Imputation of missing expenditure information in standard household income surveys

The aim of this paper is to present a new methodology for dealing with missing expenditure information in standard income surveys. Under given conditions, typical imputation procedures, such as statistical matching or regression-based models, can replicate well in the income survey both the unconditional density of household expenditure and its joint density with a set of socio-demographic variables that the two surveys have in common. However, standard imputation procedures may fail in capturing the overall relation between income and expenditure, especially if the common control variables used for the imputation have a weak correlation with the missing information. The paper suggests a two-step imputation procedure that allows reproducing the joint relation between income and expenditure observed from external sources, while maintaining the advantages of traditional imputation methods. The proposed methodology suits well for any empirical analysis that needs to relate income and consumption, such as the estimation of Engel curves or the evaluation of consumption taxes through micro-simulation models. An empirical application shows the makings of such a technique for the evaluation of the distributive effects of consumption taxes and proves that common imputation methods may produce significantly biased results in terms of policy recommendations when the control variables used for the imputation procedure are weakly correlated with the missing variable.