Internal consistency reliability across the items was high . Higher scores reflect higher levels of cumulative risk. We treated cumulative risk as a strictly between-child or time-invariant variable, given the limed within-person variability in cumulative risk over time .Informed by prior studies , we adopted a variety of child and family covariates in an attempt to adjust for potential confounding factors that could bias our estimates. As we describe in detail in the online supporting material, our control covariates included infant gender , race , temperament, primary caregiver reading ability, age at birth, marital status, maternal education, maternal hours worked per week, and research site . These control covariates were measured when the target child was 2 months of age, with the exception of temperament, which was measured at 7 months. Within- and between-person variation in saliva collection times were also controlled in all models.As indicated in prior work , there is little evidence of selective attrition in the sample through early childhood. Missing data were modest. For the cortisol measures missingness ranged between 13% and 26%, the latter representing children’s cortisol levels at 24 months. There was no clear pattern to the missingness; missingnesss was not associated with maternal education, income, or partnered status. Missing data with regard to child care quantity ranged between 3% and 9%. Of those who had a spell of child care , growing bags missing data for estimated caregiver responsivity was 19%, and missing data for those reporting center-based versus non–center-based care ranged from 23% to 31%. To adjust for biases emerging due to missing data, we fitted all models using full information likelihood estimation .
FIML estimation helps to reduce biases to the extent to which missingness is missing at random —that is, conditionally random after adjusting for the observed variables included in the likelihood function . As a robustness check, we fitted an identical taxonomy of models in which only children with 10 or more hours of child care were included in the observed covariance matrix . As the findings were substantively identical and because zero hours represents a “real” value on the childcare hours distribution, we report findings based on the less restricted sample.To address our questions, we fitted a taxonomy of multilevel models . We present a simplified exemplar model in the equations given below for parsimony and ease of explication. Here, the model tests the extent to which the respective within person effects of child care hours, age, and a within-person interaction between child care hours and age varies as function of between-person cumulative risk. We, however, tested identical series of models for non-parental caregiver responsivity and type and peer exposure. In all models, we first tested the main effects of cumulative risk and the given child care predictor. We then added the respective two- and three-way interaction terms to the model in distinct steps. In a final step, we integrated all of the child care predictors into a final model to test their unique relations. Only interaction terms that survived Benjamini-Hochberg corrections were integrated into this final specification. Sometimes called a hybrid model , this multilevel specification is more efficient than typical child-fixed effect models common to the econometrics literature while still retaining the advantages of child-fixed effect models; specifically, the estimates for the Level-1 predictors are identical those obtained using a child-fixed effects specification, such that they account for all potential observed and unobserved time-invariant confounds.
The hybrid approach also affords the ability to simultaneously model between-person relations at Level 2. Importantly, for our purposes, they also allow one to estimate cross-level interactions; specifically, as shown by parameter γ31, we tested the three-way cross-level interaction addressing our questions: Does the within-person effect of child care exposure moderate the association between cumulative risk and cortisol, and does this interactive effect vary as a function of age? As displayed, the model for the residual variances includes only within-person, time-specific residual and a random intercept —implicitly constraining the within-person effects to be identical across children. This parsimonious specification is tested and relaxed, as appropriate.Child care quantity—As displayed in Column A of Table 2, preliminary main effect models indicated that, when considered in aggregate, within-child changes in child care hours were not associated with contemporaneous changes in children’s cortisol levels. However, as evidenced by the statistically significant three-way interaction in Column B , this null main effect was explained by the fact that direction of the within-person child care effect was conditional on level of cumulative environmental risk experienced by the child, as well as child age. We display these conditional relations in Panels A through C in Figure 1; the values of all other predictors in the model were held at their respective means. Irrespective of cumulative environmental risk, no within-person child care effects were evident at 7 or 15 months of age; none of the displayed simple slopes for child care hours differed statistically from zero. Instead, the interactive effect grew over time, becoming more pronounced and statistically significant by 24 months of age. For children from high-risk contexts, increases in child care exposure at 24 months were associated with contemporaneous decreases in children’s resting cortisol levels. In contrast, for children from low-risk contexts, increases in children’s child care exposure at 24 months were associated with contemporaneous increases in children’s resting cortisol levels Tests of model constraints indicated that the absolute magnitude of the simple slopes displayed in Panel C of Figure 1 were statistically identical . In absolute terms, these relations corresponded to absolute standardized simple slopes of approximately .
Albeit somewhat modest, it is worth noting that in relative terms these simple slopes are approximately two thirds the size of the standardized within-person effect for time of day —one of the most well-understood biological mechanisms underlying normative within-person variation in cortisol. Between-child differences in children’s average weekly child care hours were predictive of only children’s cortisol growth rates , and neither this relation nor the relation between average child care hours and children’s 7-month cortisol levels were moderated by children’s level of cumulative environmental risk. Children who attended greater hours in child care on average tended to show less negative declines in their cortisol levels over, irrespective level of cumulative environmental risk . We display this relation in Figure 2 for prototypical children experiencing low versus high levels of child care exposure per week on average across infancy and toddlerhood; all other predictors are held constant at their respective means. Whereas children attending 40 hours of child care per week were estimated to maintain their 7-month cortisol levels over time , those attending 10 hours a week were predicted to show a statistically significant decline their cortisol levels between 7 and 24 months of age . Note that this decline was rather modest, approximating a 0.02-SD decrease, per month. Child care type—There was no evidence of a within-person effect with respect to child care type ; that is, nursery grow bag time-specific shifts to center-based or non–center-based child care were not associated with contemporaneous changes in cortisol. There was, however, some evidence that the positive between–person relation between attending comparatively more center-based care between 7 and 24 months and children’s cortisol levels differed for children from higher versus lower levels of cumulative risk . This two-way interaction did not vary as a function of age—that is, it was not predictive of cortisol growth. There was, however, evidence that attending greater proportions of time in center-based child care was predictive of more rapid cortisol decreases over this period . In the former case of the interaction, it is also worth noting that the raw pvalue for this estimate, .01, dropped to marginal levels of statistical significance after adjusting family-wise false discovery rate. We nonetheless display the simple slopes derived from this interaction in Figure 3, with this caveat in mind.As shown in Figure 3, we found evidence that, for children experiencing low levels of environmental risk, those who attended comparatively greater proportions of their time in center-based child care tended to have higher cortisol levels . This is most clearly illustrated by considering the vertical distance between the dashed and non dashed gray trajectories at 7 months in Figure 3; it reflects the 7-month simple slope for those attending 100% of their time in child care in center-based care, compared to those who spent 100% of their time in non–center-based child care. This conditional relation is noteworthy, approximating a standardized association of .41. In contrast, child care type was unrelated to children’s 7-month cortisol levels, for children from high-risk contexts .
As indicated by the divergence of the dashed and non dashed black trajectories over time, there was a descriptive indication that, for those from high-risk contexts, attending predominantly non–center-based care was predictive of comparatively less rapid cortisol declines between 7 and 24 months. However, the conditional effect of child care type at 24 months only approached statistical significance . Considered another way, despite the descriptive differences in the estimated cortisol values plotted at 24 months of age in Figure 3, these values were statistically identical. non-parental caregiver responsivity—Within-person changes in caregiver responsivity between 7 and 24 months were not associated with contemporaneous changes in children’s cortisol levels, irrespective of child age or level of cumulative risk. Similarly null relations were evident with respect to between-person differences in children’s average levels of caregiver sensitivity across this period. Peer exposure—Within-child changes in peer exposure were not associated with changes in children’s cortisol levels over time. However, there was some indication that between-child differences in children’s average levels peer exposure were predictive of children’s overall cortisol levels. Specifically, on average, children who typically attended child care with greater numbers of peers tended to evince lower cortisol levels . This corresponded to a moderate standardized coefficient of approximately. Adjusting for false-discovery rate, this relation did not vary over time or as a function of children’s levels of cumulative risk.There is good evidence that children’s child care experiences affect HPA axis functioning . However, it is increasingly clear that these relations may be quite different for children growing up in high-risk environmental contexts. The aims of the present study were twofold. The first aim was to test the extent to which within- and between-child differences in children’s child care experiences—quantity, type, caregiver responsivity, and peer exposure —across infancy and toddlerhood were associated with children’s cortisol levels over this period. The second aim was to consider whether these relations differed for children experiencing higher levels cumulative contextual risk outside of child care.Our results provide some support for such interactive effects. As hypothesized, for children from low-risk contexts, within-child increases in weekly child care hours were associated with contemporaneous increases in children’s resting cortisol levels. The inverse was evident for children from high-risk contexts; increases in weekly child care hours were associated with contemporaneous cortisol decreases. The magnitude of this cross-level interaction strengthened over time, reaching statistical significant between the ages of 15 and 24 months. Interactive effects between cumulative risk and attending center-based care were also evident; however, they extended to only between-child differences in overall centerbased exposure between 7 and 24 months of age . Controlling for child care hours, caregiver responsivity, and peer exposure, children from low-risk families who spent greater proportions of their time in center-based child care tended to have higher cortisol levels than their peers who attended more non– center-based childcare. In contrast, for children from high-risk contexts attending center based care was unassociated with children’s cortisol levels. With respect to children’s exposure to peers in child care, there was evidence that children exposed to greater numbers of peers between 7 and 24 months tended to show comparatively lower cortisol levels. This relation did not vary as a function of cumulative risk or age. Within-person changes in peer exposure were unrelated to changes in cortisol, irrespective of age or level of cumulative risk.Prior work conducted with lower risk samples has shown that children’s cortisol levels tend to increase over the course of the day on days when they attend child care yet show normative diurnal declines on days in which they do not attend child care . Nontrivial differences between this prior literature and the present study make direct comparisons impossible.