When: 11 november, 2020, kl. 13-14

Where: This seminar is given online. E-mail Dan Hedlin if you want to attend.

Abstract

Double blind randomized controlled trials are traditionally seen as the gold standard for causal inferences. The difference-in-means estimator is an unbiased estimator of the average treatment effect in the experiment. The fact that this estimator is unbiased over all possible randomizations does not mean that any given estimate is close to the treatment effect. One strategy to mitigate this problem is to adjust for a fixed set of covariates using regression. This paper studies the theoretical properties of both the difference-in-means and OLS estimators conditional on observed differences in covariates. By deriving the conditional distributions, we can establish guidance for how to deal with observed imbalances without falling prey to 'p-hacking’. We study both inference with OLS, as well as with a new version of Fisher's exact test, where the randomization distribution comes from a small subset of all possible assignment vectors.