Information Criterion for Nonparametric Model-Assisted Survey Estimators

• Activity Number: 90
• Type: Invited
• Date/Time: Sunday, August 9, 2015 : 8:30 PM to 9:15 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Addison Dolin James* and Lan Xue and Virginia Lesser
• Companies: Oregon State University and Oregon State University and Oregon State Univeristy
• Keywords: Nonparametric ; Survey ; Model-assisted ; Finite population ; Spline ; Variable selection

Abstract

By incorporating sampling weights, complex survey data can be summarized to obtain unbiased estimates of population parameters. Inexpensive auxiliary information may be available for the entire population. In recent years, nonparametric model-assisted methods have been proposed that incorporate auxiliary variables more efficiently than misspecified parametric models. However, careful consideration on including only relevant variables in the model is needed to obtain the most efficient estimator. Previous literature on variable selection may not perform well when the model is misspecified and the data is selected from a finite population with unequal weights. In this paper, we propose an information criterion when the superpopulation model is assumed to have an additive form and the data is collected from a finite population at a single stage with unequal sampling weights. Our proposed method is simpler to implement and is demonstrated through simulation to select the correct variables more often than a previously proposed method. An application of the method to the NHANES data is provided.