Estimation was carried out using the MLR estimator which incorporates a likelihood-based approach to the problem of missing data [27]. Any respondent with at least one repeated measure may be included under the missing at random (MAR) assumption which states that the probability of a particular variable being missing can depend on other variables in the model, but not the actual unmeasured value of the variable in question. When estimating the growth models, we compared the results obtained from a complete case (CC) analysis with those obtained in samples with one, two or three missing values to assess the robustness of the findings (i.e. the shape of the resulting latent trajectory). As the inclusion of the later alcohol outcome data led to a further reduction in the sample available for analysis, we considered an additional complementary missing data treatment when adding distal outcomes to the growth model. The results obtained when restricting the analyses to those with alcohol data were compared with a second bias-adjusted set derived using inverse probability weighting (IPW) [28]. This method was employed in preference to
