Socially oriented thinking and the biological stress response: Thinking of friends and family predicts trajectories of salivary cortisol decline

Abstract

The cortisol stress response has been related to perceived social support, but previous studies rely on self-reported social support variables. The cortisol recovery phase in particular has been theorized to serve a social coping function, but individual differences in recovery slope have not yet been examined in relation to social coping-relevant indices. This study addressed these gaps by examining the relationship of cortisol trajectories after a socioevaluative task to individual differences in covertly assessed cognitions related to close social relationships. We examined trajectories of cortisol change related to socially oriented thinking, the semi-implicit activation of cognitive representations of friends or family. Young adults (N = 64) gave salivary cortisol samples before and for 45 min after a speech task. Participants' thoughts were sampled repeatedly; the frequency of words related to friends or family was assessed to index socially oriented thinking. A free curve slope intercept latent growth curve model showed excellent fit with the cortisol data. Socially oriented thinking was unrelated to overall magnitude of cortisol response to the task (latent intercept) but predicted the latent cortisol trajectory, independently of cortisol intercept and baseline cortisol levels. Socially oriented thinkers showed more gradual cortisol declines, whereas nonsocially oriented thinkers showed a steeper downslope driven primarily by cortisol changes 45 min after the task. Individual differences in socially oriented thinking may manifest in different rates of biological changes following a performance task.

1 ∣. INTRODUCTION

All humans encounter stressful life events, but their divergent responses to such events influence outcomes for mental and physical health. Biological responses to stress (e.g., activation of the hypothalamic-pituitary-adrenal [HPA] axis increased respiration and cardiac activity) function to mobilize action (Dickerson & Kemeny, 2004; Smith & Vale, 2006). Psychological responses to stress (e.g., negative affect, heightened attention) also support coping behavior (Lazarus & Folkman, 1984). Physiological and psychological responses facilitate immediate adaptation to stressors. However, in the long term, individual differences in stress responding leave some vulnerable to psychopathology or illness, while others maintain well-being or even thrive (e.g., Chrousos & Gold, 1992; Marroquín, Tennen, & Stanton, 2017).

1.1 ∣. Social support and responses to stress

Individuals experience stress when situational demands exceed available coping resources (Lazarus & Folkman, 1984). Reaching out to others for support—perhaps by venting negative emotions, soliciting advice, enlisting instrumental help—is a way of increasing or capitalizing on available resources and is therefore common during the stress response (e.g., Rimé, 2009). Myriad social factors, such as attachment styles, connectedness, and close relationship quality, play roles in cognitive, affective, and neural responses to stress (e.g., Eisenberger & Cole, 2012; Gunnar, 2017; Ketay & Beck, 2017; Marroquín & Nolen-Hoeksema, 2015; Pietromonaco, DeBuse, & Powers, 2013; Zaki & Williams, 2013). Better social support during stress predicts far-reaching consequences, including health and longevity (e.g., Marroquín et al., 2017; Slatcher, Selcuk, & Ong, 2015; Uchino, 2009). Social connectedness and support can affect health through multiple mechanisms, including influences on stress exposure (e.g., frequency and severity of stressful life events), coping behavior (e.g., instrumental assistance and encouragement of proactive coping and emotion regulation), cognitive function (e.g., executive functioning and threat sensitivity), health behavior (e.g., social pressure toward healthy behaviors), and physiological responses to stress and repair, including immune functioning, inflammation, sleep quality, cardiovascular activity, and HPA axis regulation (Cacioppo & Hawkley, 2003, 2009; Cohen, 2004; Hawkley & Capitanio, 2015; Keicolt-Glaser, Glaser, Cacioppo, & Malarkey, 1998; Marroquín et al., 2017; Slavich & Irwin, 2014; Thoits, 2011; Uchino, 2009; White, VanderDrift, & Heffernan, 2015).

The most reliable effects of social support on well-being involve not the actual provision of support, but the perception that support is available (Cohen, 2004; Lakey & Orehek, 2011). Many challenges occur when regulatory support by others is irrelevant (e.g., a car ahead slamming on its brakes) or not immediately available (e.g., a hostile interaction at work). Fortunately, effects of social support can occur even when supports are not present, presumably via mental representations of supporters that, intentionally or not, guide affective responses, self-regulation, and behavior. Lakey and Orehek's (2011) relational regulation theory proposes that social support has its regulating effects on negative affect through the accumulation of repeated interactions that are not overtly focused on the topic of support. Repeated reductions of negative affect in the presence of a particular relationship partner help create an association between that partner and affect being regulated and, hence, the perception of social support availability (Lakey & Tanner, 2013). Even contact with “symbolic” supporters, such as public figures and TV characters, may mimic emotional support by close others, to the extent that symbolic others evoke the same mental associations (Lakey, Cooper, Cronin, & Whitaker, 2014). In their theory of the relational self, Andersen and Chen (2002) argue similarly that evoking, even subliminally, cognitive representations of specific relationship partners activates cognitive, behavioral, and emotional habits that play out as they would in the context of that relationship (for review, see Chen, Boucher, Andersen, & Saribay, 2013). We conceptualize the array of consequential responses to stressors as including such social and social-cognitive responding, alongside affective and physiological channels. With the term socially oriented thinking, we refer to the degree to which individuals tend to activate social-cognitive representations of close others, even when those others are not physically present. Given the robust relationship of social support to health, socially oriented thinking could be a pertinent dimension of comprehensive, biopsychosocial models of stress responding and could help answer the open question of how social support is internalized and ultimately integrated into individuals' coping “arsenals.”

1.2 ∣. Socially oriented thinking and cortisol response and recovery

Some stress response systems (e.g., cardiovascular) act within seconds to mobilize immediate behavioral responses. The HPA axis responds more slowly and is hypothesized to serve a longer-term coping function (Sapolsky, Romero, & Munck, 2000; Shirtcliff, Peres, Dismukes, Lee, & Phan, 2014; Thayer & Lane, 2000). The cascade of signals along the HPA axis unfolds over minutes, with cortisol peaking 20 or more minutes after the initiation of the stressor (see Lopez-Duran, Mayer, & Abelson, 2014). Cortisol responses are most reliable when personally relevant goals are threatened or obstructed (e.g., Blascovich & Tomaka, 1996; Dickerson & Kemeny, 2004; Weiner, 1992) and are especially robust under conditions that are uncontrollable or involve social threat (Dickerson & Kemeny, 2004). In other words, instead of being universal to negative situations generally, cortisol responses have some specificity to psychologically—and socially—demanding situations requiring complex coping efforts. Given its slow course and likely long-term coping function, the HPA response may be especially sensitive to the availability of resources like supportive others (Meuwly et al., 2012). Ample evidence shows that social partners reduce the magnitude of the neuroendocrine stress reactivity and may aid in its recovery (e.g., Eisenberger, Taylor, Gable, Hilmert, & Lieberman, 2007; Heinrichs, Baumgartner, Kirschbaum, & Ehlert, 2003; for review, see Gunnar, 2017).

Although most of this research examines the physical presence of a supportive person, presence may not be required to affect cortisol responding. For example, compared to a no-support condition, participants who received social support relayed by videotape demonstrated attenuated cortisol reactivity to a stressor (Thorsteinsson, James, & Gregg, 1998). Girls who received verbal support from their mothers by telephone displayed attenuated cortisol responses to a stressor, similar to girls whose mothers supported them in person (Seltzer, Ziegler, & Pollak, 2010). Thus, the social buffering effect on endocrine responding may extend to cognitive representations of supporters even when supporters are not physically present. Indeed, Gunnar (2017) has argued that social buffering works through attachment, which is itself a cognitive representation (internal working model; Bowlby, 1969). Evidence from several streams underscores the potential mechanistic importance of cognitive representations of social relationships in buffering the cortisol response. The association between high perceived social support and reduced cortisol reactivity was statistically mediated by reduced neural activation in regions involved in distress at separation from attachment figures (Eisenberger et al., 2007). Interventions targeting social-cognitive processes reduced cortisol reactivity to social-evaluative stressors, perhaps by altering appraisals of the self in relation to others (Engert, Kok, Papassotiriou, Chrousos, & Singer, 2017).

Although a link between socially oriented thinking and cortisol is likely, its direction is difficult to predict. The social buffering account of HPA responsivity would predict that socially oriented thinking would dampen stress responding. The opposite is also plausible. HPA stress responses are shaped by intersecting inherited and environmental characteristics and compensatory processes in learning and adaptation over time (for a review, see Susman, 2006). This complexity and plasticity of the HPA system create considerable heterogeneity in cortisol secretion patterns associated with seemingly “adaptive” or “maladaptive” traits (e.g., Delahanty, Raimonde, & Spoonster, 2000; Gunnar, 2017; Phillips, Ginty, & Hughes, 2013; Susman, 2006). As such, Shirtcliff and colleagues (2014) advocate a functional perspective that emphasizes the likely coping purpose of cortisol secretion. They conceptualize cortisol elevation as mobilizing individuals to engage with their environments to identify available resources (a process that may help or hinder depending on whether coping needs are ultimately met). The potential coping-related utility of cortisol elevations might explain why, while it is typically assumed that faster cortisol recovery indicates healthier or more efficient cortisol regulation (e.g., Epel, McEwen, & Ickovics, 1998), some findings contradict this. Steeper recovery following an interpersonal stressor emerged in avoidantly attached women and anxiously attached men (Powers, Pietromonaco, Gunlicks, & Sayer, 2006). For adolescents with elevated personality pathology, steeper cortisol recovery was also associated with more severe problem behavior (Tackett et al., 2014). When engaging in avoidance coping, children with higher chronic stress levels demonstrated steeper cortisol recovery (Bendezú & Wadsworth, 2017). In sum, faster recovery does not necessarily indicate healthy cortisol regulation and may in some cases indicate premature, avoidance-related termination of the stress response.

1.3 ∣. Current investigation

We examined the relationship of socially oriented thinking to the trajectory of cortisol recovery, given its potential importance for social coping (Meuwly et al., 2012). Young adults underwent a social evaluative performance task. Before and after the task, they provided salivary cortisol samples and answered thought probes, which were later used to estimate the activation of thoughts of close others (i.e., friends, family). We hypothesized that individual differences in socially oriented thinking would predict cortisol declines after the task, such that higher socially oriented thinking would relate to more gradual declines. We controlled for the effects of cognitions about people in general (i.e., not family or friends) to isolate effects of cognitions specific to close relationships, which we reasoned would be consistent with a coping-related conceptualization of cortisol recovery.

2 ∣. METHOD

2.1 ∣. Participants and procedure

Participants were recruited via posters on campus advertising a study on emotion regulation. The final sample consisted of 64 students (75% women) attending a small liberal arts college in the Midwest, with a mean age of 20.0 years (SD = 1.2 years). Their racial-ethnic backgrounds were 83% White, 5% African American, 7% Asian, 2%, Hispanic, and 5% other.

To control for typical variation in cortisol levels throughout the day, all laboratory sessions were conducted between 2 pm and 7 pm. Participants gave written consent and completed two cognitive computer tasks and a series of self-report measures unrelated to the current study. Following this and approximately 40 min after arriving, participants provided the first saliva sample (henceforth called baseline cortisol) and written thought sample, details of which are below. This time frame was modeled after the existing strategy of delaying the baseline measurement to allow participants time to habituate to the lab environment (e.g., Lopez-Duran et al., 2014). Participants were then notified about the speech task, prepared for and completed the speech, and received bogus feedback, as described below. They then immediately provided the first post-task saliva sample and a second written thought sample, and 5 min later, orally recorded a thought sample. Participants were instructed to sit by themselves quietly with a gardening and architectural magazine for 1 hr, without access to personal items, in an experimental room. During this post-task period, the experimenter entered to collect additional saliva samples at 15-min intervals, and a nearby computer prompted participants to provide thought samples at 5-min intervals, alternating between written and oral formats. This sampling regimen yielded a total of four post-task saliva samples (0 min, 15 min, 30 min, and 45 min, henceforth referred to as Times 1, 2, 3, and 4). Saliva sampling intervals range from 10–20 min in similar studies (e.g., Ketay, Welker, Beck, Thorson, & Slatcher, 2019; Lopez-Duran et al., 2014). At the end of the study, participants were debriefed about the study's purpose and use of deception. The institutional review board approved all procedures, and procedures were carried out in a manner consistent with the World Medical Association Declaration of Helsinki.

2.2 ∣. Tasks

2.2.1 ∣. Speech task

Participants completed a modification of the Trier Social Stress Test (TSST) used to elicit stress via socioevaluative threat (Birkett, 2011; Dickerson & Kemeny, 2004; Kirschbaum, Pirke, & Hellhammer, 1993). In the original TSST, adult participants perform math problems and deliver a speech in the role of a job applicant to a panel of impassive judges in the role of potential employers. The version used in the present study was a developmentally sensitive and personally relevant modification of the TSST, which translated performance to the social domain for use with adolescents and young adults and added social evaluative criticism (Hilt, Aldao, & Fischer, 2015; Leitzke, Hilt, & Pollak, 2015). Participants were informed of a surprise speech task in which they would speak for 3 min about what makes them “special and unique.” They were led to believe they would be giving this speech to four students from a peer institution, who would watch the speech live via the camera connected to a computer and then give feedback on the participant's personality. A picture of four students (of varying sex and ethnicity) appeared on the computer screen. Participants were left alone for 3 min to prepare and were then filmed giving their speeches. After the speech, participants heard identical, prerecorded feedback under the guise that it was live and intended specifically for them, consisting of a mix of neutral (e.g., “He [She] sounds like the typical college student”) and negative (e.g., “He [She] just didn't really grab my attention”) comments.¹ This variant of the task has previously demonstrated efficacy eliciting psychological and autonomic stress responses in adolescents (Leitzke et al., 2015) and similar responses, including cortisol, in young adults (Hilt et al., 2015).

2.2.2 ∣. Subjective mood

To characterize the subjective effects of the task, participants rated their subjective mood at baseline and immediately post-task using a 100-point visual analogue scale. Two items targeted common indices of stress: anxiety and calm (reverse item), and were embedded within distractor items (sad, happy, hungry).

2.2.3 ∣. Thought sampling

Nine thought samples were taken during the post-task period. Thought samples alternated between written and oral, using the following prompt: “Write [Say] whatever you are thinking about right now. Write [Say] whatever comes to mind.” Written responses were collected on a computer keyboard using E-prime Version 2.0 (Psychology Software Tools), and oral responses were recorded using a digital audio recording device (with up to 90 s to speak for each thought sample) and later transcribed by research assistants.

2.3 ∣. Measures

2.3.1 ∣. Socially oriented thinking

We assessed socially oriented thinking by coding written and transcribed oral thought samples using Linguistic Inquiry and Word Count software (LIWC; Pennebaker, Booth, & Francis, 2007). LIWC is a text processing software package that recognizes over 2,000 of the highest frequency words and word stems in the English language and separates them into over 70 categories, including grammatical (e.g., pronouns, verbs, etc.) and content-related categories (e.g., social words, emotion words, etc.). Scores are automatically computed as a percentage of each participant's total word count and thus reflect the relative presence of a particular category within that individual's language sample. For the content-based LIWC categories, such as the ones we used, the presence of themes in language is interpreted as (implicit or explicit) activation of cognitions related to that theme (for review, see Tausczik & Pennebaker, 2010). To assess socially oriented thinking, thought samples were analyzed for use of references to friends (e.g., friend, roommate, boy/girlfriend) and/or family (e.g., mom, father, brother, aunt). LIWC friends and family scores were summed to reflect the overall proportion of words from both categories in the texts. This particular combination of word categories was novel and selected specifically to fit the theme of close and potentially supportive others. This strategy emulates a previous social support-related study (Pressman & Cohen, 2007), in which frequencies of words related to social roles (e.g., friend, student, neighbor, volunteer) were selected as an index of a construct the authors called social ties, described as “the extent to which social relationships are central to one's self-concept and day-to-day existence” (p. 262).

To assess the specificity of effects to cognitions about potentially supportive others versus people more generally, the LIWC category humans was also used. The humans category, which does not overlap with friends or family categories, captures references to nonclose others (e.g., girl, guy, kid).

2.3.2 ∣. Cortisol

Saliva was collected using the passive drool collection method (i.e., spitting into a tube via a drinking straw) and frozen at −20°C until it was assayed in duplicate using an RIA commercial kit (Catalogue #CA-1549; DiaSorin Inc., Stillwater, MN). Cortisol was measured in micrograms per deciliter (μg/dl). Raw values were natural log transformed.

2.4 ∣. Data analytic plan

Given the importance of both between- and within-person differences in cortisol activity, cortisol data were analyzed using free curve slope-intercept (FCSI) latent growth curve modeling (Wood, 2011; Wood & Jackson, 2013). This approach allows estimation of separate components for between- and within-person differences in cortisol and compares their strength in accounting for the overall variance in observed cortisol. The FCSI model estimates the variance of a latent “intercept” variable (i.e., variability in total response explained by between-person elevations across all recovery time points) as well as freely estimating the error rates and factor loadings on a latent “slope” variable (i.e., variability in trajectory explained by within-person momentary changes). Slope factor loadings can then be squared to determine the variability in cortisol response explained by rates of within-person change at that time point, and these values can be directly compared to the variability in the latent intercept. The FCSI model was developed in the context of developmental research, which is similarly concerned with repeated measures over time, albeit over longer time frames than a single laboratory session. To date, we are aware only of developmental papers using this approach, for instance, to model changes in child temperament (Wood, 2011) or personality disorder symptoms (Stepp, Keenan, Hipwell, & Krueger, 2014). Fundamentally, the statistical properties of the model are preserved from one context to another, so the inferences it supports should be similar, despite the novelty of the endocrine context.

Compared to other approaches for modeling change, the FCSI model has several advantages that lend it conceptually to evaluating HPA activity. Compared to other growth curve models, the FCSI model makes the fewest assumptions about the shape of the trajectory change over time; factor loadings onto the latent slope factor are estimated freely without constraining them to a linear, quadratic, or other fixed shape (Meredith & Tisak, 1990; Wood, 2011). This is ideal, given that neither linear nor quadratic growth models are well suited to cortisol data (see Ji, Negriff, Kim, & Susman, 2016; Lopez-Duran et al., 2014). The FCSI model also does not assume that change trajectories are the same for all individuals (Wood, 2011), which is important in the case of cortisol because not all individuals experience their peak cortisol at the same measurement time point (e.g., Lopez-Duran et al., 2014; Phan et al., 2017). Because the factor loadings for the latent intercept are all fixed to a value of one, the meaning of the FCSI model's latent intercept can be understood as the portion of variance in cortisol values attributable equally to vertical elevations across all time points in the model or, in other words, as the individual variability in overall amplitude at which each cortisol curve plays out (see Wood, 2011; Wood & Jackson, 2013). Importantly, the FCSI model treats the latent intercept and latent slope as orthogonal to one another (i.e., their covariance is fixed to zero), which means that these constructs are statistically distinguished from one another fully. This full statistical separation of between-subjects levels from within-subject changes offers interpretive advantages, described by some as parsing traitlike from statelike differences (see Wood & Jackson, 2013). In the case of the present study, the orthogonal treatment of intercept and slope makes it possible to understand which portions of variance in cortisol are uniquely implicated in any predictive effects of socially oriented thinking that may emerge.

Our analyses were conducted using the full information maximum likelihood estimator in MPlus (Version 8.0.0.1; Muthén & Muthén, 1998–2011–2011). Following common practice in structural equation modeling, we began by specifying an unconditional model to confirm the suitability of the FCSI approach to the data and describe the structure of cortisol in the present sample. To improve on the unconditional model, a second model regressed the latent intercept and slope on a baseline cortisol covariate. Finally, a full model examined the relationship of post-task cortisol trajectory to socially oriented thinking by regressing the intercept and slope latent variables on socially oriented thinking, covarying baseline cortisol and references to humans in general. Models were evaluated by examining conventional indicators of good model fit: nonsignificant χ² likelihood ratio test, Comparative Fit Index (CFI) and/or Tucker-Lewis Index (TLI) ≥.95, and root mean square error of approximation (RMSEA) <.05 (McDonald & Ho, 2002). Note that, while our sample size of 64 is smaller than some recommendations, growth curve models have been successfully fitted for samples far smaller; the number of repeated measures in this study, which exceeds minimum recommendations, mitigates sample size concerns (see Curran, Obeidat, & Losardo, 2010).

3 ∣. RESULTS

3.1 ∣. Descriptive results and zero-order correlations

Descriptive statistics and bivariate correlations for all study variables appear in Table 1. As the table shows, there is evidence for an association between socially oriented thinking and cortisol elevations late in the recovery period (T3, r = .27, p = .047; T4, r = .37, p = .007).

TABLE 1.

Descriptive information and zero-order correlations among study variables

	Mean or proportion	SD	1	2	3	4	5	6	7
1. Female gender	48 (75%)		1
2. Minority race	10 (15.6 %)		−0.27*	1
3. SOT	0.39	0.44	0.11	0.02	1
4. Cortisol BL	−0.31	0.18	−0.21	−0.03	0.18	1
5. Cortisol T1	−0.35	0.18	−0.08	−0.06	0.12	0.62*	1
6. Cortisol T2	−0.38	0.16	−0.09	−0.02	0.17	0.60*	0.95*	1
7. Cortisol T3	−0.44	0.16	0.03	−0.18	0.27*	0.56*	0.86*	0.92*	1
8. Cortisol T4	−0.48	0.18	0.12	−0.13	0.37*	0.48*	0.77*	0.85*	0.92*

Abbreviation: SOT, socially oriented thinking (family or friend words as a percentage of total word count).

^*p < .05.

3.2 ∣. Task response

3.2.1 ∣. Subjective mood

Compared to baseline, after the speech task (post-task T1), participants reported feeling more anxious, F(1, 56) = 7.75, p = .007 (M_B = 29.70, SD_B = 23.93; M_T1 = 37.36, SD_T1 = 26.91), and less calm F(1, 62) = 22.54, p < .001 (M_B = 64.97, SD_B = 22.72; M_T1 = 52.31, SD_T1 = 25.75). Responses to the distractor items did not change from baseline to post-task (Fs 0.00 to 0.62, ps .956 to .435).

3.2.2 ∣. Cortisol

The sample as a whole did not show a further increase in cortisol levels at T1 compared to the baseline measurement, with mean cortisol levels at baseline and T1 being nonsignificantly different from one another, F(1, 54) = 0.78, p = .382 (see Table 1 for all Ms and SDs). This pattern is known to occur when a pretask cortisol sample is taken after the participants have been in the lab for some time, as the novelty of the lab situation initiates the stress response for some individuals, causing feedback inhibition that can suppress a later response to the applied stressor (e.g., Duan et al., 2015; Phan et al., 2017; Ruttle, Shirtcliff, Armstrong, Klein, & Essex, 2015). Further probing indicated that a potentially nontrivial minority (n = 19, 30%) of participants demonstrated a cortisol increase from their pretask baseline compared to some point in the post-task period (n = 15 immediately from BL to T1; n = 3 peaking later at T2, T3; n = 1 peaking at T4). The modal nature of decreases to post-task, along with a notable amount of heterogeneity in responses, warrants caution in the interpretation of any forthcoming findings related to cortisol slope, as the meaning of the post-task slope cannot fairly be characterized as “recovery” from a task-related activation for the sample as a whole. An independent samples t test with equal variances not assumed showed that participants who showed increases (n = 19) versus decreases (n = 45) from baseline to any time point post-task did not differ in socially oriented thinking, t(33.80) = −.493, p = .625 (M_increasers = 0.37, SD_increasers = .44; M_decreasers = 0.43, SD_decreasers = .44).

3.3 ∣. Post-task cortisol slope

We first fitted an unconditional FCSI latent growth model to understand the structure of participants' cortisol after that task. Log-transformed cortisol measurements at T1 (immediately poststressor; 0 min) through T4 (45 min) were treated as the observed variables to estimate latent intercept and latent slope factors. The unconditional FCSI model had excellent fit with the data, indicated in part by nonsignificant χ² likelihood ratio test (p = .477; for additional fit statistics, see Model 1 in Table 2).

TABLE 2.

Fit statistics for FCSI models

	N	χ2	df	χ2 p value	CFI	RMSEA (90% CI)	AIC	BIC
1. Unconditional FCSI model	57	2.49	3	.477	1.00	0.00 (0.00–0.21)	−454.18	−431.71
2. Model 1, adjusted for baseline cortisol	57	3.98	5	.553	1.00	0.00 (0.00–0.16)	−476.45	−449.89
3. Model 2, regressed on SOT variables*	64	5.91	9	.749	1.00	0.00 (0.00–0.10)	−392.68	−336.55

Abbreviations: FCSI, free curve slope intercept; SOT, socially oriented thinking.

^*Final model, represented in Figure 1.

To understand the relative amounts of between-subjects and trajectory-related (within-subject) variance in cortisol responding across the recovery period, we examined the unstandardized parameter estimates for intercept variance and slope factor loadings (reported in Figure 1). The small but statistically significant intercept variance (σ_i = 0.03, p < .001) indicates the presence of individual differences in overall cortisol levels. Nonsignificant factor loadings for T1, T2, and T3 cortisol (−.05, −.03, .02, respectively; ps > .05) indicated that there was not a meaningful amount of variability in our data at these times and that cortisol measured at these times did not contribute to the latent change pattern of cortisol recovery. The T4 loading on the latent slope factor (.05, p = .047) indicated significant variability in cortisol at this time point contributing to the latent change trajectory. In other words, the latent slope (i.e., the total variability of within-person trajectories) was most strongly characterized by within-person changes occurring at T4.

FIGURE 1.

This figure illustrates the unstandardized parameter estimates for the free curve slope intercept (FCSI) model of cortisol recovery as a function of socially oriented thinking as indicated by references to friends and family. Covariates are baseline cortisol and references to humans more broadly, as indicated by references to nonclose others. Circles indicate latent variables, and boxes represent manifest (observed) variables. Slings (small, rounded, double-headed arrows) represent variances, straight arrows represent regression paths, and arcs (larger double-headed arrows) represent correlations that were allowed between manifest variables. Parameters appearing as “1” were fixed to one. The covariance between latent intercept and slope variables is not depicted because this parameter was fixed to zero. *p < .05; **p < .01; ***p < .001

Squaring the slope factor loadings (i.e., 0.0025 at T1, 0.0009 at T2, 0.0004 at T3, and 0.0025 at T4) allowed direct comparison to the intercept variance (i.e., .03) to determine the relative proportions of variance due to overall cortisol levels (intercept) versus latent change trajectories at each time point. The small proportion of these squared slope loadings relative to the intercept variance indicated that the majority of the variability in observed cortisol was explained by individual differences in the overall amplitude at which cortisol curves played out, rather than within-subject changes from point to point. In other words, between-subjects differences in overall levels had a larger effect size than within-person changes from point to point, when it comes to explaining the total structure of the cortisol data.

The next version of the model controlled for baseline cortisol (Model 2). Because the two models were not fully nested, a χ² difference test to compare models would not have been informative. However, fit statistics (Table 2) show that the model including baseline cortisol provided even better fit with the data. Baseline cortisol was strongly related to cortisol intercept (B = 0.53, SE = 0.10, p < .001), justifying its retention in the model. Baseline cortisol was not related to latent cortisol slope (B = −0.96, SE = 0.87, p = .271). Taken together, these associations suggest that baseline cortisol was more strongly associated with between-persons differences in overall cortisol levels than to within-person change trajectories.²

3.4 ∣. Socially oriented thinking as a predictor of cortisol regulation

To examine the relationship of socially oriented thinking to the cortisol response, we elaborated the FCSI model so that the latent intercept and slope variables for cortisol were regressed on socially oriented thinking. Baseline cortisol was retained as a covariate and was allowed to correlate with socially oriented thinking to help isolate effects of socially oriented thinking on distinct aspects of cortisol responding. The final model, depicted in Figure 1, showed excellent fit to the data (Model 3 in Table 2).³

We first examined the final model for the predictive role of covariates on cortisol responding. As in the previous model, the baseline cortisol covariate was related to the latent cortisol intercept (B = 0.52, SE = .010, p < .001) and marginally related to latent cortisol slope (B = −1.66, SE = 0.94, p < .076). The LIWC score for references to humans in general was unrelated to both the latent intercept (B = 0.07, SE = 0.05, p = .159) and latent slope (B = −0.16, SE = 0.52, p = .755). In other words, within-person changes in cortisol varied significantly and independently as a function of socially oriented thinking, above and beyond effects of covariates, whereas overall differences in level did not.

Consistent with our hypothesis, socially oriented thinking, while unrelated to both baseline cortisol (B = 0.02, SE = 0.01, p = .176) and latent cortisol intercept (B = 0.05, SE = 0.05, p = .351), was significantly related to latent cortisol slope (B = 1.48, SE = 0.43, p = .001). Figure 2 depicts cortisol trajectories for participants high and low in socially oriented thinking, based on a median split and all covariates from the final model. Individuals low in socially oriented thinking showed a downturn in their cortisol trajectory during Minutes 15–45 post-task, while individuals high in socially oriented thinking showed a more gradual recovery trajectory. This effect was statistically significant by the 45-min post-task mark, as indicated by the significant factor loading of the 45-min cortisol time point on the latent slope.

FIGURE 2.

Average growth curve for FCSI model of cortisol recovery and average curves for individuals high and low in socially oriented thinking (SOT). High and low SOT profiles were created for visualization purposes by adjusting the means of the FCSI curve based on a median split of frequency of references to friends and family, with all covariates from the final model included. Recovery time points are names relative to the offset of the stressor, such that “0 min” reflects cortisol immediately poststressor

4 ∣. DISCUSSION

Findings showed that trajectories of cortisol after a performance task differed as a function of individual differences in socially oriented thinking, the activation of cognitive representation of friends and family. Specifically, low levels of socially oriented thinking were associated with a steeper downward trajectory in cortisol during a 45-min post-task period. By contrast, high socially oriented thinking, which may represent an internalized facet of social support coping, predicted a more gradual decline. This effect involving the steepness of the cortisol trajectory was independent of the overall amplitude of participants' cortisol curves, which varied at the individual difference level and was controlled statistically in the FCSI model. This effect of socially oriented thinking on cortisol slope was also robust against all covariates tested, which included gender and race/ethnicity in preliminary models and baseline cortisol and references to humans in general in the final model.

The relationship between socially oriented thinking and cortisol trajectories is broadly consistent with prior evidence that social processes may shape the biological stress response even when social supporters are not present (Eisenberger et al., 2007; Engert et al., 2017; Lakey & Orehek, 2011). However, the direction of this effect seems perhaps counter-intuitive, in that processes related to protective elements of social cognition and perceived social support are often found to reduce, rather than amplify or extend, cortisol responses (Gunnar, 2017). Interpreting cortisol outcomes has proven challenging, given the absence of known “adaptive” shapes of cortisol response curves (Susman, 2006). Thus, given the preliminary nature of the present study, we can only speculate on the possible reasons why socially oriented thinking was associated with a more gradual downward cortisol trajectory. Consistent with the specific notion that prolonged cortisol activations may in some cases help mobilize coping resources (Shirtcliff et al., 2014), one possible interpretation is that the delayed cortisol decline was somehow involved in mobilizing a social cognitive orientation. Alternatively, perhaps thoughts of friends and family prolonged cortisol elevation by themselves acting as stressors; when close relationships are abrasive or conflictual, they fail to show a social buffering effect and may instead provoke increased stress response (e.g., Keicolt-Glaser et al., 1998). Post hoc review of participants' responses suggests this was likely not so, however, as most friends or family words occurred in neutral or pleasant contexts, including statements indicating desire to connect (e.g., “I'm currently thinking about one of my friends from back home, and his grandma's house, because when we'd go there and help out or hang out or whatever;” “There are friends back home that I haven't talked to in a while, too, that I should contact. I miss them;” “I should probably call my parents, uh, because I haven't talked to them in a really long time.”). Another source of stress that could have differentially affected socially oriented thinkers is the post-task experimental experience itself. Participants were relatively isolated during the post-task phase, with the only human interactions being scripted, task-oriented entrances by the lab assistant to collect saliva samples. Replication studies would benefit from incorporating additional trait and state measures to understand potential systematic differences in the impact of recovery period protocols on socially oriented thinkers' social experiences (e.g., feelings of loneliness, isolation, and perceived stress).

Inferring psychological processes from language is complicated by the imperfect correspondence of words with mental content (e.g., “the mother of all tasks” is not literally about mothers). Thus, our socially oriented thinking scores should not be taken as indicators of participants' explicit or conscious thoughts. Rather, these scores should be taken as indirect estimates of the presence of themes related to friends and family, activated explicitly or implicitly. This semi-implicit nature of LIWC allowed us to characterize thought samples without drawing participants' awareness to the socially oriented nature of our research question, thereby reducing demand characteristics, reducing participant bias, and providing a more ecologically valid measure of social responding. The linguistic approach to the study of individual differences is well established (Boyd & Pennebaker, 2017), and the use of social words in particular to index facets of social cognition has an existing precedent in the social support literature (Pressman & Cohen, 2007). Nonetheless, the particular construct of socially oriented thinking was defined and operationalized here for the first time and thus requires future replication and ongoing construct validation.

In addition to its limitations, this preliminary study also had several strengths, including the rigorous analytical approach, including testing of covariates, which yielded results with a high degree of precision. Overall, the FCSI model provided exceptional fit to the data despite our sample size. The FCSI approach allowed us to locate sources of variance in the cortisol data precisely and without imposing a particular geometric shape (e.g., linear, quadratic) on the curve a priori. Because the FCSI model statistically parses point-to-point changes in trajectory (latent slope) from vertical elevations across all time points (latent intercept) and treats these as statistically independent (i.e., orthogonal), the model is able to fully separate momentary, within-person fluctuations from potentially confounding effects of level (see Wood, 2011; Wood & Jackson, 2013). Even before the predictors of interest were evaluated, the unconditional model alone provided several insights into the structure of the cortisol activity observed in this sample. The variance term for the latent intercept revealed considerable variability in cortisol elevations across all recovery points, which can be thought of as vertical spread in cortisol levels, meaning that some individuals' post-task cortisol curves played out at higher amplitudes. This variance was proportionally many times greater than the trajectory-related variance at any time point, meaning that the relative size of trajectory, or momentary changes, was relatively small. The freely estimated factor loadings further characterized the sources of this—small in size—variability in trajectory as having stemmed mostly from cortisol sampled late in the protocol, 45 min post-task at T4.

Moving beyond the unconditional model, the inclusion of covariates makes the incremental effects observed especially noteworthy. Not surprisingly, elevation across all post-task points (i.e., latent intercept) was related to the baseline cortisol covariate, meaning simply that starting levels predicted subsequent levels. Yet, even while accounting for such strong intercept- or elevation-related effects, cortisol recovery trajectories nevertheless differed uniquely as a function of socially oriented thinking. Furthermore, it is striking that references to humans in general were neither related to the cortisol slope term nor interfered with the significance of the friend/family effect. Rather, the more gradual downward trajectory appeared uniquely related to the activation of social cognitions about close relationships. It has been noted elsewhere that cortisol responses and their regulation may be sensitive to the degree of closeness in supportive relationships (Hostinar, Johnson & Gunnar, 2015; Smith, Loving, Crockett & Campbell, 2009). Our covariate analyses thus contribute further evidence that certain social effects on cortisol are confined to closer relationship contexts.

As with any cross-sectional design, the direction of effects is unclear. Perhaps prolonged elevation of cortisol amplified socially oriented thinking as a means of coping with the task, consistent with functional accounts of cortisol elevation (Shirtcliff et al., 2014). Conversely, thoughts of friends and family may have prolonged cortisol activation, perhaps by activating neurocognitive (Eisenberger et al., 2007) or appraisal-based (Engert et al., 2017) effects of social support. Untangling these possibilities will require greater temporal specificity regarding when in time socially oriented thoughts occur, which was difficult in the present study due to low base rates of family/friend words. Future studies using more intensive thought sampling could model trajectories of socially oriented thinking and cortisol regulation in parallel and identify the temporal influences between them. State-level measures of socially oriented thinking could also clarify whether thoughts of close others tend to arise automatically, as part of the organism's mounting of a coping response. If social representations activate reliably after stressors onset, it might implicate socially oriented thinking as a dimension of the stress response itself, alongside more established stress response channels.

Results of this study are inextricable from important limitations and caveats. Although covarying gender did not alter the main finding (Footnote 3), the disproportionately female nature of the sample may limit generalizability and require follow-up. The absence of an overall cortisol response to the speech task, both for the modal participant and the aggregated sample, means that the downward trajectories observed in the post-task period cannot be described as cortisol recovery. Responses to psychosocial stress tasks are notoriously inconsistent, with some prior studies similarly failing to show group-level responses (e.g., Linnen, Ellenbogen, Cardoso, & Joober, 2012). In the present study, absence of a group-level task activation may be due to cortisol elevations already in effect at the pretask baseline, perhaps in response to the novelty of the lab situation. Thus, the meaning of downward slopes observed in the present study should be considered more mixed than in typical studies. For the approximately 30% of participants who showed a peak after the speech task, the decline may potentially represent cortisol recovery toward baseline levels. However, for the sample as a whole, and especially for the modal participant, the decline should be considered more conservatively as changes over time. These changes could have been multiply determined by a host of plausible factors and/or their interactions, including biochemical processes unrelated to acute stress recovery, such as tonic regulation of cortisol by mineralocorticoid receptors (de Kloet et al., 2016, Gesing et al., 2001; Joëls, Karst, DeRijk, & de Kloet, 2008), diurnal cortisol variability, which may also encode social effects (Adam et al., 2017; Chida & Steptoe, 2009) or regulatory effects of other neuropeptides such as oxytocin (e.g., Heinrichs et al., 2003; Linnen et al., 2012). Furthermore, although it did not appear that socially oriented thinking was associated with differential cortisol responses to the speech task, it remains possible that rates of cortisol decline were confounded with the amplitude of initial peaks from lab arrival levels, which this sampling regimen did not allow us to measure thoroughly.

In sum, the present findings, while remaining in need of replication and further interpretive work, can be described with a high degree of precision: cortisol trajectories or rates of change over time, but not overall levels, were related to thinking about friends and family, but not about humans in general. This is the first demonstration to our knowledge that trajectories of cortisol change over time differ as a function of mere activation of thoughts of close relationships, in the absence of explicitly interpersonally themed study elements. This finding underscores the need to understand the internalization of social support representations (e.g., Lakey & Orehek, 2011; Marroquín & Nolen-Hoeksema, 2015) and to consider further the possibly heightened importance of social resources during cortisol recovery periods (Meuwly et al., 2012). Broadly, results echo notions that stress can prompt the drive to affiliate (Rimé, 2009) and that neurocognitive systems may have evolved to depend on social resources (Coan & Sbarra, 2015). Future studies of cortisol activity might consider applying our statistical approach, which is equipped to model change trajectories without a specific geometric shape and isolate them from overall activation levels, and thus appears promising for capturing the complex dynamics of cortisol changes.