Far from normal: Multiple imputation of missing values in a German establishment survey

Abstract

"In this paper we suggest some adjustments for the standard multiple imputation routines to handle the real data problems mentioned above and illustrate a successful implementation of multiple imputation for complex surveys by describing the multiple imputation project for a German establishment survey, the IAB Establishment Panel.<br> The remainder of the paper is organized as follows: In Section II we recapitulate the basic concept of multiple imputation. In Section III we briefly discuss the two main approaches for multiple imputation, joint modeling and sequential regression, and discuss their advantages and disadvantages. In Section IV we present some adjustments for standard multiple imputation routines to handle problems that often arise with real data. We don't claim that the ideas presented in this Section are new. They have been suggested in several other papers. The main aim of this Section is to give an overview of potential problems that are likely to arise in real data applications and to provide a summary of possible solutions in one paper to free the potential multiple imputation user from the burden of a complete literature review in hopes of finding a solution to his specific problem. In Section V we describe results from the multiple imputation project at the German Institute for Employment Research (IAB) that heavily relies on the methods described in Section IV: The multiple imputation of missing values in the German IAB Establishment Panel. In this Section we also discuss the methods we used to evaluate the quality of the imputations. The paper concludes with some final remarks." (Text excerpt, IAB-Doku) ((en))

Cite article

Drechsler, J. (2009): Far from normal: Multiple imputation of missing values in a German establishment survey. (United Nations, Economic Commission for Europe. Working paper 21), New York, 13 p.