Abstract by Josh Meyers
Matching by Balancing Margins: The RCBalance algorithm
For bio-actuarial and health studies, matching cases with controls is a useful way to get expected differences in health outcomes, mortality, and care required. When matching on several variables one often exhausts the number of observations because of the number of combinatorial cells available. The RCBalance algorithm, developed by Samuel Pimentel, synthesizes this matching criterion by matching only on margins rather than matching person to person. My presentation reviews this algorithm and examines some of its properties. Of particular interest is the sensitivity of the resulting match to the input values of the parameters. We illustrate one parameter that is important in controlling the number of matches, which in turn appears to have an extremely large influence on matching outcome. The algorithm and the concerns presented here are illustrated with simulated examples.