Using Imputation to Reduce the Cost of Survey Collection in the Current Population Survey

John S. Dixon

Abstract

Survey cost has been of concern due to the increasing cost of interviewing, where increased effort has been needed to maintain sample size with more households being reluctant to respond. The reluctance is seen in the increase in nonresponse in most household surveys. This study will investigate whether imputing survey responses, rather than collecting the data, for select respondents in a longitudinal survey can reduce costs while maintaining estimate quality. The Current Population Survey (CPS) is designed to measure labor force characteristics of the United States. Subgroups which are likely to have stable labor force status will be of special interest for imputing. Some examples could be older retired people (where the entire household would be not-in-labor-force), parents involved in childcare for young children (where one parent is employed. The effect of imputing will be simulated by removing the responses of those predicted to have stable labor force status, and then imputing their responses. The difference in estimates between the original data and the data with imputations will show the effectiveness of the model.