of population counts consistent with census projections. The final weights incorporate auxiliary information such as census population estimates and the common sample between two consecutive months of survey data. For the multivariate analysis in the paper, sub-weights are used as the sampling weights instead of the final weights. In fact, the sub-weights and the final weights can both be used in the multivariate analyses. Both of these weights take into account the complex design features in the LFS. However, since there is no closed-form solution to adjust for standard errors by means of the final weights when using general statistics software (such as STATA), one may use the sub-weights, given the complexity of using the final weights. The use of sub-weights in the analysis would yield, in general, more conservative estimates (in terms of larger standard errors).
Standard statistical packages like STATA can produce standard errors that account for complex survey design. This can be done by using the SVY commands when sampling weights, stratification, and clustering scheme are identified. Since 1976, LFS data have been collected by means of four different sample designs, drawing on updated information from decennial censuses. These different sample designs cover the periods 1975-to-1984, 1985-to-1994, 1995-to-2004, and 2005-to-present. While sample design identifiers are available on the LFS data files for the years 1997 to present, they are not available on earlier files and must be derived. See Chan (2011) for a more detailed discussion of how the appropriate standard errors can be computed by using the LFS data.
- Date modified: