LANS Seminar

Seminar Title: Estimating Treatment Quantiles Without Assumptions

Speaker: Siddharth Bhandari, Research Assistant Professor, Toyota Technological Institute Chicago

Date/Time: June 19, 2024/ 10:30 AM-1:30 AM
Location: See Meeting URL on the cels-seminars website which will require an Argonne login.

Description: Estimating the Average Treatment Effect (ATE) is a well-studied problem in causal inference. However, it does not necessarily capture the heterogeneity in the outcomes of a Randomized Controlled Trial (A/B test). Several approaches have been proposed to address this issue, including estimating the Quantile Treatment Effects. In this talk, we will focus on the special case of the median. In a finite population setting containing n individuals, with treatment and control values denoted by the potential outcome vectors a and b, much of the prior work has focused on estimating median(a)−median(b), or making additional assumptions about the potential outcomes. However, none of the works address the estimation of median(a−b), called the Median Treatment Effect (MTE), without making assumptions. The fundamental problem of causal inference—where for every individual i, we can only observe one of the potential outcome values, i.e., either ai or bi, but not both—makes estimating the MTE particularly challenging.  In this talk, we will demonstrate that the MTE is not estimable. Instead, we will focus on a notion of quantile approximation of the MTE. We will prove lower bounds on the approximation factor and, by drawing connections to the notions of instance-optimality studied in theoretical computer science, we will present an instance-optimal linear-time algorithm for approximating the MTE without assumptions.  This work is to appear at the Conference on Learning Theory (COLT 2024) and is based on joint work with Raghavendra Addanki at Adobe Research (2403.10618.pdf)

Bio: Siddharth Bhandari is currently serving as a Research Assistant Professor at the Toyota Technological Institute in Chicago. Prior to this, he was a Simons-Berkeley Fellow at the Simons Institute for the Theory of Computing. Siddharth completed his Ph.D. at the School of Technology and Computer Science at the Tata Institute of Fundamental Research, Mumbai. His research interests lie in exploring interdisciplinary directions with a focus on developing techniques from computer science. He has worked and continues to work in the areas of coding/information theory, MCMC sampling algorithms, and causal inference.  Siddharth’s work has been recognized with several awards. His dissertation work was awarded the ACM India Dissertation Award. His paper “Improved Bounds for Perfect Sampling of k-Colorings in Graphs” won the Danny Lewin Best Student Paper Award at STOC 2020. He also won the Jack Keil Wolf Student Paper Award at ISIT 2018 for his work “Bounds on the Zero-Error List-Decoding Capacity of the q/(q−1) Channel.”