Priors for Bayesian model selection: difficulties and remedies.

[Autom. eng. transl.] The choice of the model represents an area of great importance in statistical inference. I believe that the Bayesian approach is particularly suitable for this purpose, since it can effectively overcome some difficulties inherent in that frequentist. However, the Bayesian approach in this field also presents critical aspects. In particular, I argue that it is not appropriate to use conventional distributions, valid for the conditional inference to a model, when considering a plurality of models. I develop my treatment using a simple example, which however highlights the main problematic issues. I end with a discussion of some projects that see me involved and with some issues that deserve further study and consideration.