2
Statistical Matching and Microsimulation Models
Michael L.Cohen
INTRODUCTION
Microsimulation models are large, computationally demanding models that make use of data at the level of the decision-making unit in order to determine the impact of program changes by separately evaluating the effect of those changes for each unit. In order to do such an evaluation, it is necessary to have a data file with enough information available so that the actions of each decision-making unit can be simulated. When using an existing sample survey for input into a microsimulation model (e.g., the Current Population Survey [CPS]), the information needed to establish eligibility or the size of the award is often not available. This is especially true for microsimulation models in the areas of long-term health care, pensions, and taxes.
Statistical matching is used when one needs two groups of variables that do not exist simultaneously on any one data set. For example, suppose a subset of the variables available from one sample survey includes financial information and that variables from another data set provide information on health. Often, exact matching is impossible either because identifiers do not exist on both files or, more commonly, because both data files are small samples so that the two files have an essentially empty intersection. The collection of a data set
Michael L.Cohen is assistant professor in the School of Public Affairs at the University of Maryland; he served as a consultant to the Panel to Evaluate Microsimulation Models for Social Welfare Programs.
