Abstract:
|
Missing Not at Random (MNAR) is a difficult problem for nonresponses in surveys. In this talk, we first describe commonly used methods for handling survey nonresponses. Secondly, we identify a connection between complex survey logistic regression and weighted GEE models. Then we demonstrate that by exploiting this connection, the missing data models for longitudinal studies developed in biostatistics can be applied to complex survey data. Specifically, we describe the use of pattern mixture weighted GEE (PMWGEE) models for assessing potentially MNAR data arising from survey nonresponses. Finally, we use some public domain survey data sets to demonstrate how to use latent class analysis of missing data patterns and PMWGEE models for handling MNAR survey nonresponses.
|