Nonparametric Regression in Some Nonstandard Sampling Situations

• Activity Number: 551
• Type: Invited
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Alan H. Dorfman*+
• Companies: Bureau of Labor Statistics
• Keywords: local linear regression ; asymptotics ; bandwidth selection ; stratification ; design based ; discontinuities

Abstract

Standard sampling methods require randomization. Estimators typically rely on the inclusion probabilities inherent in the sample design. We consider here a variant of adaptive sampling where, auxiliary information being available for all population units, sampling proceeds in stages, initially taking a partial sample to get a preliminary picture of the relationship between the variable of interest and the auxiliary, and using the information gained to decide where collecting further data would be most useful. It is necessary to appraise the relative importance of competing determinants of where sample is needed, for example, indications of heteroscedasticity, of peaks or valleys, and of discontinuities. Calculating inclusion probabilities in such a scheme is forbidding. We aim at robust inference by employing estimators based on non-parametric regression.