Disentangling substantive responses from faking: Statistical and practical performance of the Bayesian Retrieve-Edit-Select model