Pinpointing outliers in experimental data: the Hat matrix in Anova for fixed and mixed effects models