Judicial Drift? 

 In light of Jim's post below, it is worth pointing out an ongoing conversation at the  Northwestern Law Review  on ideological change on the Supreme Court.  The discussion was prompted by a forthcoming article entitled " Ideological Drift among Supreme Court Justices:  Who, When, and How Important? ", authored by a who's who of empirical court scholars: Lee Epstein, Andrew Martin, Jeffrey Segal, and our own Kevin Quinn.  In addition to their comments on the article, there is a response by Linda Greenhouse, who covers the Supreme Court for the  New York Times .  (It also got a plug in the  Washington Post  this morning). 

 I'm more sympathetic to the project of modelling judicial decisions than I take Jim to be; I think that the ideal point framework gives us a useful way of thinking about the preferences of political actors, including judges.  On the other hand, his points about precedent and interference across units are well-taken.  Consider the following graph, which appears in the Epstein et al. paper: 

   

 It is explained as the estimated probability of a "liberal" vote by Justice O'Connor on two of the key social policy cases decided by the court in the past few years:  Lawrence  (which struck down Texas' anti-sodomy law) and  Grutter  (upholding the University of Michigan's law school admissions policy; the undergraduate policy was struck down in  Gratz v. Bollinger ).  I assume that these probabilities were calculated using the posterior distribution of the case parameters in Lawrence and Grutter and combining them with the posterior distribution for O'Connor's ideal points in each year.  Fair enough, but what does this actually mean?  If  Grutter  had come before the court in 1985, it would not have been  Grutter . I don't say this to be flippant; the University of Michigan used different admissions policies in the 1980s (in fact, when I went to Michigan as an undergrad, I was admitted under a different policy than the procedure struck down in  Gratz );  Adarand ,  Hopwood , and related cases would not have been on the books,  etc.  I just don't see how the implied counterfactual ("What is the probability that O'Conner would cast a liberal vote if Grutter had been decided in year X") makes any sense.