Although research on adolescent violence has yielded robust information on risk and protective factors pertaining to perpetration of violence, we know little about how these factors may differentially configure across youth. Evidence from multiple fronts, such as research on trajectories of violence and comparisons of violence types, argues the need for fuller investigation of heterogeneity among violent and at-risk youth. Correlation-based tools that assume relative sample homogeneity (eg, regression and SEM) are not ideally suited to detect etiological heterogeneity. By contrast, person-oriented techniques, such as latent class or profile analysis, seek meaningful subgroups based on distinct clustering patterns on theoretically salient variables (Muthen, 2002). This paper provides 1) a conceptual basis for application of person-oriented methods to systematically search for structure within heterogeneity of important risk factors and 2) findings of subgroup structures that hold implications for practice.
This NIMH-funded study surveyed an ethnically diverse school-based sample (n=849) of youth across two US regions, recruited on the basis of school drop-out risk, with demonstrated elevated histories of violence exposure (Nurius et al., 2008). These interviewer-administered interviews used established measurement scales across multiple sociodemographic, psychological, and interpersonal domains. Latent profile analysis (LPA) is a cluster analytic type modeling technique that uses maximum likelihood estimation to seek and test for significantly distinct subgroups based on continuous variables. LPA was conducted using four scales selected on the basis of strong empirical and theoretical links to violence perpetration: family conflict, deviant peer associations, anxiety, and violent victimization. These constructs were assessed multidimensionally and have demonstrated good to excellent psychometric properties. Analyses were run using M plus version 5.2.
Multiple measures of model fit statistics were examined, including the LMRT, which indicated that a 4-class solution fit best (LMRT= 153.26, p=0.019) with probability statistics (akin to factor loadings) of 0.89, 0.85, 0.98, and 0.95. Group 1 reported high deviant peers and moderate levels of victimization, but low risk otherwise; they reported high perpetration. Group 2 youth had high anxiety, deviant peers, and family conflict, but low victimization and perpetration. Group 3 youth were low on all risk factors as well as perpetration. Group 4 youth were high on all risk factors and perpetration, with victimization exceptionally high. Planned analyses will examine these four groups in more detail relative to sociodemographics, protective factors, and other risk factors, such as alcohol and drug use.
These results provide strong evidence of complex interrelationships among risk factors and their etiological implications for adolescent violence perpetration. Distinctly different risk profiles are evident for both youth with elevated perpetration histories as well as those with limited histories. These findings provide a basis for subsequent analysis, such as stability of groups over time implications of these risk clusters for impaired psychosocial development, and guidance in developing tailored preventive and remedial interventions reflective of distinctive needs. Implications of these specific findings are integrated with discussion of person-oriented methods augmenting social work's capacity for discerning and effectively responding to complex diversity.