The intensity prediction equations, also called intensity attenuation relations, for Italy have been evaluated statistically by Mak et al. (2015). A main result is a ranking of the different models. I show that this evaluation is not in line with the rules of statistics. For example, the number of estimated parameters is not considered in the retrospective testing, which contradicts the classical statistical criteria for model selection such as the Akaike information criterion or the Bayesian information criterion. A criterion (score) of statistical model evaluation and selection should be well established in the community of statisticians to ensure the appropriateness of the criterion. I have also discovered further weak points in the analysis, such as the inhomogeneous numerical precision of epicentral intensity. Additionally, it is demonstrated that the mean square error is a better distance measure for the selection than the mean absolute error of Mak et al. (2015). The theoretical basics of this fact are also explained.