Many statistical analyses are performed by means of a regression model. These models investigate the effect of covariates on an outcome variable. A basic assumption of these regression models is independence between the subjects of the study. This assumption, however, is often violated, such that special models need to be used. In practice, the random effects model is a common choice. Frequently, the focus of the analysis lies on the prediction of the outcome variable and not on the significance of the covariate effects. To this end, many measures of the predictive ability have been developed to gauge the quality of the predictions. These measures were mainly developed for models of univariate data, hereby inadequate for the special nature of clustered data and the random effects model. We developed a general methodology to extend existing measures of the predictive ability of models for univariate data to their corresponding random effects models. This methodology consists of defining existing measures either by means of conditional predictions, which are predictions that contain both covariate and cluster information, or by means of marginal predictions, which contain covariate information only. As such, the impact of clustering on the quality of predictions can be evaluated by comparing these different versions of the measure. Moreover, a general strategy is developed to investigate the predictive ability of a model in full detail. As such, an internal validation procedure, a procedure to calculate interval estimates and an evaluation of the individual contributions to the measure of interest are proposed.This methodology was developed for the frailty and the multilevel binary regression model. The Brier score and the concordance probability were chosen as measures of the predictive ability, because they are the most widely used measures for both models and because they are known to be complementary. In addition, a new time-dependent definition was proposed for the concordance probability of the survival model.