Estimating Treatment Effects from Text using Deconfounding Factors

• In Event: Advances in Methods for Causal Inference

Brief Overview

We apply the deconfounder (Wang-Blei 2018) to texts to recover causal effects of features on outcomes. Applications are effects of judicial opinion text on reversal rates, and effects of politician speech text on re-election rates.

Authors

• Elena Labzina, ETH_Zürich

• Elliott Ash, ETH Zurich