R Statistics Cookbook

How it works...

The tighter the prior is (more concentrated around) a value is, the more concentrated the posterior is. It is always preferable to choose loose priors to avoid forcing the coefficients to be near a specific value.

Priors should always be defined before looking at the data in order to avoid setting a prior that is almost confirmed by the data.

Bayesian models are, in general, much more robust to overfitting than classical methods. The reason is that when we assign a prior, we are imposing a structure on the model that is not dictated by the data. Another way of posing this is that Bayesian models will generally fit not as well into the data, because of the additional structure we are imposing (the model won’t have the ...