If an event of interest is correlated with text data, we can learn models of text that predict the event outcome. For example, researchers have predicted financial risk with regression models that use the text of company financial disclosures.2 Topic models can predict outcomes as a function of the proportions of a document that are devoted to the automatically discovered topics,3 and this technique has been used to develop, for example, a topic model that forecasts roll call votes using the text of congressional bills.4 An advantage of the topic model prediction approach is that the model learns interpretable topics and the relationships between the learned topics and outcomes. A disadvantage of the topic model approach is that other, less interpretable text models often exhibit higher predictive power.