In this talk, I cover two related pieces of work.  In “Adjusting for Confounding with Text Matching” (joint with Roberts and Nielsen), we identify situations in which conditioning on text can address confounding in observational studies. Our proposed solution is to estimate a low-dimensional summary of the text and condition on this summary via matching. We propose a method of text matching, topical inverse regression matching, that allows the analyst to match both on the topical content of confounding documents and the probability that each of these documents is treated. Using an application studying censorship in China, we demonstrate the promise of the approach.  However, any approach that aims to address latent confounding is going to come with strong assumptions.  In a second paper, “Naïve regression requires weaker assumptions than factor models to adjust for multiple cause confounding” (joint with Grimmer and Knox), I explore some of the limits of using factor model based approaches to address latent confounding by re-examining a recently proposed approach for the multiple cause confounding setting. I conclude with implications for future work.

Short Bio:

Brandon Stewart is an Assistant Professor of Sociology at Princeton University and is also affiliated with the Politics Department, the Office of Population Research, the Princeton Institute for Computational Science and Engineering, The Center for Information Technology Policy and the Center for the Digital Humanities. He develops new quantitative statistical methods for applications across computational social science, including the Structural Topic Model, one of the most popular approaches to topic modeling in the social sciences.  His recent work focuses on text as data, causal inference, and the intersection of the two.  He completed his master’s degree in Statistics (2014) and PhD in Government (2015) at Harvard. His new book (with Justin Grimmer and Molly Roberts), Text as Data: A New Framework for Machine Learning and the Social Sciences comes out from Princeton University Press at the end of March.