Abstract:
|
In observational studies, matched samples are created so that a treated group is similar to a matched control group on observed covariates. Often matched samples consist of matched pairs. If a pair match fails to make treated and control units sufficiently comparable, alternate strategies include (1) matching a variable number of controls to each treated unit (2) adopting fine balance constraints. Under fine balance, a nominal covariate is exactly balanced, but individual treated and control units may not be comparable on this variable. We propose a method that allows fine balance constraints while matching treated units to variable numbers of controls, which is not possible using existing network-based matching algorithms. We use the entire number to determine the optimal number of controls for each treated unit. Within entire-number strata, we then apply fine balance constraints. We apply our method in an evaluation of Peer Health Exchange, an intervention in high schools designed to decrease risky health behaviors. We find that pair matching produces unsatisfactory balance, then demonstrate that a variable-ratio match with fine balance outperforms a variable-ratio match alone.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.