Table 2.
Changes over time in adolescent alcohol use using event-based approach, as a function of demographic and personality characteristics.
| Estimate (s.e.) | Estimate (s.e.) | Estimate (s.e.) | |
|---|---|---|---|
| Model 1 | Model 2 | Model 3 | |
| Intercept | .27*** (.02) | −.06 (.05) | −.02 (.04) |
| Time before HS | .10*** (.04) ↑ | .13*** (.04) ↑ | .03 (.04) |
| Time after HS | .26*** (.01) ↑ | .28*** (.01) ↑ | .28*** (.01) ↑ |
| Sex (boy) | .04 (.04) | .04 (.04) | |
| White | .19*** (.04) | .18*** (.04) | |
| Delinquency | .27*** (.01) | .24*** (.02) | |
| Delinquency × Before HS | .14*** (.03) ↑ | ||
| Delinquency × After HS | .002 (.01) | ||
| Fit statistics | |||
| AIC/BIC LL |
16,733/16,773 −8,360 |
16,416/16,475 −8,199 |
16,398/16,470 −8,188 |
Note:
N = 891.
p ≤ .05;
p ≤ .01,
p ≤ .001.
Arrows in all models indicate terms associated with statistically significant changes in adolescent alcohol use over time (a log-transformed Frequency × Quantity measure of past month alcohol use). Smaller AIC/BIC fit indices suggest a better model fit.
In the estimated spline models, parameter estimates for “Before HS” and “After HS” represent individual slopes for pre- and post-HS intervals (default coding by STATA mkspline command, without invoking the ‘marginal’ option), and the associated p-values show whether these individual slopes significantly differ from zero, or whether there is a significant growth in alcohol use over those distinct time periods. Additional probing of these effects was conducted, indicating a significant difference between these slopes for every ‘event-based’ model as well: parameter estimate β (s.e.) = −.16 (.04), p < .001 for Model 1; parameter estimate β (s.e.) = −.14 (.04), p < .001 for Model 2, and parameter estimate β (s.e.) = −.24 (.05), p < .001 for Model 3.