Machine Learning Methods for Causal Inference from Complex Observational Data

• Activity Number: 576 - Matching Methods for Causal Inference with Emerging Data and Statistical Challenges
• Type: Topic Contributed
• Date/Time: Thursday, August 6, 2020 : 3:00 PM to 4:50 PM
• Sponsor: Biometrics Section
• Author(s): Alexander Volfovsky*
• Companies: Duke University
• Keywords: observational study; networks; causal inference; matching

Abstract

A classical problem in causal inference is that of matching treatment units to control units in an observational study. This problem is distinct from simple estimation of treatment effects as it provides additional practical interpretability of the underlying causal mechanisms. Many matching methods require expert input into the choice of distance metric that guides which covariates to match on and how to match on them. This task becomes impractical for huge modern datasets simply because humans are not naturally adept at constructing high dimensional functions by hand. We design a novel matching framework to tackle these problems: this framework proposes an optimization objective for match quality that captures covariates that are integral for making causal statements while encouraging as many matches as possible. We demonstrate this framework's ability to construct good matched groups on relevant covariates and leverage these high quality matches to estimate individualized effects. When data are entangled by a network we demonstrate that traditional approaches are likely to produce misleading results and so we propose a modeling solution to resolve these inconsistencies.