Means and standard deviations for the two antisocial behavior subscales are presented in . T-tests showed differences by drinking onset group for Wave 4 aggression and for delinquent behavior at three time periods: Waves 1, 2, and 4. Early drinkers engaged in more delinquent activities at most time periods – with both more aggression and more delinquent behavior in the transition from late childhood to early adolescence.
Means (and Standard Deviations) by Drinking Onset Group for Aggression and Delinquent Behavior
To more precisely specify developmental differences in antisocial behavior between early drinkers and non-drinkers, stacked structural equation models were used to examine the influences of early antisocial behavior on later antisocial behavior. All parameters were first allowed to vary across drinking onset groups. When a tenable model was obtained, these parameters were constrained to be equal in an invariant model to test for group differences.
For aggression, the first model in which paths exist between each adjacent wave (without a path from Wave 1 to 4) produced a poor fit: χ2 = 16.41, df = 6, p = .01, RMSEA = .126, NFI = .96, CFI = .97. The path from Wave 1 to 4 aggression was then added, resulting in a model with acceptable fit statistics: χ2 = 3.79, df = 4, p = .43, RMSEA = .000, NFI = .99, CFI = 1.00. This specification was a significant improvement (Δχ2 = 12.62, Δdf = 2, p < .01). All paths were significant regardless of drinking onset status. Adding the remaining two pathways for a full model did not help (Δχ2 = 3.79, Δdf = 4, p > .05); paths from Wave 1 to Wave 3, and Wave 2 to Wave 4, aggression were non-significant for both groups.
For delinquent behavior, the first model in which paths exist between each adjacent wave (again without the path from Wave 1 to 4) also provided a poor model fit: χ2 = 36.17, df = 6, p < .001, RMSEA = .215, NFI = .87, CFI = .89. The path from Wave 1 to 4 delinquent behavior was added, resulting in a model with significant improvement (Δχ2 = 11.36, Δdf = 2, p < .01), but additional changes were needed to meet acceptable fit criteria. Modification indices suggested a direct path between Wave 2 and Wave 4 delinquent behavior. A tenable model was achieved when this parameter was included (χ2 = 2.34, df = 2, p = .31, RMSEA = .039, NFI = .99, CFI = 1.00). It was also a significant improvement in fit with Δχ2 = 22.47, Δdf = 2, p < .01. (See for the common metric completely standardized solution for aggression and for the delinquent behavior solution. Parameter estimates for the nonEFD group are above each arrow; parameter estimates for the EFD group are below.) The path from Wave 1 to 4 delinquent behavior, reflecting the influence of early predisposition on adolescent outcomes, was significant for early drinkers only (whereas the path from Wave 3 to 4 delinquent behavior was significant for the nonEFD group only). Adding the remaining pathway for a full model did not further improve the fit (Δχ2 = 2.34, Δdf = 2, p > .05); the path from Wave 1 to Wave 3 delinquent behavior was non-significant for both groups.
Common metric standardized solution for aggression by drinking onset group. Note. Parameter estimates are above arrows for the nonEFD group and below arrows for the EFD group; EFD = Early First Drinking
Common metric standardized solution for delinquent behavior by drinking onset group. Note. Parameter estimates are above arrows for the nonEFD group and below arrows for the EFD group; EFD = Early First Drinking
Comparing each model to an invariance model (parameters between groups constrained to be equal) revealed differences in the covariance structure by first drink onset status for delinquent behavior but not aggression. For aggression, the invariance model yielded χ2 = 12.35, df = 12, p = .42, RMSEA = .016, NFI = .97, CFI = 1.00 which was not significantly different from the free parameter model, Δχ2 = 8.56, Δdf = 8, p > .05. For delinquent behavior, the invariance model yielded χ2 = 68.99, df = 11, p < .001, RMSEA = .220, NFI = .82, CFI = .85. The free parameter model for delinquent behavior showed significantly better fit: Δχ2 = 66.65, Δdf = 9, p < .01.
CFA was used to test whether antisocial problems in preschool and early adolescence could be used to parsimoniously categorize individuals into groups that differ in likelihood of early drinking onset. The CFA was not significant for aggression (Pearson’s χ2 = 6.31 for df = 3, p > .05), but was significant for delinquent behavior (Pearson’s χ2 = 18.38 for df = 3, p < .001) (see ).
Configurations for Delinquent Behavior Group by First Drink Onset
For delinquent behavior there were more cases than expected by chance for those low in delinquent behavior during both preschool and early adolescence who had not used alcohol by 12–14 years of age (i.e., the “11” type). There were fewer cases than expected by chance for low-low early drinkers (i.e., the “12” antitype). The opposite pattern was found for those who were high in delinquent behavior during both time periods, with more cases than expected by chance for the high-high group with drinking onset (i.e., the “42” type) and fewer cases than expected by chance for high-high children who had not yet tried alcohol (i.e., the “41” antitype).
Given the striking findings of a relationship between EFD and only delinquent behavior, additional post hoc analyses were conducted to identify the preschool CBCL delinquent behavior items most predictive of EFD. Significant Spearman’s rho correlations were found for hanging around with others who get into trouble (r=.221, p < .01), lying or cheating (r=.259, p < .01), setting fires (r=.204, p < .01), swearing (r=.168, p < .05), and thinking about sex too much (r=.161, p < .05).