Reassessing the role of constituency in congressional voting

Poole and Rosenthal (1997) argue that most congressional voting can be understood in terms of a low-dimensional spatial model. This paper uses their model to assess the importance of the two mechanisms that could contribute to the vote-predicting power of constituency variables: (i) constituency variables may predict where legislators fall along one or two dimensions in the vote-predicting spatial model and (ii) constituency variables may account for errors in the spatial model's predictions. The paper compares different methods of using a basic set of constituency variables to generate out-of-sample predictions for representatives' votes. The analysis covers a large number of recent House roll call votes, considering Democrats and Republicans separately and using Poole and Rosenthal's W-NOMINATE scores to measure legislators' locations in vote-predicting space. The results show that the predictive power of a basic set of constituency variables arises principally from its ability to predict representatives' locations in Poole and Rosenthal's space, not from its ability to explain errors in the predictions based on that space. This holds true to a remarkable extent, consistent with Poole and Rosenthal's argument that the influence of constituent interests occurs largely through logrolling mechanisms reflected in their spatial model.