The feasibility of matching and quota sampling in epidemiologic studies

Matching is often used to eliminate the effects of potential confounders. The practical execution of a matched design will usually involve additional sampling effort to find appropriate quotas of persons who satisfy the matching criteria. If this effort is substantial, it may render the matched design less attractive than using unmatched samples with covariance adjustment in the analysis. This paper presents numerical results on the sampling effort needed to complete various matched designs: The effort is measured by the number of candidates (N) that must be sampled to identify the matches for analysis. Feasibility of matched sampling is described by the ratio of the expected value of N to its theoretical minimum, and by its coefficient of variation. Feasibility must be weighed against the resulting quality of matches and the possibility of residual confounding. The results show that designs with smaller matching quotas in larger numbers of categories are more difficult to execute. However, the relative variation in N is mainly a function of the category quotas and is only weakly dependent on the number of categories. Recommendations are made regarding the number of categories and the quota sizes that lead to acceptable matched designs. The results are applicable to both observational and experimental studies in which matching or other quota sampling is used.