In defense of matching.

This paper discusses the simplified situation of an epidemiologic study involving disease, exposure, and a single (possibly confounding) extraneous factor, all of which are dichotomous. The question is: In studying the association between disease and exposure, should the comparison group be selected by random sampling or by matching on the extraneous factor? An example is used to demonstrate the general principle that matching controls confounding in estimating the risk ratio in a follow-up study, but not in estimating the exposure odds ratio in a case-control study. Calculations based on a probability model show that, despite the possible reduction in sample size which may be associated with matching, matching will often lead to a more precise estimate of the effect measure than random sampling and is not likely to result in a significant loss in precision in situations of practical importance. Therefore, selection of the referent group by matching should be given serious consideration for both follow-up and case-control studies.