We first examined the means, standard deviations, and bivariate correlations among study variables. Next, we performed SEM to test the hypothesis that BM characteristics would interact with adoptive parent affective state to predict infant heightened attention during the frustration task. As a subsidiary analysis, we examined this hypothesis using a regression approach with the BF data. Finally, we explored prenatal substance use effects as a potential mediating pathway.
Descriptive Statistics and Correlations
The analytic sample consisted of 348 linked triads for whom data were available on at least two of the three members of the triad. Means, standard deviations, correlations among study variables, and n sizes for each variable are summarized in . Infant attention during the frustration trials was moderately associated with infant attention during the neutral trials (r = .33, p < .001). The correlation matrix also indicated significant associations among the three BM externalizing measures, significant associations between two of the three BF externalizing measures, and significant associations among about half of the BM–BF externalizing behavior cross correlations. In addition, there were several modest but significant correlations among the remaining variables. For example, BM prenatal ATOD use was correlated with the BM externalizing measures and with two of the BF externalizing measures. BM delinquency was also modestly associated with attention during the frustration trials (r = .13, p < .05) and was inversely associated with AF affective state (r = –.12, p < .05), and AM affective state was associated with openness (r = .12, p < .05). Otherwise, the study variables were no more associated than would be expected by chance.
Means, Standard Deviations, and Correlations Among Study Variables
Evaluation of BM x Adoptive Parent Effects
The SEM analyses used full information maximum likelihood, which has been shown to provide unbiased estimates when data are missing at random (Arbuckle, 1996
) and to offer greater statistical efficiency compared to mean-imputation, list-wise, and pair-wise deletion methods (Wothke, 2000
). The first series of models tested infant attention during the frustration trials regressed on the latent BM externalizing behavior variable (ATOD dependence, delinquency, and novelty seeking), AM affective state, and their interaction; AF affective state, adoption openness, infant attention during neutral trials, and infant sex were statistically controlled.
The first of these models (prior to entry of the interaction term) showed an acceptable fit to the data. Higher levels of AM anxious/depressive symptoms were modestly associated with higher levels of infant attention during the frustration trials (p < .05). Higher levels of BM externalizing behavior indicated a trend toward being significantly associated with higher levels of infant attention during the frustration trials (p < .10). Other than infant attention during the neutral trials (p < .001), none of the control variables were associated with infant attention to frustration in this model. Next, a model in which the interaction between BM externalizing problems and AM affective state was fitted to the data. The results are summarized in . Compared to the first model, in which this path was fixed at zero, this model fit the data significantly better, Satorra-Bentler χ² difference (df = 1) = 6.67, p < .05. More importantly, the interaction term also was significant (p < .05). The AM affective state remained significant (β = .10, p < .05), and the BM externalizing behavior path became significant in the interaction model (β = .17, p < .05).
Figure 1 Standardized path coefficients from study variables to infants’ attention to frustration. Note. Also depicted are paths from birth mother (BM) externalizing behaviors and explained variance in outcome above and below the mean on the adoptive mother (more ...)
Decomposition of Interaction Between BM Externalizing Behavior and AM Affective State
To examine the nature of this interaction, the model was run within the subsamples that were above and below the mean on the AM affective state variable; nonsignificant covariates (AF affective state, infant sex, and adoption openness) were omitted given the reduced subsample sizes. In the subsample above the mean on the AM affective state variable (n = 184), the model provided a good fit to the data, χ² = 5.17, df = 6, p = .523; CFI = 1.00; RMSEA = .00. The path from BM externalizing problems to the infant attention to frustration was significant (β = .35, p < .01). This model was then rerun with the subsample below the mean on the AM affective state variable (n = 164). This model also fit the data well, χ² = 6.45, df = 6, p = .38; CFI = .99; RMSEA = .02. The path from BM externalizing problems to infant attention was not significant (β = –.01).
The interaction decomposition models were rerun to examine whether the pattern of effects remained when the sample was split at 1 SD above and below the mean and at 0.5 SD above and below the mean on the AM affective state variable. Due to the reduced sample sizes, these models were considered preliminary. The standardized path coefficients from BM externalizing to infant attention to frustration were in the same direction as the mean-split analyses for both models: .34 at 1 SD above the mean of AM affective state (n = 62); .36 at 0.5 SD above the mean (n = 121, p < .05); –.04 at 1 SD below the mean (n = 68); and –.02 at 0.5 SD below the mean (n = 121). These follow-up analyses support the effects found with the mean-split approach. Infant attention to frustration was regressed on BM externalizing problem factor scores for the two mean-split groups (see ), reflecting the linear regression results using observed (versus latent) variables and list-wise deletion of cases with missing data.
Relations between birth mother (BM) externalizing problems and infants’ attention to frustration above and below the mean on the measure of adoptive mother (AM) affective state.
Finally, we ran a similar series of models to evaluate whether AF affective state interacted with BM externalizing to predict infant attention during the frustration task. In these models, AM affective state and the other covariates (described above) were controlled. All main effect and interaction paths were nonsignificant, suggesting that the effects described above are specific to AM affective state.
Linear Regression: Evaluation of BF x Adoptive Parent Effects
The GxE hypotheses were examined in a parallel way with BF data using linear regression techniques due to the nonconvergence of the latent factor and the smaller sample size. Externalizing behavior problems were represented by the factor score composed of BF ATOD dependence, delinquency, and novelty-seeking indicators. Infant attention during the frustration trials was not predicted by this BF composite alone or in interaction with the AM affective state variable, F(3, 81) = .587, ns, or the AF affective state variable, F(3, 78) = .255, ns.
Exploratory Models Controlling for BM Prenatal ATOD Use
Next, exploratory models were run to test whether the effect of the interaction between BM externalizing problems and AM affective state on infants’ attention during the frustration trials was better accounted for by BM prenatal ATOD use. In general, all models and path coefficients were regarded with caution in these analyses due to the functional relationship between general and prenatal ATOD use (i.e., mothers who report never using will also report never using during pregnancy, and mothers who use during pregnancy typically use outside of pregnancy).
BM prenatal ATOD use was entered in the full model as a mediator of interaction effects. That is, BM prenatal ATOD use was regressed on BM externalizing problems, AM affective state, and their interaction, and infant attention to frustration was regressed on BM prenatal ATOD use, attention during neutral trials, infant sex, AF affective state, and adoption openness but not directly on terms involved in the interaction. BM externalizing was significantly associated with BM prenatal ATOD use, but the interaction term was not significant. Additionally, BM prenatal ATOD use did not significantly predict infant attention to frustration. This model was compared with a model that allowed direct predictive paths from BM externalizing, AM affective state, and their interaction to infant attention during the frustration trials. The latter model fit the data better, Satorra-Bentler χ² difference (df = 3) = 15.46, p < .05, and the path from the interaction term to the infant attention outcomes was significant (p < .01). Similarly, within the subsample that was above the mean on AM affective state, the reduced model (i.e., exclusion of infant sex, AF affective state, and openness in adoption) provided a marginal fit to the data, χ² = 14.924, df = 9, p = .093; CFI = .960; RMSEA = .060. BM externalizing problems predicted BM prenatal ATOD use (β = .72, p < .001) and infant attention during the frustration trials (β = .48, p < .05), whereas BM prenatal ATOD use did not predict infant attention during the frustration trials (β = –.19, p = ns). The model that fixed the direct path from BM externalizing problems to infant attention during the frustration trials did not fit the data adequately, χ² = 25.381, df = 10, p = .005; CFI = .896; RMSEA = .091. Therefore, the findings did not support BM prenatal ATOD use as an explanation for the GxE findings described above.