[Date Prev][Date Next]
[Chronological]
[Thread]
[Top]
abstracts
CON PREGHIERA DI DIFFUSIONE
********************************************************************************
AVVISO SEMINARIO
Si comunica che il 21 Maggio, ore 12, il professor James Berger, Department
of Statistics, Purdue University, USA, terra', presso il Dipartimento di
Economia, Universita' di Roma III, ( Via Ostiense, 139 ) aula 7, il seguente
seminario:
DEFAULT BAYESIAN ANALYSIS OF MIXTURE MODELS
One of the difficulties in Bayesian analysis of mixture models is that
sophisticated noninformative priors (such as reference or Jeffreys priors)
are very difficult to determine, while simple choices (such as the constant
density) will typically lead to improper posterior distributions. It also
is not effective to simply choose proper priors that are diffuse, as the
answers can depend markedly on the degree of diffuseness.
A natural line of attack upon this problem is through the "intrinsic
Bayes factor" or "fractional Bayes factor" methodologies. These methodologies
were designed for scenarios in which the same difficulty - that standard
noninformative priors cannot be used - is encountered. The application of
these approaches to the mixture model problem will be discussed, with
computational issues and examples being discussed.
__________________________________________________________
Per informazioni rivolgersi a: Julia Mortera tel. (06)5737-4206
(mortera@uniroma3.it)