Power in text : extracting institutional relationships from natural language
MetadataShow full item record
How do legislators allocate policy-making authority? At least in the legal context, distribution-of-power arrangements are usually articulated in written documents. Unfortunately, extracting these relationships is difficult, leading scholars to restrict themselves to studies of single policy areas or to a small set of high-visibility laws. In this project, I address this limitation through a neural network-based approach that extracts power relationships from legal language in a scalable, valid fashion. I then apply this approach to study institutional design in enacted US legislation. Substantively, I demonstrate that policy preferences of executive and legislative actors exert surprisingly little influence on formal institutional design choices. For all but the most politically salient laws, implementation arrangements are structured by the policy area and issue under consideration rather than elite political preferences. This argument - which would not have been possible to test without the measurement tools I develop - highlights both the importance of the tools I develop and the need for scalable measurement techniques in political science.