Latent Class Analysis of Individual-Level Characteristics Predictive of Intervention Outcomes in Urban Male Adolescents

Abstract

Preventive intervention research dictates that new techniques are needed to elucidate what types of interventions work best for whom to prevent behavioral problems. The current investigation applies a latent class modeling structure to identify the constellation of characteristics—or profile—in urban male adolescents (n = 125, aged 15) that interrelatedly predict responses to a brief administration of an evidence-based program, Positive Adolescent Choices Training (PACT). Individual-level characteristics were selected as predictors on the basis of their association with risk behaviors and their implication in intervention outcomes (e.g., mental health, stress exposure, temperament, cognitive function, stress reactivity and emotion perception). Outcome measures included virtual reality vignettes and questionnaire-style role play scenarios to gauge orientations around aggressive conflict resolution, communication, emotional control, beliefs supporting aggression and hostility. A three-class model was found to best fit the data: “NORMative” (NORM), with relatively low symptomatology; “Mental Health” problems (MH-I) with elevated internalizing symptoms; and “Mental Health-E + Cognitive Deficit” (MH-E + Cog) with elevated mental health symptoms paired with cognitive decrements. The NORM class had positive PACT effects for communication, conflict resolution, and aggressive beliefs. Moderation was evidenced by lack of positive PACT effects for the MH-I and MH-E + Cog groups. Also, PACT classes with MH issues showed marginally significant worsening of aggressive beliefs compared to control students in the same class. Results suggest that a latent class model may identify “signatures” or profiles of traits, experiences and other influences that collectively—and more realistically—predict variable intervention outcomes with implications for more effectively targeting interventions than singular factors.