On the overestimation of widely applicable Bayesian information criterion