IMPERIAL COLLEGE STATISTIC SECTION SEMINAR
This weeks seminar is given by Chris Holmes (Imperial College).
2pm -- Friday 12th November
Issues in Bayesian prediction
We consider predictive inference for the class of Bayes linear models.
Prediction often involves taking account of model uncertianty
in the set of predictors (or covariates) that are used in the regression.
A common approach in recent years has been to introduce random Bernoulli
indicator variables, one for each covariate, that indicate whether the
predictor is "in" the model. We highlight some problems with this method
and offer a possible solution using hierarchical covariance priors for the
regression coefficients that induce variable selection as
a special case.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|