ABSTRACT
Over the past two decades, political scientists have made great advances in the empirical estimation of ideal points (Bafumi et al., 2005; Bailey and Chang, 2001; Clinton et al., 2000; HeckmanandSnyder, 1997; Jackman, 2001; Londregan, 2000;Martin andQuinn, 2002; Poole and Rosenthal, 1997). An ideal point, or preference, is a foundational theoretical concept for explaining the choices a political actor makes. For example, in simple unidimensional spatialmodels of voting, a legislator’s vote choice ismodeled as a rational decision based on a (Euclidean geometric) calculation of differences in utility values between the legislator’s ideal point, a proposed bill, and the status quo. Although an ideal point is often assumed to be static for theoretical convenience, dynam-
ics in ideal points pose an important theoretical and empirical puzzle to researchers. For example, examining the judicial opinion writing of 16 US Supreme Court justices, Epstein et al. (1998) conclude that there is enough evidence to invalidate the assumption of preference stability over time.∗ They also go on to claim that any inference about a justice’s “revealed preference” that is based on the stable preference assumption can be misleading if the justice actually underwent several preference changes over a lifetime. However, the development of statistical methods for dynamic ideal point estimation has been limited to a few published works (Martin and Quinn, 2002; McCarty et al., 1997). Also, the existing methods for dynamic ideal point estimation fail to distinguish fundamental changes from randomdrifts. In this paper I propose amethod to detect sharp, discontinuous changes in ideal points. The approach I take in this paper is to combine Chib’s (1998) hidden Markov model
(HMM) with the two-parameter item response theory (IRT) model. In this model, the dynamics in ideal points are modeled as agent-specific hidden regime changes. I demonstrate the utility of the hidden Markov IRT model by analyzing changes in ideal points among the 43 US Supreme Court justices serving between 1937 and 2006, and conclude that the model provides an effective benchmark for making probabilistic inferences about the timing of preference changes.
