The Computerized Construction of a Matched Sample

Abstract

The use of matched sampling in experimental and nonexperimental research has been frustrated by the problems of attrition and incomplete matching. The amount of attrition and incomplete matching. The amount of attrition of sample size and incomplete matching are shown to be a function of various aspects of the research, including the level of measurement of the matching variables, the sampling situation, the definition of a match, and the means used to construct matches. Several basic and heretofore neglected considerations in the conceptualization of a "match" are discussed. These considerations and a rarallel set of desiderata became apparent during the computerized construction of two matched samples of black and white college graduates. Three approaches to the computerization of matching are described, the last of which-"guaranteed variable caliper matching"- meets the desiderata. We describe our clerical and field procedures from the computerized matching to the final matching based on completed interviews. The degree to which both computer matching and these procedures controlled attition and incomplete matching is assessed. The quality of matches is found to be fairly high and attrition has been virtually elliminated, thus demonstrating the possibilities of our approach.