On Weighting of Propensity Models When Analyzing Nonresponse

• Activity Number: 116
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Government Statistics
• Author(s): Daniell Toth*+ and John Eltinge and Polly Phipps
• Companies: Bureau of Labor Statistics and Bureau of Labor Statistics and Bureau of Labor Statistics
• Keywords: conditional distributions ; nonresponse analysis ; paradata ; survey redesign ; survey weights

Abstract

In any large scale survey there are usually selected sample units that provide no responses, and the missing information from these nonrespondents could have a negative impact on data quality. Response propensity models estimate the likelihood that a given unit will respond to the survey based on known characteristics. These models are important in understanding survey response and the impact that intentional or unintentional treatments or actions will have on response. These models usually incorporate data available from the sample frame, as well as administrative data obtained during the attempts to collect the survey data.

This paper explores the prospective impact of weights in estimation of these models, with emphasis on (a) the intended use of the models; (b) adequacy of fit for the proposed model; and (c) the degree of association between the available model predictors and the underlying design variables. The principal ideas are illustrated with data from BLS surveys.