I present my own perspective on the philosophy of Bayesian statistics, based on my experiences doing applied statistics in the social sciences and elsewhere. My motivation for this project is dissatisfaction with what I perceive as the standard view of the philosophical foundations of Bayesian statistics, a view in which Bayesian inference is inductive and scientific learning proceeds via the computation of the posterior probability of hypotheses. In contrast, I view Bayesian inference as deductive and as part of a larger Bayesian data-analytic process, different parts of which I believe can be usefully understood in light of the philosophical frameworks of Popper, Kuhn, and Lakatos. The practical implication of my philosophy is to push Bayesian data analysis toward a continual creative-destruction process of model building, inference, and model-checking rather than to aim for an overarching framework of scientific learning via posterior probabilities of hypotheses. This work is joint with Cosma Shalizi.
Paris Diderot Philmath Seminar
Lundi 15 février 2010 à 14h-15h30
Université Paris Diderot – Site Rive Gauche, Bâtiment Condorcet
4 rue Elsa Morante
75205 PARIS CEDEX 13
Salle Klee (454A)