Causal Inference with Free-Text in Both Randomized and Observational Settings

• 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): Michael Baiocchi* and Jordan Rodu and Joshua Kravitz
• Companies: Stanford University and University of Virginia and Stanford University
• Keywords: causal inference; free-text; measurement; behavior; matching

Abstract

Free-text is a natural way for social scientists to gain insight into human thought and behavior. We develop a framework for estimating the behavioral effect of an intervention as measured by free text. In contrast to existing (model-based) methods our framework for inference with free-text relies solely on the randomized controlled trial (RCT) framework, in which the researcher-directed randomization is all that is required for causal inference. Our goal is to be able to detect differences in how people are articulating their beliefs, relevant to the intervention. There is currently no known statistically valid means for analyzing free-text for detecting, and estimating, the effect of an intervention. Once establishing this framework in the RCT framework, we extend it to cover matching with free-text as the outcome of interest.