Imputation Techniques in Microsimulation

Abstract

Microsimulation has gained attention for its use in analyzing and forecasting the individual impacts of alternative economic and social policy measures. In practice, however, microsimulation cannot be carried out from a single data source, since it requires far more information than any single data source can provide. This paper discusses ways to combine separate data sources when there are no identical key variables, using imputation techniques, to make a large but synthetic data source for microsimulation. A new approach based on propensity score matching is suggested and discussed.