Optimal caliper widths for propensity‐score matching when estimating differences in means and differences in proportions in observational studies

• DOI: 10.1002/pst.433
• Publication Date: 2010-04-28T16:43:43Z
• Authors: Peter C. Austin
• Keywords: Calipers

In a study comparing the effects of two treatments, propensity score is probability assignment to one treatment conditional on subject's measured baseline covariates. Propensity-score matching increasingly being used estimate exposures using observational data. most common implementation propensity-score matching, pairs treated and untreated subjects are formed whose scores differ by at pre-specified amount (the caliper width). There has been little research into optimal width. We conducted an extensive series Monte Carlo simulations determine width for estimating differences in means (for continuous outcomes) risk binary outcomes). When or differences, we recommend that researchers match logit calipers equal 0.2 standard deviation score. least some covariates were continuous, then either this value, close it, minimized mean square error resultant estimated effect. It also eliminated 98% bias crude estimator, it resulted confidence intervals with approximately correct coverage rates. Furthermore, empirical type I rate was correct. all binary, choice had much smaller impact performance estimation means.