Mixed Posterior Predictive checks for the Latin square data set. The upper plots present the distribution of the mixed predictive p-values for the two conditions fitting the calibration model presented in (3) which assumes that the variability is specified by a polynomial function of the level of expression. They show a uniform behavior for both conditions, indicating an excellent model fit. The values for λ1jk, λ2jkand λ3jkare presented in Table 6 in Additional file 1. The bottom plots present the distribution of the p-values for the two conditions after fitting the simpler model assuming λ2jk= λ3jk= 0. They show a lack of fit, in particular for condition 1, indicating that the simple variance model is not appropriate for this dataset.