| Session Title: Testing the Robustness of Propensity Scores That Violate Balance Criteria |
| Expert Lecture Session 585 to be held in Panzacola Section H2 on Friday, Nov 13, 3:35 PM to 4:20 PM |
| Sponsored by the Quantitative Methods: Theory and Design TIG |
| Chair(s): |
| MH Clark, Southern Illinois University at Carbondale, mhclark@siu.edu |
| Presenter(s): |
| MH Clark, Southern Illinois University at Carbondale, mhclark@siu.edu |
| Abstract: The effectiveness of propensity scores is determined by how well they balance treatment and control groups, making them as similar as possible prior to an intervention. In 2001, Rubin established three criteria that should be met to conclude that propensity scores are balanced across covariates. These balancing criteria serve as statistical assumptions for the use of propensity scores when making adjustments to non-randomized experiments. Unfortunately, it is not unusual for one or more of these assumptions to be violated. Therefore, it is useful to know which statistical adjustment method (matching, subclassification, weighting or ANCOVA) is least affected by these violations. The present study used computer simulations to create various data sets that violate those statistical assumptions for having balanced propensity scores. A two-factor design examined the reduction of bias in treatment effects from quasi-experiments depending on the type of statistical adjustment and the type of statistical assumption that was violated. |