Sample. Participants consisted of the younger cohort of the Minnesota Twin Family Study (MTFS), a population-based, longitudinal study of same-sex adolescent twins born in the state of Minnesota and their parents. Though often characterized as a single sample for the sake of brevity, the MTFS in fact consists of three samples: male twins assessed at ages 11, 14, 17, and beyond; female twins assessed at ages 11, 14, 17, and beyond; and another cohort of same-sex male and female twins assessed at ages 11, 14, and currently participating in the age 17 assessment. Unfortunately, the Adolescent Opinions and Attitudes Scale was not administered to the original sample of male twins at intake. The current sample of males thus functions as an accelerated longitudinal twin design. Specific sample sizes are presented in Table 4.3. Further information regarding all aspects of MTFS recruitment and zygosity assessment is detailed elsewhere (Iacono, Carlson, Taylor, Elkins, & McGue, 1999).

Measures. I again made use of the ANTI and AGO scales within the Adolescent Opinions and Attitudes survey. Average levels of ANTI and AGO do not differ between the original and augmented samples (independent sample t-tests, p > 0.3, two-tailed), suggesting that the two samples are comparable.

Analyses. Fitting a triangular, or Cholesky decomposition, model allowed me to examine overall genetic and environmental contributions to ANTI and AGO at ages 11, 14, and 17 years. The Cholesky approach parses the phenotypic variance at each age into that which is due to genetic, shared environmental, and non-shared environmental factors. I then employed constraint parameters to equate unstandardized variances across age within this model. I was thus able to explicitly evaluate whether, for example, the genetic influences at age 14 were larger than those at age 11. Note that, because phenotypic variances change with development (particularly for ANTI), I focused here on unstandardized, or absolute, changes in genetic and environmental influences, as proportional influences can mask or otherwise distort relevant changes (as described in Purcell, 2002).

Because of missing data, I made use of Full-Information Maximum-Likelihood raw data techniques (FIML), which produce less biased and more efficient and consistent estimates than other techniques, such as pairwise or listwise deletion, in the face of missing data (Little & Rubin, 1987). To adjust for positive skew in the data while still allowing for changes in the mean and variance with age, the measures were log-transformed prior to model-fitting to better approximate normality (skew following transformation: 1.65, 0.35, and −0.26 for ANTI at 11, 14, and 17, respectively; 0.71, 0.14, and 0.29 for AGO at 11, 14, and 17, respectively). Mx (Neale et al., 2003) was again used to fit models to the transformed raw data. The possible moderating effects of sex were also examined in these analyses.

Results. Mean levels of each scale, along with their standard deviations, are presented in Fig. 4.3, separately across assessments. Paired sample t-tests revealed that, in both males and females, mean levels of ANTI increased from 11 to 14 and from 14 to 17 (all p < 0.001; note that although the male comparison across ages 11 and 14 was restricted to males in the augmented sample, all other comparisons made use of the full sample). Mean levels of AGO increased significantly from 11 to 14 (p < 0.001) and decreased from 14 to 17 (p < 0.05). Further, as compared to female youth, male youth reported significantly higher levels of both AGO (Cohen’s d = 0.65, 0.98, and 1.03 at ages 11, 14, and 17, respectively; all p < 0.001) and ANTI (Cohen’s d = 0.21, 0.20, and 0.35 at ages 11, 14, and 17, respectively; all p < 0.01), although standardized effect sizes appeared notably larger for AGO than for ANTI.

I initially tested an ACE Cholesky model, in which the variances attributable to genetic (A), shared environmental (C), and unique environmental plus measurement error (E) factors were all estimated. The model was fit both allowing for sex differences in variance parameter estimates (ANTI: −2lnL = 211.39 on 4,473 df, AIC = −8,734.61; AGO: −2lnL = 434.28 on 4,478 df, AIC = −8,521.72) and constraining the parameter estimates to be equal across sex (ANTI: −2lnL = 225.422 on 4,491 df, AIC = −8,756.56; AGO: −2lnL = 468.49 on 4,496 df, AIC = −8,523.51). In both cases, the no-sex-differences model provided a better fit to the data, as indicated by a lower AIC value. These results thus suggest that the genetic and environmental architecture underlying ANTI and AGO does not vary across sex.

I next statistically tested for changes in the magnitude of genetic and environmental parameters with age (see Table 4.4), constraining variances to be equal across age and evaluating the subsequent changes in model fit. A significant change in chi-square indicates that the unstandardized variances are significantly different across age. For AGO, the magnitude of genetic variance remained constant across age (see Fig. 4.4). In contrast, non-shared environmental variance in AGO decreased slightly but significantly with age. Shared environmental variance did not change significantly with age. For ANTI, by contrast, there was evidence of significant change across age for all three parameters. The best-fitting model thus contained no constraint parameters, allowing the magnitude of all variance components to vary freely across age.⁶ Consistent with the earlier results, the magnitude of genetic influences increased across adolescence, increasing from 11 to 14 (∆χ ² = 6.03 on 1 df) and from 11 to 17 (∆χ ² = 10.9 on 1 df), although this increase was not significant from age 14 to age 17. Such findings again highlight fundamental distinctions in the developmental course of genetic influences on AGO and ANTI during adolescence.⁷