we observed with ACFS, a continuous measure, is less susceptible to this source of error. Given concerns about the sensitivity of inferences from G × E interactions to scaling, we ran an additional analysis using Huber robust regression which down-weights outlier observations: the G × E interaction term remained significant (p=0.026) with a modest reduction in effect size (beta= -0.37 versus beta= -0.41 in the linear regression analyses). The association with alcohol-related phenotypes could be bidirectional, however, the mean age (11.0y) of first CSA occurrence suggests that CSA typically precedes alcohol problems. The use of a binary variable for CSA combines diverse abuse experiences encompassing a wide range of severity; a stronger association might have been observed if a quantitative covariate incorporating information on abuse severity and duration were available. Retrospective assessment of CSA in adults may raise concerns regarding the introduction of bias. The CSA measure used in this study has been found to have reasonable concordance within female like-sex twin pairs and significant association with psychiatric sequelae and parental risk factors (Dinwiddie et al. 2000; McLaughlin et al. 2000). Examination of discordant pairs found no evidence for retrospective bias in the association of CSA with parental rejection (McLaughlin
