Studies confirm that juvenile offenders have higher rates of past year illicit drug use (ranges from 38-85%) and psychological disorders (estimates from 70-100%) than normative youth populations. These mental health issues are complicated by issues of ethnicity and gender; studies of juvenile offenders indicate that drug use prevalence is higher among males and also among Caucasian than African American youth. Despite these high rates of substance use and mental health challenges, considerable heterogeneity exists among these youth. Identifying and testing new techniques to model this heterogeneity can yield important insights above and beyond standard analytic techniques. This study compares multiple regression with latent class mixture regression to examine three research questions: 1) What are the strongest predictors of past year substance use among juvenile offenders who are identified as moderate versus severe substance users? 2) Do youth in these latent classes vary by race and gender? 3) Does latent class mixture regression analysis yield information beyond standard regression?
Methods
Juvenile offenders (n=713) were recruited from juvenile correctional facilities following admission by one of 45 juvenile courts across a mid-western state. The dependent variable, past year substance use, was measured by a multi-item poly-substance use matrix that included heroin, ecstasy, marijuana, hallucinogens, cocaine, amphetamines, and inhalants. Independent variables (ethnicity, gender, impulsivity, fearlessness, psychological distress, delinquency level, and trauma experiences) were measured utilizing various standardized scales with good to excellent reliability. Latent class mixture regression was utilized, and results were compared to linear regression analyses. Latent class mixture regression uses independent variables and covariates to estimate a model that assigns cases (i.e., persons) to each latent class, without the conventional regression assumptions of normality and homogeneity of variances. Bayesian Information Criterion (BIC) determined the best-fitting model.
Results
Comparative statistics for class models suggested a two-class solution was the best fitting model. Class 1 (N = 401) included moderate past year drug abuse where independent variables accounted for 21% of the explained variance; Class 2 (N=312) included severe past year drug abuse where the same independent variables accounted for 47% of explained variance. Youth identified in Class 1 were significantly more likely to be African-American, but less likely to be psychologically distressed, impulsive, and highly delinquent. Impulsivity, fearlessness, and psychological distress were significantly associated with severe past year drug abuse among the largely Caucasian Class 2. Compared to standard regression, latent class mixture regression identified different predictors of substance abuse.
Conclusions and Implications
The differing latent classes reveal that youth who use illicit drugs prior to incarceration significantly differ by ethnicity and level of mental and behavioral health problems. Latent class mixture regression led to more refined results than standard multiple regression. Findings suggest that utilizing a more advanced statistical methodology can improve identification of the heterogeneity among juvenile offenders and better predict key outcome variables. As drug use persists among high-risk youth, it is important that researchers consider using these more refined types of analyses to inform social work practice, address barriers to effective treatment, and inform future prevention.
