Dynamic Weighting of Panel Data

Washington University, Saint Louis MO

Investigators

Abstract

General Although it is not always the case, surveys are generally used to make inferences about a larger population using a relatively small sample. In order to do that, each respondent in a survey is given a weight to describe the relationship between the proportion of individuals like them in the survey relative to those in the population. The construction of these weights is critically important as they ensure the survey sample can accurately reflect the population. This project develops a tool for users to estimate weights for several of the most commonly used panel surveys in political science. By enabling researchers to estimate weights that are appropriate for their individual research questions, we increase the reliability of studies that use these panel surveys. Technical Because of the issues that panel weighting involves, we develop an R package that delivers a single weight variable based on the sub-samples obtained after attrition and item non-response. The motivation behind the creation of such a statistical package is the need for a tool that can deliver a set of weights suitable for the analyses desired by researchers dependent on the outcome variables, covariates and waves specified. Moreover, we develop program to run a diagnosis of the socio-demographic characteristics of the remaining sample after accounting for attrition and item non-response in order to flag potential complications such as over-weight of respondents that belong to groups with high levels of attrition. Finally, we generate an accessible and user-friendly program that facilitates the analysis of panel survey data. Countless scholars rely upon nationally-representative and publicly-available panel surveys as a data resource to draw inferences about national opinion. This project improves those inferences by providing researchers with the tools they need to weight their data appropriately.