Inverse Probability Weighting for Clustered Nonresponse

• 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): Chris Skinner*+
• Companies: University of Southampton
• Keywords: missing data ; nonresponse ; propensity score ; weighting

Abstract

Inverse probability weighting is a widely-used tool in estimation under selection. In survey applications, it is used to handle unequal probability sampling and non-response. This paper considers its use when the selection of different units may be dependent because of clustering. The main motivation is adjustment for nonresponse which exhibits clustering. It is assumed that nonresponse follows a logistic regression model which includes a cluster effect, which would conventionally be taken as random. The question addressed in the paper is how to 'estimate' this cluster effect when constructing inverse probability weights. Approaches considered include a marginal approach (in which the effect is integrated out), BLUP prediction of the random effect and a fixed effect approach.