Abstract Model selection is often implicit: when performing an ANOVA, one assumes that the
normal distribution is a good model of the data; fitting a tuning curve implies that an additive …