An Evaluation of Statistical Matching

The validity of findings based on statistically matched data sets depends on the accuracy of underlying assumptions about relationships between variables that are unique to each input file. Simulations of statistical matching procedures on samples from populations with known characteristics provide the basis for an evaluation of the usefulness of statistical matching, and for choosing among various matching techniques. In three such simulations frequent and substantial errors occur in estimates of bivariate and multivariate relationships between variables taken from two matched files. Alternative procedures for meeting the objectives of statistical matching with more solid theoretical justification and empirical support are proposed.