Abstract

Randomized controlled trials are the gold standard for estimating the causal effects of new interventions. In a trial, we want to randomly assign experimental units into two groups so that certain unit-specific pre-treatment variables, called covariates, are balanced across different groups. We formulate this task as a discrepancy minimization question. By exploiting recent advances in algorithmic discrepancy theory, we obtain random assignments with a nearly optimal tradeoff between the gain we have if covariates are predictive of treatment outcomes and the loss we suffer if covariates are not. This is based on joint work with Harshaw, Sävje, and Spielman (https://arxiv.org/pdf/1911.03071.pdf).

Time

2023-06-17  17:00 - 17:30   

Speaker

Peng Zhang, Rutgers University

Room

Guangdong Hotel