Research projects

Project CEMAPRE internal

TitleBayesian model selection and point null hypotheses
ParticipantsPedro Miguel Fonseca, Rui Paulo (Principal Investigator)
SummaryBerger and Pericchi (1996, JASA) argued extensively that model selection should have a Bayesian
basis. It's practical implementation is however difficult, not only because of all the ingredients
needed to compute posterior model probabilities, but also for strictly computational reasons.
Automatic methods for model selection abound, but mainly in the the context of variable selection
and prediction in the Gaussian linear regression model.

This project is devoted to explore the problem of testing point null hypotheses from a default,
automatic Bayesian model selection perspective. This of particular importance because in this
context classical methods are known to over-reject the null in large samples. We are particularly
interested in point null hypotheses testing problems arising in the context of digit analysis, where
testing adherence to Benford's Law is central.