Nonparametric Estimation of Response Probabilities in Survey Sampling: An Empirical Investigation

• Activity Number: 685
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Audrey Beliveau*+ and David Haziza
• Companies: Université de Montréal and Université de Montréal
• Keywords: unit nonresponse ; nonparametric estimation ; response probability ; weighting classes

Abstract

Nonresponse bias occurs when respondents and nonrespondents differ with respect to characteristics of interest. A common way of reducing the nonresponse bias is to adjust the design weight of the respondents. A popular weight adjustment procedure consists of dividing the design weight of respondents by their estimated response probability, which can be estimated using parametric methods (e.g., logistic regression model) or nonparametric methods. Nonparametric methods include kernel smoothing methods and local polynomial regression (Da Silva & Opsomer, 2006, 2009) or weight adjustment procedures applied within weighting classes (which is commonly used in practice). In this talk, results of a Monte Carlo simulation study comparing different nonparametric methods in terms of bias and mean square error will be presented.