How to Make Causal Inferences Using Texts

• In Event: 4377 - Computational Sociology

Abstract

New techniques in the text as data literature offer a great promise: the ability to inductively discover measures of interest from large collections of text. We introduce a new conceptual framework to understand how to use these discovered measures as a treatment or outcome in a causal inference research design. This allows researchers to inductively discover effective high-dimensional interventions and capture the ways that observed treatments affect text-based outcomes. We demonstrate that nearly all text-based causal inferences depend upon a latent representation of the text and we provide a framework to learn the latent representation. Estimating this latent representation, however, creates new risks: we may unintentionally introduce an identification problem or produce overfitting. To address these risks, we introduce a split sample framework and apply it to estimate causal effects from an experiment on immigration attitudes and a study on bureaucratic responsiveness. Our work provides a rigorous foundation for text-based causal inferences, connecting two previous disparate literatures.

Authors

• Naoki Egami, Princeton

• Christian Fong, Stanford University

• Justin Grimmer, University of Chicago

• Margaret Roberts, University of California-San Diego

• Brandon Michael Stewart, Princeton University