Towards Principled Representation Learning to Improve Overlap in Treatment Effect Estimation
Keywords: representation, overlap, treatment effect
TL;DR: We propose quantifying how much representations used as adjustment sets improve overlap using the chi-2 divergence and show that improvement in overlap as such is completely determined by prediction of treatment assignment
Abstract: A common approach to mitigate undesirable effects of poor overlap is to use well-crafted representations of covariates as adjustment sets. In this abstract, we motivate quantifying the overlap induced by a representation using the $\chi^2$-divergence, show that the overlap improvement under this metric is precisely how much the representation does not predict the propensity score, which confirms intuitions in previous work, and discuss next steps.
Submission Number: 30
Loading