The preceding chapter showed that press releases are a useful source for measuring how senators present their work to constituents. But the large collection of texts introduces a new problem: analyzing what senators say in the thousands of press releases issued each year. Applying the standard content-analysis technique – hand coding – is foreboding. Indeed, this cost is substantial enough that it leads most previous studies of congressional communication to focus on a subset of legislators or a set of case-study policy debates. But limiting the data to make analysis tractable also limits the questions the study can address – an unattractive option.

Rather than use hand coding, in this chapter I introduce a statistical model to measure senators' expressed priorities. The statistical model that I introduce simultaneously estimates the topics that senators discuss and the attention allocated to those topics. Along the way, the model also classifies each press release into a topic and classifies senators' expressed priorities into a broader type of presentational style. Using an array of validations, I show that the model is able to provide comprehensive, systematic, and verifiable measures of how senators present their work to constituents.

A skeptical reader is likely to wonder how useful statistical models are for analyzing the content of texts. After all, language is complex, subtle, and often difficult for humans to decipher – let alone a machine. But a growing literature suggests that automated and statistical models for texts are extremely useful.