Since empirical studies have shown that cannabis users are much more likely to initiate hard drug use, a causal linkage has been suggested (?gateway hypothesis?). However, individual differences in proneness and accessibility to drugs provide alternative non-causal explanations for the observed drug use pattern. We propose a Bayesian estimation and predictive framework to analyze the effects and relative importance of previous cannabis use, proneness and accessibility factors on hard drug initiation and to explore possible policy implications. We employ a novel model specification, motivated by four gateway transmission channels, to analyze data from a recent Norwegian survey of young adults.