Effects of manipulating the amount of social-evaluative threat on the cortisol stress response in young healthy women

Abstract

Psychological stress is known to activate the hypothalamic–pituitary–adrenal axis, resulting in the release of cortisol from the adrenal cortex into the bloodstream. Cortisol is the major human stress hormone and its health correlates continue to be investigated by laboratories around the world. One line of research suggests that specific situational variables play a role in the creation of a stressful situation. The current study examined the effects of systematically varying several situational characteristics on the cortisol stress response in 80 healthy young women exposed to a public speaking task. Three main factors and its interactions were investigated by locating the expert panel either inside or outside of the room, having the subjects speak either about themselves or somebody else, and by asking half of the subjects to perform a distractor task in addition to performing the public speaking. We interpreted these manipulations as variations of social evaluative threat, ego-involvement, and divided attention. We hypothesized that the variations and their interactions would cause differences in endocrine stress responses.

The results showed that only the manipulation of social-evaluative threat had a significant main effect on the cortisol stress response in women. There was a further trend (p = 0.07) for a four-way interaction effect. No other main or interaction effects could be observed. We conclude that in women, social-evaluative threat affects the endocrine stress response. This is in contrast to a previous study showing no effects of this variation in men. Thus, future studies should more closely investigate sex or gender effects that might be interacting with the situational aspects of a stressful task.

Keywords::

• Cortisol
• ego involvement
• gender
• social-evaluative threat
• stress
• TSST

Introduction

Psychological stress contributes both directly and indirectly to the etiology of many diseases including depression (McEwen 2005), hypertension, truncal obesity (Peeke and Chrousos 1995; Kudielka and Kirschbaum 2003), diabetes mellitus type 2 and related thoracic disorders (Kudielka and Kirschbaum 2003). Thus, many laboratories investigate the mechanisms by which stress has these health impairing consequences. It is well known that stress causes the activation of the hypothalamic–pituitary–adrenal (HPA) axis. The HPA axis is the major stress axis of the organism, producing and releasing a number of stress hormones with the final product cortisol released into the blood stream (Mason 1968; Sapolsky et al. 2000). Important interindividual differences in the hormonal response to stress exist, at least in part mediated at the level of the central nervous system, in interaction with the perception and processing of the stressor. One line of research investigates which situational elements modulate the perception of stress, and the subsequent initiation of the hormonal stress response. This line of research was originally begun by Mason (1968), who pointed out the importance of novelty, unpredictability, uncontrollability, ego-involvement (EI), and anticipation of negative consequences as situational characteristics that contribute to the perception of a situation as stressful. Recently, Dickerson and Kemeny (2004) performed a meta-analysis and showed that social-evaluative threat is another key situational characteristic that if present, will contribute to the perception of a situation as stressful. Similarly, Kajantie and Phillips (2006) proposed that the amount of personal involvement by an individual, or EI, in a social interaction influences the cortisol secretion in response to stress.

Finally, at least one study has investigated the effect of having to perform several tasks simultaneously on stress perception and neuroendocrine responses (Elsemore 1990). This situational condition is probably similar to many modern work environments, where subjects are engaged in multiple responsibilities at the same time. In one study, using a multitasking stressor, results suggested that this type of task induces significant elevations in immunoglobulin-A levels in at least some of the subjects (Wetherell et al. 2004).

To our knowledge, only a few studies have attempted to systematically manipulate several components of psychological stress simultaneously in the same design. However, experimenting with a combination of several situational characteristics is likely more representative of real-world situations as compared to single-stress situations. Therefore, it is important to investigate if and how situational elements differentially contribute to, and interact with each other, in the perception and processing of stress.

One study (Gruenewald et al. 2004) has investigated the manipulation of social evaluative threat on the cortisol stress response. In this study, the Trier Social Stress test (TSST) was performed in two conditions, either in front of a regular audience (evaluation condition) or alone in a room (non-evaluative condition). The authors observed a significant cortisol stress response only in the evaluative group, while no significant cortisol release was found in the condition where the subject was alone in the room (Gruenewald et al. 2004). Although this manipulation does not investigate proportional stress responses by implementing gradients of threat, it does show that social evaluative threat is a necessary component of the TSST.

Rohleder et al. (2007) also investigated the impact of social-evaluative threat in an evaluative group setting, by varying the number of participants (couple vs. group dancing) in a “natural” stress setting (dance competition), with the number of judges remaining the same. Kelly et al. (2007) tested the effects of panel in/panel out condition in a virtual stress environment and demonstrated effects of this manipulation on the cortisol stress response as well.

Our group has recently added to this line of work by manipulating the number of panel members and the location of the panel in a public speaking stress task (Andrews et al. 2007), known to reliably induce stress including physiological and endocrine stress responses (Kirschbaum et al. 1993). In our modification of this task, subjects were exposed to an audience consisting of either one or two people, sitting in front of or behind a one-way mirror. Interestingly, we could not observe any endocrine differences with these manipulations, that is, all groups showed similar cortisol stress responses. However, this study investigated only men, and recent findings suggest that women might be more sensitive to social types of stress (Stroud et al. 2002). Thus in the current study, we wanted to test the effects of manipulating situational characteristics on the stress response in women.

Given the recent findings in the literature, in addition to manipulating the amount of social-evaluative threat by locating the expert panel either in front of or behind a one-way mirror, we further manipulated the stressful situation by letting the subjects either speak about themselves, or a different person. We speculated that this manipulation might result in a variation of EI, and hypothesized that greater EI (speaking about oneself) could lead to higher stress responses. Finally, given the recent studies on multitasking and the importance of multitasking in typical work environments, we introduced a distraction task over public speaking. Subjects were asked to keep track of time and inform the observers whenever they thought a minute had passed. We interpreted this manipulation as “divided attention” (ATT-d) and hypothesized that the ATT-d condition would elicit a greater cortisol stress response. We implemented these conditions by modifying the widely used “TSST” (Kirschbaum et al. 1993), and used a nested design to test main and interaction effects.

Methods

Subjects

Eighty women between the ages of 18–35 years (mean age+SEM = 22.63 ± 3.0 years) were recruited from McGill University. Medical history was ascertained through a telephone interview and those who reported a presence or history of neurological or psychiatric conditions, diabetes, asthma, respiratory disease, infectious illness, thyroid dysfunction, adrenal dysfunction, or excessive drug or alcohol use were excluded from the study. To control for the effects of the menstrual cycle on HPA axis regulation, only women using estrogen containing oral contraceptive pills were included, to minimize variations of estrogen levels in our study population (Kirschbaum et al. 1995; 1999). Smoking was a further exclusion criterion, due to its known effect on HPA axis regulation (Kirschbaum et al. 1993). The study was approved by the Ethics Research Board of the Douglas Institute and all subjects gave written informed consent prior to participation.

General procedure

All subjects were randomly assigned to eight groups, which contrasted the situational variations interpreted as reflecting: high (h) vs. low (l) social-evaluation (S-Ev), high (h) vs. low (l) ego-involvement (EI), and divided (d) or undivided (u) attention (ATT), in a completely balanced design. This led to 40 subjects per group for all first-order (main effects) conditions. Thus, we had forty subjects assigned to the high social-evaluation condition (S-Ev-h) and 40 subjects to the low social-evaluation condition (S-Ev-l), and similarly for the ego-involvement (EI-l or -h) and the attention (ATT-d or -u) conditions. For the second order conditions (simple interaction effects), we had 20 subjects assigned to all possible combinations of two variables: S-Ev-h and EI-h, S-Ev-l and EI-h, S-Ev-h and EI-l, S-Ev-l and EI-l, and similarly for all combinations of S-Ev and ATT, and EI and ATT. At the level of the third-order (multiple interaction effects), our design then resulted in ten subjects per cell for the different combinations of S-Ev, EI, and ATT. This setup resulted in a total of eight groups, in all possible iterations of S-Ev, EI, and ATT.

Owing to the circadian rhythm of cortisol secretion all subjects were tested in the afternoon between 14.00 h and 17.00 h. On the testing day, participants were met by the investigator upon arrival at the Institute and brought to a room where they were asked to rest for 25 min. They were then asked to fill out consent forms and questionnaires, followed by the modified version of the TSST. After the testing had been completed and subjects had filled out all questionnaires, they were debriefed and informed about the study purpose.

Psychological questionnaires

All participants were instructed to complete a number of questionnaires designed to assess personality variables known to affect the cortisol stress response such as self-esteem, locus of control and parental care (Pruessner et al. 2005). To assess these factors we used the parental bonding index (Parker et al. 1979), the Rosenberg self esteem scale (Rosenberg 1965), the NEO-five factor inventory (Costa et al. 1992) and the questionnaire for competency and control (Greve et al. 2001). There were no specific hypotheses associated with this selection, rather the desire to control for possible effects of variations in these measures.

Psychological stress

The original TSST protocol was modified to allow the manipulation of the location of the panel (interpreted as variation of social-evaluative threat), the topic of the speech about oneself or another person (variation of EI), and keeping track of time in addition to the speech (ATT-d). While these manipulations were guided by previous publications from our and other laboratories, no direct manipulation check was performed. Ten minutes prior to entering the test room and the beginning of their performance, subjects were instructed that they had to prepare a 10-min speech. They were also informed that two behavioural analysts would evaluate their verbal and paraverbal skills throughout their performance and that they would be video recorded during the task for further behavioural analysis.

In the testing area, the two panelists were either located in the presentation room together with the subject [“panel in” (S-Ev-h) condition] or on the other side of a one-way mirror where they could see the subject without being seen by the subject [“panel out” (S-Ev-l) condition]. In the panel-out condition, the panelists communicated with the subject via an intercom system. The equipment was tested before each speech to ensure effective communication between subject and panelists. Upon entering the testing room, subjects were instructed to either talk about themselves [present own strengths and weaknesses by giving examples of themselves in real-life situations; high EI (EI-h) condition] or to talk about the same characteristics in another person that they knew well [low EI (EI-l) condition]. During the speech, subjects were asked to keep track of time and inform the panel whenever they thought a minute had passed (ATT-d condition). Failure to inform the panel within a two minute period about the time passed would result in a panel member interrupting the speech and reminding the subject to keep track of the elapsed time. In contrast, the undivided attention (ATT-u) group could perform their presentation without any need to keep track of time (ATT-u condition).

During the task, the subject could speak freely for 10 min, but was prompted by prepared questions if a 20 s pause arose, for example, “Are there any other strengths (weaknesses) that you can think of?” After the end of the testing subjects were escorted to a private room where they could rest, complete questionnaires, and were asked to provide saliva samples according to the experimental design.

Salivary cortisol measurement and analysis

A total of seven cortisol samples was taken throughout the testing session for each subject. The first two samples were baseline samples and were taken 20 and 10 min before the TSST ( − 20, − 10). The third and forth salivary samples were obtained prior to, and immediately after the testing (0, 10). After the TSST, three additional samples were obtained at 10-min intervals (20, 30, 40). SARSTEDT cotton swabs (salivettes) were used to collect all saliva samples (Sarstedt, Quebec City, Quebec, Canada). After use, salivettes were stored in a − 20°C freezer until analysis. All samples were analyzed using a time-resolved fluorescence immunoassay (Dressendorfer et al. 1992). Intra and interassay variability were below 6 and 10%, respectively.

Statistical analysis

Serial one-way ANOVAs with group as independent variable, and age and personality variables as dependent variables were conducted to test whether there were any age or personality differences between the groups. Since cortisol values were not normally distributed, we performed a logarithmic conversion prior to ANOVA. For the analysis of the main and interaction effects of our manipulations, we performed a four-way (time × S-Ev × EI × ATT) mixed design ANOVA using the seven logged cortisol measures as dependent variables. In line with earlier studies from our group showing the importance of global self-esteem in association with the cortisol stress response in psychological stress paradigms (Pruessner et al. 1999), we covaried for global self-esteem levels derived from the global self-concept scale of the questionnaire for competency and control (Greve et al. 2001).

To test for violations of sphericity assumptions, we performed a χ²-test. In cases of violations, we performed Greenhouse Geisser corrections. In cases of significant effects in the ANOVA, we chose Tukey's HSD post hoc tests for follow-up. Significant effects were further analysed for their effect sizes using the formulae provided by Cohen (1988). All statistical analyses were completed using the SPSS software package 11.0.4 on Apple Macintosh computers running OS X.

Results

The serial one-way (group) ANOVAs using personality variables and age as the dependent variables did not show any significant differences for these variables between the eight groups (all F < 1.57, p>0.16). The four-way (time × S-Ev × EI × ATT) repeated measures mixed design ANOVA using the seven logged cortisol measures as dependent variables first showed that the assumption of sphericity was violated for the within subjects effects (χ² = 139.4; p < 0.001). Thus, for the repeated measures main and interaction effects, Greenhouse Geisser corrections were performed. Further analyses revealed a main effect of time, F(3.6, 253) = 2.79, p = 0.032, with mean levels of cortisol decreasing prior to the onset of stress, and increasing in response to the TSST. Power analysis showed a small effect size of f² = 0.02, with rather small amounts of variance explained (ω² = 0.02) by the time factor.

Next, we found an interaction effect of time × S-Ev, F(3.6, 253) = 3.3, p = 0.015 (Figure 1), f² = 0.03, ω² = 0.03. Post hoc tests confirmed that the EI-h condition led to greater cortisol stress responses after the stress task as compared to the EI-l condition (Figure 1).

Figure 1.  Effect of the public speaking task on the salivary cortisol response in the two groups exposed to either high or low social-evaluation. The x-axis shows time in minutes in relation to the onset of the speaking task (time 0), the y-axis shows the log transformed mean cortisol values (nmol/l). S-EV, social evaluation. The error bars shown are standard errors of mean. The number of subjects in each group was n = 40.

Finally, there was a trend for an interaction effect of time × S-Ev × EI × ATT, F(3.6, 253) = 2.22, p = 0.074 (Figure 2), f² = 0.01, ω² = 0.01. It appeared that within the S-Ev-h condition, ATT-d led to a greater cortisol stress response if the ego involvement was also high; this pattern was opposite (ATT-u resulting in larger cortisol stress responses) if the EI was low (see Figure 2(a) for the S-Ev-h condition and Figure 2(b) for the S-Ev-l condition).

Figure 2.  (a) and (b) Effect of public speaking on salivary cortisol concentration in relation to the eight experimental groups. The x-axis shows time in minutes in relation to the onset of the speaking task (time 0), the y-axis shows the log transformed mean cortisol values (nmol/l). To increase readability, (a) shows the effects within the S-Ev-h condition, while (b) shows the effects within the S-Ev-l condition. S-EV, social evaluation; EI, ego involvement; ATT, attention; h, high; l, low; u, undivided; d, divided. To increase readability, no error bars are shown. The number of subjects in each group is shown next to the group label.

None of the other interaction effects were found to be significant, all F < 1.11, p>0.3. Also, there was a main effect of S-Ev, F(1, 70) = 4.17, p = 0.045, f² = 0.05, ω² = 0.04. To further investigate the interaction effect of time × S-Ev a Tukey HSD post hoc test was performed. No differences were found for the pre-TSST measures ( − 20, − 10, and 0 min) between the groups (p>0.20). Saliva samples of the S-Ev-h group at +10,+20, and +30 min, were significantly higher than of the S-Ev-l group, p < 0.05, while differences for the recovery sample (+40 min) showed a trend, p = 0.09.

A Tukey HSD post hoc was also carried out to explore the main effect of time on the cortisol response. Significant differences in salivary cortisol concentration were found between samples at − 10, 0, and +10 min compared to timepoints +20 and +30 min (stress reactivity period). Moreover, sample two differed from sample at 0 min, and this timepoint also differed from samples at +10 and +40 min.

Discussion

The aim of this study was to investigate if social-evaluative threat, ego involvement and ATT-d independently contribute to the cortisol stress response as measured through salivary cortisol in 80 young healthy women using oral contraceptives. This was achieved by using a modified version of the TSST and by varying the position of the panel (panel in vs. panel out), the topic of the speech (self-relevant vs. other-relevant) and keeping track of time or not (divided versus undivided attention). We hypothesized that these factors would independently affect the cortisol stress response.

Our hypotheses were found to be partially supported as we did find significantly different salivary cortisol levels between the two social-evaluative threat conditions. Further, we found a significant interaction of S-Ev by time, suggesting that the cortisol response to the stress task over time was significantly different between the two S-Ev groups. Finally, we found a trend for the complex four-way interaction effect of time by S-Ev by EI by ATT, with ATT-d and ATT-u showing opposite patterns of cortisol stress responses over time depending on EI, but only if S-Ev was high. Altogether, it has to be noted that the explained amounts of variability for the different significant effects were rather low, ranging from 1 to 4%.

The major finding for this study was the observed effect on the stress response for the S-Ev condition, operationalized by the “panel in” and “panel out” groups. Speaking in front of the panel led to overall higher salivary cortisol levels as compared to presenting in front of the one-way mirror. If the presence of the panel is indeed associated with the amount of S-Ev, this result would indicate that the cortisol stress response in women is proportional to the perceived degree of S-Ev. Aside from the visual presence, the “panel in” condition also allowed for non-verbal feedback from the panel members. Although the panel was instructed and is trained to be neutral and to give no verbal or nonverbal feedback (other than the prepared questions), even a face showing a neutral expression can be considered to communicate at least indifference, and thus some sort of evaluation. This communication was not present in the “panel out” condition. Further contributing to this differential response is the verbal communication; in the “panel out” condition, communication occurred via an intercom system, and this more anonymous procedure might have allowed the subject to further disengage from the task. A number of previous studies have also shown effects of social-evaluative threat on the cortisol stress response, although none is directly comparable with the current experiment. One study did compare the presence of a panel to the complete absence of the panel (Gruenewald et al. 2004), and the other manipulated the social-evaluative threat by the number of other people in the room, while keeping the number of panel members constant (Rohleder et al. 2007). A third study tested the effects of panel in/panel out condition in a virtual stress environment (Kelly et al. 2007) and demonstrated effects of this manipulation on the cortisol stress response as well. Taken together, the current findings together with the previous studies point to the importance of social evaluative threat as a factor in determining the extent of the cortisol stress response.

No significant effect of ego involvement was observed, indicating that speaking about oneself or someone else in public does not affect the cortisol stress response. Perhaps this can be interpreted to mean that the stressfulness originates from the simple fact of giving a speech in public. Future studies could follow up on this initial null result by comparing cortisol responses between subjects talking about themselves or listening to a friend talk about them. A post hoc power analysis revealed that assuming a moderate effect size of at least (f² = 0.05) with the current number of subjects, we had a power of at least (1 − β) = 0.90 to find differences in the current study.

Moreover, no significant differences in the cortisol response were observed between our ATT-d and ATT-u groups. This manipulation was based on previous findings which suggest that multitasking is a source of stress, although evidence to support this premise is sparse (Wetherell et al. 2004). Of course, our way of introducing multitasking was very different from previous implementations, which asked subjects to solve four tasks on the computer simultaneously. Perhaps the addition of only one task, performed every minute, was not stressful enough to lead to a differential stress response with respect to our ATT groups. Again, these being negative results we need to take into account the beta error. A power analysis established that we had a power of at least (1 − β) = 0.90 to find differences in the current study assuming an effect size of f² = 0.05. Finally, we observed a trend for a four-way interaction effect of time by S-Ev by EI by ATT, suggesting that the subgroups showed a tendency to be different from each other. Here, it could be speculated whether the combination of our different conditions led to some combined effect on the salivary cortisol levels over time. It appeared that ATT-d and EI only exerted effects if the S-Ev was high; within this subgroup, ATT-d led to a greater cortisol stress response if the ego involvement was high; however, ATT-u resulted in larger cortisol stress responses if EI was low. Unfortunately, this interaction failed to reach statistical significance and is only a trend, thus we have to refrain from discussing this further. Here, it also has to be noted that although the completely nested design maximized our chances of revealing effects on the first (main effects) and second (simple interaction effects) order computations, with this most complex four way interaction, our power was compromised by only having ten subjects per group. A power analysis revealed that we only had a power of at least (1 − β) = 0.50 to find differences in the current study assuming an effect size of f² = 0.05, for the three-way interaction.

Our results are in stark contrast to Andrews et al. (2007), which did not find a differential stress response to varying degrees of social-evaluation in men (including a S-Ev condition). This suggests the possibility of a sex or gender difference, and although not tested directly in the current study, this emerges as one of the important suggestions for future studies. The results of the current study combined with the results of Andrews et al. (2007) allow speculation about whether social-evaluative threat has a differential effect in men and women. It might be that women react differentially to the amount of perceived social-evaluation while men do not. This is an important hypothesis for future studies; and while no conclusions can be drawn from the previous studies that either tested men or women, they point to the necessity of investigating this effect in a study with both men and women.

These results are directly in line with Stroud et al. (2002) who argued that women are more likely to be attuned to social rules and they would be more responsive to social-evaluation in comparison to men for whom stressors are traditionally deemed to be more achievement based. Moreover, the idea that women are more sensitive to social rejection compared to men further supports the hypothesis that a difference in the perception of stress is at the root of the sex or gender based difference in depression rates.

Limitations

Finally, there are a number of limitations with respect to the current study. We only included women in the current study, so we are unable to speculate on the outcome in men. However, situational characteristics and the team of investigators were identical to the Andrews et al. (2007) study investigating men, allowing some comparisons to be drawn with that previously published study.

Furthermore, only women on oral contraceptives in early adulthood were tested, thus we can not conclude anything about possible effects of our manipulations on women not on oral contraceptives at different times in their menstrual cycle, or after the menopause. Also, no data are available about the exact type of oral contraceptive pill and their hormonal contents or effects, which could have further affected the results.

Next, we need to emphasize that our situational manipulations were interpreted by us as manipulations of social-evaluative threat, EI, and ATT-d, without having conducted an independent manipulation check. Whether the evaluation panel inside the room really leads to a perceived increase in social-evaluative threat is thus speculative, but we were interested in the effects of our manipulation on the cortisol stress response. While we cannot conclude with certainty that our experimental design was a manipulation of these concepts, we can at least conclude that the location of the panel has an effect on the cortisol stress response, while the topic of the speech and the keeping track of time do not seem to have any discernible effect.

As a final point of consideration, we did not record any measures of subjective stress experience. While this might be regarded as a limitation, we have failed to see systematic associations between the amounts of perceived stress and the cortisol stress response whenever we recorded them. Thus, it is questionable whether this would have added much to the current design. It seems that personality variables like locus of control and self-esteem–independent trait markers–are more systematically associated with the cortisol stress response. These variables were included in the current study to control for possible effects that they might have had on the cortisol stress response.

Acknowledgements

We would like to thank Mr Rafael Languay, who assisted in study coordination and data management. This study was supported in part by a Natural Sciences and Engineering Research Council of Canada (NSERC) Research Grant to JCP.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.