Representative Wealth Data for Germany from the German SOEP: The Impact of Methodological Decisions Around Imputation and the Choice of the Aggregation Unit

Abstract

The definition and operationalization of wealth information in population surveys and the corresponding microdata requires a wide range of more or less normative assumptions. However, the decisions made in both the pre- and post-data-collection stage may interfere considerably with the substantive research question. Looking at wealth data from the German SOEP, this paper focuses on the impact of collecting information at the individual rather than household level, and on “imputation and editing” as a means of dealing with measurement error. First, we assess how the choice of unit of aggregation or unit of analysis affects wealth distribution and inequality analysis. Obviously, when measured in “per capita household” terms, wealth is less unequally distributed than at the individual level. This is the result of significant redistribution within households, and also provides evidence of a significant persisting gender wealth gap. Secondly, we find multiple imputation to be an effective means of coping with selective nonresponse. There is a significant impact of imputation on the share of wealth holders (increasing on average by 15%) and also on aggregate wealth (plus 30%). However, with respect to inequality, the results are ambiguous. Looking at the major outcome variable for the whole population—net worth—the Gini coefficient decreases, whereas a top-sensitive measure doubles. The non-random selectivity built into the missing process and the consideration of this selectivity in the imputation process clearly contribute to this finding. Obviously, the treatment of measurement errors after data collection, especially with respect to the imputation of missing values, affects cross-national comparability and thus may require some cross-national harmonization of the imputation strategies applied to the various national datasets.