Journal of Official Statistics, Vol.7, No.3, 1991. pp. 325–337
Weighting via Response Modeling in the Finnish Household Budget Survey
Anders Ekholm and Seppo Laaksonen
Abstract:The purpose of the Finnish Household Budget Survey is to monitor the consumption of the total population and of segments of the population. The data are collected by interviews. The sampling frame is a person-based register, but the unit of observation is the household. Nonresponse has been increasing in recent rounds. For the estimation of totals and means from the 1985 data, we implemented a “model based Horvitz-Thompson estimator.” We model the response propensity by logistic regression on explanatory factors, most important of them being the household structure. The estimator weights by the inverses of the estimated response probabilities from the fitted model. We state the assumption under which the estimator is unbiased and derive its appropriately conditioned variance. We compare the new estimator with the one previously used, theoretically and by selected empirical results. We discuss some alternatives to weighting by response propensity modeling.
Keywords:adjustment cell; estimation of mean; estimation of total; model based Horvitz-Thompson estimator; logistic regression; response probability; selection probability; unit nonresponse.
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group