Propensity Score Matching:Moving Towards Causal Inference with Observational Data

INSTRUCTOR:	Jennifer L Hill
DATES:	Tuesday, March 21 and continuing on Wednesday, March 22, 2006
TIME:	1pm-4pm
LOCATION:	Please note the course takes place in two different locations. March 22, 2006 Irving Cancer Research Center, Room 115 1130 St. Nicholas Ave West 167th and St. Nicholas (one very short block east of Broadway) March 21, 2006 from 1-5 pm Hammer Health Sciences Room 408 701 W. 168th Street at the corner of Fort Washington (one block west of Broadway)

Topic
The vast majority of public policy research focuses on causal questions. Researchers want to know what the impact would be on participants if we changed an existing policy or implemented a new policy. At the same time, most researchers are forced to rely on observational data that is not well suited to answering causal questions. If we want to know the impact of participating in Program A versus Program B and have collected data on people in both situations, simple comparisons of the people across programs generally will not lead to estimates of the causal effect of Program A relative to Program B because the people in each programs may differ from each other in other important ways. In this situation the study is said to suffer from selection bias.

Many statistical methods have been proposed to correct for selection bias ranging from simple OLS corrections to complicated econometric modeling. This course will focus on a set of techniques broadly referred to as "propensity score matching" that have the advantage of being intuitively straightforward and relatively easy to implement. While propensity score matching does assume that the researcher has measured all important covariates, it does not rely on many of the strong parametric and structural assumptions necessary for the validity of some competing techniques.

The short course will combine theory regarding propensity score matching and causal inference with practical instruction for implementation of the technique both in general and specifically within the context of a computer package available in Stata. Real-data examples will be used to illustrate how models are fit, chosen, and interpreted. A "homework assignment" will provide participants with practice using the methods and software; time will be devoted during the second lecture to trouble-shooting problems and questions that arose during implementation. The second session will also provide an opportunity to explore extensions to the basic techniques and to discuss some slightly more technical issues regarding variance estimation.

Audience
This short course is targeted at researchers from the social and behavioral sciences and medicine who investigate questions that are causal in nature. The course will assume the participant has the mathematical and statistical sophistication typical of graduates of Ph.D. programs in psychology, political science and sociology.

Instructor
Jennifer Hill is an Assistant Professor of Public Affairs at Columbia University's School of International and Public Affairs. She earned her PhD in Statistics at Harvard University and completed a post-doctoral fellowship in child and family policy at Columbia University's School of Social Work. Professor Hill's methodological research focuses on causal inference and missing data, statistical problems that plague public policy research. Her applied public policy research, which takes advantage of the methods she researches, currently focuses on child and family policy (e.g. the effect of maternal employment on child development, the effect of high quality child care) and education policy (the effect of holding children back in school) though she has worked on a wide range of social science problems. Her work has appeared in both Statistics journals (e.g. Journal of the American Statistical Association, Journal of Educational and Behavioral Statistics) and social science, public health, and public policy journals (e.g. American Political Science Review, Developmental Psychology, Journal of Policy Analysis and Management, American Journal of Public Health).

To register
This short course is open free of charge to faculty, postdoctoral fellows, and graduate students at Columbia University as well as faculty and postdoctoral fellows at other sites of the Robert Wood Johnson Health & Society Scholars (H&SS) Program. Enrollment is limited to 20; H&SS affiliates will have priority.

To register, please send an email to: chssp@columbia.edu. Please include your mailing address, as readings will be sent to course participants. Also, include a few sentences

The Health & Society Scholars Program at Columbia University is a postdoctoral program funded by the Robert Wood Johnson Foundation. It is a joint initiative of the Mailman School of Public Health and the Institute for Social and Economic Research and Policy (ISERP) at Columbia, and is co-directed by Bruce Link and Peter Bearman. For more information call 212-854-3694 or email chssp@columbia.edu.