Abstract
The causal analysis of observational data plays a central role in essentially every scientific field. There is a large body of work developing identification and estimation strategies to estimate various causal estimands in complex settings. On the other hand, for even the most basic causal functional (the average treatment effect), we have a very rudimentary understanding of the fundamental limits of statistical estimation. We will review some (now) classical ideas, before presenting three vignettes which aim to probe the intrinsic difficulty of causal effect estimation under a spectrum of structural assumptions. The first vignette will consider the fundamental limits of estimating personalized causal effects. The second vignette will derive rates for estimating causal effects eschewing smoothness assumptions entirely. The final vignette will characterize the intrinsic difficulty of estimation in a discrete setting. No background on causal inference will be assumed.