Strengths of the present prospective study are the relatively large sample and the use of structured interviews conducted by well-trained staff to assess conduct problems and substance use. A feature that may limit the generalizability of the findings to the general population is the modest oversampling for familial alcoholism risk. However, given that the oversampling had no systematic effect on the estimates of the variance components for several risk-relevant outcomes in early adolescence [17], we do not expect it has strongly affected our results. In addition, although examining levels of use of the various substances as opposed to lifetime incidence is more informative in young adulthood, this approach assumes that the same factors influence no versus any use and any versus more frequent use. Multi-stage modelling showed that there is a substantial overlap in genetic and environmental factors influencing earlier and later stages of substance use, but also that there are factors that contribute specifically to later stages of use [34–35]. Finally, although our dataset is relatively large, the power to discriminate between the different variance components was limited due to the use of a threshold model (as necessitated by the highly skewed data) [36].
