Date/Time
Date(s) - February 10, 2020
11:45 am - 12:35 pm
Location
406 Weil Hall
Categories
Ruoxuan Xiong
Stanford University
Abstract: Panel Data Models with Staggered Treatment: From Observational Data to Experimental Design
In online marketplace, personalized medicine, policy-making, we seek to take the right action at the right target. However, actions and decisions have consequences. It is important to understand the causal effects of an action or a policy in order to make the right decisions. How can we study the causal effect? The gold standard is to run experiments. Nowadays, tech firms run thousands of experiments every year, that can involve millions of users. However, in some domains, such as healthcare and public policy, it can be impractical, unethical or costly to run experiments. An alternative approach is to draw inferences from observational data. In the first part of my talk, I will present a new approach that can answer a broad range of causal questions on panel data, which is based on the paper “Large Dimensional Latent Factor Modeling with Missing Observations and Applications to Causal Inference”. In the second part of my talk, I will present how we can design a multi-period experiment to efficiently learn causal effect, which is based on the paper “Optimal Experimental Design for Staggered Rollouts.”