Does Power Posing Work? What Fifteen Years of Replications Found

In June 2012 an American social psychologist walked onto a stage in Edinburgh and told the room that two minutes could change their body chemistry. Amy Cuddy, then an associate professor at Harvard Business School, was describing a study she had published two years earlier with Dana Carney and Andy Yap, both then at Columbia University. Stand like a superhero for two minutes before you walk into a job interview, she said, and your testosterone rises, your stress hormone falls, and you become the kind of person who takes the shot. As of September 2026 the talk has been played 77,356,075 times on TED’s own site. A bestselling book followed. After the book came a whole genre of interview advice about finding an empty bathroom and putting your hands on your hips.

Fifteen years later most of that claim has been taken apart, and one piece of it has not. Awkwardly, the surviving piece is the one most people were after.

Four promises, graded separately

What the 2010 paper promised	What happened when it was tested again	Grade
Testosterone rises	No effect in a blinded replication of 200 people, none in seven preregistered studies, and two 2022 meta-analyses find no posture effect on any physiological measure	R
Cortisol falls	Same replications, same nothing. A 2024 study of 92 adults reported a link between falling cortisol and risk tolerance inside the expansive group, which is not a difference between groups	R
You take more risks	The original result rested on a single $2 gamble and a test whose p value was .052 by the more common method. No preregistered study has reproduced a behavioural effect	R
You feel more powerful	Holds up. Replicates in large samples and in preregistered work, with a small effect. How much of it is people knowing what the pose is for remains open	C

What the 42 people were asked to do

The original experiment is small enough to describe in full. Forty-two university students, 26 women and 16 men, were told the study was about how the placement of heart-monitoring electrodes affects readings. An experimenter physically arranged each person’s arms and legs into one of two configurations: open and taking up space, or closed and folded in. Two poses, one minute each. While holding them, participants formed impressions of nine photographed faces, a filler task.

Then came the four measurements. Each person was handed $2 and offered a choice: keep it, or roll a die with even odds of turning it into $4 or losing it. They rated how “powerful” and “in charge” they felt from one to four. Saliva was collected before the posing and again about 17 minutes afterwards, frozen, and shipped to a commercial laboratory for testosterone and cortisol assays.

All four came out as predicted. In the expansive group 86% took the gamble, against 60% of the contracted group. Testosterone rose in one group and fell in the other; cortisol did the reverse; the expansive group reported feeling more powerful. The paper closed by saying the finding had “real-world, actionable implications” (Carney, Cuddy & Yap, 2010).

A claim that names a hormone borrows the authority of a blood test. You cannot argue with your own cortisol, and that is a large part of why the idea travelled as far as it did. The same move sits underneath polyvagal theory and its vagus nerve: attach a physiological mechanism to an ordinary piece of advice and the advice stops sounding like advice.

Zurich, 2015: same design, five times the people

A team of economists led by Eva Ranehill, then at the University of Zurich, working with colleagues at the Stockholm School of Economics, ran the design again on 200 participants: 98 women and 102 men. Three changes matter. Poses were held for three minutes instead of one. Instructions arrived on a computer screen, so the experimenter never learned who was in which condition. And the consent form said plainly that the study was about whether physical position affects hormone levels and behaviour.

Their statistical power to detect an effect the size of the original was above 95%. Put plainly: if the effect were as large as the 2010 paper reported, this study would almost certainly have caught it.

It caught one of the four. People in the expansive condition reported feeling more powerful: a mean difference of 0.245 on that four-point scale, 95% CI [0.044, 0.446], Cohen’s d = 0.344. Testosterone: d = -0.200, p = .162. Cortisol: d = -0.157, p = .272. Risk-taking in the gambling task: d = -0.176, p = .215. Two extra behavioural tasks the team had added, risk-taking when money was on the line to lose and willingness to enter a competition, showed nothing either (Ranehill et al., 2015).

Translate those numbers. A d of 0.34 is small: line up a hundred posers and a hundred slouchers, ask them how powerful they feel, and the two sets of answers overlap almost entirely. You need the arithmetic to see the gap. A d of -0.2 attached to a p value of .16 is a statistical shrug, and it points in the wrong direction anyway.

The first author writes down what happened

In September 2016 Dana Carney, by then an associate professor at the Haas School of Business at the University of California, Berkeley, posted a two-page document on her faculty page called “My position on ‘Power Poses’”. It is a strange thing to read: a first author taking her own paper apart item by item, in public, with nobody forcing her to.

Her list of admissions includes:

• The sample grew while the results were being watched. “We ran subjects in chunks and checked the effect along the way. It was something like 25 subjects run, then 10, then 7, then 5.” She adds that at the time this felt like saving money rather than bending a result.
• A choice between two tests. For the gambling result, “One p-value for a Pearson chi square was .052 and for the Likelihood ratio it was .05. The smaller of the two was reported.” She calls this “clearly using a researcher degree of freedom”.
• The feelings measure was picked after the fact. “Many different power questions were asked and those chosen were the ones that ‘worked’.”
• The experimenters knew the hypothesis while running the sessions.
• Winners were told they had won. Participants learned the outcome of the gamble immediately, and winning a bet is itself known to raise testosterone. So the hormone result “may merely be a winning effect, not an expansive posture effect”.

Her conclusion: “I do not believe that ‘power pose’ effects are real.” She also wrote that she discourages others from studying them, has stopped teaching them, and has stopped talking about them in the media (Carney, 2016). Her two co-authors did not join her, and the disagreement between them has never been resolved.

Two teams point the same tool at the same pile

After the failed replication, Carney, Cuddy and Yap published a review listing 33 studies they read as supporting the effect. Joseph Simmons and Uri Simonsohn, both at the University of Pennsylvania, ran those 33 through a technique called p-curve analysis.

The idea behind it fits in two sentences. A real effect leaves a fingerprint in a literature: plenty of very small p values, few that just scrape under .05. Selective reporting leaves the opposite fingerprint, bunched up against the threshold. Their reading of the power-posing pile: the distribution “is indistinguishable from what would be expected if (a) the average effect size were zero and (b) selective reporting (of studies or analyses) were solely responsible for the significant effects that were published”. Direct replications of those studies, they wrote, “would not be expected to succeed” (Simmons & Simonsohn, 2017).

Cuddy replied the following year with a wider sample of the same literature, 55 studies, and a p-curve broken up by type of outcome. On feelings and moods she and her co-authors found strong evidential value. On the rest, their own text concedes the split: removing the feelings measures “flattens the curve for the remaining effects, which could indicate that evidential value for behaviors and hormones is weak” (Cuddy, Schultz & Fosse, 2018). Two teams, opposite headlines, and on the hormones they were far closer than the headlines let on.

Seven laboratories that promised their results in advance

The cleanest test of a contested finding is to have people commit to their design before they see any data. In 2017 the journal Comprehensive Results in Social Psychology did exactly that: seven teams submitted protocols for peer review before collecting a single measurement, and the studies were accepted on the strength of the design rather than the outcome. Dana Carney reviewed all of them. Because acceptance came first, no result could be shelved later for being dull.

What came back, in the editors’ own summary: “there was virtually zero effect of power poses on any of the behavioral or hormonal measures”. Risky behaviour, dominance in a simulated interview, how persuasive a message seemed, self-concept, testosterone as the mediating step: none of them moved. What did move was the feeling. “Looking across the studies, it is clear that an effect on felt power was observed,” the editors wrote, before adding that whether this is more than participants sensing what is expected of them “is still unclear” (Jonas et al., 2017).

A Bayesian meta-analysis of six of those preregistered studies put a number on the surviving piece and found very strong evidence for an effect on felt power. The same authors add a qualification themselves: restricted to participants who had never heard of power posing, the evidence drops to “only moderate” (Gronau et al., 2017). Part of the feeling, then, may be people recognising the pose and supplying the expected answer.

Where the posture literature stands now

Two meta-analyses published in 2022 stepped back from power posing and looked at body position in general.

Robert Körner and colleagues at the University of Bamberg preregistered a review covering 313 effects from 88 studies and 9,779 participants. Overall effect of body position: g = 0.35, 95% CI [0.28, 0.42]. Split by outcome, self-report and behaviour showed effects while physiological measures showed none. Their own sensitivity checks then chipped away at the behavioural half: those effects were influenced by publication bias, by outliers, or both. Effects were also larger in studies with no cover story and in studies where the same person tried both poses, which is what you would expect if participants were working out the purpose and obliging (Körner et al., 2022).

Emma Elkjær’s team at Aarhus University screened 5,819 records down to 73 studies and asked a question the field had mostly skipped: compared with what? Nearly every power-posing experiment sets a wide pose against a slumped one and reports the gap. Add a neutral posture as a third reference point and the gap moves house. Contracted versus neutral: Hedges’s g = 0.45. Expansive versus neutral: g = 0.06 (Elkjær et al., 2022). Sitting hunched makes people feel measurably worse. Standing wide does close to nothing. A g of 0.06 is a difference you would never spot in a room of thirty people.

Newer single studies point the same way. In 2024 a team in Israel ran the hormone protocol on 92 young adults and failed to reproduce the testosterone effect; unusually for this literature, they also failed to reproduce the feelings-of-power effect. They did report a link between falling cortisol and risk tolerance among the expansive posers, which is a correlation inside one condition rather than a difference between conditions (Barel et al., 2024).

In 2025 a Danish group took the question out of the laboratory. Over twelve weeks, 127 adults each named a personally difficult action they wanted to take that week, then were randomly assigned to hold an expansive, contracted, neutral or walking-on-the-spot body display before attempting it. Effect on whether they went through with the action: d = 0.06. Effect on how they felt while doing it: d = 0.02 to 0.06 (Mikkelsen et al., 2025). Fifteen years after the superhero stance, that is what it does to a real week.

The 2010 study belongs to a cohort of small, elegant social-psychology findings from that era that shrank once bigger samples arrived, alongside self-control as a fuel tank that runs dry. The pattern repeats: a striking result from a few dozen undergraduates, a memorable metaphor, a decade of citation, then a proper sample and a much smaller number at the end of it.

Another idea from the same period went through the same scrutiny and came out with more of itself intact, though less than the slogans promise: what large trials found about growth mindset.

Sources

• Cuddy, A. J. C (2012). Your body language may shape who you are. TED (TEDGlobal 2012). Talk recorded 26 June 2012, posted 1 October 2012; 77,356,075 plays on ted.com as of 5 September 2026 — the reach number for the claim ted.com
• Carney, D. R., Cuddy, A. J. C., & Yap, A. J (2010). Power Posing: Brief Nonverbal Displays Affect Neuroendocrine Levels and Risk Tolerance. Psychological Science. 21(10), 1363-1368; N = 42 (26 women, 16 men), two 1-min poses, saliva at baseline and ~17 min after; testosterone F(1,39) = 4.29, cortisol F(1,38) = 7.45, gamble taken by 86.36% vs 60%, felt power F(1,41) = 9.53 doi:10.1177/0956797610383437
• Ranehill, E., Dreber, A., Johannesson, M., Leiberg, S., Sul, S., & Weber, R. A (2015). Assessing the Robustness of Power Posing: No Effect on Hormones and Risk Tolerance in a Large Sample of Men and Women. Psychological Science. 26(5), 653-656; N = 200, experimenter blind to condition, power > 95% for the original effect size; felt power mean difference 0.245, 95% CI [0.044, 0.446], d = 0.344, p = .017; testosterone d = -0.200 (p = .162), cortisol d = -0.157 (p = .272), risk d = -0.176 (p = .215) doi:10.1177/0956797614553946
• Carney, D. R (2016). My position on "Power Poses". Author statement, Haas School of Business, University of California, Berkeley. Two-page public statement by the first author: "I do not believe that 'power pose' effects are real"; lists data checked in batches (25, then 10, then 7, then 5 subjects), reporting the smaller of two chi-square p values (.05 vs .052), p-hacked self-report, unblinded experimenters, and the possibility that the testosterone result was a winning effect faculty.haas.berkeley.edu
• Simmons, J. P., & Simonsohn, U (2017). Power Posing: P-Curving the Evidence. Psychological Science. 28(5), 687-693; p-curve of the 33 studies cited by Carney, Cuddy & Yap (2015); distribution "indistinguishable from what would be expected if (a) the average effect size were zero and (b) selective reporting … were solely responsible"; direct replications "would not be expected to succeed" doi:10.1177/0956797616658563
• Cuddy, A. J. C., Schultz, S. J., & Fosse, N. E (2018). P-Curving a More Comprehensive Body of Research on Postural Feedback Reveals Clear Evidential Value for Power-Posing Effects: Reply to Simmons and Simonsohn (2017). Psychological Science. 29(4), 656-666; p-curve of 55 studies; strong evidential value for feelings of power and other affective outcomes, while the authors' own discussion states that removing those measures "flattens the curve" and that "evidential value for behaviors and hormones is weak" doi:10.1177/0956797617746749
• Jonas, K. J., Cesario, J., Alger, M., Bailey, A. H., Bombari, D., Carney, D., et al (2017). Power poses – where do we stand?. Comprehensive Results in Social Psychology. 2(1), 139-141; editors' summary of the special issue of seven preregistered replications: "there was virtually zero effect of power poses on any of the behavioral or hormonal measures", while "it is clear that an effect on felt power was observed" doi:10.1080/23743603.2017.1342447
• Gronau, Q. F., Van Erp, S., Heck, D. W., Cesario, J., Jonas, K. J., & Wagenmakers, E.-J (2017). A Bayesian model-averaged meta-analysis of the power pose effect with informed and default priors: the case of felt power. Comprehensive Results in Social Psychology. 2(1), 123-138; six preregistered studies; "very strong evidence for an effect of power posing on felt power", dropping to "only moderate" when restricted to participants unfamiliar with the effect doi:10.1080/23743603.2017.1326760
• Körner, R., Röseler, L., Schütz, A., & Bushman, B. J (2022). Dominance and prestige: Meta-analytic review of experimentally induced body position effects on behavioral, self-report, and physiological dependent variables. Psychological Bulletin. 148(1-2), 67-85; preregistered meta-analysis of 313 effects from 88 studies and 9,779 participants; overall g = 0.35, 95% CI [0.28, 0.42]; effects for self-report and behaviour but not for physiological measures, with behavioural effects influenced by publication bias and/or outliers and larger without cover stories doi:10.1037/bul0000356
• Elkjær, E., Mikkelsen, M. B., Michalak, J., Mennin, D. S., & O'Toole, M. S (2022). Expansive and Contractive Postures and Movement: A Systematic Review and Meta-Analysis of the Effect of Motor Displays on Affective and Behavioral Responses. Perspectives on Psychological Science. 17(1), 276-304; 73 studies from 5,819 records; effects driven by the absence of contractive displays (contractive vs neutral Hedges's g = 0.45) rather than the presence of expansive ones (expansive vs neutral g = 0.06) doi:10.1177/1745691620919358
• Barel, E., Shahrabani, S., Mahagna, L., Massalha, R., Colodner, R., & Tzischinsky, O (2024). The effects of power posing on neuroendocrine levels and risk-taking. BMC Psychology. 12(1), article 726; 92 young adults; no replication of the testosterone effect or of feelings of power; among high-power posers a decrease in cortisol was associated with risk tolerance, a within-condition correlation rather than a group difference doi:10.1186/s40359-024-02194-7
• Mikkelsen, M. B., Elkjær, E., Mennin, D. S., Michalak, J., & O'Toole, M. S (2025). The Power of Posing: Do Body Display Instructions Have an Impact on Behavior in Daily Life?. Brain and Behavior. 15(6), article e70643; 127 adults over 12 weeks, randomly assigned to expansive, contractive, neutral or active-control body displays before a self-chosen difficult action; no effect on behaviour (d = 0.06) or affect (ds = 0.02-0.06) doi:10.1002/brb3.70643