Data-adaptive Approaches to Causal Effect Estimation

Presented by Dr. Debashis Ghosh, Professor of Statistics at Penn State University

Monday, February 14th, 2013
Time: 10:30 AM – 12:00 PM Pacific / 1:30 PM – 3:00 PM Eastern
Host Location: Santa Monica, conference room 5312
Other Locations: Pittsburgh (room 6202) & Washington, DC (room 4302)

Abstract

In many studies, investigators wish to make causal statements about the effect of an intervention on some outcome measure. For a well-designed randomized study, we may assume that any covariates that may influence the outcome are distributed the same among different treatment groups. However, in an observational study, where the treatment assignment is not controlled by a randomization scheme, there usually exists a set of confounders that may influence both the outcome and the treatment assignment. For this case, any causal inference failing to account for the confounders will lead to biased estimates of the treatment effect. In this talk, we study the role of propensity score models for causal effect estimation. We find that this framework leads to a nonstandard model selection problem. We propose the use of data-adaptive modeling procedures that are shown to lead to improved causal effect estimation. Extensions to the mediation setting will also be described. We will illustrate the use of the methods using both simulated and observational data. This is a joint work with Yeying Zhu, Donna Coffman, Nandita Mitra and Bhramar Mukherjee.

About the Presenter

Debashis Ghosh is Professor of Statistics at Penn State University. He received his Ph.D. in biostatistics from the University of Washington in 2000. Prior to joining Penn State, Professor Ghosh spent six years as a professor in the Department of Biostatistics at the University of Michigan. Professor Ghosh has worked on a variety of application areas, ranging from cancer genomics to epidemiological studies in diabetes. His methodological interests center around high-dimensional data analysis and the application of machine learning methodologies. He was recently elected Fellow of the American Statistical Association and has over 140 publications in leading scientific and statistical journals.

To Attend

Visitors to RAND’s Santa Monica and Pittsburgh locations are welcome to attend & must RSVP at least one day prior to the seminar. To ensure attendance please, contact Carolyn Higgins with your name, company or affiliation & national citizenship (for security purposes).

Sponsored by the RAND Statistics Group