People tend to judge repeated information as more valid than new information, a phenomenon known as the truth effect. While this effect is robust on average, recent research shows that it is not universal: some individuals show strong truth effects, whereas others show none. Understanding these differences is key to revealing the cognitive mechanisms underlying belief formation and developing interventions that promote more accurate truth judgments. To capture this variability, recent methodological advances have introduced Bayesian hierarchical latent-mixture models to classify individuals into latent classes that differ qualitatively in their behavior. We extend this approach by including covariates as predictors of class membership and demonstrate the validity and robustness through simulations. We applied the model to two datasets examining how the truth effect relates to Need for Cognition (NfC), the tendency to enjoy and engage in effortful cognitive activities. Individuals higher in NfC process information more deeply, which should increase processing fluency and, in turn, strengthen the truth effect - yet prior research has been inconclusive. Across both datasets, we replicated the previously reported two-class structure (positive and null truth effects) and found that individuals higher in NfC had a higher probability of belonging to the positive truth-effect class, even in a dataset where traditional analyses failed to detect a relationship. We thus make both a methodological and substantive contribution by (a) demonstrating Bayesian hierarchical latent-mixture modeling as a powerful framework for investigating qualitative individual differences, and (b) providing clearer evidence for the relationship between NfC and the truth effect.