Nevertheless, there are a number of limitations to these analyses which should be considered. First, as with any longitudinal study, missing data becomes more extensive at each successive stage of data collection. Attrition rates frequently differ across socio-demographic variables, and could have had a modest impact on our results. Furthermore, although we did observe differences in demographic measures depending on data availability (as described in Table 1), the results presented were comparable to results from analyses that used only individuals for whom complete data were available. Secondly, the sample used in these analyses is primarily of European ethnicity, all of whom were born within a specified geographic area; results may not generalize to regions with a markedly different ethnic make-up. An additional limitation is the use of only one reporter (in this case self-report) of depressive symptoms and alcohol use, which could hence result in reporter bias. Finally, due to the timing of data collection as well as the longitudinal consistency of the alcohol measures administered, we were unable to account for concurrent alcohol in these models. It is likely,
