Model Fitness and Predictive Accuracy in Linear Mixed-Effects Models with Latent Clusters