Examining Infants’ Cortisol Responses to Laboratory Tasks Among Children Varying in Attachment Disorganization: Stress Reactivity or Return to Baseline?

Abstract

Cortisol is a hormone involved in mounting a stress response in humans. The evidence of stress reactivity among young children has been mixed, however. In the present study, the order of two laboratory tasks (i.e., Strange Situation and play) was counterbalanced, and home saliva samples were obtained. Saliva samples were also collected upon the children's arrival at the laboratory and at 40, 65, and 80 min after arrival. The authors examined changes in cortisol using piecewise hierarchical linear modeling, testing whether observed increases reflected a return to baseline or stress reactivity. An interaction between attachment disorganization and task emerged, such that disorganized infants showed increases in cortisol in response to the stressor compared with play, whereas organized infants did not show cortisol reactivity to either task. Implications for the buffering effects of maternal care on stress reactivity are discussed.

Following a stressor, glucocorticoids (i.e., cortisol in humans) are released as an end product of the hypothalamus–pituitary–adrenal (HPA) axis. The production of glucocorticoids promotes immediate survival by preparing the organism for a fight or flight response, diverting energy away from processes less critical to immediate survival, such as immune functioning, growth, digestion, and reproduction (Gunnar & Cheatham, 2003; Gunnar & Quevedo, 2007). In infancy, however, adequate maternal care may serve to buffer stress reactivity (Gunnar & Donzella, 2002). In the present study, we examined cortisol responses among infants with varying levels of attachment disorganization by counterbalancing the order of exposure to a stressor, thus ruling out alternative explanations for differential stress reactivity, such as a return to baseline levels.

Developmental Changes in Response to Stress

Elevations in cortisol among adults are elicited fairly consistently when situations include elements of unpredictability, uncontrollability, and social–evaluative threat (for a review, see Dickerson & Kemeny, 2004). Studies of cortisol reactivity in children, however, suggest a far more complex story. Various paradigms have been used to examine children's cortisol responses to stress, including mild pain tasks, fear and anger tasks, and maternal separation tasks. From birth until about 3 or 4 months of age, infants reliably demonstrate a rise in cortisol following various stressors, but by the time that infants are 6 months old, similar stressors elicit increases in cortisol less reliably, and by 12 months, it is even more difficult to elicit an increase in cortisol (for reviews, see Gunnar, Talge, & Hererra, 2009; Jansen, Beijers, Riksen-Walraven, & de Weerth, 2010). Across tasks, however, maternal presence appears to buffer infants’ stress reactivity.

Effects of Quality of Care on Cortisol Response

The quality of maternal care has been suggested to serve an important role in cortisol reactivity in infants. Attachment quality, assessed with the Strange Situation procedure, reflects the quality of maternal care received by infants (Ainsworth, Blehar, Waters, & Walls, 1978). Whereas most infants develop organized strategies for managing stressful circumstances in the presence of a parent, some infants lack a coherent strategy. Disorganized attachment reflects a breakdown of a strategy and is associated with specific parenting behaviors, such as maltreatment and frightening behavior (Main & Solomon, 1990).

Associations between attachment disorganization and cortisol reactivity have been reported in a few studies. Spangler and Grossman (1993) examined cortisol reactivity to the Strange Situation in a low-risk sample (N = 41) of 12-month-old infants. Relative to infants with secure attachments, infants with insecure attachments—especially those with disorganized attachments—showed higher cortisol levels following the Strange Situation. In a similar study (N = 38) of 19-month-old infants, Hertsgaard, Gunnar, Farrell, Erickson, and Nachmias (1995) found that infants with disorganized attachments had significantly higher levels of cortisol after the Strange Situation than infants with organized attachments. Other studies have shown associations between insecurity and cortisol reactivity, but only when insecure children were also fearful (Gunnar, Brodersen, Nachmias, Buss, & Rigatuso, 1996) or temperamentally inhibited (Nachmias, Gunnar, Mangelsdorf, Parritz, & Buss, 1996).

These studies offer exciting preliminary evidence of the effects of quality of maternal care on the stress response. However, their results are difficult to interpret in and of themselves, primarily due to the lack of baseline or control measures of cortisol activity. Of the four studies in which associations between attachment classification and cortisol reactivity were reported, three relied on a pretest measure of cortisol obtained immediately before the stressor (i.e., Gunnar et al., 1996; Nachmias et al., 1996; Spangler & Grossman, 1993), and one did not include a baseline measure of cortisol (i.e., Hertsgaard et al., 1995). Although a pretest measure may seem sufficient, Gunnar, Mangelsdorf, Larson, and Hertsgaard (1989) found that cortisol levels of laboratory baseline samples were significantly lower than those of home baseline samples collected at the same time on a different day. Larson, Gunnar, and Hertsgaard (1991) also reported a significant decrease in cortisol levels following a car ride, resulting in baseline levels at laboratory arrival that were significantly lower than baseline levels at home. Thus, increases in cortisol in the laboratory may be imposed on already lowered baselines, making findings ambiguous and difficult to interpret. Without adequate comparison samples, it is unclear whether increases in cortisol following a laboratory task represent cortisol reactivity or simply return to baseline (i.e., at-home) levels.

The Present Study

In the present study, we examined the association between attachment disorganization and cortisol reactivity when the order of a stressor (i.e., Strange Situation) and play was varied. The order of tasks was counterbalanced such that half of the participants experienced the stressor first followed by free play, and half experienced free play followed by the stressor. The design of the study allowed for examination of competing hypotheses. If the increases in cortisol reflect a return to baseline levels (presumably following a drop in cortisol prior to arrival), we would expect that infants’ cortisol levels would increase following the first task, regardless of whether it was a stressor or period of play. On the other hand, if the increases in cortisol reflect reactivity to a stressor, we would expect levels of cortisol to increase following the stressor but not following the play, regardless of the episode order.

Given findings of Spangler and Grossman (1993) and Hertsgaard et al. (1995), we expected attachment disorganization to moderate the association between the type of task and increases in cortisol, with children higher on attachment disorganization showing cortisol reactivity to the stressor compared with play and children lower on attachment disorganization showing lack of cortisol reactivity in both contexts.

Method

Participants

Participants included 32 infants who ranged in age from 11.3 to 20 months (M = 15.2, SD = 2.3). Participants were recruited from community day-care centers, from local mothers’ groups, and through announcements posted on a university web site. Nineteen (59%) of the children were girls. Twenty-two (69%) of the children were White/non-Hispanic, five (16%) were African American, two (6%) were biracial, two (6%) were Hispanic, and one (3%) was Asian American. Parents ranged in age from 21.0 to 42.9 years (M = 32.6, SD = 5.3). All parents were mothers, except for one father. Most parents were married (88%), had completed college or earned an advanced degree (84%), and were employed outside the home (69%). Family income ranged from the lowest category (<$10,000) to the highest (>$100,000), with most (53%) earning more than $100,000.

Procedure

During an initial visit, a research assistant reviewed the consent form with the parents and gave instructions and materials for taking a home saliva sample. Laboratory visits were scheduled for the morning in an attempt to control for diurnal fluctuations in cortisol production. Arrival times ranged from 7:58 a.m. to 10:00 a.m. (M = 9:13 a.m.). The 80-min laboratory visits were divided into two videotaped 40-min episodes: free play and Strange Situation. Participants were randomly assigned to the order of the tasks. For the free play task, parents were instructed to play with their children as they normally would in a playroom with a number of attractive toys, while the experimenter waited in an adjacent room. The Strange Situation procedure was conducted in a different room that resembled a waiting room at a doctor's office. Participants changed rooms between activities, such that each episode took place in its respective room.

Saliva samples were obtained from each child five times, including at the child's home before the parent and child left for the lab, upon the child's arrival at the lab, and at 40, 65, and 80 min postarrival (see Figure 1). These intervals were chosen because peak levels of cortisol are detectable in peripheral measurements about 20 min after a stressor (Pollard & Ice, 2007). For infants who experienced the Strange Situation first, the sample obtained at the 40-min point was expected to capture cortisol reactivity. The timing of this postsample (i.e., about 20 min after the final episode) is consistent with procedures in previous studies (e.g., Spangler & Grossman, 2003) and was expected to capture effects of the mounting stress, rather than stress experienced during a particular episode (e.g., reunion vs. separation). For those who experienced the Strange Situation second, the sample obtained at the 80-min point was expected to capture cortisol reactivity. Between the end of the Strange Situation and collection of the cortisol sample (approximately 20 min), the parent completed demographic and infant health status questionnaires while remaining in the room with the child. The inclusion of multiple samples (e.g., 65 min) further increased the reliability of the estimates of changes in cortisol during the visit.

Figure 1.

Timeline for laboratory tasks and saliva samples.

Strange Situation

The Strange Situation (Ainsworth et al., 1978) is divided into eight increasingly stressful episodes (lasting 15–25 min total). The standard protocol was followed, with infants experiencing a novel setting, interaction with a stranger, and brief separations from their parents. Attachment quality was coded from digital recordings. Infants were classified as secure (B), avoidant (A), or resistant (C) according to procedural guidelines of Ainsworth et al. (1978) and received a continuous score for disorganization, ranging from 1 to 9. Infants could receive a primary classification of disorganized (D) as per procedures described by Main and Solomon (1990). A primary coder, blind to other participant information, coded all videotapes. The primary coder (the second author) had attended a training course and passed reliability tests for classifying organized and disorganized attachment strategies. A second blind coder, who also attended the Strange Situation coding training course, coded 20% of tapes for reliability. Reliability for the major classifications (including D) for this subset of tapes was 100%. The Spearman correlation for interrater agreement on the continuous scale of disorganization was .82. Continuous disorganization scores were averaged for double-coded cases. Coding of additional cases was discussed in conference when the primary coder was unsure of classification.

Saliva Sampling

Procedures for saliva sample collection were reviewed with parents at the initial consent visit. Parents were instructed to hold one end of a dental cotton roll and moisten the other end in the child's mouth. After briefly moistening the cotton in the child's mouth, parents dipped the cotton into a cup containing 0.03 g of flavored drink crystals (Pathmark cherry-flavored drink mix; Path-mark Stores, Inc., Montvale, NJ) to facilitate sampling. The cotton was then placed back in the child's mouth until it was sufficiently wet with the child's saliva. In recent controlled studies, effects of sweeteners on radioimmunoassay values have been found to be minimal (Gordon, Peloso, Auker, & Dozier, 2005; Talge, Donzella, Kryzer, Gierens, & Gunnar, 2005). Parents were instructed not to give the child anything to eat or drink 30 min prior to sampling. Parents completed questionnaires about infant health status variables, such as whether children were teething or sick or had eaten prior to sampling.

Cortisol Assay

All saliva samples were stored in a freezer at –20 °C prior to assay procedures. Samples were assayed with the High Sensitivity Salivary Cortisol Enzyme Immunoassay Kit (Salimetrics, LLC, State College, PA). To minimize variability of results, researchers assayed all samples for one child in duplicate on the same plate. Pairs of samples with coefficients of variation greater than 10% were retested in a later assay. For this study, inter- and intraassay coefficients of variation fell below 7% and 14%, respectively.

Results

Data Preparation

Of the 32 children, all had at least three samples, with 29 having four samples, and 25 having all five samples. Five home and two arrival-at-lab samples were excluded because parents reported that their children had ingested food or liquids within the 30 min prior to sampling. No parent identified significant illness in his or her child on the day of the visit, although 61% reported that the child had a runny nose or cold and 39% reported that the child was teething. Cortisol values were not associated with either a cold or teething and thus were not excluded for these conditions. For each sample time (home, arrival, and at 40, 65, and 80 min postarrival), cortisol values of 3 standard deviations above the mean were considered outliers and excluded prior to further analyses, consistent with procedures used in similar studies (e.g., Dettling, Gunnar, & Donzella, 1999). This resulted in the exclusion of one home sample, one arrival sample, and one 65-min sample. Finally, because cortisol values were positively skewed, log-10 transformation was used to normalize the distributions, consistent with procedures used in similar studies (Dettling, Parker, Lane, Sebanc, & Gunnar, 2000). Descriptive statistics for salivary cortisol values are presented in Table 1.

Table 1.

Descriptive Statistics for Raw Cortisol Levels

		Raw cortisol level (measured in micrograms per deciliter)
Sample obtained	n	M	SD	Minimum	Maximum
Home	26	0.20	0.19	0.03	0.93
Arrival at laboratory	28	0.28	0.38	0.02	1.37
40 min postarrival	31	0.33	0.43	0.02	1.57
65 min postarrival	31	0.24	0.37	0.03	1.57
80 min postarrival	30	0.22	0.23	0.04	1.35

Preliminary Analyses

We examined demographic variables to determine whether care-giver or child characteristics were associated with cortisol values. Caregiver age, ethnicity, marital status, income, and education level and child age and ethnicity were not associated with cortisol values at any of the time points. Cortisol was not associated with time of arrival at the laboratory, likely because we had restricted the range in arrival time to control expected time-of-day effects. As a result, time of day was not included in further analyses.

Associations between attachment disorganization and demographic variables were also examined. Approximately half of the children (44%) were classified as secure in the Strange Situation, with most others classified as disorganized (25%) or resistant (25%). Two children (6%) were classified as avoidant. Of those classified as disorganized, half had a secondary classification of resistant, 25% had secondary classifications of secure, and 25% had secondary classifications of avoidant. Given that hypotheses related to attachment disorganization, primary analyses included the continuous measure of attachment disorganization, rather than categorical classifications of security. The disorganized versus organized categorization was used in follow-up analyses. It is expected that the continuous scale of disorganization may better represent the dimensional nature of attachment behaviors (Fraley & Speiker, 2003) and has been used previously in examinations of correlates of this construct (e.g., Carlson, 1998). Attachment disorganization scores ranged from 1 to 6 (M = 3.6, SD = 1.77). Disorganization was not associated with child or caregiver demographic variables.

Primary Analyses

We used piecewise linear growth modeling to examine changes in cortisol levels across three time periods: car ride to the lab, first episode in lab, and second episode in lab. In hierarchical linear modeling (Raudenbush & Bryk, 2002), multiple observations are treated as nested within persons, which allows for variability in the number and spacing of time points. Thus, the approach allows for inclusion of participants who are missing one or more points of data, if data are assumed to be missing at random (MAR; Shafer & Graham, 2002). For this study, MAR refers to “missingness” that would be related to previous measurements of cortisol or other measured covariates, rather than future unobserved measurements of cortisol. Furthermore, piecewise linear growth modeling allows for the division of growth trajectories (i.e., patterns of cortisol production) into separate linear components. Rather than estimating the rate of change in cortisol (i.e., slope) across the entire visit, we used this approach to estimate slopes for each distinct period. Therefore, it allowed for the examination of between-individual correlates of change during each period, such as type of task. A piecewise approach is a common strategy for simultaneously modeling separate components of a process, such as reactivity and recovery (e.g., Llabre, Spitzer, Saab, & Schneiderman, 2001).

The dependent variable was the log-10 transformed cortisol value measured at each time point. Time was recoded into three separate Level-1 predictors to form a three-piece linear model, as depicted in Table 2. The first linear component (Episode 1) captured change between the home sample and arrival at lab sample. The second linear component (Episode 2) captured change between the arrival at lab sample and the 40-min sample. The third linear component (Episode 3) captured change across the 40-, 65-, and 80-min postarrival samples. Thus, the Level-1 model was of the following form:

$${\text{Log cort}}_{ti}={\pi }_{0i}+{\pi }_{1i}\left(\text{Episode}1\right)+{\pi }_{2i}\left(\text{Episode}2\right)+{\pi }_{3i}\left(\text{Episode}3\right)+e_{ti}$$

where π_0i represents child i's log cort upon arrival at lab (coded as zero); π_1i, π_2i, and π_3i represent the rate of linear change in log cort over Episode 1 (car ride), Episode 2 (first episode in lab), and Episode 3 (second episode in lab) for child i; and e_ti represents the within-individual error in child i's log cort that cannot be accounted for by initial cort (π_0i) or by linear change in log cort over time.

Table 2.

Coding Scheme for Three-Piece Linear Model

	Sample
			Postarrival
Variable	Home	Arrival	40 min	65 min	80 min	Interpretation of πs
Episode 1	–0.47a	0	0	0	0	Rate of change during car ride
Episode 2	0	0	1	1	1	Rate of change during first task
Episode 3	0	0	0	0.63	1	Rate of change during second task

Note. One unit of time is equal to 40 min.

^aThe length of the car ride varied between participants, such that the recoded time of the home sample ranged from –1.25 to –0.125 (from 50 min to 5 min before arrival). The average amount of time spent in the car (M = 18.6 min, recoded as –.47 units) is presented above; however, in hierarchical linear modeling analyses, Time 1 duration was allowed to vary for each individual.

Return to Baseline Hypothesis

First, we examined an unconditional Level-2 model with π_0i, π_1i, π_2i, and π_3i random. This allowed for examination of mean linear rates of change in cortisol during each episode. If infants’ cortisol levels increased following the first episode, regardless of the type of task, this would support the “return to baseline” hypothesis. Results for the unconditional model are presented in Table 3. Mean rates of change were not significantly different than zero at any of the time periods (p > .05). These results do not provide evidence of a decrease in baseline levels following a car ride or an associated return to baseline levels, at least when averaged across individuals. However, variance components were significant for the slope of each episode, indicating significant variability between individuals.

Table 3.

Three-Piece Unconditional Linear Model for Changes in Salivary Cortisol

Effect	Coefficient	SE	t	df	p
Intercept, β00	–.79	.11	–7.48	31	.00
Slope, β10	–.11	.17	–0.63	31	.53
Slope, β20	–.03	.05	–0.51	31	.61
Slope, β30	–.01	.08	–0.11	31	.92

Note. β₀₀ represents the baseline measure of cortisol obtained at child's arrival to lab; β₁₀, β₂₀, and β₃₀ represent the changes in salivary cortisol across Episode 1 (car ride), Episode 2 (first lab task), and Episode 3 (second lab task), respectively.

Stress Reactivity Hypothesis

Next, we examined between-individual predictors of rates of change at Level 2 to explore individual differences in the slopes for each episode. This model allowed us to examine whether the type of task (i.e., Strange Situation vs. play), attachment disorganization, or an interaction of the two was associated with changes in cortisol levels. The Level-1 model remained the same as that described earlier. At Level 2, the model was expanded to include several between-subject predictors, including task (TASK, a dummy variable indicating lab task: 1 = Strange Situation, 0 = play), attachment disorganization (DISORG, the continuous disorganization score), and the interaction of task and attachment disorganization (TASK × DISORG). As Episode 1 represented the rate of change during the car ride, minutes in the car (CARTIME) was included as a predictor in this equation, and predictors associated with lab tasks (i.e., TASK and TASK × DISORG) were excluded. Due to concern that changes in cortisol levels may depend on baseline cortisol levels (laws of initial values or LIV; Wilder, 1958), we included baseline cortisol values to control for possible LIV effects on slopes. The cortisol sample collected at home was included as the baseline measure in the slope equation for the car ride episode (i.e., Episode 1, π_1i), and the cortisol sample collected at arrival to lab was included as the baseline measure in the slope equations for the laboratory episodes (i.e., Episode 2, π_2i, and Episode 3, π_3i). The resulting Level-2 model can be represented as

$$\begin{array}{cc}\hfill {\pi }_{0i}& ={\beta }_{00}+{\beta }_{01}\left(\text{TASK}\right)+{\beta }_{02}\left(\text{DISORG}\right)+{\beta }_{03}(\text{TASK}\times \text{DISORG})+r_{0i}\hfill \\ \hfill {\pi }_{1i}& ={\beta }_{10}+{\beta }_{11}\left(\text{BASELINE}\right)+{\beta }_{12}\left(\text{CARTIME}\right)+{\beta }_{13}\left(\text{DISORG}\right)+r_{1i}\hfill \\ \hfill {\pi }_{2i}& ={\beta }_{20}+{\beta }_{21}\left(\text{BASELINE}\right)+{\beta }_{22}\left(\text{TASK}\right)+{\beta }_{23}\left(\text{DISORG}\right)+{\beta }_{24}(\text{TASK}\times \text{DISORG})+r_{2i}\hfill \\ \hfill {\pi }_{3i}& ={\beta }_{30}+{\beta }_{31}\left(\text{BASELINE}\right)+{\beta }_{32}\left(\text{TASK}\right)+{\beta }_{33}\left(\text{DISORG}\right)+{\beta }_{34}(\text{TASK}\times \text{DISORG})+r_{3i}\hfill \end{array}$$

where π_0i represents the initial value of log cort at an individual's arrival to the lab, and π_1i, π_2i, and π_3i represent individual rates of linear change in log cort over time during each separate episode. The term β₀₀ estimates the mean lab arrival log cort value for participants when other predictors equal zero. The term β₀₁ represents the difference in initial log cort value between infants who had the play episode first versus those that had the Strange Situation first (i.e., main effect of task on intercept). The term β₀₂ represents the change in initial log cort value as scores of attachment disorganization increase (i.e., main effect of disorganization on intercept). The term β₀₃ represents the interaction of task and attachment disorganization in prediction of the initial cortisol value. The equations for linear change (i.e., π_1i, π_2i, π_3i) can be similarly broken down in order to understand the relative contributions of each predictor on each episode's slope.

Initial cortisol values, defined as the values obtained upon the child's arrival at the lab, differed significantly with respect to attachment disorganization. Specifically, as attachment disorganization increased, the initial cortisol value decreased significantly (p < .05). As this finding was unanticipated, we conducted post hoc analyses to test whether attachment disorganization was also negatively correlated with the home sample. Indeed, although the findings were not significant, attachment disorganization tended to be inversely correlated with the home sample of cortisol (r = –.34, p < .08). Type of task (i.e., play or Strange Situation) was not associated with the initial cortisol value obtained upon a child's arrival at the lab, suggesting that there were no pretask differences between groups assigned to each order. Individual differences in slope were not associated with time spent in the car, attachment disorganization, or baseline levels obtained at home.

The primary effects of interest were those associated with the slopes of the laboratory episodes (i.e., Episode 2 and Episode 3). The interaction of task and attachment disorganization emerged as a predictor of the rate of cortisol change for Episode 3 (p < .01) and approached significance for Episode 2 (p = .06; see Table 4). To further examine this effect, we analyzed separate models for children with a primary classification of disorganized attachment and children with a primary classification of organized attachment (i.e., secure, avoidant, resistant). As seen in Figure 2, attachment disorganization (grouped according to disorganized or organized classification) moderated the association between task and cortisol reactivity. Results for children with disorganized attachment and those with organized attachment are presented together in Table 5. For infants with a disorganized classification, the change in cortisol (slope) differed depending on the task, which reached statistical significance for Episode 3 (β₂₁ = .91, p < .01) and approached significance for Episode 2 (β₁₁ = .37, p = .08). Specifically, the Strange Situation elicited an increase in cortisol for these infants compared with the cortisol level for the period of play. For infants with an organized classification, differences in cortisol response between the tasks did not approach significance.

Table 4.

Linear Piecewise Modeling Coefficients of Between-Individual Effects on Salivary Cortisol

Effect	Coefficient	SE	t	df	p
Intercept, β00	–0.51	0.18	–2.91	28	.01
TASK, β01	–0.21	0.14	–1.45	28	.16
DISORG, β02	–0.14	0.06	–2.44	28	.02
TASK × DISORG, β03	0.07	0.05	1.43	28	.16
Slope, β10	–1.32	0.37	–3.62	28	.00
BASELINE, β11	–1.19	0.16	–7.28	28	.00
CARTIME, β12	0.01	0.01	2.78	28	.36
DISORG, β13	–0.12	0.09	–1.43	28	.16
Slope, β20	0.56	0.28	1.99	27	.06
BASELINE, β21	1.15	0.08	14.74	27	.00
TASK, β22	–0.08	0.25	–0.32	27	.75
DISORG, β23	0.12	0.09	1.46	27	.16
TASK × DISORG, β24	0.16	0.08	1.89	27	.06
Slope, β30	–0.12	0.16	–0.77	27	.45
BASELINE, β31	–0.40	0.07	3.33	27	.00
TASK, β32	–0.14	0.21	–0.67	27	.51
DISORG, β33	–0.18	0.05	–3.52	27	.00
TASK × DISORG, β34	0.22	0.07	–3.55	27	.00

Note. β₀₀ represents the baseline measure of Cortisol obtained at child's arrival to lab; β₁₀, β₂₀, and β₃₀ represent the changes in salivary cortisol across Episode 1 (car ride), Episode 2 (first lab task), and Episode 3 (second lab task), respectively.

Figure 2.

Changes in cortisol levels as a function of type of task and attachment disorganization. SS = Strange Situation.

Table 5.

Linear Piecewise Modeling Coefficients of Salivary Cortisol During Laboratory Tasks for Children With Disorganized and Organized Attachment Classifications

Effect	Coefficient	SE	t	df	p
Disorganized attachment classification
Intercept, β00	–1.25	0.25	–4.95	5	.00
TASK, β01	0.19	0.39	.49	5	.64
Slope, β10	–0.01	0.11	–.04	5	.97
TASK, β11	0.37	0.17	2.09	5	.08
Slope, β20	–0.42	0.12	–3.70	5	.02
TASK, β21	0.91	0.15	5.99	5	.00
Organized attachment classification
Intercept, β00	–0.85	0.16	–5.38	22	.00
TASK, β01	0.25	0.23	1.12	22	.27
Slope, β10	–0.18	0.09	–2.10	22	.05
TASK, β11	0.21	0.13	1.60	22	.12
Slope, β20	–0.14	0.14	–.97	22	.34
TASK, β21	0.31	0.20	1.60	22	.12

Note. β₀₀ represents the baseline measure of cortisol obtained at child's arrival to lab; β₁₀ and β₂₀ represent the changes in salivary cortisol across Episode 2 (first lab task) and Episode 3 (second lab task), respectively.

Discussion

In the present study, we examined children's levels of cortisol before and after two tasks (i.e., Strange Situation and free play). Attachment disorganization moderated the association between task and cortisol response. For children with disorganized attachment classifications, the Strange Situation elicited increases in cortisol that were significantly different than changes in cortisol elicited during the play episode. For children with organized attachment classifications, there were no significant differences in cortisol changes associated with the type of task. For these children, neither the Strange Situation nor the play elicited an increase in cortisol. These findings support the hypothesis that adequate maternal care buffers infants’ stress reactivity as assessed through cortisol levels. Similar findings regarding the association between disorganized attachment and stress reactivity have been reported in previous studies (i.e., Hertsgaard et al., 1995; Spangler & Grossman, 1993). The findings of the present study are exciting because, to our knowledge, it is the first study of cortisol reactivity in which the order of the Strange Situation and a comparison laboratory task (i.e., play) has been counterbalanced, which allowed for more systematic investigation of observed changes in cortisol.

In previous studies of infants’ cortisol reactivity to the Strange Situation, researchers have relied on baseline measures of cortisol collected upon arrival at the lab immediately before the stressor (e.g., Spangler & Grossman, 1993). As demonstrated by Larson et al. (1991), these baseline measurements may not be the most appropriate measure of typical cortisol levels. However, we did not find evidence of a drop in cortisol following a child's car ride to the laboratory. Cortisol levels at home and at arrival to lab were comparable, and the change in cortisol during the car ride was not significantly different from zero. Given that we did not set out to test this effect but rather to control for home values, we did not directly replicate the experimental conditions of Larson et al. (i.e., we did not manipulate the length of the car ride). Thus, the failure to replicate the findings is difficult to interpret, given the methodological differences.

A significant main effect of attachment disorganization on the baseline level of cortisol (intercept) also emerged, such that as attachment disorganization increased, arrival-at-lab cortisol levels decreased. Although this finding was not hypothesized, similar effects have been reported in previous studies. Lowered basal levels of cortisol may be the result of down-regulation of the HPA system, serving a protective adaptation to elevated levels (Gunnar & Vasquez, 2001). A growing body of literature suggests that atypically low levels of basal cortisol and flattened daytime rhythms may be indicators of risk for later health problems (Heim, Ehlert, & Hellhammer, 2000), antisocial behavior (Vanyukov et al., 1993), aggression (McBurnett, Lahey, Rathouz, & Loeber, 2000), and anxiety disorders (Yehuda et al., 2000). Thus, the mechanisms underlying the development of lowered basal cortisol levels should be examined in future research.

The present study had a relatively small sample size. Given that our sample represented a middle-class, nonclinical group, there were a higher percentage of children classified as disorganized (25%) than expected from the rates in comparable samples (15%). The rates of disorganized attachment were significantly higher in the present study compared with distributions for comparable samples reported in a recent meta-analysis (van IJzendoorn, Schuengel, & Bakermans-Kranenburg, 1999), χ² = 7.84, p < .01. The atypically high proportion of children with disorganized attachment may have increased our ability to detect the reported effect within a relatively small sample size. Nevertheless, given the small sample size, replication of the findings will be important.

Our findings should be considered in the context of typical diurnal fluctuations in cortisol. In the present study, children's cortisol levels were measured in the morning, a time when decreases in cortisol are typically observed. Thus, a slight increase or flat pattern of cortisol production might reflect cortisol reactivity at this time of day. Although counterbalancing the presentation of tasks allowed us to control for order effects, this issue could be further addressed by collection of time-matched cortisol samples at home. Additionally, obtaining parents’ reports of infants’ wake-up times would be helpful for determining how findings fit within the expected diurnal rhythms of cortisol production.

Taken together, results of this study add to our understanding of the neurobiology of the human stress response during infancy. Several methodological strengths facilitated interpretability of findings, including restricting the time of laboratory visits, using multiple baseline samples, including a comparison (nonstress) laboratory task, and counterbalancing the order of laboratory tasks. Many questions remain regarding HPA functioning in the human infant. Given support for the moderating effects of attachment disorganization on cortisol reactivity, it will be important to examine what aspects of maternal care (e.g., sensitivity, frightening behavior, intrusiveness) are associated with the hypo- versus hyperreactivity to stressors. Associations between cortisol reactivity and diurnal regulation of cortisol within individuals also should be investigated. Although these functions are considered to be relatively orthogonal, developmental changes may reflect more interdependent processes. Finally, the present study highlights the need for longitudinal studies in which the effects of early cortisol regulation on later outcomes are examined.