For the current study, adolescents with missing data significantly differed by race. That is, adolescents without European American ancestry were more likely to have missing temperament and substance use data. This is most likely due to the design of the study, where non-White families were added in later waves. To account for this pattern of missingness and to avoid population stratification, we decided to present models for only participants with European American ancestry (n = 398). The remaining missing values were missing completely at random (34.417[23], p = .06). We imputed missing data using the expectation maximization algorithm in IBM SPSS 22. Given that the sample included siblings, preliminary multilevel linear modeling analyses were used to assess family clustering. A significant clustering effect of families on resiliency, externalizing behavior, alcohol use and cigarette use (range of Interclass Correlations = .07 to .31) was found. Accordingly, multilevel analyses accounting for family clustering were estimated.
