Tuesday 5 March 2013

Bayes2013

Last year Julien invited me to the Bayes-Pharma workshop in Aachen, which turned out to be quite nice. 

This year, I've been invited in the scientific board for the new edition, which will be held in Rotterdam this coming May. So far, we've had a very good response, with many people sending in their abstracts (but I think there is still time to do so!). We have, I think, a very interesting programme lined up. 

I'll give a short introduction to INLA but, perhaps even more than that, I'm quite looking forward the session on Bayesian methods for missing data on the second day. Nicky Best and Alexina Mason (both from Imperial College) will present their work, which is based on a fully Bayesian approach.

This contrasts with more "standard" methods, like multiple imputation (MI, eg as in this), which are Bayesian in nature, but which people normally apply and evaluate in terms of their frequentist properties. While I think that the general framework upon which MI is built is quite neat, I have always found it a bit confusing that it starts as a Bayesian procedure, only to them move towards something not Bayesian at all. 

I don't really know the history of this all too well, but I guess this may be an indirect effect of the fact that when MI methods were originally developed Bayesian computations were still very difficult to make. 

I'll post more on this as the actual meeting approaches.

No comments:

Post a Comment