In a previous post, we introduced Eric Posner and Adrian Vermeule’s paper “The Votes of Other Judges,” which was published in volume 105 of the Georgetown Law Journal. To sum up, Posner and Vermeule present a theory of interdependent judicial voting, i.e. judges on multi-member panels should engage in Bayesian updating when they decide issues of law or questions of interpretation. Specifically, a judge sitting on a multi-member panel should do two things: (i) he should first take into account the way in which his fellow judges have voted, and (ii) he should be willing to change his vote accordingly. Although we agree with the spirit of Posner and Vermeule’s Bayesian approach to judging, their paper is short on specifics. After thirty dense, single-spaced pages devoted to peer disagreement and interdependent voting, their paper concludes with a timid whimper. Talk about an anti-climax! In particular, Posner and Vermeule propose a simple two-stage method of voting–“in the first stage, each judge votes; in the second stage, the judges may change their votes in light of what they learned from the first stage” (p. 189)–but they don’t formalize this two-step method any further. As a result, Posner and Vermeule’s approach to judging is too crude to be of any practical use. They offer a general exhortation–judges should be willing to change their votes (in light of the way their fellow judges have voted)–but they don’t specify the precise conditions under which a judge should actually change his vote.
Nevertheless, Posner and Vermeule are right that judges’ votes contain information (independent of whatever reasons judges may give to justify their votes) and that judges should always update their priors before casting their final and decisive votes, especially in close cases, but is there any way of operationalizing Posner and Vermeule’s theory of interdependent voting? In a word: yes! Let judges engage in Bayesian voting, i.e. let judges express their degrees of belief when voting. (As an aside, this is the main reason why Posner and Vermeule’s exhortation is too crude. In their model, judges are still emitting binary votes, i.e. judges must vote all or nothing: either “for” or “against” the moving party.) Ironically, Posner and Vermeule discuss the importance of degrees of belief (or “confidence levels”) in their paper, but their approach to judicial voting makes no use of confidence levels. In our next post, we will propose an alternative method of Bayesian voting by judges, one in which judges are allowed to disclose their degrees of belief instead of voting up or down.