Yi Ding: Introduction to Causal Inference and Minimax Crossover Designs

Monday, November 11

12:30-1:00pm (note shorter time)

JCL 390

Introduction to Causal Inference and Minimax Crossover Designs

Food today is from Modern Chinese Cook Book.

Despite normally occurring every other week, this week we have another Student Seminar talk, and next week another! Come and eat some free food!

This week Yi Ding is giving a practice talk about statistical experimental design,

Introduction to Causal Inference and Minimax Crossover Designs

Causal inference, or the problem of causality in general, is one of the most important topics in statistics, while not many people are aware of it, especially those in computer science. Causal inference has received a lot of attention in recent years. The question is simple, is correlation enough for inference? You may not be sure about the answer, but you may have heard this: correlation does not imply causation.

One related topic to causal inference is design of experiments. In this talk, I will discuss crossover experiments, where two broad classes of treatment effects are typically considered: direct effects that capture the instantaneous impact of the treatment, and carryover effects that capture the lagged impact of past treatments. Existing approaches to optimal crossover design usually minimize a criterion that relies on an outcome model, and on assumptions that carryover effects are limited, say, to one or two time periods. The latter assumption is problematic when long-range carryover effects are expected, and are of primary interest. We derive minimax optimal designs for estimating both direct and carryover effects simultaneously. In contrast to prior work, our minimax designs do not require specifying a model for the outcomes, relying instead on invariance assumptions. This allows us to address problems with arbitrary carryover structure, such as those encountered in digital experimentation.