Dealing with Incomplete Household Panel Data in Inequality Research

Population surveys around the world face the problem of declining cooperation and participation rates of respondents. Not only can item nonresponse and unit nonresponse impair important outcome measures for inequality research such as total household disposable income; there is also a further case of missingness confronting household panel surveys that potentially biases results. The approach commonly used in such surveys of interviewing all adult household members and aggregating their individual incomes to yield a final outcome measure for welfare analyses often suffers from partial unit non-response (PUNR), i.e., the non-response of at least one unit, or member, of an otherwise participating household. In these cases, the aggregate income of all household members lacks at least one individual's income. These processes are typically not random and require appropriate correction. Using data from more than twenty waves of the German Socio-Economic Panel (SOEP) we evaluate four different strategies to deal with this phenomenon: (a) Ignorance, i.e., assuming the missing individual's income to be zero. (b) Adjustment of the equivalence scale to account for differences in household size and composition. (c) Elimination of all households observed to suffer PUNR, and re-weighting of households observed to be at risk of but not affected by PUNR. (d) Longitudinal imputation of the missing income components. The aim of this paper is to show how the choice of technique affects substantive results in the inequality research. We find indications of substantial bias on income inequality and poverty as well as on income mobility. These findings are obviously even more important in cross-national comparative analyses if the data providers in the individual countries deal differently with PUNR in the underlying data.