Effect on Prediction When Modeling Covariates in Bayesian Nonparametric Models