Using Latent Profile Analysis to Identify Response Style Subgroups on the Personality Assessment Inventory (PAI): Implications for Predictive Validity

Many self-report multiscale personality measures incorporate validity scales to allow clinicians and researchers to assess the trustworthiness of evaluee responses. However, the conventional scale-by-scale cut-score approach to validity scale interpretation is limiting because (1) there are multiple validity scales on the most popular measures, leading to the possibility of many different validity profiles across scales, and (2) it presumes a strong—but typically untested—moderating effect of response style on the association between substantive clinical scales and outcomes. In the current study, I performed a Latent Profile Analysis with the Personality Assessment Inventory (PAI; Morey, 1991) validity scale scores from 1,506 male Sexually Violent Predator (SVP) evaluees to sort offenders into response-style subgroups. The best fitting model was a four-class model: Honest Responders (n = 405, 26.89%), Positive Impression (n = 777, 51.59%), Negative Impression (n = 122, 8.10%) and Disengagement/Inattention (n = 202, 13.41%). I then examined whether response-style group membership moderated the association between select PAI scales (ANT, AGG, VPI, BOR, INT and EXT) and post-release recidivism. There were five statistically significant interaction effects across the 24 models, but the pattern of effects differed across models and there was no evidence of a consistent pattern of moderating effects. Overall, this study lends support to the utility of the PAI validity scales to delineate four theoretically consistent response styles. However, results from this study only minimally support the presumed strong moderating effect of response style which has primarily driven the conventional cut-score approach to validity scale interpretation.