Is cannabis a gateway to hard drugs?

The gateway hypothesis proposes that use of cannabis directly increases the risk of consuming hard drugs. We test this controversial, but influential, hypothesis on a sample of cannabis users, exploiting a unique set of drug price data. A flexible approach is developed to identify the causal gateway effect using a bivariate survival model with shared frailty estimated using a latent class approach. The model suggests two distinct groups; a smaller group of "troubled youths" for whom there is a statistically significant gateway effect that more than doubles the hazard of starting to use hard drugs and a larger fraction of youths for whom previous cannabis use has less impact.