Censoring of outcomes and regressors due to survey nonresponse: Identification and estimation using weights and imputations

Abstract

Survey nonresponse makes identification of population parameters problematic. Except in special cases, identification is possible only if one makes untestable assumptions about the distribution of the missing data. However, nonresponse does not preclude identification of bounds on parameters. This paper shows how identified bounds on unidentified population parameters can be obtained under several forms of nonresponse. Organizations conducting major surveys commonly release public-use data files that provide nonresponse weights or imputations to be used for estimating population parameters. The paper shows how to bound the asymptotic bias of estimates using weights and imputations. The results are illustrated with empirical examples based on the National Longitudinal Survey of Youth.