Textual Predictors of Bill Survival in Congressional Committees

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 793–802, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics Textual Predictors of Bill Survival in Congressional Committees Tae Yano Noah A. Smith Language Technologies Institute Carnegie Mellon University Pittsburgh, PA 15213, USA {taey,nasmith}@cs.cmu.edu John D. Wilkerson Department of Political Science University of Washington Seattle, WA 98195, USA jwilker@u.washington.edu Abstract A U.S. Congressional bill is a textual artifact that must pass through a series of hurdles to become a law. In this paper, we focus on one of the most precarious and least understood stages in a bill’s life: its consideration, behind closed doors, by a Congressional committee. We construct predictive models of whether a bill will survive committee, starting with a strong, novel baseline that uses features of the bill’s sponsor and the committee it is referred to. We augment the model with information from the contents of bills, comparing different hypotheses about how a committee decides a bill’s fate. These models give significant re- ductions in prediction error and highlight the importance of bill substance in explanations of policy-making and agenda-setting. 1 Introduction In representative governments, laws result from a complex social process. Central to that process is language. Text data emerging from the process in- clude debates among legislators (Laver et al., 2003; Quinn et al., 2010; Beigman Klebanov et al., 2008), press releases (Grimmer, 2010), accounts of these debates in the press, policy proposals, and laws. In the work reported here, we seek to exploit text data—specifically, the text of Congressional bills— to understand the lawmaking process. We consider an especially murky part of that process that is dif- ficult to study because it happens largely behind closed doors: the handling of bills by Congressional committees. This early stage of a bill’s life is precar- ious: roughly 85% of bills do not survive commit- tee. By contrast, nearly 90% of bills that are recom- mended by a committee (i.e., survive the committee and are introduced for debate on the floor) will sur- vive a roll call vote by the legislature. Because fil- tering by these powerful Congressional committees is both more opaque and more selective than the ac- tions of the legislature as a whole, we believe that text-based models can play a central role in under- standing this stage of lawmaking. This paper’s contributions are: (i) We formu- late computationally the prediction of which bills will a survive Congressional committee, presenting a (baseline) model based on observable features as- sociated with a bill, the committee(s) it is assigned to, members of that committee, the Congress as a whole, and expert combinations of those features. The task formulation and baseline model are novel. (ii) We propose several extensions of that strong baseline with information derived from the text of a bill. (iii) We validate our models on a hard predic- tive task: predicting which bills will survive com- mittee. Text is shown to be highly beneficial. (iv) We …

Content truncated to 500 words