Speaker 1: Yun-Ju Cheng

Title

A review and case study on matching methods: full matching

Abstract

Randomized experiment is the golden standard in estimating the effect of treatment in scientific studies. It ensures a minimum systematic difference between the treatment and control groups at baseline. However, randomized experiment may not be feasible for ethical or political issues, or from the economy aspect, and observational data are more handy. With observational data, there’s pre-existing difference between treatment and control groups, and it cause a problem when estimating the causal effect and making causal inference. To replicate a randomized experiment as closely as possible from the observational study, that is, make treatment and control groups have similar covariate distribution, matching method is in need to achieve this goal. The matching methods have been developed since 1970s. The presentation will be focus on one of the matching methods - the full matching.

Speaker 2: Wenxin Lu

Title

Doubly Robust Estimator for a Population Mean

Abstract

Inverse probability of treatment weighting is widely used among the weighting method, which is a good way to obtain causal effect. Based on the inverse probability weighting, the augmented approach guarantees the double robustness property, because it contains the propensity score model and outcome regression model. The estimator remains consistent if either the propensity score model or the outcome regression model is incorrect. Considering efficiency, usual double robust estimator does not have the smallest variance if the outcome regression model is wrong. For alternative double robust estimator, it has the smallest variance even when the outcome regression model is wrong. This talk aims to review the process on constructing doubly robust estimator.