For each outcome (maximum and usual quantities, usual frequencies, and frequency of consuming 4+ drinks), we performed a 2 LR group by 3 education group by 4 racial/ethnic group (EA, Asian, White Hispanic, and “others”) by 3 time-point mixed design ANOVA, with time as the repeated measure, including an evaluation of the LR by education group by time 3-way interaction. Four- and 3-way interactions regarding LR by education groups by sex by time were also conducted. Race/ethnicity and sex were included in the 4-way ANOVAs because both LR and alcohol use patterns can differ across these subgroups, and two separate evaluations were done rather than one 5-way analysis to optimize interpreting results while avoiding extremely small cells. Linear contrasts were used because, consistent with our prior approaches (Schuckit et al., 2012a), we projected linear decreases in alcohol use over time. Data were evaluated using log transformations because of the skewed distributions often seen with alcohol-related data where many subjects have low scores and fewer report high values. SPSS multiple imputation was used for the 4 subjects with some missing data. Effect sizes were determined using partial eta squared (np2).
