Multiple Imputation Concerning Item Nonresponse Using the Example of an Analysis Model for Direct Democratic Voting Behaviour

Abstract

In this article, we deal with problems of two different natures. On the first level, the paper handles with direct democratic voting behavior on local level, concretely via logistic regression models which indentify issue oriented variables as the most important determinants. The second level, which is the actual core of this article, discusses two approaches for dealing with missing values: case-reducing versus imputation of the missing data. The linkage between a real scientific and a methodological question shows the importance of a sensitive handling of missing data – not least in social sciences. Our sample is used for the comparison of Completes-Cases-Analyses with Multiple Imputation. By means of a procedure which combines the advantages of a simulation with those of real data we create a valid base for our comparison. The analytical and graphical results show clearly that it’s appropriate to use Multiple Imputation instead of using a Complete-Cases-Analysis.