
240 Current Trends in Bayesian Methodology with Applications
This point has been noted and discussed well in [18]. It appears to us that
this is a problem involving more than one loss function, recently brought
to our no tice by Yaacov Ritov and Peter Bickel at the Joint Statistical
Meeting, 2013. It also s e e ms clear that an optimal objective prior would
be different for these two loss functions. Our personal v iew is that one
should decide what is the more impor tant loss function; in most cases it
will be model selection, but in principle it could be otherwise too. We favor
choosing the Robust prior for the more important model selection problem ...