Time-Varying Covariates of Distinctive Patterns of Youth Violence Exposure Across Time: A Latent Transition Analysis

Several types of violence exposure have been described in the literature but few studies have examined the structural stability of distinctive patterns of youth violence exposure over time and their relations with time-varying covariates. Drawing on conceptual frameworks such as the general theory of crime (e.g., Gottfredson and Hirschi, 1990) and lifestyles theory (e.g., Hindelang, Gottfredson, & Garofalo, 1978), this longitudinal study has three goals: First, the identification of latent classes of youth violence exposure at each assessment time; second, examining the structural stability and change of latent classes of youth violence exposure across three assessment times; third, investigating their relations with time-varying covariates (i.e., male gender, race/ethnicity, family poverty level, major depressive disorder (MDD) symptoms, ADHD symptoms, delinquent behavior, low informal social control in neighborhood).

Three waves of data from the Healthy Passages study are used to address these study goals. A representative community sample of 4,824 children and their primary caregivers was assessed at ages 10, 12, and 15 years. A cluster probability sampling procedure was used to recruit the children from three geographic areas in the United States.  About 49% of the children are boys; about 37% are African-American, 37% Latino, and 27% White. Direct and indirect youth violence exposure was assessed with nine items from the Youth Risk Behavior Surveillance Survey; e.g., how often over the past 12 months has someone threatened or injured you with a knife. Family poverty level was indexed by an annual household income at or below 100% of the Federal Poverty Level. MDD (6 items) and ADHD (8 items) symptoms were assessed with two subscales from the Diagnostic Interview Schedule for Children Predictive Scales, delinquent behavior with seven items adapted from YRBS Survey, low informal social control in neighborhood with five items from the PHDCN project.

Latent class and latent transition analysis were conducted in Mplus version 7 using robust maximum likelihood estimation. Sampling weights, clustering, and stratification were accounted for in all analyses. Results supported a three-class solution (i.e., low, moderate, high violence exposure) for each assessment time. Second, stability of class membership across time was considerably higher for the low and moderate violence exposure classes compared to the high violence exposure class. Third, African-American and Latino race/ethnicity, family poverty level, and ADHD symptoms were consistently linked to membership in the moderate and high violence exposure classes relative to reference category low violence exposure class. Implications of the findings for theory and applied purposes will be discussed.