RAND Statistics Seminar Series
Propensity Score Matching to Recover Latent Experiments
Presented by Ben B. Hansen, Ph.D., Assistant Professor, University of Michigan
Wednesday, December 9, 2009
1:30 p.m. – 3:00 p.m. ET
Conference Room 6202, RAND Corporation, Pittsburgh, PA
Washington, D.C., Conf. Rm. 4132 1:30 p.m. ET
Santa Monica, CA, Conf. Rm. 3312 10:30 a.m. PT
Please contact Denise Miller if you would like to attend this seminar.
Propensity score matching aims to gain for an observational study various benefits characteristic of experiments. It tries to rid treatment versus control comparisons of selection bias, as random assignment does and as competing methods of adjustment for nonrandomized data also try to do; but in contrast with alternate adjustment methods, it aims explicitly to mimic experiments in others of their workings as well, parroting habits of randomization mechanisms that may be less essential for causal inference but can more easily be observed. When it is shown to have met these observable goals, one gains confidence that it also has succeeded at removing bias from estimates of intervention effects. Be that as it may, existing theory offers only vague links from propensity scores' diagnostics to the soundness of causal inferences made with them. The theory seems to require exact matching on true propensity scores, even in the absence of hidden bias, whereas in practice the best one can do is to match approximately, on estimated scores.
I propose a novel large sample theory to address this gap in the literature. Rather than relying on a specific matching technique, the new theory puts the more nearly verifiable of propensity matching’s aims in a central role, clarifying their contributions to the integrity of inferences about treatment effects and illuminating certain methodological debates. It reveals the sense and extent to which propensity matched inferences can, given non-confounding assumptions, be likened to randomization-based inferences in randomized experiments.
Ben B. Hansen is an Assistant Professor of Statistics and Faculty Associate of the Survey Research Center, Institute for Social Research, at the University of Michigan. His research interests center on causal inference for comparative studies, including randomization-based inference, optimal matching, propensity scores and related techniques; applications of these methods to education, political science and the social sciences broadly; and optimal expected-length confidence interval estimation. He is a developer of several add-on libraries for the statistical package R, an associate editor of the Journal of the American Statistical Association and an editorial consultant of the Sage Quantitative Applications in the Social Sciences series of monographs.
Attending a Seminar
RAND visitors are welcome to attend and must RSVP at least one day prior to the seminar. To ensure your attendance please contact Denise Miller at email@example.com with your name, company (or university) affiliation, and national citizenship (for security purposes).
For parking and directions to RAND's Santa Monica office, please see: http://www.rand.org/about/locations/santa-monica.html.
For parking and directions to RAND's Pittsburgh office, please see: http://www.rand.org/about/locations/pittsburgh.html.
For further information and to be added to the mailing list contact Denise Miller at firstname.lastname@example.org.