Further analyses were conducted to adjust for potential confounding by adolescent behavior problems. Paths from the teacher-rated externalizing behavior score at age 12 to school performance and substance use variables at ages 12 and 14 were included in these models. For the drinking variables, this adjustment had only minor effects on the cross-lagged path coefficients and did not affect the overall results. In the models for smoking behaviors, the path from school performance at age 12 to daily smoking at age 14 was non-significant after the adjustment (standardized coefficient = −.08, p = .117), and similarly the path from student status at age 17 to any smoking in young adulthood was no longer significant after the adjustment (standardized coefficient = −.04, p = .129). The adjustment also reduced the residual correlations between school performance and substance use at ages 12 and 14 to non-significance with the exception of school performance and daily smoking at 14. The adjusted models had good fit to data (CFI=.997–1.000, RMSEA=.009–.024, full results available by request).
