Reassessing the Role of Constituency in Congressional Voting

Abstract

Poole and Rosenthal (1997) argue that mostcongressional voting can be understood in terms of alow-dimensional spatial model. This paper uses their model toassess the importance of the two mechanisms that couldcontribute to the vote-predicting power of constituencyvariables: (i) constituency variables may predict wherelegislators fall along one or two dimensions in thevote-predicting spatial model and (ii) constituency variablesmay account for errors in the spatial model's predictions. Thepaper compares different methods of using a basicset of constituency variables to generate out-of-sample predictionsfor representatives' votes. The analysis covers a large numberof recent House roll call votes, considering Democrats andRepublicans separately and using Poole and Rosenthal'sW-NOMINATE scores to measure legislators' locations invote-predicting space. The results show that the predictivepower of a basic set of constituency variables arisesprincipally from its ability to predict representatives'locations in Poole and Rosenthal's space, not from its abilityto explain errors in the predictions based on that space. Thisholds true to a remarkable extent, consistent with Poole andRosenthal's argument that the influence of constituentinterests occurs largely through logrolling mechanismsreflected in their spatial model.