The proportion of the sample endorsing early, average, and late age at first alcohol use and the prevalence of problem use in each of these categories was calculated by ethnicity, using SAS (SAS Institute, 2008). To estimate the risk for problem use associated with early and late (vs. average) age at first drink, logistic regression analyses were conducted separately for the AA and EA subsamples using Stata (StataCorp, 2007). The Huber-White correction was applied to adjust for the non-independence of observations in twins.