Several limitations of our analyses merit consideration. First, all data in Add Health were self-reported and thus may be subject to recall bias or social desirability bias. Second, during the first 3 waves of Add Health, HED was defined as 5+ drinks, whereas at Wave 4, the modern gender-specific definition of HED was implemented (4+ drinks for females, 5+ drinks for males). Thus, changes in the association with HED over time may partially reflect this change in measurement. Third, it is possible that cohort effects may be present; yet, given the relatively small age range (approximately 6 years) of individuals enrolled at Wave 1, these effects should be minimal. Fourth, we considered dichotomous measures of substance use behaviors in order to increase our statistical power at young ages when substance use is rare; however, use of categorical or continuous measures could provide more nuanced information regarding these associations. Fifth, lower base rates of behavior and lower variances at younger ages may limit statistical power during this age period; however, both the large sample sizes and the numerous significant associations we
