Treating Matching as a Variable in a Sociological Experiment

The matching of experimental and control subjects on critical variables, particularly when used in conjunction with random assignment, is widely recognized as a valuable element in research design. Both in discussions of its usefulness and in applications, matching tends to be treated as an attribute. In fact, however, it is a variable, with a range determined by the number and precision of measurement of the items taken into account in the matching process, and by the size and homogeneity of the pool from which subjects are drawn. Failure to distinguish between well-matched and poorly-matched pairs can lead to significant errors of interpretation. But treating the quality of the match itself as a variable, and introducing it into the interpretation can, it is suggested, contribute to the measurement of the effects of the experimental stimulus. An index of congruence, designed to measure the quality of matching in an experiment in progress, is described. We propose to use this new method as a complement to established statistical techniques, whose comparative advantages and disadvantages are here briefly discussed