A1856
Title: Direct sampling from conditional distributions of discrete exponential families
Authors: Shuhei Mano - The Institute of Statistical Mathematics (Japan) [presenting]
Abstract: Sampling contingency tables with fixed marginal totals is a classic problem to which MCMC has been applied. Seminal work showed that an irreducible Markov chain can be constructed by using computational algebra. It is demonstrated that discrete exponential families are associated with a system of hypergeometric partial differential equations and that the Groebner basis can guarantee that direct sampling is possible in principle. Moreover, sequential MLEs may be used if approximations are allowed, and the approach is exact if the model is rational.
