To minimise the likelihood that the effect of childhood covariates where the specified time periods were ages 6 – 13 (parental conflict, single parent family, strict parenting, frequent childhood religious attendance) may have occurred after the point of cannabis use opportunity, any individuals who reported use opportunity before the age of 6 were removed from model one. This resulted in the observations of only one participant being removed from the model. Huber-White analysis for clustered data was implemented to adjust for the non-independence of observations from members of a twin pair. The assumption of proportional hazards was assessed through tests of Schoenfeld residuals and modelling of the interaction of covariates with time in the analysis (represented as ‘_t’) (P=≤0.05). Any variables found to violate the proportional hazards assumption were reparameterized via modelling interactions between the variable and time in the analysis, resulting in an extended Cox Proportional Hazards model.
