There were several divergences from the plans as specified in the preregistration. Firstly, the plan was to use latent variables for all the substance use outcomes. Although model fit for the latent alcohol use factor was good, the use of dichotomous and continuous variables within the same model led to non-convergence in some of the models. For the cannabis outcome we wanted to use cannabis initiation as well as Cannabis Use Problems Identification Test (CUPIT) scores, but due to high missingness and floor effects we could not use the latter. As specified in the preregistration we resorted to using the outcomes that were most similar to those in the GWAS used for the computation of the PGS. Secondly, the preregistered power calculations were based on larger sample size (N=1,842) than we had available in the analyses, due to the exclusion of relatives, non-European ancestry individuals, and individuals with high levels of phenotype missingness (final N=1,649). Thirdly, we could not estimate the SEMs including all latent variable interactions in the same model because the model did not converge. We had taken this possibility into account in the preregistration and followed the plan to present the models separately per parent factor.