Sample Weighting Methods and Estimation of Totals in the Consumer Expenditure Survey

Abstract

The widely used Principal Person method of weighting households in federal government surveys uses external post-Censal information on population to improve survey sample weights by a form of poststratification. While the Principal Person Methodology can be viewed as part of a procedure to adjust for nonresponse and undercoverage, it is not oriented for efficiently incorporating ancillary information or combining information from multiple surveys into survey estimates of subdomain totals. In this article a generalized least squares adjustment algorithm is shown to incorporate ancillary information in a way that, in principle, reduces the design variance of estimated survey totals. The flexibility of the method is exploited in an application to the Consumer Expenditure Survey that makes use of its “weighting control” and “composition” features.

Key Words:

• Poststratification
• Generalized regression estimators