A Markov Chain Monte Carlo Multiple Imputation Procedure for Dealing with Item Nonresponse in the German SAVE Survey

Content Important empirical information on household behavior is obtained from surveys. However, various interdependent factors that can only be controlled to a limited extent lead to unit and item nonresponse, and missing data on certain items is a frequent source of difficulties in statistical practice. This paper presents the theoretical underpinnings of a Markov Chain Monte Carlo multiple imputation procedure and applies this procedure to a socio-economic survey of German households, the SAVE survey. I discuss convergence properties and results of the iterative multiple imputation method and I compare them briefly with other imputation approaches. Concerning missing data in the SAVE survey, the results suggest that item nonresponse is not occurring randomly but is related to the included covariates. The analysis further indicates that there might be differences in the character of nonresponse across asset types. Concerning the methodology of imputation, the paper underlines that it would be of particular interest to apply different imputation methods to the same dataset and to compare the findings.