Cross-validation and information criteria help us in three ways. First, they provide useful expectations of predictive accuracy, rather than merely fit to sample. So they compare models where it matters. Second, they give us an estimate of the tendency of a model to
