Medical School Office Building (MSOB)
|DATE:||January 11, 2018|
|TIME:||1:30 - 2:50 pm|
|TITLE:||Matching, Balancing, and Inverse-Propensity Weighting|
Assistant Professor of Operations, Information and Technology, Graduate School of Business and,
by courtesy, of Statistics, Stanford
Variants of matching and weighting are frequently used to adjust for confounders when estimating average treatment effects in observational studies. In this talk, I'll first briefly review pros and cons of state-of-the-art methods for average treatment effect estimation. I'll then show how some limitations of these methods can be addressed by using numerical optimization tools to directly bound minimax error, and discuss both formal (asymptotic) and empirical advantages of the resulting approach.
For background reading on average treatment effect estimation, see:
- Imbens, Guido W. "Nonparametric estimation of average treatment effects under exogeneity: A review." The Review of Economics and Statistics 86.1 (2004): 4-29.
The talk will draw from the preprint:
- Hirshberg, David A., and W. "Balancing Out Regression Error: Efficient Treatment Effect Estimation without Smooth Propensities." arXiv preprint arXiv:1712.00038 (2017).