Variance estimation for Public Use Microdata Files / Wesley Yung and Joseph Duggan. : CS11-617/00-3E-PDF

"Statistics Canada produces many Public Use Microdata Files (PUMF) which are used by analysts wishing to perform their own analyses of data gathered by Statistics Canada's surveys. However, sample design information necessary to calculate proper design based variances, can not be included on the PUMF due to confidentiality constraints. As part of the PUMF documentation, approximate sampling variability tables are included which allow the user to calculate an approximate coefficient of variation for totals and ratios of categorical variables. Unfortunately, these tables can not be used for continuous data nor for complex statistics such as regression coefficients. In this paper, we investigate two possibilities for replacing the approximate sampling variability tables. The first method, the Generalized Variance Function approach, is easier to use than the existing method but similarly it suffers from a lack of versatility since it can also only be used for totals and ratios. The second method is based on the bootstrap variance estimator and is much more versatile but does have some confidentiality concerns. The two methods are described and empirical results are obtained using two of Statistics Canada's surveys"--Abstract.

Lien permanent pour cette publication :
publications.gc.ca/pub?id=9.840385&sl=1

• Format MARC XML
• Format MARC HTML

Demander des formats alternatifs