Partially Linear Models in Dual-Frame Surveys

• Activity Number: 249
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Yan Lu*+ and Yang Cheng
• Companies: University of New Mexico and U.S. Census Bureau
• Keywords: Dual frame surveys ; Partially linear models ; difference-based variance estimator ; Combined inference ; Simulations

Abstract

As the population and methods used to collect survey data change, the traditional single frame surveys may miss parts of the population. In order to obtain better coverage of the population of interest and cost less, a number of surveys employ dual frame design, in which independent samples are taken from two overlapping sampling frames. In this research, we propose partially linear models for regression structure in dual frame surveys. We intend to model the domain (the nonoverlapping components of the union of two frames) effect (or called treatment effects) parametrically and to model the effect of the covariate nonparametrically. We incorporate the dual frame survey weights into the partially linear model by constructing a weighting parameter from the overlap domain through cross validation method. An extended difference-based variance estimator is used to approximate the noise level of the model. A combined inference frame work is used to derive the estimators' properties. Simulation studies are conducted to investigate the finite sample properties of the proposed estimators.