We used multilevel modeling (Bryk & Raudenbush, 1992) with HLM 6.08 software to examine the data. Two levels of analysis were conducted. First, there were 337 × 4 observations, accounting for the repeated measures within individuals. Second, there were 337 individuals. Because the outcome variable of past-month substance use consisted of count data (number of times using substances) with many zeros (non-consumption) and had a sample variance (.69) greater than the sample mean (.32), we conducted nonlinear analysis using a negative binomial regression for this dependent variable. Models examined the impact of youth genotype at both DRD4 and 5HTT, intervention condition, and the interaction of each genetic polymorphism with treatment. We used a similar approach to examine effects on intervention-targeted parenting but did not use the negative binomial distribution to model the data because the parenting measure was not composed of count data and did not include any zeroes. In the follow-up analyses to rule out passive effects, we focus only on parent DRD4.
