A state-trait model of cortisol in early childhood: Contextual and parental predictors of stable and time-varying effects

Abstract

This study examined state-trait models of diurnal cortisol (morning level and diurnal slope), and whether income, cumulative risk and parenting behaviors predicted variance in trait and state levels of cortisol. The sample of 306 mothers and their preschool children included 29% families at or near poverty, 27% families below the median income, and the remaining families at middle and upper income. Diurnal cortisol, income, cumulative risk, and parenting were measured at 4 time points, once every 9 months, starting when children were 36-40 months. State-trait models fit the data, suggesting significant state but not trait variance in cortisol. Low income and cumulative risk were related to trait levels of diurnal cortisol with little evidence of time-varying or state effects. Stable levels of parenting predicted trait levels of diurnal cortisol and time-varying levels of parenting predicted time-varying state levels of diurnal cortisol. Findings highlight the allostatic process of adaptation to risk as well as time-specific reactivity to variability in experience.

The stress response system, including the hypothalamic-pituitary-adrenal (HPA) axis, has been described as an “environmentally sensitive” physiological system (Granger et al., 1998) and is associated with psychosocial adjustment outcomes across development (Granger et al., 1996; Gunnar, Sebanc, Tout, Donzella, & van Dulmen, 2003; Yehuda et al., 1996). Evidence points to both stable aspects of HPA activity, attributed in part to genetic factors and individual differences (Schreiber et al., 2006), and a significant degree of change in the activity of the system over time, even within the same developmental period (Ross, Murphy, Adam, Chen & Miller, 2014; Shirtcliff, Allison, Armstrong, Slattery, Kalin & Essex, 2012). The stability in individuals’ HPA activity may be a function of stable environmental contexts known to relate to HPA activity, including poverty and contexts characterized by multiple or “cumulative” risk factors (Dowd, Simanek, & Aiello, 2009; Miller, Chen, & Zhou, 2007). These associations are in line with the theory of allostasis, wherein exposure to chronic stress would lead to new “set points” or stable parameters of the stress system. While these distal and relatively stable factors relate to HPA activity, so too do contemporaneous person and situational factors, potentially predicting time-specific variations from stable levels of HPA activity. Few studies have examined whether more proximal and variable ecological factors influence individual’s HPA activity to the point of predicting changes in an individual’s “set point” functioning (Berry, Blair, Granger, & The Family Life Project Key Investigators, 2016). That is, few studies of HPA axis activity include repeated measures across time, and fewer still have explored how proximal and distal factors relate to stability and change in the system over time (Laurent et al., 2013). Understanding the predictors of early life alterations to the HPA axis is of particular salience when considering the developmental origins of disease (Gluckman & Hanson, 2006) and potential for biological embedding (Miller, Chen, & Parker, 2011), which argue that childhood stress is programmed into stress-sensitive systems early in life.

The latent state-trait (LST) framework is a useful analytic approach to the identification of stable individual cortisol traits (the commonality of an individual’s cortisol activity over time) as well as time-specific, or state, variations from trait levels (c.f. Steyer, Schmitt, & Eid, 1999). The measurement approach may smooth the confluence of person and situational factors impacting cortisol levels sampled at any given moment on the trait level, allowing for a test of whether an accumulation of environmental or time-specific situational factors exert influence on an individuals’ set point, trait HPA axis activity (e.g. Hellhammer, Fries, Schweisthal, Schlotz, & Hagermann, 2007; Kirschbaum, Steyer, Eid, Patalla, Schwenkmezger, & Hellhammer, 1990). The model further decomposes variance by identifying time (or epoch) specific variance in states. In a model with additional latent variables (e.g. income, cumulative risk, and parenting), it is possible to examine the association between stable aspects of cortisol activity and environmental factors (e.g. if characteristic levels of cumulative risk are associated with characteristic levels of diurnal cortisol), and also to examine if time-specific (state) variations in cortisol are associated with time-specific changes in the environment (e.g. if the experience of a change in parental negativity relative to what is typical for the individual is associated with a contemporaneous change in cortisol from what is typical for the individual).

The present study utilized cortisol samples from four time points, each separated by 9 months to characterize trait diurnal cortisol levels, with three consecutive morning and evening samples obtained at each time point to capture state values of cortisol morning level and diurnal slope to address the following questions: 1) Do contextual factors, such as income, cumulative risk and parenting, predict both trait and state variance in preschool-age children’s diurnal cortisol (morning levels and diurnal slope)? 2) Do stable, distal factors, such as income and cumulative risk, predict trait variance in diurnal cortisol measures, whereas more time-varying and proximal factors, such as parenting, predict state variance? The current study models the stable and time-varying relations between cortisol patterns (measured as diurnal morning level and slope) with stable and time-varying contextual factors including income, cumulative risk and parenting.

Low Income, Cumulative Risk and HPA Axis Function

Studies of rodents and non-human primates indicate that early exposure a range of adverse environments result in immediate and long-term changes to the HPA axis activity and reactivity, as well as animal behavior (c.f. Sanchez, 2006). In humans, aberrant profiles may take the form of hypocortisolism, or low cortisol levels or responses, and hypercortisolism, in which an individual’s cortisol levels start elevated and remain elevated across the day. Both profiles are observed among individuals experiencing chronically stressful environments (c.f. Davies, Sturge-Apple, Cicchetti, & Cummings, 2007) and types and nature of environmental stressors that predict elevated versus blunted cortisol are not clearly understood.

One measure of environmental stress is a cumulative risk index, which captures the number of risk factors that an individual experiences across demographic, psychosocial, and environmental domains. Although there are studies that show high levels of economic or cumulative risk are associated with higher levels of cortisol in children (Blair, Granger, & Razza, 2005; Evans & English, 2002; Evans & Kim, 2007), both a review and a meta-analysis of the relation of risk to cortisol show that low income and high cumulative risk are associated with lower or blunted cortisol levels, whereas acute stress is associated with elevated diurnal levels (Dowd, Simanek, & Aiello, 2009; Miller, Chen, & Zhou, 2007). This finding is consistent with the allostatic load theory; after prolonged exposure to stress common to low-income, high-cumulative risk contexts, an individual, unable to generate the energy required to regulate and maintain an acute stress response, may shift from hyper-responding to a downregulated response. Of note, these reviews were comprised of studies from early childhood through adulthood. Therefore, while there is some evidence of these observed patterns in children (e.g. Evans & Kim, 2007; Badanes, Watamura, & Hankin 2011), the review findings are driven by adult studies (e.g. M_age = 38.39 in Miller et al., 2007) and it is less clear, developmentally, when these patterns may emerge in children.

The literature does not clearly address the extent to which risk exerts a stable influence over HPA axis functioning versus a more transient disruption the stress response system. There is some evidence that socioeconomic risk is exceedingly taxing to the HPA axis over time and related to its long-term stability (Evans & Kim, 2007), evidence that risk relates to the dynamic change of the stress response system, rather than its stability (Fernald & Gunnar, 2009) and research to suggest that stable, distal risk factors relate to stable, characteristics of HPA activity while more time-specific, proximal risk factors relate to contemporaneous states of HPA activity (Laurent et al., 2012). Further study on the contributioFns of environmental risk to the stability, as well as the variability, of cortisol over time, are needed. The current study seeks to address this gap in the literature, using four time points of cortisol data to examine whether income and cumulative risk predict stable, trait aspects of HPA axis functioning.

Parenting and HPA Axis Function

Parenting reflects a proximal environmental influence upon stress hormone functioning. Studies have suggested that maternal withdrawal and insensitivity relates to elevated baseline cortisol among infants (Bugental, Martorell, & Barraza, 2003; Haley & Stansbury, 2003), maternal negative affect relates to blunted diurnal slope among preschoolers (Zalewski et al., 2012), and decreased parenting quality, as indexed by lower involvement and lower warmth, relates to flatter diurnal slope in children in kindergarten and adolescence (Pendry & Adam, 2007). Early parenting appears to be a long-term predictor of HPA axis functioning, with studies finding, for example, that higher levels of maternal engagement during infancy predicted lower basal cortisol in toddlerhood (Blair et al., 2008) and higher levels of maternal insensitivity in the first 3 years of life predicted lower awakening cortisol at age 15 (Roisman et al., 2009).

While parenting behaviors have long-term predictive value, they are also thought to be variable over time (Dallaire & Weinraub, 2005). Providing some support to the notion that state parenting contributes to children’s contemporaneous or state cortisol, one study found that inconsistency in parenting, not parenting behavior itself, interacted with genetic and prenatal risk factors to predict cortisol variability across the first 4.5 years of life (Marceau et al., 2013). The current study examined whether parenting behaviors relate to stable, trait HPA axis functioning in preschool age children and/or explain time-varying variance in HPA axis functioning—that is, whether variations from one’s “typical” parenting experiences can account for deviation from one’s own “typical” HPA axis functioning.

State-trait Models of Cortisol

Few studies of children’s cortisol function have utilized state-trait models, which may have limited the field’s ability to evaluate predictors of diurnal cortisol. As highlighted by Kirschbaum and colleagues (1990), there are three sources of variance in cortisol: (1) trait factors related to internal personal factors, and we note, that these might also reflect stable aspects of an individual’s environment or context, (2) state factors related to time varying situational and/or person-situational interactions, and (3) error variance, which includes measurement error introduced by assay unreliability and unmeasured sources of variance. Collapsing across these sources of variance may not only obscure relations by reducing the power to detect meaningful differences, but also ignores potential specificity between the predictor and the portion of cortisol variance explained.

When it comes to prior LST studies, the time between assessments vary significantly, with some studies considering traits over days and weeks (Kirschbaum et al., 1990), as well over years (Wang et al., 2014). On balance, as is characteristic of a variable’s diminishing correlation over time, the longer the time between time points of assessment, the lower the trait-effect estimates. Further, stability estimates vary depending on the aspect of cortisol (e.g., morning v. evening level) being examined (Doane, Chen, Sladek, Van Lenten, & Granger, 2015; Stroud, Chen, Doane, & Granger, 2016; Wang et al., 2014).

Notably, few studies have examined stability and change in the HPA among children. On balance, in mid-childhood and adolescence, models suggest the majority of morning cortisol is attributed to state factors, with estimates including 50% of variance to 70% of variance, and a significant but smaller percent of variance (e.g. 41% and 28%, respectively) attributed to trait factors (Kertes & van Dulmen, 2012; Shirtcliff, Granger, Booth, & Johnson, 2005). Estimates of trait variance may be inflated by time frame, such that these estimates have captured epoch specific rather than true trait variance in cortisol. In terms of diurnal slope, state variance estimates have ranged widely, with estimates ranging from 35% to 76% of variance and non-significant trait factors have been reported (Ross et al., 2014). In a study of children followed from the third to ninth grade, comprising, developmentally, the sample nearest the present study, situation-specific influences comprised 51.9% of variability in cortisol levels, short-term or epoch-specific stability comprised 35% of the variance in cortisol levels, and cortisol that was stable across days and years (i.e., trait-like) comprised 13.2% of total cortisol level variability (Shirtcliff et al., 2012). It is important to note that few if any of studies have examined LST models in preschool age children, where proportions of variance accounted for by trait v. state factors might differ from older children given the shorter duration of contextual/environmental exposure to risk.

Few studies have gone beyond descriptive LST to examine predictors and correlates of stability and change to cortisol activity. In a sample of girls in early adolescence, assessed cortisol over 3 days, greater early adversity was related to a lower latent trait cortisol level Stroud et al., 2016). In a sample of adolescents over the transition to college, early life experiences of abuse were associated with higher latent trait cortisol (Doane et al., 2015).

A small number of studies have utilized hierarchical linear modeling to investigate within-individual predictors of change in cortisol level. Among infants and toddlers, within-child increase in chaos predicted contemporaneous increases in resting salivary cortisol (Berry et al., 2016). In a study of children at 6, 12, 24, and 36 months, children exposed to high levels of cumulative risk exhibited higher levels of cortisol across age and higher within-age cortisol reactivity (Holchwost, et al., 2017). Among children assessed from age 9 to 15 years old, exposure to typical levels of early life stress predicted moderately high trait-like morning cortisol and modest developmental change (Essex et al., 2011). In comparison, children exposed to higher levels of early life adversity demonstrated altered trait cortisol, especially in morning level, and wider swings in state HPA activity which also coincided with mental health symptoms. Finally, as referenced earlier, the experience of variability in parenting, rather than parenting itself, has been shown to interact with genetic and prenatal risk factors to predict cortisol variability in the first 4.5 years of life (Marceau et al., 2013). Taken together, these studies provide evidence for both a role of distal factors such as poverty and cumulative risk and proximal family factors such as parenting in predicting stable cortisol traits as well as time-specific deviations in HPA axis activity. The energy required to regulate and counter-regulate these challenges may contribute to a wear and tear on that body that results in not only altered ‘set points’ of the system, but contemporaneous mental health symptoms and illness possibly by way of allostatic load (Essex et al., 2011; Lupien et al., 2011).

No studies were identified that utilized LST modeling of preschool-age children’s diurnal cortisol to identify predictors of trait and state activity. This dearth of research within the preschool period is surprising given that exposure to high levels of cortisol early in development confers enduring alterations to activity and reactivity of the system. Indeed, the developmental origins of disease (Gluckman & Hanson, 2006) and biological embedding (Miller et al., 2011) theories argue that childhood stress is programmed into stress-sensitive systems early in life. In turn, these early life stress experiences endow an individual with responses that tend toward exaggerated or decreased stress responses and ultimately disease across the lifespan. Cortisol, which can cross the blood-brain barrier, is a likely candidate for sensitizing the developing brain (Gunnar & Quevedo, 2007). While alterations to the HPA axis are considered among the primary mechanisms through which stress is related to disease, nature and degree of influences associated with these alterations, particularly across development, remains unclear.

Current Study

This study sought to build upon existing studies linking children’s cortisol with known avenues of environmental influence—income, cumulative risk and parenting. This study is unique in investigating the extent to which the relations of the HPA axis and the environment may be understood as enduring associations among stable factors. Further, this study investigates whether deviations from an individual’s trait cortisol functioning may be understood by time-specific deviations in the individual’s experience of income, cumulative risk and parenting. Given the relative stability of experiences of low-income and cumulative risk and prior studies linking early life stress with trait cortisol (Doane et al., 2015; Stroud et al; 2016), we theorized that these contextual factors might relate to trait variance in cortisol. Given the relative malleability of aspects of parenting behaviors and their more proximal influence on a preschooler’s functioning, in combination with prior research showing variability in parenting predicts cortisol variability (Marceau et al., 2013), we theorized that time-varying parenting behaviors might explain time-varying variance in children’s cortisol. Based on prior review of the literature, we predicted that low income and chronic cumulative risk would be related to lower morning levels and flatter diurnal slope cortisol patterns (Dowd et al., 2009; Miller et al., 2007). Based on the mixed literature of cortisol and lower-quality parenting relating to both elevated levels of cortisol (Bugental, Martorell, & Barraza, 2003; Haley & Stansbury, 2003) and lower/flatter levels of cortisol (Pendry & Adam, 2007; Zalewski et al., 2012), we did not hold directional hypotheses with regard to the relation of warmth, negativity, responsiveness, scaffolding, and limit setting to children’s diurnal cortisol. We utilize a sample including families across the full range of income, oversampling families in lower income categories to provide a robust test of the effects of income and the accompanying cumulative risk factors.

Using a longitudinal, prospective design, we explored: (1) whether income and cumulative risk related to trait cortisol (as indexed by morning level and diurnal pattern across models); (2) whether time-specific variance in income and cumulative risk predicted time-specific or state variance in diurnal cortisol; (3) whether parenting factors predicted trait levels of diurnal cortisol; and (4) whether time-specific variation in parenting behaviors related to time-specific or state variance in diurnal cortisol.

Method

Participants

Study participants were 306 mothers and their 36-39 month old children (M=37, SD=0.84 mos.) who were recruited from various public- and privately-funded sources, including a university hospital birth register, daycares, preschools, libraries, health clinics, and charitable agencies and organizations serving low-income families. Families at these sites received information forms about the study and could indicate their interest in participating in the study on the information forms returned through their organization or mailed directly to the research project in postage-paid envelopes. Recruitment sites received an honorarium of $100 when 90% or more of their families returned the forms, regardless of the number of families indicating interest in participating. If a site returned 75% or 50% of the forms, the site received $75 or $50, respectively.

Families were recruited for participation so that there was equal representation across income levels to be able to rigorously test the effects of income. For recruitment, poverty status was determined using the 2009/2010 HHS Poverty Guidelines (Department of Health, n.d.) in place at the start of the study, which is an income-to-needs ratio based on the number of people in the home. The sample was evenly distributed across income levels, with 29% of the sample at or near poverty (N = 90 at or below 150% of the federal poverty threshold), 28% lower income (N = 84 above 150% of poverty threshold and below the local median income of $58K), 25% middle- to upper-income (N = 77 above the median income to $100K), and 18% affluent (N = 54 above $100K). To participate, families required reasonable proficiency in English to comprehend the assessment procedures, and children diagnosed with a developmental disability were excluded. Participants included 50% girls. According to mothers’ reports, the racial and ethnic composition of the sample of children included 64% European American, 9% African American, 3% Asian American, 10% Latino or Hispanic, 2% Native or American Indian, and 12% multiple racial and ethnic backgrounds or other. Mothers’ educational distribution included 3% mothers with some high school attainment, 6% completed high school, 34% with some college, technical school or professional school, 30% college graduates, and 27% with post-graduate education. Eighty-one percent of mothers were married or had long-time partners, 12% were never married, 7% were separated, divorced or widowed and were the single heads of household.

Procedures

Families were assessed in research offices on the university campus. They were assessed at four time points separated by nine months when children were 36-40, 45-49, 54-58, and 63-67 months. At the beginning of each assessment, following the guidelines stipulated by the Institutional Review Board, both active parental consent and child assent were secured prior to data collection. A team of trained experimenters administered questionnaire measures. Mothers and their child engaged in parent-child interaction tasks. Families were compensated $70 for their first visit to our research offices, and compensation increased by $20 for each subsequent visit.

Across all four time points, mothers were oriented or retrained in the collection of child cortisol and were given a home collection kit and instructions to collect the saliva samples at home. Mothers were instructed to collect their child’s saliva 30 minutes after the child woke in the morning and 30 minutes prior to bedtime for three consecutive days. A staff member called families on the first night to ensure proper collection and answer questions. A reminder call was placed on the third evening to prompt mothers to return the packets via the mail. Mailing saliva has been shown not to influence saliva collection (Clements & Parker, 1998) and this method has been successfully used in childhood samples (Bruce, Davis, & Gunnar, 2002). Parents were paid an additional $30 for all cortisol packets returned. Families were invited to collect cortisol regardless of their ability to attend the laboratory visit at that time point.

Measures

Income

At each time point, mothers reported on household income from all sources on a 14-point Likert scale that provided a fine-grained breakdown of income at the lower levels facilitating identification of families at the federal poverty cutoff using an income to means ratio (e.g. 1 = $14,570 or less, 2 = $14,571-$18,310, 3 = $18,311-22,050, etc.). The 14-point variable representing the full range of income was used. The mean income at time 1 was 8.75 (SD = 3.93, Range = 1.00 – 14.00, 8 = $35,601-$39,200).

Cumulative risk

At time 1, cumulative risk included eight factors: Low maternal education, adolescent parent status, single parent status, family structure transitions, household density, residential instability, negative life events, and maternal depression. At times 2 – 4, the cumulative risk score excluded factors that were immutable across time (i.e. adolescent parent status). At each time point, a total cumulative risk score was the sum of all of the component factors. Dichotomous scores were scored 0 = not present, 1 = present. Continuous scores were converted into proportions of the total possible score so that they ranged from 0 to 1, and thus, were weighted equally with the dichotomous variables.

Mothers reported on their educational attainment. Risk was indicated by mothers’ not graduating from high school. Mothers reported their age at the time of the study child’s birth, and were considered an adolescent parent if they were 19 years or younger when the child was born. At each visit, mothers reported on their marital status, and families were identified as single parent families if the mother indicated she was never married, currently widowed, separated or divorced, or living for less than 1 year with a live-in partner. Mothers’ report of divorce was scored as family transition risk. Household density was the ratio of number of people in the home to the number of rooms in the home. Mothers reported on the number of moves for a measure of residential instability. At time 1, residential instability risk was 2+ moves in the child’s lifetime. At time 2-4, the presence of a move since the last assessment constituted residential instability. Negative life events were assessed with parent report on the General Life Events Schedule for Children (Sandler, Ramirez & Reynolds, 1986), previously shown to have significant associations with child adjustment (Lengua & Long, 2002). The 29 events include a range of moderate to major negative events including changing schools, death of a family member or friend, parental arrest, loss of friends or pets. Parents reported the occurrence of events within the previous 9 months, and total scores were the number of events that occurred. The total score was converted into a proportion of the possible 29 events.

Mother reported on their own depressive symptoms over the previous month using the 20-item Center for Epidemiological Studies–Depression Scale (CES-D, Radloff, 1977), a widely used self-report scale designed to measure depressive symptoms in the general population. Participants indicate whether each symptom was present on a scale of 0 (rarely or never) to 3 (most of the time), and the items were summed for a total score, with higher scores indicating higher levels of depression. Internal consistency ranged from .88-.91 across the 4 time points. The total score was converted into a proportion of the total possible score of 60.

Parenting

At time 1 mothers and children engaged in 4 activities (7 minutes restricted play, 7 minutes unrestricted, 7 minutes instructional activity, 3 minutes clean-up; Kerig & Lindahl, 2000). In restricted play, mothers were instructed to allow children to play with toys in the room except those in a specified place, an accessible shelf of highly desirable toys. This was followed by free play in which mothers and children were informed that they could now play with the previously restricted toys. Next mothers were instructed to help children build a challenging Lego figure. Finally, mothers were to obtain children’s assistance in cleaning up. At assessment time points 2-4, the first two activities were modified to be 7 minutes of child-directed play and 7 minutes of mom-directed play. During the child directed play segment, mothers were encouraged to take their child’s lead. This was followed by mom-directed play in which mothers were asked to change and have the child follow their lead.

Coded behaviors were selected a priori based on existing literature. Warmth, negativity, limit setting, scaffolding, and responsiveness were coded in 1-minute epochs for all segments, and then averaged across epochs and across segments. Parenting was coded from videotapes by advanced undergraduates using a coding system that was adapted from established coding systems: the System for Coding Interactions and Family Functioning (SCIFF: Lindahl & Malik, 2000), the Parenting Style Ratings Manual (Cowan & Cowan, 1992), and the Parental Warmth and Control Scale–Revised (Rubin & Cheah, 2000), and used previously by this research team (Lengua, Kiff, Moran, Zalewski, Thompson, Cortes, & Ruberry, 2014). All behaviors were rated on 6-point scales (0=absent/lowest, 5=highest). Positive affect captured the frequency and level of behavioral and verbal expressions of happiness, comfort, connection, and warmth toward the child. Interactiveness assessed the quantity of verbal and non-verbal engagement. Positive affect and interactiveness were combined into a measure of warmth. Negativity assessed the negative tone or tension expressed by the mother and included verbal and non-verbal expressions of irritation or frustration with the child that were critical, rejecting or invalidating. Limit setting assessed mothers’ clarity, consistency, and follow-through of directives coded contingently when children were noncompliant, oppositional, or disruptive. Scaffolding was a combination of guidance/structuring, encouragement of autonomy, and low negative or intrusive control. Responsiveness to children’s expressions of negative affect were coded contingently as indicator of mothers’ sensitivity to cues of the child. Inter-rater reliability was assessed by independent recoding of 20% of the interactions. For each parenting dimension, the average and range of intra-class correlations (ICCs) across time points were: warmth (ICC_mean = .85, ICC_range = .80-.90), negativity (ICC_mean = .80, ICC_range = .75-.86), scaffolding (ICC_mean = .84, ICC_range = .81-.89), limit setting (ICC_mean = .75, ICC_range = .63-.83) and responsiveness (ICC_mean = .76, ICC_range = .67-.86). Of note, the two scales with ICCs <.80 were the two contingently-coded scales (limit setting and responsiveness), coded only in cases of child non-compliance (limit setting) and child negative affect (responsiveness). While the moderate reliability of these scales should be noted, the ICCs are based on fewer cases within subject. With regard to stability/variability of the parenting dimensions: The lowest and highest correlations with the observed parenting variables were r = .00 to .30 among responsiveness scores (lowest correlations over time) and r = .29 to .50 in negativity scores (highest correlations over time).

Cortisol

Saliva samples were sent to the University’s Biobehavioral and Nursing Systems Laboratory for processing (for detail, see Zalewski, Lengua, Thompson, & Kiff, 2016). Samples were assayed using the High-Sensitivity Cortisol Salivary Enzyme Immunoassay Kit provided by Salimetrics LLC (State College, PA). The sensitivity of this kit ranges from 0.005 to 2.5 mg/dl. All samples from the same subject for each set of saliva were included in the same assay batch to minimize interassay within-subject variability. Each time point was assayed after all cortisol had been collected. At Time 1, the intraassay cortisol value (CV) was 3.98% and the interassay CV was 2.78%; Time 2 intraassay CV ¼ 3.82% and interassay CV ¼ 4.9%; Time 3 intraassay CV ¼ 3.35% and interassay CV ¼ 4.15%; Time 4 intraassay CV ¼ 3.73% and interassay CV ¼ 4.0%, all acceptable values. Assay results that were over 2.0 μg/dL were deemed biologically implausible and the values were not used, consistent with methods used in other studies (Ashman, Dawson, Panagiotides, Yamada, & Wilkinson, 2002). Only one case was fully discarded because all cortisol values for the time point were over 2.0 μg/dL. At each time point, the assay results for all three mornings (T1: M = 0.29, SD = 0.21; T2: M = 0.31, SD = 0.18; T3: M = 0.26, SD = 0.18; T4: M = 0.25, SD = 0.15) and all three evenings (T1: M = 0.13, SD = 0.18; T2: M = 0.08, SD = 0.10; T3: M = 0.08, SD = 0.12; T4: M = 0.06, SD = 0.09) were utilized. As is common with cortisol data, values were positively skewed, and log transformations were applied to the raw morning values for the measure of morning cortisol. A diurnal slope was computed by subtracting the raw evening value from the raw morning value and dividing this number by the time between samples. This rate of change variable was then log transformed. Data were one, morning cortisol value and one, diurnal slope value for three days per each of the four time points. Values in samples that had been collected 90 minutes after wake up or prior to bedtime were discarded, with 74 samples excluded, were a sample represents a single instance of sampling on 1 day of any of the 3 days across the 4 timepoints. Of note, models presented here did not vary in pattern or magnitude from models using a more conservative sample time cut-off of ± 3SD from mean sample time, which excluded 141 instances of cortisol collection. As such, models where presented reflect the 90-minute cut-off to maintain a larger portion of the sample and consistency with prior research with this sample.

Analytic Plan

State-trait models decompose variance of observed measures, in this case, cortisol samples, as variance that is consistent across observations that is accounted for by a latent, trait factor, and variance in individual observations that is specific to that time point and accounted for by a latent, state factor. In other words, a state-trait model captures both the stability of a construct within an individual (the trait) and occasion-specific variation around one’s own average (the state). In this study, both trait and state diurnal cortisol factors were predicted by both characteristic factors of cumulative risk and parenting and their time-specific indicators.

Study models were tested using Mplus 6.0 (Muthén & Muthén, 2010) and Full Information Maximum Likelihood (FIML) estimation to handle missing data. For the income, cumulative risk, and parenting latent factor models, there was only one indicator per time-point. As such, for the latent model, variances were fixed. For the cortisol latent state-trait models, there were three indicators per time-point, allowing the latent factors and residual variances to be freed and estimated. The time-point factor loadings were set to 1.0 such that the metric and contribution of each indictor was equal, as is appropriate to the specification of latent state-trait models (Kenny & Zautra, 2001). Residual covariances among the factor indicators were fixed to zero. The chi-square statistic was utilized to test model fit. However, this statistic is sensitive to both sample size and violations of normality. Therefore, as is common practice, Root Mean Square Error Approximation (RMSEA) and the Comparative Fit Index (CFI) were used to supplement chi-square statistics of model fit (Hu & Bentler, 1999).

Missing Data

As a conservative estimate of missing cortisol data, we computed missing data for any family not returning a cortisol sample, even if the family did not attend their laboratory visit. For 306 families, at time 1, 33 families (10.78%) did not return any samples, at time 2, 42 (13.73%) families did not return samples, at time 3, 38 (12.42%) families did not return samples, and at time 4, 47 (15.36%) families did not return samples. Therefore, we had usable cortisol values for 287 preschoolers.

Attrition across the study was very low, with 95% of participants remaining between Time 2-4. At T1 all participants had complete income data, 96% of participants had complete cumulative risk scores, and 94% had parenting data. Analyses were conducted to assess the degree of bias introduced by missing data. We compared children with complete cortisol data to children missing cortisol data on measures of time 1 income, cumulative risk, and the parenting dimensions. Participants missing cortisol data differed from those not missing cortisol data in that they had higher cumulative risk (M missing = 1.00, M no missing = 0.77, t[304] = −2.53, p > .001), lower responsiveness (M missing = 4.28, M no missing = 4.55, t[287] = 3.02, p > .001) and lower scaffolding (M missing = 3.41, M no missing = 3.62, t[286] = 3.08, p > .01). The effect sizes of missingness were modest (r = .14, r = −.18, and r = −.18, respectively) and therefore were not likely to introduce substantial bias (c.f. Collins, Schafer, & Kam, 2001). Participants missing parenting data did not differ from those with complete parenting data on measures of time 1 income and cumulative risk. Thus, it appears that little bias was introduced due to missing data.

Results

Preliminary Analyses of Potential Covariates

Variables indicating the time of sampling and the latency from children’s wake time to morning collection and from evening collection to bedtime were calculated from mother’s reports. The questionnaires were reviewed to ensure compliance. In addition, mothers were given a phone call on the first evening of collection to review the collection procedures and answer any questions. Average morning collection times ranged between 8 a.m. to 8:12 a.m. across Time 1–Time 4. Average latencies to collect these morning samples ranged from 27 to 40 min after awakening across Time 1–Time 4 (SD_range = 10-22 minutes). Average evening collection times were 8:19 p.m. to 8:27 p.m. across Time 1–Time 4. On average, these samples were collected between 31 and 49 min before bed across Time 1–Time 4 (SD_range = 12-19 minutes). The sampling times and latencies on each day were correlated with the cortisol morning value and diurnal slope from the corresponding day. There were few and modest associations among collection times and cortisol levels. The correlations of morning sample time of day with morning cortisol and diurnal cortisol and evening sample time of day with diurnal cortisol, were examined resulting in 36 correlations. Five correlations were significant. T1 day 1 morning sample time of day was associated with morning cortisol value r = −.17, p = .01 and diurnal cortisol value r = −.15, p = .02. T1 day 2 morning sample time of day was associated with morning cortisol value r = −.16, p = .01. T2 day 2 morning sample time of day was associated with morning cortisol value r = −.19, p = .003. T3 day 3 morning sample time of day was associated with morning cortisol value r = −.14, p = .03. As such, these 5 time of day variables were included in the models as potential covariates. Including only the significant compliance variables in the models is consistent with the latent state-trait cortisol measurement approach of other authors (c.f. Stroud et al., 2016). To assess correlations with latency to collect sample from wake/prior to bed, we examined the correlations of the latency to collect morning cortisol with morning and diurnal cortisol values and latency to collect evening sample with diurnal slope, resulting in 36 correlations. Three correlations were significant. T1 day 3 latency to collect morning sample was associated with morning cortisol value r = −.16, p = .02, T3 day 1 latency to collect evening sample was associated with diurnal cortisol value r = −.17, p = .02, and T3 day 3 latency to collect evening sample was associated with diurnal cortisol value r = .20, p = .003. As such, these three latency variables were included in the models as potential covariates.

Similarly, the use of steroid medications or inhaler and health status has been shown to affect cortisol levels. Mothers completed a daily questionnaire regarding their children’s health and medication use sampling days. Mothers were instructed to avoid sampling when their children were using steroid-based medications or were ill. Mothers were mailed additional materials if they accidently sampled when the child was ill. Regardless, the following health and medication variables were examined in relation to cortisol values: inhaler/steroid medication use, non-steroid medication use, and 13 health symptoms. Each possible covariate was examined with the corresponding morning and diurnal cortisol values, resulting in 408 correlations. There were 20 significant correlations, roughly the frequency of chance. There were no systematic patterns to the significant correlations and they were modest in magnitudes (r = .14 to .24). Therefore, no health or medication compliance variables were included as covariates.

Next, given the research suggesting that there may be sex differences in diurnal cortisol, even among pre-pubertal children, (Rosmalen, Oldehinkel, Ormel, Winter, Buitelaar, & Verhulst, 2005), child sex was examined in relation to diurnal cortisol. Child sex was correlated with a single morning sample, day 2 time 2 (r = −0.13, p = .04) and unrelated to any diurnal slope value. These correlations are in line with other studies that have found cortisol to be unrelated to sex in pre-pubertal children (e.g. Klimes-Dougan et al., 2001). Nevertheless, given covariance of child sex and parenting, child sex was retained as a variable in all models.

State-trait Models

We examined the latent models of income, cumulative risk and each dimension of parenting behavior to assess the feasibility of decomposing these variables into latent factors. Model fit and factor loadings are presented in Table 1. Fit statistics indicated that the initial models fit the data well and that each factor loading was significant, with the exception of the latent variable of income, where the overall model did not fit the data. This is not surprising given that income was highly stable across the study (correlations ranging from .80 to .89 across time points). Income, when measured using the T1 variable, correlated with few cortisol variables: income correlated with a single morning sample, day 1 time 4, (r = 0.16, p = .01). Income was related to diurnal slope for each day at the fourth time point (r = 0.14, p = .04, r = 0.15, p = .04, r = 0.14, p = .04, respectively) but at no other time points. Low income related to a flatter diurnal slope at this time point. As such, observed T1 income rather than a latent income factor was included as a covariate in all subsequent models.

Table 1.

Factor loading and fit statistics for state-trait models

	Factor Loading	χ2	df	RMSEA	CFI
Morning Level		142.12	117	0.03	0.93
Time 1	0.62
Time 2	0.43
Time 3	0.54
Time 4	0.36
Diurnal Slope		104.07*	75	0.04	0.90
Time 1	0.68
Time 2	0.59
Time 3	0.67
Time 4	0.70
Income		33.44***	5	0.14	.98
Time 1	.90
Time 2	.95
Time 3	.92
Time 4	.93
Cumulative Risk		6.04	5	0.03	1.00
Time 1	0.69
Time 2	0.87
Time 3	0.88
Time 4	0.90
Warmth		6.79	5	0.03	0.99
Time 1	0.52
Time 2	0.64
Time 3	0.69
Time 4	0.65
Negativity		0.50	5	0.00	1.00
Time 1	0.61
Time 2	0.54
Time 3	0.58
Time 4	0.64
Scaffolding		1.37	5	0.00	1.00
Time 1	0.47
Time 2	0.58
Time 3	0.62
Time 4	0.70
Responsiveness		9.88	5	0.05	0.89
Time 1	0.33
Time 2	0.51
Time 3	0.45
Time 4	0.44
Limit Setting		9.96	5	0.05	0.92
Time 1	0.41
Time 2	0.42
Time 3	0.53
Time 4	0.45

Next we examined the latent state-trait models of cortisol to decompose these variables into state, trait, and error variance. The state-trait models for morning level χ² (117 = 142.12, p = .06; RMSEA = 0.03, CFI = 0.93) including the compliance variables (latency to collect and time of day), child gender and income demonstrated good model fit. Compliance variables loaded significantly on their respective cortisol value and were retained. Child gender was unrelated to morning and diurnal cortisol factors but was retained in the model given the association with predictor variables. Similarly, low income did not relate to lower trait morning level (β = 0.17, p = 0.08), but was retained in the model given the known associations with cumulative risk and parenting. For morning level, T1-T4 state factors explained between 13% and 38% of variance (R² = 0.38, 0.18, 0.29, and 0.13, respectively, all p ≤ .05), and the trait factor accounted for 5.8% of variance (ns).

In the full, latent state-trait model of diurnal slope, putative compliance covariates failed to significantly predict their respective cortisol values and were therefore trimmed from the model. Low income was related to trait lower diurnal cortisol (β = 0.23, p = 0.01). The diurnal slope model including child gender and income demonstrated good fit, χ² (75 = 104.07, p = .01; RMSEA = 0.04, CFI = 0.90). For diurnal slope, T1-T4 state factors accounted for between 35% to 49% of variance (R² = 0.46, 0.35, 0.44, and 0.49, respectively, all p ≤ .05), and the trait factor accounted for 5.3% of variance (ns). The adequate overall fit of these preliminary models supported the plausibility of testing the state relations among cortisol, cumulative risk, and parenting and examining if the addition of parameters to the model would increase the power to detect and predict systematic variance in the trait factor.

Predictors of Trait Cortisol Factors

To test the trait and factor-level relations of cortisol with cumulative risk and parenting, a series of models were tested. First, factor-level cumulative risk was tested as a predictor of trait morning cortisol level. Next, factors of each parenting dimension were tested as a predictor of trait morning level above the effects of factor-level cumulative risk. This resulted in six models, one for cumulative risk alone, and one for each of 5 parenting dimensions, which were repeated with the trait measure of diurnal slope. Of note, with the introduction of the additional parenting and risk factors, the variance accounted for by trait cortisol, while still small (e.g. R²= .20, p = .04 for trait variance in morning cortisol including income, risk and parenting; R²= .25, p < .01 for trait variance in diurnal cortisol slope), reached significance and supported the viability of testing predictors of trait variance. Model fit statistics for each model are presented in Table 2 (morning level) and Table 3 (diurnal slope). Standardized coefficients for models with significant prediction of trait and state diurnal cortisol are presented in Table 4.

Table 2.

Fit statistics for trait and state models predicting morning level

Model	df	χ2	RMSEA	CFI
Morning Level: Trait Models
Cortisol & CR	191	247.53**	0.03	0.96
Cortisol, CR & Maternal Warmth	281	353.82**	0.03	0.95
Cortisol, CR & Maternal Negativity	281	434.49***	0.04	0.90
Cortisol, CR & Scaffolding	281	402.07***	0.04	0.92
Cortisol, CR & Responsiveness	281	401.69***	0.04	0.92
Cortisol, CR, & Limit Setting	281	382.70***	0.03	0.93
Morning Level: State Models
Cortisol & CR	187	241.02***	0.03	0.96
Cortisol, CR & Maternal Warmth	277	339.82**	0.03	0.96
Cortisol, CR & Maternal Negativity	277	427.88***	0.04	0.91
Cortisol, CR & Scaffolding	277	398.55***	0.04	0.93
Cortisol, CR & Responsiveness	277	399.13***	0.04	0.91
Cortisol, CR, & Limit Setting	277	374.96***	0.03	0.93

Table 3.

Fit statistics for trait and state models predicting diurnal cortisol

Model	df	χ2	RMSEA	CFI
Diurnal Slope: Trait Models
Cortisol & CR	133	187.99***	0.04	0.95
Cortisol, CR & Maternal Warmth	207	268.51**	0.03	0.96
Cortisol, CR & Maternal Negativity	207	345.33***	0.05	0.91
Cortisol, CR & Scaffolding	207	338.41***	0.05	0.91
Cortisol, CR & Responsiveness	207	297.53***	0.04	0.93
Cortisol, CR, & Limit Setting	207	302.744***	0.04	0.93
Diurnal Slope: State Model
Cortisol & CR	129	179.05***	0.04	0.96
Cortisol, CR & Maternal Warmth	203	261.47**	0.03	0.96
Cortisol, CR & Maternal Negativity	203	343.53***	0.05	0.91
Cortisol, CR & Scaffolding	203	329.96***	0.05	0.92
Cortisol, CR & Responsiveness	203	293.73***	0.04	0.93
Cortisol, CR, & Limit Setting	203	296.53***	0.04	0.93

Note.

^*p < .05,

^**p < .01,

^***p < .001. CR = Cumulative risk. Bolded text reflects the variable predicting state variance.

Table 4.

Standardized coefficients for models with significant prediction of trait and state cortisol morning level

Morning Level: Trait Models	CR	CR & Responsiveness	CR & Negativity	CR & Scaffolding
Income on Trait CR	−0.65***	−0.65***	−0.65***	−0.65***
Child Sex on Trait CR	−0.07	−0.07	−0.07	−0.07
Income on Trait Cortisol	−0.01	−0.13	−0.14	−0.19
Child Sex on Trait Cortisol	−0.18	−0.18t	−0.19t	−0.16t
Income on Trait Parenting	–	0.45***	−.50***	0.58***
Child Sex on Trait Parenting	–	−0.03	0.00	−0.05
Trait CR on Trait Cortisol	−0.32*	−0.25t	−0.19	−0.19
Trait Parenting on Trait Cortisol	–	0.31t	−0.38*	0.42*
Morning Level: State Models	CR & Warmth	CR & Negativity	CR & Limit Setting
Income on Trait CR	−0.65***	−0.65***	−0.65***
Child Sex on Trait CR	−0.07	−0.07	−0.07
Income on Trait Cortisol	−0.03	−0.11	−0.02
Child Sex on Trait Cortisol	−0.17t	−0.18t	−0.17t
Income on Trait Parenting	0.38***	−0.50***	0.31***
Child Sex on Trait Parenting	0.04	0.01	−0.11
Trait CR on Trait Cortisol	−0.25t	−0.20	−0.28t
Trait Parenting on Trait Cortisol	−0.25t	−0.25	0.15
T1 State Parenting on T1 State Cortisol	−0.24**	0.00	−0.10
T2 State Parenting on T2 State Cortisol	−0.21*	0.01	−0.17*
T3 State Parenting on T3 State Cortisol	0.02	−0.13	0.14
T4 State Parenting on T4 State Cortisol	0.09	−0.22*	−0.04

Note.

^t<.10,

^*p < .05,

^**p < .01,

^***p < .001. CR = Cumulative Risk.

Income strongly predicted factor-level cumulative risk (β = −0.65, p > 0.001), and with factor-level cumulative risk in the model, income did not predict trait cortisol morning level (β = −0.03, p = .85). Higher factor-level cumulative risk predicted lower trait cortisol morning level above the effects of income (β = −0.30, p = 0.03). Above the effects of income and factor-level cumulative risk, lower factor-level parent negativity (β = −0.38 p = .01) and higher characteristic scaffolding (β = 0.42, p = 0.01) predicted higher trait morning cortisol level.

Higher factor-level cumulative risk predicted lower trait diurnal slope (β = −0.53, p < 0.001). A higher value for slope indicates a diurnal pattern characterized by a steep decline or decrease in cortisol level over the course of the day, whereas a lower value of slope indicates a diurnal pattern with less decline over the day. When accounting for factor-level cumulative risk, income no longer predicted trait diurnal slope (β = −0.10, ns). Of the factor-level parenting dimensions, higher characteristic maternal responsiveness (β = 0.30, p < 0.05) and lower characteristic negativity (β = −0.35 p = 0.01) predicted higher trait diurnal slope.

Predictors of State Cortisol Factors

The next series of models tested the relations of state-level cortisol with time-specific cumulative risk and parenting. First, time-specific cumulative risk was tested as a predictor of time-specific state variance in cortisol morning level above the variance accounted for by the cumulative risk factor and trait morning level. Next, time-specific parenting was tested as a predictor of state variance in morning level above the effects explained by time-specific cumulative risk and the three (cortisol, cumulative risk, and parenting) factors. This resulted in six models tested for cortisol morning level, and another six for diurnal slope. Model fit statistics for each state model are presented in Table 2 (morning level) and Table 3 (diurnal slope).

Above the effects of income, trait cortisol, and the cumulative risk factor, time-specific T1 maternal warmth (β = −0.24, p = 0.01) predicted T1 state morning level. T2 time-specific maternal warmth as well as T2 maternal limit setting predicted T2 state morning level (β = −0.21, p = 0.02, and β = −0.17, p < 0.05, respectively). T4 time-specific maternal negativity predicted T4 state morning cortisol level (β = −0.22, p = 0.02). Please note that these are not classically directional findings. Rather, these predictions of state level variance should be thought of as a residual from an individual’s trait experiences. For example, a child’s time-specific experience of maternal warmth at T1, over their experience of T1-T4 characteristic warmth, predicted their T1 state morning cortisol level, over their T1-T4 trait morning level. The negative beta indicates that more warmth at T1, relative to an individual’s characteristic experience of warmth, predicted a lower morning cortisol level at that point, relative to the individual’s trait morning level.

Above the effects of income, trait cortisol, and factor-level cumulative risk, time-specific T1 maternal warmth (β = −0.22, p = 0.03) and scaffolding (β = −0.23, p = 0.01) significantly predicted T1 state diurnal slope. T2 time-specific maternal limit setting predicted T2 state diurnal slope (β = −0.20, p = 0.02). State diurnal slope values reflect the slope or rate of decline of the individual’s cortisol at that time point relative to an individual’s typical slope or rate of diurnal decline.

Discussion

This study sought to model both the stable, trait variance and time-varying, state variance in diurnal cortisol. Further, this study tested contextual factors (income, cumulative risk) and parenting as predictors of both state and trait variance in the measures of diurnal cortisol. The state-trait models of children’s cortisol morning level and diurnal slope, measured at four time points across 9-month intervals, fit the data well. This framework adequately decomposed children’s HPA axis functioning into variance attributable to trait variance (that is, variance attributable to stability across the longitudinal study) and state variance (that is, variance attributable to time-varying change across observations). State variance accounted for between 13-38% of morning level and 32-53% of diurnal slope. Trait variance did not account for significant variance in morning level (5.8%) or diurnal slope (4.6%). Few studies have examined within-person change in cortisol longitudinally using freely estimated latent state-trait modeling, and no studies were identified that examined these components of cortisol activity within the preschool period. Shirtcliff (2012), in a longitudinal model of 3^rd graders followed through the 9^th grader, found similar levels of cortisol state stability (~35%) to those in the present study, and somewhat greater trait variance of cortisol (13.2%). The relatively lower trait value observed in the current study may be explained by a developmental difference, as the literature on predictors of stability and change in cortisol in this period of development remains sparse. The trait value may alternatively be more consistent with other studies finding little evidence for stable trait-like variance in cortisol (Ross et al., 2014). Finally, the smaller trait variance in cortisol may be the result of the adversity represented in the sample, which captures a greater range in chaos and instability and might manifest in a decreased trait component of cortisol levels. Many prior studies of LST cortisol model are limited by having only 2 cortisol samples at each time point. Latent factors are ideally specified by a minimum of 3 indicators allowing variances of the factors and errors to be estimated. In models with 2 indicators there is no error variance, which is why the studies account for 100% of variance and may explain why state variance in the current study is lower than that reported by studies using 2 indicators. Notably, although state factors represented a more substantial and significant proportion of the variance of preschoolers’ cortisol values, trait factors, while accounting for a smaller percentage of variance, might have more relevance in explaining the effects of contextual risk factors on children’s adjustment (Shirtcliff et al., 2005). Further research unpacking the state and trait components of cortisol, as well as predictors of within-person stability and change, are needed.

With regard to correlates of state and trait variance, we found evidence that income and cumulative risk, both indicators of stable, pervasive contextual risk, were related to trait levels of morning and diurnal cortisol, with little evidence of time-varying effects. Conversely, for parenting, factor levels of parenting predicted trait levels of diurnal cortisol, and in addition, time-varying levels of parenting predicted time-varying state levels of diurnal cortisol. Although that was the general observed pattern, there was little consistency across time points in the relations of specific measure of diurnal cortisol (morning level or diurnal slope) and specific parenting behaviors. On balance, the findings suggest that characteristic experiences of cumulative risk and parenting contribute to children’s stable or trait-HPA axis function while time-specific deviations of parenting experiences contribute to contemporaneous variations in HPA activity.

With regard to trait cortisol and risk, income accounted for significant variance in preschooler’s diurnal cortisol, and above the effects of income, the cumulative risk factor accounted for significant variance in preschoolers’ morning level of cortisol as well as their diurnal cortisol. Notably, when accounting for characteristic cumulative risk, income was no longer a significant predictor of trait diurnal slope. This finding is in line with research showing that cumulative risk mediates the effects of income on child outcomes (Lengua, Moran, Zalewski, Ruberry, Kiff, & Thompson, 2014). Stable experiences of high cumulative risk accounted for a characteristic low morning levels and lower diurnal slope. In general, a strong morning cortisol level is thought to be essential for mobilizing metabolic and cognitive processes (e.g. de Kloet, 1991). Similarly, a strong diurnal rhythm (high morning levels and a negative slope to low evening levels) is considered normative, and weak or absent diurnal rhythm is considered a sign of a dysregulated system (Adam & Gunnar, 2001). Overall, the pattern of findings observed in the present study (low morning levels, lower diurnal slope) is consistent with prior research showing that income and cumulative risk are related to a flatter diurnal slope (Dowd, Simanek, & Aiello, 2009; Miller, Chen, & Zhou, 2007) and further, that chronic exposure to poverty and the stability of cumulative risk experiences predict children’s cortisol levels (Evans & Kim, 2007; Laurent et al., 2013). These relations might reflect children being unable to mount a full cortisol response or allostatic load (McEwen, 2004). It cannot, however, be ruled out that the flattened slope may represent high, stable levels over low “blunted” cortisol in some children in this sample. However, the relations of low income and high cumulative risk with low morning levels suggest this would be a less common observation. The findings that risk and parenting (discussed below) predict characteristic, trait cortisol activity as early as the preschool period is of concern when considering to potential for the biological embedding of stress or the developmental origins of disease. From a practical standpoint, the findings highlight the benefit of early intervention when attempting to address long-term effects of contextual risk.

With regard to trait cortisol and parenting, parenting factors predicted children’s characteristic morning level and diurnal slope. Specifically, above the effect of income and cumulative risk, stable, high maternal negativity predicted stable, low morning cortisol, whereas stable, high maternal scaffolding predicted stable, high morning cortisol. Stable, low maternal responsiveness and stable, high maternal negativity predicted a characteristic, flattened diurnal slope. The patterns of associations for factor-level parenting correspond with previous findings showing lower-quality parenting relates to low morning cortisol levels and blunted diurnal slopes (Pendry & Adam, 2007; Zalewski et al., 2012), although those findings were not framed as tests of stable levels of parenting. In the current study, mothers who characteristically had difficulty with scaffolding (i.e., providing guidance and structure while respecting her child’s autonomy), mothers with characteristically low responsiveness (i.e., contingent, timely, and appropriate responding to children’s cues), and mothers with higher levels of negativity (i.e., negative affect, invalidation, as well as harsh and critical behaviors), had children with stable low morning cortisol levels and flattened diurnal slopes. These findings suggest that stable patterns of parental behaviors predict trait-like or stable functioning of the HPA-axis system. Further, when the factor of parenting was entered into the morning cortisol models, the magnitude of the effect of the cumulative risk factor decreased and became nonsignificant. This finding, while not an explicit test of mediation, is in line with research suggesting that parenting mediates the relations of risk and cortisol activity (Zalewski et al., 2012).

With regard to cortisol state findings, variability in children’s HPA axis functioning was not predicted by time-specific cumulative risk above the effects of T1 income, trait cortisol, and the cumulative risk factor. It is likely that the high cross-time correlations of cumulative risk resulted in most of the effects of cumulative risk being accounted at the factor level. However, time-varying effects of parenting were observed. Time-specific maternal responsiveness, negativity, and scaffolding predicted children’s state morning cortisol. Time specific maternal warmth, scaffolding, and limit setting predicted children’s state diurnal slope. That is, deviations from characteristic experiences of these parenting behaviors predicted deviations from characteristic HPA activity. These results should be interpreted cautiously as each of these time-specific effects were observed at only one time point. These findings suggest that characteristic patterns of parenting largely shape children’s HPA-axis functioning, however, variations in parenting behaviors from characteristic levels might account for state level fluctuations in HPA-axis functioning. It may be that marked or notable deviations in parenting from typical levels are stressful to children, who thrive on consistency and predictability. Highlighting the value children place on consistent parenting, the predictability hypothesis contends that children in inconsistent parenting contexts will engage in oppositional and defiant behaviors designed to elicit predictable parental responses (Whaler & Dumas, 1986). While this theory is generally applied to the inconsistent application of discipline strategies, it is possible that any deviation or inconsistency in a parents’ behavior is distressing to a child (e.g. Marceau et al., 2013), and distress is captured by this study as deviations to typical diurnal cortisol. It is further possible that chronic instability in parenting experiences, captured in this study as a time-specific change in parenting, contributes to allostatic load. Further study of the coupling of parenting and children’s cortisol functioning over time is warranted.

The potential for variability in parenting to exert contemporaneous influence on children’s cortisol can also be interpreted positively: This study found that parenting behaviors, even in the context of low income and high cumulative risk, had the ability to exert stable influence on preschoolers’ HPA axis functioning, and the ability to have time-specific influence on preschoolers’ morning and diurnal cortisol. The current study findings contribute to a larger literature on parenting behaviors as predictors, mediators, and moderators the effects of risk on young children’s stress neurobiology (e.g. Hagan, Roubinov, Gress-Smith, Luecken, Sandler, & Wolchik, 2011; Zalewski et al., 2012). In combination with research noting that parent effects may attenuate over development as parents become less central and exclusive influences on children (e.g Eisenbarth, 2017; Goodman et al., 2011), the current study suggests that parenting behaviors may be a promising target early for interventions for children in contexts characterized by pervasive contextual risk where ‘wear and tear’ on stress neurobiology and the embedding of risk is more likely.

Strengths and Limitations

Using a large sample and a flat income distribution that overrepresented low-income families, facilitated a rigorous test of the effects that income and cumulative risk may have on children’s HPA-axis was facilitated. Further, the present study maintained a higher cortisol collection rate compared to other preschool samples in which parents also collected morning and evening values at home (Dougherty, Klein, Olino, Dyson, & Rose, 2009). The longitudinal design and measurement of cortisol afforded observation of children’s morning and diurnal cortisol on three consecutive days, across four time points occurring at 9-month intervals. This permitted the use of latent state-trait modeling and facilitated a measure of stable, trait cortisol among preschool-age children. In addition, the use of observational parenting measures afforded independent assessment of parenting, thus strengthening the study conclusions. Together, these strengths made this study suited to address gaps in the current literature regarding the decomposition of trait-and state cortisol, and the contextual factors that contribute to the stability and time-varying variance in HPA-axis functioning.

This study was limited by several factors. First, the most rigorous studies of diurnal cortisol patterns use MEMS caps, which are used to monitor sampling times and adherence to collection protocols. We attempted to address this limitation by training mothers on collection techniques, by calling mothers on intended collection days to reiterate instructions and troubleshoot, by carefully reviewing the returned diary cards and collection tubes, and finally, by emphasizing that mothers could report incorrect sampling without loss of benefit. Of note, the current study did not include measures of the cortisol awakening response, which is distinct from diurnal variation, thought to be malleable by psychosocial factors, and implicated in adjustment. Additionally, our measure of diurnal slope is based on two samples per day, while a more rigorous measure of diurnal slope would utilize 3+ samples (Adam & Kumari, 2009). The current study has limited ability to interpret flatter diurnal slopes. A near-zero or flattened diurnal slope indicates nearly equivalent AM and PM levels, but does not represent whether the levels are high or low. Previous profile research utilizing this sample (Zalewski et al., 2016) would suggest that flat slopes are driven by low AM-low PM levels. It is nevertheless possible that some individuals with flat but high diurnal slopes influence the study. Due to power constraints, sex was not examined as a potential moderator of the relations of context and diurnal cortisol activity. However, there is evidence to support such moderation (e.g. Polk, Cohen, Doyle, Skoner, & Kirschbaum, 2005) and tests of moderation represent an important area of future study. In general, there is a dearth of research in this age group examining models similar to the current study, highlighting that the current findings be replicated in further study.

A future direction suggested by the findings of this study would be to test how trait and state cortisol, that is, children’s trait cortisol functioning and their variability from their own trait level (state cortisol), relate to the emergence of child adjustment problems. Hypercortisolism, hypocortisolism, and flatter diurnal slopes are thought to confer maladjustment. Yet few studies have tested a full model in which risk influences HPA axis functioning, which in turn impacts child adjustment, especially while considering how changes to the system may predict contemporaneous or downstream changes in well-being. As a notable exception, Essex (2014) found that children exposed to higher levels of early life stress experienced alterations to trait-level cortisol, especially in morning level, and also demonstrated wider swings in epoch-specific HPA activity. Of particular salience, and underscoring the need to understand predictors of stability and change in the HPA axis, epoch specific change in HPA activity was observed with contemporaneous increases in mental health symptoms. While it is tempting to conclude that lower morning levels or flatter diurnal slopes observed in this study represent HPA-axis dysregulation, the true meaning of such morning and diurnal cortisol states and traits to a preschooler’s adjustment has not been clearly elucidated and studies that model child adjustment merit further study.

Overall, these findings suggest that the stability of children’s experiences of cumulative risk predicts their cortisol function. Namely, it appears that consistent exposure to high levels of cumulative risk or lower-quality parenting relates to consistently low morning level and flattened diurnal slope. These relations might be indicative of the allostatic process of adaptation to stable and chronic risk or disadvantage, with children eventually being unable to mount a full cortisol response, possibly reflecting a developmental origin or embedding of long-term health risk. Simultaneously, there was some evidence to suggest that time-specific variation in children’s cortisol was predicted by time-specific variations in parenting behaviors, suggesting time-specific reactivity to variability in experience. These relations indicate that, during the preschool period, experiences of variability in maternal responsiveness, negativity, scaffolding, and limit setting are registered contemporaneously with variation in HPA activity. These proximal disruptions in parenting may prompt contemporaneous, acute stress responses in children. Together, these results add to the growing literature suggesting environmental influences on children’s HPA axis functioning, and point to the relevant influence of parenting and consistency in parenting on preschoolers’ HPA axis activity, even in high cumulative risk contexts, in predicting cortisol stability and change.

Figure 1.

Estimated measurement model of latent trait cortisol for morning level (1a) and diurnal slope (1b). Standardized loadings are presented (ps <.001). The proportion of variance accounted for by each state and trait are presented in parentheses. For ease of presentation, covariates are not presented in figure.

Table 5.

Standardized coefficients for models with significant prediction of trait and state cortisol diurnal slope

Diurnal Slope: Trait Models	CR	CR & Responsiveness	CR & Negativity
Income on Trait CR	−0.65***	−0.65***	−0.65***
Child Sex on Trait CR	−0.07	−0.07	−0.07
Income on Trait Cortisol	−0.10	−0.21	−0.21
Child Sex on Trait Cortisol	−0.07	−0.06	−0.07
Income on Trait Parenting	–	0.45***	0.50***
Child Sex on Trait Parenting	–	−0.03	−0.00
Trait CR on Trait Cortisol	−0.53***	−0.48**	−0.2T
Trait Parenting on Trait Cortisol	–	0.31*	0.42*
Diurnal Slope: State Models	CR & Warmth	CR & Scaffolding	CR & Limit Setting
Income on Trait CR	−0.65***	−0.65***	−0.65***
Child Sex on Trait CR	−0.07	−0.07	−0.07
Income on Trait Cortisol	−0.07	−0.23t	−0.13
Child Sex on Trait Cortisol	−0.06	−0.05	−0.05
Income on Trait Parenting	0.38***	0.58***	0.31***
Child Sex on Trait Parenting	0.04	−0.05	−0.12
Trait CR on Trait Cortisol	−0.54***	−0.43**	−0.51***
Trait Parenting on Trait Cortisol	−0.01	0.40*	0.16
T1 State Parenting on T1 State Cortisol	−0.22*	−0.23*	−0.05
T2 State Parenting on T2 State Cortisol	−0.16t	−0.15	−0.20*
T3 State Parenting on T3 State Cortisol	−0.06	−0.11	0.05
T4 State Parenting on T4 State Cortisol	−0.02	−0.01	−0.08

Note.

^t<.10,

^*p < .05,

^**p < .01,

^***p < .001. CR = Cumulative Risk.

Highlights.

• Preschoolers’ diurnal cortisol (morning level and diurnal slope) showed significant time-specific stability.

• Low income and cumulative risk predicted stable, diurnal cortisol traits only.

• Parenting behaviors predicted both stable, diurnal cortisol traits and time-specific, state changes in diurnal cortisol.

• Characteristic experiences of cumulative risk and parenting contribute to children’s stable or trait-HPA axis function while time-specific deviations of parenting experiences contribute to contemporaneous variations in children’s HPA activity.