IMS
Session Slot: 10:30-12:20 Wednesday
Estimated Audience Size:
AudioVisual Request: Two Overheads
Session Title: Wald Lecture III
Theme Session: No
Applied Session: No
Session Organizer: Lindsay, Bruce The Pennsylvania State University
Address: 422 Thomas Building, Department of Statistics, University Park, PA16802
Phone: (814)865-1220
Fax: (814) 863-7114
Email: bgl@psu.edu
Session Timing: 110 minutes total (Sorry about format):
Opening Remarks by Chair - 5 First Speaker - 90 minutesFloor Discussion - 10 minutes
Session Chair: Owen, Art Stanford University
Address: Department of Statistics Stanford University Stanford, CA 94305
Phone:
Fax:
Email: art@stat.Stanford.EDU
1. De Finetti's Theorem
Freedman, David, University of California at Berkeley
Address: Department of Statistics University of California, Berkeley, CA 94720-4735
Phone: 510-642-2781
Fax:
Email: freedman@stat.berkeley.edu
Abstract: Consider an infinite sequence of independent, identically distributed (IID) random variables; for simplicity, suppose each variable takes only finitely many values. Plainly, the variables are ``exchangeable,'' in the sense that permuting them does not affect their joint distribution. (Multiplication is, after all, commutative.) Consider next a Bayesian statistician whose predictive distribution--in advance of data collection--can be described as follows: (i) a marginal distribution is chosen at random, (ii) a sequence of independent random draws are made from this marginal. Such a predictive distribution is called a ``mixture of IID processes.'' And any such mixture must also be exchangeable. (The expected value of a constant is constant.) A celebrated theorem of de Finetti's asserts the converse: any exchangeable distribution is a mixture of IID processes. De Finetti's theorem has been generalized in many different directions. For example, the state space can be the real line (or any complete, separable metric space). Mixtures of specific parametric families of distributions can be characterized. So can mixtures of Markov chains. Bounds can be given for finite sequences of random variables. I will review some of the work on de Finetti's theorem.
List of speakers who are nonmembers: None