Online Program

Identifying Factors Affecting Nonresponse Rates of Complex Surveys
MoonJung Cho, U.S. Bureau of Labor Statistics 
Jenny FitzGerald, Bureau of Labor Statistics 
*Lily Wang, University of Georgia 

Keywords: Clustered data, Generalized estimating equations, Group variable selection, Semiparametric, Shrinkage estimation, U.S. International Price Program.

Nonresponse is a problem for almost every survey. An ongoing goal for all surveys is to reduce the rates of nonresponse. Identifying characteristics of potential nonrespondents is beneficial in reducing the nonresponse rate and in evaluating the nonresponse bias. For surveys that use multiple stages of sampling, incomplete data patterns may be associated with characteristics of primary sample units (PSUs); of secondary or finer level sample units; of interviewers; or PSU-level workload. In this talk, we will describe a semi-parametric marginal mean logistic regression model to analyze the nonresponse patterns related to the predictors listed above. We represent the nonparametric components by a linear combination of spline basis functions, and develop a factor selection procedure for complex surveys to identify important explanatory factors that affect the nonresponse rate. When categorical predictors or basis functions are present in the model, the traditional variable selection is not satisfactory as it only selects individual derived variables instead of whole factors. The proposed factor (group variable) selection overcomes this problem and is able to select the correct grouped variables consistently. We adopted this methodology to analyze the U.S. International Price Program (IPP) Surveys to explore the relationship between response rate and establishment factors as well as other characteristics. We will illustrate the superior performance of our method using data sets from the IPP nonresponse study. The proposed method is applicable to any standard nonresponse rate study for surveys with multiple-stage sample designs.