Associations Between Dopamine D2 Receptor (DRD2) Gene, Maternal Positive Parenting and Trajectories of Depressive Symptoms from Early to Mid-Adolescence

The genetics subsample (N = 1090) did not differ from the full sample (N = 2164) on the vast majority of variables: gender and age of the child, age and educational levels of mother and father, family income, maternal positive parenting and depressive symptoms at ages 12, 13, 14, 15 and 16. Only slightly lower levels of depressive symptoms at age 11 existed in the genetics subsample versus the full sample (p < 0.05, mean difference − 0.34). In order to maintain consistency in sample size throughout the analyses, we chose to restrict all analyses (i.e., analyses about trajectories of depressive symptoms and predictors of trajectories) to the genetics subsample. In a follow-up analysis, we tested the consistency of our findings across the full sample (N = 2164) and genetics subsample (N = 1090). Three main findings are highlighted. First, the three-trajectory model identified in the genetics subsample was confirmed in the full sample: (i) low-stable (40.2%; B _intercept = 0.86, SE = 0.11, p < 0.001), (ii) moderate-increasing (40.8%; B _intercept  = 1.64, SE = 0.06, p < 0.001; B _linear  = 0.12, SE = 0.01, p < 0.001), and (iii) high-increasing (19.0%; B _intercept  = 2.56, SE = 0.07, p < 0.001; B _linear  = 0.13, SE = 0.03, p < 0.001; B _quadratic  = −0.01, SE = 0.01, p < 0.01). The entropy was 0.90, and the mean posterior probabilities of trajectories ranged from 0.95 to 0.96, suggesting good classification precision. Second, we identified no significant differences between the availability of genetic data across the three trajectories (χ² = 4.81, df = 2, p > 0.05). Third, the pattern of regression findings was replicated when we used the trajectories as identified in the full sample. Specifically, the meta-analytically derived A241G × parenting interactions remained statistically significant (high-increasing vs. moderate-increasing: OR = 1.85, 95% CI [1.22, 2.81], p < 0.01; moderate-increasing vs. low-stable: OR = 0.67, 95% CI [0.49, 0.93], p < 0.05), and they were similar in magnitude to the effects with trajectories obtained in the genetics subsample.

Given no within class variances in the LCGA model, the LCGA is less complex than the GGMM (General Growth Mixture Modeling). The LCGA is more likely to avoid convergence issues and increase overall model stability versus the GGMM (Jung and Wickrama 2008). Therefore, in line with previous research (Dekker et al. 2007; Fanti and Henrich 2010), the LCGA was selected.

Again, mothers of girls had both higher levels of maternal positive parenting than mothers of boys at ages 10 (t (998) = 3.31, p < 0.001) and 11 (t (1051) = 2.10, p < 0.05), which is in accordance with previous research in similar age groups (i.e., 10 and 11 ages) (Oldehinkel et al. 2006).

The multinomial logistic regressions in this study were estimated on assigning classes to their most likely class. This is mainly because the mean posterior probabilities in the three trajectories ranged from 0.96 to 0.98, suggesting very high estimate probabilities for belonging to each class (Brendgen et al. 2005; Castelao and Kröner-Herwig 2013; Mezulis et al. 2014; Stoolmiller et al. 2005).