Model fit remained excellent when the AO by delay interaction was added, χ2(df = 78) = 335.09, p < 0.001; CFI = 0.97; TLI = 0.94; RMSEA = 0.05. All main effects of AO and delay remained. The AO by delay interaction was significant for the slope (β = −0.13, p = 0.02) but not the intercept (β = 0.04, p = 0.22). To decompose the interaction, the model was tested separately for participants with an AO by age 16 years (48.6% of the sample) and an AO after the age of 16 years. Within the earlier AO sample, a longer delay was associated with larger increases in heavy drinking over time (β = 0.28, p < 0.001). In contrast, within the later AO sample, delay was unrelated to changes in heavy drinking (β = 0.01, p = 0.92).
