Abstract

Although many scholars take parsimony for granted today, Elliott Sober shows in his latest book, Ockham’s Razors, that they might not be rationally justified to do so. In particular, he claims that the famous Ockham’s Razor, the heuristic that says one should not postulate more entities than necessary, rests on some implicit assumptions that go back to Newton and his rules of reasoning. The problem is that Newton justified those basic rules on theological grounds, that is, the world is parsimonious because God is orderly. All is not lost : Sober suggests that two contemporary perspectives from probability theory do justify parsimony. The first one is related to Bayesianism, and the fact that Ockham’s Razor is embedded in Bayes’ theorem. Sober criticizes this view and argues for an alternative, one in which predictive accuracy is more fundamental. I suggest that Sober might be right about the unseen role of predictive accuracy, but that this does not entail that Bayesians should adhere to Sober’s framework. It is my contention that Sober’s case against Bayesian model selection has more to do with the Bayesian worldview than the methodology per se.