Estimating Party Influence on Roll Call Voting: Regression Coefficients versus Classification Success

Abstract

Contrary to the claims of McCarty, Poole, and Rosenthal, our method does not estimate the ideal points of moderates significantly less accurately than the ideal points of extremists. This is true for at least two reasons: (1) there is significant randomness in voting; as a consequence, on a lopsided vote moderates often vote with the extremists; and (2) our data set includes some roll calls that require a supermajority for passage; for these we define a 50%–50% roll call as lopsided. We also show that the classification-success method of McCarty, Poole, and Rosenthal severely understates the presence of party influence. Furthermore, we show that a proper interpretation of some of their results reveals a significant amount of party influence in Congress.