View Submission - CMStatistics

B0371
**Title: **Collaborative inference for causal effect estimation and general missing data problems
**Authors: **David Benkeser - Emory University (United States) **[presenting]**

**Abstract: **Doubly robust estimators are a popular means of estimating causal effects. Such estimators combine an estimate of the conditional mean of the study outcome given treatment and confounders (the so-called outcome regression) with an estimate of the conditional probability of treatment given confounders (the propensity score) to generate an estimate of the effect of interest. They have several desirable features: they are consistent so long as at least one of these two regressions is consistently estimated; they are also often efficient and achieve the lower bound on the variance of regular, asymptotically linear estimators. Moreover, they facilitate the use of modern machine learning methods in the estimation of the outcome regression and propensity score. However, in problems where causal parameters are weakly identified, doubly robust estimators may behave erratically. We propose a new framework for inference in these challenging settings. We introduce the idea of collaborative asymptotically linear estimators. These estimators use doubly robust frameworks; however, rather than using an estimate of the propensity score, they opt for an alternative quantity with reduced dimension relative to the true propensity score. We will discuss these issues in the context of estimating the causal effect of a binary treatment on an outcome, and discuss extensions to a general setting where the observed data are the result of a conditionally random coarsening of a full data structure.