ISBA 2012 [#1]

The empirical likelihood session was the first one I attended in the morning. As I had slept fairly little the past night, I had alas trouble (more than usual!) to stay awake during the talks! They covered the application of a mix of empirical likelihood and Bayesian tools to missing data and survey data. My overall impression is however that there was not much and not enough discussion about the validation of the approach, i.e.~in connection with my lecture of the previous day, whether or not it was Bayesian, and whether or not it leads to a coherent type of inference, albeit asymptotically.

The following session was ABC #1 that I organised, with Scott Sisson chairing. As a biased organiser, I though it went on well and presented some current viewpoints on model choice and summary statistic selection, all of which have been discussed on this ‘Og! In particular, Dennis Prangle exposed his extension of the Read Paper to the model choice issue, raising interesting questions about the notion of sufficiency in this setting. Similarly, Chris Drovandi discussed the choice of pseudo-model in indirect inference, using a model fit as the selection tool, which does not seem an obvious solution to me as what matters is rather the different behaviour of the corresponding estimator in a collection of models…

After a very pleasant lunch with Ed George in what sounded like a very local and secluded restaurant, where we prepared tomorrow’s memorial session by mostly exchanging stories and memories about George Casella, I went to the adaptive Monte Carlo session, where exhaustion got the better of my genuine interest in the topic (despite Pierre prodding me awake from time to time!)… The more relevant my call for contributions to those impressions of ISBA 2012 from all volunters!