evaluate additional possible mediators as well as moderators. Fourth, for reasons described in the Methods, only one externalizing measure, the Barratt, was available for use in these analyses, and it is important to evaluate additional impulsivity measures and other externalizing characteristics in future work. Similarly, DSM-IV was used for alcohol problems, and the fact that 10 of the 11 criterion items are the same in DSM-IV and DSM-5 make it likely similar results would be seen for DSM-5, but this needs to be directly tested (Hasin et al., 2013). Next, all Baseline and follow-up information involved self-reports without corroboration. Finally, the PEER latent variable was based on the participants’ perception of drinking in close friends, and different results might be observed if those peers had been directly interviewed. Similarly, while our group considered other approaches (e.g., a longitudinal change model) we felt that the potential clinical implications of our work were best served by the SEM in Figure 3, and to optimize clarity of the results in Figure 3, the analyses presented here did not test for reciprocal influences.
