In 3 experiments, we offer evidence for the theoretical position that having power leads to a reduction in the stress response—which can have both positive and negative consequences. Experiment 1 tested the power-buffers-stress hypothesis in the context of a high-pressure mock job interview. Experiment 2 extended further the power-buffers-stress hypothesis by testing the effect on a physical stressor—an ice water submersion task. Experiment 3 tested the hypothesis in the context of an interrogation about theft in a high-stakes mock crime. Across three experiments, high-power individuals (vs. low-power) exhibited less of a stress response across measures of emotion, cognition, physiology, and nonverbal behavior. We suggest some specific biological trajectories through which power may buffer individuals from the stress response.
Power Buffers Stress – For Better and For Worse
Vladimir Putin lives an intense life. Decisions, which for him, are routine – carry great local and international consequence and affect the social and economic livelihood of millions. He, like many leaders, wakes early and retires late. He meets with other international luminaries frequently and makes high-stakes public speeches which are broadcast to millions (sometimes billions) of people. To meet the demands of his position, he necessarily pushes his mind and body to the outer limits of mental and physical well-being. We might imagine that Vladimir Putin’s daily life would be become overwhelmingly, chronically stressful. Yet, he persists and even thrives. How does Putin – or any other powerful person—meet these overwhelming psychological and physical challenges? Vladimir Putin possesses a great deal of power as he resides in his current, third, non-consecutive term as president. Does Putin’s power help him respond to and manage these stressors? In other words, instead of increasing stress, could possessing power actually buffer us from the stress response? This question is the focus of the current research.
Early research harvested from the animal literature in the 1950’s led to the idea that power is associated with increases in the stress experience. These findings came to be known as the "executive stress syndrome" (Brady, Porter, Conrad, & Mason, 1958). The conclusion drawn from this work was that those at the top of a hierarchy (vs. those at the bottom) demonstrate relatively higher chronic levels of the stress hormone cortisol. This animal research was applied widely toward the understanding of both human and nonhuman primates. Although the “executive stress syndrome” was certainly a provocative hypothesis, more recent research in both nonhuman animals and humans has failed to provide strong empirical support for it. If anything, it seems that the “executive stress syndrome” occurs only in a very narrow kind of social structure: rigid societies in which hierarchies can be threatened or are unstable and power can be lost. In these cases, power is maintained through continuous aggression and intimidation (Hellhammer, Buchtal, Gutberlet, & Kirschbaum, 1997; Sapolsky, 1990; Sapolsky, 2005; Sapolsky, 2011; Jordan, Sivanathan, & Galinsky, 2011). However, the findings on power and stress continue to be mixed, and their precise association still remains a hot topic of debate.
While research converges on the suggestion that power increases the number and severity of stressors in one’s life (Dansereau, Graen & Haga, 1975; Hogg, 2001), whether power increases (or decreases) the stress response is largely an open question. There is some direct evidence to suggest that power is linked to less—and not more—of a stress response. For example, in nonhuman primates, power-holders show lower levels of basal cortisol, and cortisol sometimes drops as power is acquired (Abbott, Keverne, Bercovitch, Shively, Mendoza, Saltzman, Snowdon, Ziegler, Banjevic, Garland, & Sapolsky, 2003; Callaway, Marriott, & Esser, 1985; Coe, Mendoza, & Levine, 1979; Rejeski, Gagne, Parker, & Koritnik, 1989; Sapolsky, Alberts, & Altmann, 1997; Winberg, Overli, Lepage, 2001). As such, powerful primates (including humans) live longer, are more disease-resistant and recover more quickly from psychological and physical stressors (Sapolsky, 1990; Sapolsky, 2005; Sapolsky, Alberts, & Altmann, 1997). Research in public health and medicine also provides evidence for the hypothesis that power may buffer the stress response. Individuals from high-ranking social groups (e.g., whites; those with wealth and education) do not suffer from chronically elevated cortisol and associated diseases as much as individuals from lower-ranking social groups (e.g., US-born African Americans; individuals without wealth or education; Adler, Boyce, Chesney, Folkman, & Syme, 1993; Cohen, Schwartz, Epel, Kirschbaum, Sidney, & Seeman, 2006; Gravlee, 2009; Krieger, Rowley, Herman, Avery, & Phillips, 1993; Williams & Collins, 1995).
If power does indeed buffer the stress response, the phenomenon would help to further unite a large body of literature, which seems to be of two minds. On the one hand, many papers have demonstrated all of the ways in which power leads to behavior with positive social consequences. On the other hand, an equally large body of work has demonstrated how power leads to behavior with negative social consequences. How can it be that power leads to behavior with both good and bad social consequences? Based on previous research and theorizing, and the simplicity afforded by the idea that power buffers stress, we predicted that possessing (versus lacking) power would buffer individuals from the stress response—for better and for worse.
For Better: The Positive Consequences of Power. Power has been shown to promote behaviors with many positive consequences. High- versus low-power leads to successfully obtaining resources (de Waal, 1998; Keltner, Gruenfeld, & Anderson, 2003). The powerful (vs. the powerless) are more action-oriented and willing to approach a difficult task (Galinsky, Gruenfeld, & Magee, 2003). They are more optimistic (Anderson & Galinsky, 2006), reward-focused (Anderson & Galinsky, 2006; Carney, Cuddy, & Yap, 2010; Ronay & von Hippel, 2010), and they feel more agency and control over their own body and mind (Anderson & Berdahl, 2002; Galinsky et al; Keltner et al.). The powerful are more pro-social (DeCremer & Van Knippenberg, 2002; Griskevicius, Tybur, & Van den Bergh, 2010; Harbaugh, 1998), they generally feel good (Anderson & Berdahl, 2002; Keltner, Gruenfeld, & Anderson, 2003) and possessing power helps us express positive emotions (Berdahl & Martorana, 2006). The powerful are more behaviorally consistent (a mark of psychological health; Chen, Lee-Chai, & Bargh, 2001; Keltner, Young, Heerey, Oemig, & Monarch, 1998) and enjoy higher self-esteem, better physical health, and increased longevity (Adler, Epel, Castellazzo, & Ickovics, 2000; Barkow et al., 1975; Bugental & Cortez, 1988). The powerful think carefully about and find unique value in individuals (Overbeck & Park, 2001) and are more likely to focus on interpersonally rewarding aspects of social interaction (Anderson & Berdahl, 2002). All of these experiences enjoyed by the powerful reflect and support the idea that power may buffer the stress response; but for every behavior with a positive social consequence, there seems to be a behavior with an equally impactful negative consequence. News headlines are replete with examples of people in positions of power – business (Madoff), academic (Stapel), and political leaders (Nixon) - who fall from their position of power and grace because of societal, ethical or moral transgressions, including stealing, cheating and lying.
For Worse: The Negative Consequences of Power. High- versus low-power has been shown to increase behaviors with negative social consequences. Power leads to a general reduction in empathy and empathic responses (Ronay & Carney, 2013; van Honk & Schutter, 2007; van Kleef, Oveis, van der Löwe, LuoKogan, Goetz, & Keltner, 2008). Likely stemming from a reduction in empathy, Kipnis (1972) demonstrated that people with power objectified the powerless and treated them more poorly—a finding echoed in Gruenfeld, Inesi, Magee, & Galinsky (2008). These findings are also consistent with the link between higher testosterone (a dominance hormone) and increases in perceiving others as purely a means to an end (Carney & Mason, 2010). In negotiation contexts, high-power negotiators were found to bluff more and exchanged less information with their low-power counterparts (Boles, Croson, & Murnighan, 2000; Crott, Kayser, & Lamm, 1980). The powerful (vs. the powerless) are more likely to be moral hypocrites (Lammers, Stapel, & Galinsky, 2010), engage in more adversarial behaviors (Mazur & Booth, 1998; Dabbs & Morris, 1990), cheat (Lammers, Stapel, & Galinsky, 2010; Yap, Wazlawek, Lucas, Cuddy, & Carney, 2013), engage in infidelity (Lammers, Stoker, Jordan, Pollmann, & Stapel, 2011), steal (Yap et al.), stereotype (Fiske, 1993), and violate traffic laws (Piff, Stancatoa, Côté, Mendoza-Denton, & Keltner, 2012; Yap et al.). The powerful also engage in more serious criminal behavior including sexual harassment (Pryor, LaVite, & Stoller, 1993), hate crimes (Green, Strolovitch, & Wong, 1998), violent crimes (Dabbs, Carr, Frady, & Riad, 1995) and child abuse (Bugental & Cortez, 1988). How is it possible that power can simultaneously lead to behavior with both positive and negative social consequences?
Current Research: Power Buffers the Stress Response
We propose that a common stress-buffering mechanism may underlie the influence of power on both pro- and anti-social behaviors. In three experiments, we test the hypothesis that power buffers individuals from the stress response – for better and for worse. Experiment 1 tested the power-buffers-stress hypothesis using a stress-eliciting laboratory paradigm—the Trier Social Stress Test in which participants must deliver a speech in front of an evaluative audience (e.g., Foley & Kirschbaum, 2010). Experiment 2 used an ice water submersion task to test the power-buffers-stress hypothesis in the domain of physical stress (Hines & Brown, 1932). Experiment 3 tested the power-buffers-stress hypothesis in a domain with potentially negative social consequences – lying. Borrowing a naturalistic theft paradigm from the criminal justice literature, we tested whether power would buffer the stress response during a high-stakes interrogation about a crime. All experiments tested the stress-buffering effects of power on emotion, cognition, physiological stress response, and nonverbal displays of stress.
Operational Definitions of: (1) The Stress Response and (2) Power
The Stress Response. Hans Selye, in 1936, studying the endocrinological systems of animals, borrowed the term “stress” from engineering to describe phenomena he was observing in his lab. The term was used to describe the body’s nonspecific response to an “insult.” An insult can be a real or imagined psychological or physical stressor which results in the mobilization of physiological resources so that the organism can either “fight” or “flee.” One of the primary stress pathways, the hypothalamic-pituitary-adrenocortical (HPA) axis, is activated in response to a stressor. This activation results in the secretion of the hormone cortisol, which is detectable in saliva approximately 20-30 mins after the stressor occurs (Dickerson & Kemeney, 2004). In the current research we define the stress response as cortisol reactivity to a physical or psychological stressor. We also examined self-reported stress and nonverbal indications of stress (specific nonverbal behaviors defined in the context of each experiment).
Power. In the current research we define power as the psychological experience and sense of power, which is linked to the ability to control one’s own and others’ access to resources (including people, information, and instrumentalities) without social interference (e.g., Anderson & Berdahl, 2002; Galinsky, Gruenfeld, & McGee, 2003; Galinsky, Magee, Inesi, & Gruenfeld, 2006; Keltner et al., 2003).
Experiment 1: Power Buffers the Stress Response During a Public Speech
To directly test the hypothesis that power buffers the stress response, Experiment 1 used the Trier Social Stress Test (e.g., Foley & Kirschbaum, 2010). Participants prepared for and delivered an employment-related speech to two judgmental evaluators and a video camera (in the control condition only a camera was present). The stress response was measured by examining nonverbal behaviors indicative of feeling stressed, self-reported stress, and cortisol activity at three points in time: baseline, reactivity, and recovery. The third cortisol sample allowed for a test of an alternative hypothesis – that power doesn’t buffer the stress response; it leads to a faster recovery. We predicted that power would exert its effects through a buffering (not recovery) mechanism rendering the powerful less stressed than the powerless as evidenced in physiological, emotional, cognitive and nonverbal indicators of stress.
Participants and Design
Fifty-five paid participants (37 female) from Columbia University were recruited and randomly assigned to one of four conditions in a 2 (power: low vs. high) x 2 (stress: low-stress vs. high-stress) factorial design.
Manipulation of Power
Following Carney et al. (2010; and also Bohns & Wiltermuth, 2012; Huang, Galinsky, Gruenfeld, & Guillory, 2011; Yap et al., 2013), power was manipulated by inducing powerful feelings and a physiological and nonverbal profile of power by configuring participant’s bodies into either (a) expansive and open body postures or (b) contractive and closed body postures (i.e., so-called “power poses”). This particular power manipulation is relatively quicker to implement, psychologically impactful, and comparable to lengthier role-assignment manipulations (Huang et al., 2011). Following Carney et al., experimenters instructed participants to position his/her body into two poses representative of either high or low power. To correct poses, verbal instructions were accompanied by light touches of arms and legs to properly configure bodies into the poses. A manipulation check administered an hour into the experiment (after the stress manipulation) found support for a difference in sense of power between the high-power (M = 2.84; SD = .64) and low-power condition (M = 2.46; SD = .76; F(1, 52) = 3.79, p < .06, effect size r = .27) on 8-items: dominant, in control, in charge, high status, like a leader, powerful, and two additional reverse-scored items: subordinate and submissive (α = .84).
The Trier Social Stress Test
Next, participants were either exposed to a high-stress or a low-stress condition. Stress was manipulated with the Trier Social Stress Test (TSST; Kirschbaum, Pirke, & Hellhemmer, 1993). The TSST produces a two to threefold rise in salivary cortisol levels in 70–85% of participants (Dickerson & Kemeney, 2004; Kudielka, Bellingrath, & Hellhammer, 2007).In the current version of the TSST participants prepared for and then gave a speech in front of 2 evaluators who were trained to express neutrality (which comes off as moderate negativity) in facial expression.1 Evaluators were dressed in white lab coats holding clipboards with notepads. The high-stress condition contained evaluators and a video camera; the low stress condition contained only a video camera. After a 5 min preparation phase (during which power was manipulated with expansive versus contractive postures), participants delivered a 5 min speech about why s/he should be hired for their dream job.
Two manipulation checks confirmed that high-stress participants felt more stressed on a 5-point scale (see scale description in following paragraphs; M = 2.78; SD = .86) than low-stress participants (M = 2.28; SD = .67), F(1, 52) = 6.60, p < .02; effect size r = .36. Participants’ nonverbal indications of stress were also coded on a 1-5 scale (variable descriptions below); evaluated participants expressed more nonverbal stress (M = 3.81; SD = .32) than non-evaluated participants (M = 2.26; SD = .32), F(1, 52) = 11.64, p < .01; effect size r = .47.
Hormone Sampling and Assays
Standard salivary hormone-collection procedures were used (Kirschbaum & Hellhammer, 1994; Schultheiss & Stanton, 2009). Before providing saliva samples, participants did not eat, drink, or brush their teeth for at least one hour. Participants were tested in the afternoon (12:00-5:00pm) to control for diurnal rhythms in hormones (e.g., Kudielka, et al., 2004). Each participant first rinsed his/her mouth with water and provided approximately 1.5 mL of saliva through a straw into a sterile polypropylene microtubule. Saliva samples were immediately frozen to avoid hormone degradation and to precipitate mucins. Two weeks after the end of the study, frozen samples were packed in dry ice and shipped for analysis to Salimetrics in State College, Pennsylvania. At Salimetrics, samples were assayed in duplicate for salivary cortisol, using a highly sensitive enzyme immunoassay. Three saliva samples were taken (which brought the total testing time to 1.5 hours). The first sample was taken approximately 10 mins after arrival; the second was taken approximately 22 mins after the end of the TSST (M = 22.16; SD = 3.21; range was 17 to 33 mins). The purpose of the third sample was to test the degree of recovery from stress; this was taken at the very end of the experiment, approximately 13 mins after the second sample (M = 12.51; SD = 4.08; range was 4 to 26 mins).2 Across the three points in time, the intra-assay coefficient of variation (CV; calculated on the 10% of the sample that was assayed in duplicate) was 5.45 %. Cortisol levels at time 1 were in the normal range (M = 0.17 µg/dL, SD = .10). To illustrate the change in cortisol over time, all three points in time were plotted as a function of the stress and power variables. The amount of time that passed between the TSST manipulation and the second saliva sample did not differ across conditions: a one-way ANOVA across the 4 conditions revealed no significant variance from the grand mean of 22.16 mins, F(3, 54) = 0.18, p > .90. The same was true for the difference in time between the 2nd and 3rd saliva sample (Mg = 12.51 mins), F(3, 54) = 0.53, p > .66.
Additional Measures of the Stress Response
Self-reported stress response. To assess participants’ level of self-reported stress following the job interview with two discerning evaluators, we asked participants to indicate how anxious, nervous, shaky, at ease, clam, and relaxed they felt (the latter 3 items were reverse-scored) on a scale from 1 (not at all) to 5 (extremely). Items were averaged together to form a composite variable of self-reported stress response (α = .86).
Measure of cognitive load. The Stroop was administered on the computer (Stroop, 1935). This task is an index of how cognitively taxed a person is (MacLeod, 1991; Swick & Jovanovic, 2002). We will refer to performance on this measure as cognitive load. In the Stroop task participants indicated “as quickly and accurately as possible” whether each of a series of letter strings was written in red or blue (ignoring the meaning of the words; the key-press version was taken from Smith, Jostmann, Galinsky, & van Dijk, 2008). Trials began with a 1-s fixation point located in the center of the computer screen. Fixation points were immediately followed by a red or blue-colored letter string. Participants responded to the string by indicating if it was blue or red by pressing a designated key on a computer keyboard. A 2-s blank screen appeared in between trials. In total the Stroop task consisted of 120 trials (no feedback about whether responses were correct or incorrect was offered). There were 40 congruent trials, 40 neutral trials, and 40 incongruent trials presented randomly. Reaction times to the incongruent trials were subtracted from reaction times to the congruent trials. As is typical, the distribution was skewed and was therefore transformed using a reciprocal transformation; higher scores indicate more cognitive load.
Nonverbal behavior. Three nonverbal behaviors known to be robustly associated with the stress response were coded: self-touches (Harrigan, Lucic, Kay, McLaney, & Rosenthal, 1990; Harrigan, Wilson, & Rosenthal, 2004), anxious smiles (Hall, Carney, & Murphy, 2002; Harrigan & Taing, 1997; Harrigan, Wilson, & Rosenthal, 2004) and lip bites (Ekman & Friesen, 1978; Ekman & Rosenberg, 1997). These variables were coded as frequency counts. All variables were coded reliably by two coders who overlapped on a subset of 10% of the videos. Only the first min of each 5 min speech was coded (Carney, Colvin, & Hall, 2007; Murphy, 2005). Inter-rater reliabilities for all variables were sufficient (self-touches were r = .95; anxious smiles were r = .99; lip bites were r = 1.0).
We predicted that power would buffer individuals from the stress response typical of social evaluative situations. We tested this hypothesis by examining cortisol reactivity and recovery, self-reported stress response, cognitive load, and nonverbal indications of the stress response. The a priori analytical approach was to test the planned contrast sequence expecting that when stressed, high-power individuals would appear as if they were not experiencing stress—that they would “look” like unstressed individuals across all measures. As such, we specified a contrast weight sequence of 3, -1, -1, -1 (low-power+stress = 3; high-power+stress = -1; low-power+unstressed = -1; high-power+unstressed = -1). The only exception was with the cortisol data, which included 3 points in time. Thus, to examine the stress-buffering effect of power on cortisol reactivity to a social stressor, we conducted a 2 (power) x 2 (stress) x 3 (cortisol at 3 points in time: baseline, reactivity, recovery) mixed-model ANOVA.
Did Power Buffer Cortisol Reactivity after the Stressor?
As hypothesized, there was a small but statistically significant 3-way interaction, F(2, 48) = 3.23, p < .05, η2 = .12. Figure 1 contains two panels in which the raw cortisol means are displayed. Panel A displays the results for the low-stress condition in which no evaluators were present. Cortisol at 3 points in time is graphed separately for the low-power and high-power participants. Consistent with expectations, when not stressed, normal downward sloping of cortisol across time is observed (due to typical diurnal variation in cortisol across the day; Kudielka, et al., 2004).3 The means in the high-stress condition (Panel B) suggests a pattern in which low-power individuals showed cortisol reactivity from Time 1 to Time 2 but then recovered. The high-power individuals, on the other hand, did not show cortisol reactivity—the means across the three points in time suggest they may have been buffered from the stress response. No support was found for the alternative hypothesis that power leads to a faster return to homeostasis after a stressor. No pair-wise comparisons were statistically significant (ps > .20).
Did Power Buffer Self-Reported Feelings of the Stress Response?
There was a significant effect of the planned contrast suggesting that the low-power individuals in the high-stress condition reported feeling more stressed than other participants, t(48) = 2.64, p < .02, effect size r = .38. Figure 2 displays the means for the low- and high-power participants across both stress (low-stress and high-stress) conditions. While the means and the overall effect was consistent with the hypothesis, pairwise analyses revealed some complexities: low-power+high-stress individuals did not report significantly more stress than those in the high-power+high-stress condition (p < .14) or those in the low-power+low-stress condition. (p < .16) but did report more stress than those in the high-power+low-stress condition (p < .001). High-power+high-stress individuals reported less overall stress than both low- and high-power low-stress individuals (ps < .03). No other tests were significant (p > .94).