Partisan Knowledge Claims in Congressional Oversight

ABSTRACT Congress negotiates while uncertain about the effects of proposed policies, and often relies on committees to conduct oversight that reports about the past and projects the future. Little is known about their conclusions, their accuracy, and how politics seeps into what they choose to report. We apply a typology of knowledge claims contained in congressional oversight, which classifies them based on whether they are prospective or retrospective, and whether they contain an inference. We use a large language model (LLM) to extract findings, evaluations, predictions, and recommendations from a new dataset of committee reports issued in the 103rd–118th Congresses. We argue that partisan investigation teams reveal more raw findings and make fewer recommendations, relative to bipartisan teams, which is consistent with their need to demonstrate they have acquired expertise. Our study pilots a general approach to studying the acquisition and sharing of information in legislative oversight.