This paper uses dimension asymptotics to study why overfit linear regression models shouldbe compared out-of-sample; we let the number of predictors used by the larger model increasewith the number of observations so that their ratio remains uniformly positive. Under this limittheory, the naive...