Third, although the majority of the sample was ascertained in the peak years for cannabis initiation and early-to-peak years for problem onset, not all participants had passed through the age of risk for the cannabis outcomes. Fourth, although we attempted to minimize potential retrospective reporting bias by adjusting for participant age in the analyses, retrospective assessments are inherently vulnerable to reporting bias. Fifth, findings may not generalize to other populations, other definitions of maltreatment, or more severe cannabis outcomes (e.g., cannabis use disorder). Relatedly, it is possible that certain forms of maltreatment (e.g., physical vs. sexual abuse) are more closely related to cannabis involvement in EAs vs. AAs (as indicated by Werner et al., 2016). Sixth, our analyses do not address which specific aspects of initiation or problem use differ in their relationship with maltreatment. Our broad definitions of initiation and problems may also have limited a more specific understanding of the effects of maltreatment. Seventh, we did not test whether the heritability of cannabis initiation and problems, and their covariance, was modified by the presence or absence of maltreatment (i.e., a putative gene × environment interaction, with maltreatment representing the environmental modifier). Larger sample sizes in the AAs would
