• Spencer Greenberg and Clare Harris

Shifting Your Stance on the Psychological Impacts of Power Posing: From Real to Fake to Small

Updated: 3 days ago


In 2010, a group of researchers made headlines after publishing a study claiming to have found a simple but effective technique through which people could easily make themselves feel more powerful. The strategy involved adopting specific expansive postures, known since then as "power poses." The researchers found that people could "embody power and instantly become more powerful" after adopting these poses for just one minute each.


Their study became progressively more famous in the years that followed, but not necessarily for the right reasons. The original paper had multiple methodological flaws and swiftly became a well-known casualty of the replication crisis. Here, we outline how this shift in perspective occurred, then we explain what we found when we ran the largest single study (to our knowledge) on the psychological effects of power posing. We randomized 1002 participants to spend 60 seconds adopting either a power pose (as shown in Figure One (1-A, 1-B)), a contractive posture (1-C), or a neutral posture, then asked them to rate their mood. We found that the psychological effects of power posing are small - so of debatable importance - but real. Our results support skepticism regarding the large effects in the original research paper, but they also suggest it's probably a mistake to dismiss the psychological effects of power posing as “fake” or “non-existent.”


(1-A) strongPerson:	(1-B) Akimbo:	(1-C) Crumpled:Figure One: High-power (1-A and 1-B) and low-power (1-C) poses. For the "strongPerson," "Akimbo," and "Crumpled" intervention groups, participants were instructed on how to hold the pose to which they had been assigned via a written description accompanied by a photograph. The photographs shown for the three conditions are depicted in (1-A), (1-B), and (1-C), respectively. Participants were asked to hold their assigned posture for 60 seconds.

Summary of Our Study Results


• In contradiction to the increasingly popular view that power posing does not work, our pre-registered [1] randomized controlled trial on 1002 participants found that people who were randomized to adopt a power pose for 60 seconds reported slightly greater positive mood and greater feelings of power (hereafter referred to as “felt power”) compared to people who adopted a neutral pose and also compared to people who adopted a contractive (or low power) pose.

• Though statistically significant, these effects were very small, so the practical significance of power posing remains debatable.

• The effects of power posing tended to differ depending on whether (at the end of the study) participants reported that, prior to the experiment, they had already believed that posture affects mood - those who reported that they did already believe this tended to report greater increases in felt power than those who reported they had not already believed this. This raises the question of whether the effects of power posing could be (partially or wholly) a placebo effect or due to some form of reporting bias.

• However, even when we eliminated the participants who reported believing in the effects of power posing before the study, there was still a tiny, but statistically significant, increase in reported feelings of power following power poses versus neutral poses.

• Participants may vary in the extent to which power posing impacts their mood and feelings of power. (Approximately 17.7% of participants in the power pose groups reported no change in any of the eight mood-related questions after adopting their assigned power pose; and 44.7% of those participants reported no change in subjective feelings of power after adopting their assigned power pose. However, 17.7% of them reported an increase in felt power of 2 points or more on the 7-point scale, suggesting that it might have been useful for those people - though these measurement changes could also be the result of noise.)

So what’s our current stance on power posing?


Since power posing costs nothing and takes almost no effort, the power posing technique may be worth experimenting with to see if you notice a large enough effect to make it worth applying in your own life, in situations where greater feelings of power or a slight mood boost are helpful (e.g., by adopting either of the power poses (depicted in Figure 1-A and 1-B) for one minute, in private, prior to an important meeting). We expect many people will find that there are no useful effects of power posing, but some people may find that the technique has a large enough effect to be worth using.


From large and real to (seemingly) fake: How and why did the effects of power posing become so contentious in the first place?


Before skepticism about power posing set in, the original 2010 study had been met with excitement - especially among the original research team, journalists, and the public. The study authors suggested that "over time and in aggregate, these minimal postural changes and their outcomes potentially could improve a person’s general health and well-being." In 2012, one of the authors gave a TED Global talk discussing the findings. Not long after, power posing was mentioned in a best-selling book, which helped its popularity spread. To date, the talk (titled, "Your body language may shape who you are") remains one of the most-watched TED videos of all time, with 62 million views.


Within four years of the blockbuster TED talk, however, the 2010 study had become, according to some, "a shorthand for flashy social psychological work that could not be replicated." By then, some of the findings had failed to replicate across multiple subsequent studies, and the methodological problems in the original study had been described in detail. For example, the original "power posing" study was relatively under-powered (that is, it had few participants - there were only n = 42 participants). Smaller sample sizes not only increase the rate of false negative results (type II errors), but also increase the risk of false positives (type I errors) (for example, see the explanations here and here). In addition to the small sample size, there were many other methodological problems with the 2010 study, such as the researchers not being blinded to the hypothesis or experimental conditions, and the fact that the analyses did not adequately account for gender. There were also issues with the way in which the sample size was chosen (with Carney reporting p-hacking, although it was not seen as p-hacking at the time), as well as with the choice and reporting of statistical analyses. [2]


The original study investigated not only the reported psychological effects of power posing, but also sought to investigate the effects on salivary testosterone and cortisol levels, as well as on risk-taking behavior (tested via a gambling task); unfortunately, there were problems with how the salivary testosterone and cortisol were measured and with the gambling task administration: for example, the gambling task occurred prior to the second salival sampling (for the hormonal assays) and it included immediate feedback on whether the participant had won, leaving open the possibility that any testosterone increases could have been due to a "winning" effect rather than a power posing effect. We are not investigating any hormonal or risk-taking effects here, but we note that these have failed to replicate elsewhere. [3] By 2016, the lead author (Carney) changed her mind about her own work, publicly stating that she did "not believe that 'power pose' effects are real."


Hearing about the 2010 study and its failed replications, you might ask (as Cesario, Jonas, and Carney did in 2017): "What was the point, and what did we learn?" Well, firstly, a number of researchers have taken the paper as a case study in practices that we want to move beyond in social science (e.g., see here, here, and here). Moreover, the reactions to the paper, including the many pre-registered, carefully designed studies investigating power posing, provide promising examples of psychology’s growing drive to self-correct for its previous mistakes.


From the criticisms that were levied against the original power posing work, a number of social scientists came to believe that power posing has no psychological effects. But did the critics go too far?


Updating from fake to real? - the surprising effects of a Bayesian meta-analysis on the psychological effects of power


In 2017, Gronau and colleagues published a Bayesian model-averaged meta-analysis using data from six pre-registered studies that assessed the effect of power posing on felt power across a total of 1071 participants. Carney and other researchers (introducing the special issue in which the meta-analysis was published [4]) had expected the meta-analysis to confirm the non-existence of the effects of power posing, but were surprised to see that this was not the case. Instead, Gronau and colleagues found strong evidence supporting a positive effect of power posing on felt power. Interestingly, however, when they excluded people who indicated that they were already aware of the concept of power posing (leaving a sample of 809 people), the evidence was only moderately supportive of an effect. We will discuss the implications of these results later.


What postures have researchers been comparing power poses to?


Unfortunately, for all of the studies in that meta-analysis cited above (and for most other studies published on power posing to date, too), researchers measured the effects of high-power posing in comparison to low-power (or contractive) poses. In other words, they did not include neutral postures as a control condition. This means that, for these prior studies, it is impossible to distinguish between the following possibilities: (1) power posing increases feelings of power, (2) contractive postures decrease feelings of power, or (3) the seemingly positive effects of power posing are arising from a combination of both of these effects. In contrast to most previous studies, our study explicitly includes a neutral sitting and a neutral standing posture, enabling us to examine the possible effects of power posing and contractive posing separately.


Our Study: Methods


Participants


We recruited 1090 people to participate in the study, via a combination of Positly.com (formerly taskrecruiter.com) and Facebook. Following the (pre-planned) removal of participants who didn’t complete the study or who reported not having followed the study instructions, the final sample size was n=1002 participants. Participants gave written informed consent prior to participating.


Procedures


This online randomized controlled trial included two "power pose" intervention subgroups and three comparison subgroups, and was pre-registered using aspredicted.org/. The pre-registration can be found here. You can view the study exactly as study participants experienced it by visiting this link. Please note that the link will randomize you to only one of the conditions at a time, so to see every condition you will have to go to the link repeatedly. Prior to this study, a pilot was run in order to plan which poses to use for the current pre-registered, randomized controlled trial. (The pilot study was aimed at uncovering which power posing postures were most likely to create effects - see Supplementary Figure Four for some of the results from the pilot.)


Following the informed consent process, participants in the main study reported their current posture (e.g., "seated" or "standing"), answered eight questions regarding their mood (the order of which was randomized each time), then were provided with written instructions on a specific posture to adopt. In the case of the neutral groups, participants were told to stand (the "noPoseStanding" group) or sit (the "noPoseSeated" group), while in the case of the high- and low-power groups, participants were given detailed instructions accompanied by a photograph of someone adopting that posture (as depicted in Figure One). After 60 seconds of adopting their assigned posture, all participants resumed their previous position, and answered the same eight questions regarding their mood.


At the end, participants answered some final questions, including (1) a question about whether, before the study, they believed holding their body in specific positions for a short period of time could have an effect on their mood, and (2) whether they had followed the experiment instructions. Anyone who had not finished the study or who reported not having followed the instructions was excluded from our analyses. To encourage participants to answer honestly, immediately before people were asked if they had followed instructions, those who were in the paid participant pool (i.e., people recruited via a paid participant portal rather than via Facebook) were reminded that they would still be paid for participation regardless of their answer to the question about following instructions.


For the "power pose" intervention group, people were randomly assigned to receive instructions on adopting either a "strongPerson" pose (n=194) or an "akimboHandsOnHips" (hereafter “akimbo”) pose (n=195). As illustrated in Figures 1-A and 1-B, both postures involve standing straight, without any bodily tension, with feet comfortably far apart from each other (i.e., comfortably open), and with the person looking ahead with their shoulders back and chin pointing upwards (i.e., an expansive posture). The difference between the two types of power pose lies in the positioning of the hands and arms. In this study, the strongPerson describes a pose in which one’s hands are closed, forming fists, with their elbows at 90 degrees such that the fists are pointing straight up at the ceiling. On the other hand, akimbo (as its name suggests) describes a posture in which one has their hands on their hips, with their elbows pointing out to the sides.


For the comparison groups, people were randomly assigned to receive instructions on adopting one of three poses: a noPosingSeated posture (n=206), which involved sitting in whatever posture the participant considered normal, a noPosingStanding posture (n=207), which involved standing in whatever posture the participant considered normal, or a "Crumpled," otherwise known as "contractive" or "powerless pose," (Figure 1-C), in which the participant was instructed to stand with their feet together, their arms crossed (with their hands clasping the opposite arm), their chin tilted downward, and their gaze directed down toward the ground (n=207).


Outcome Measures and Analysis Methods


The analyses were conducted as described in the pre-registration document here. For the pre-registered analyses, the main effect of interest was the change in positive and negative feelings experienced by participants after (versus before) the one-minute posture intervention (or lack thereof). Participants were asked the degree to which they would currently describe themselves as powerful, confident, and so on, using a seven-point Likert scale (from "totally agree," which conferred a score of 3, to "totally disagree," which conferred a score of -3 on the scale).


In addition to changes in felt power, a "total positive feeling" value was also calculated; this was done by taking the sum of the scores for the four positive emotions - "powerful," "confident," "good," and "happy" - and the reversed-scores for the four negative emotions - "anxious or worried," "down, depressed or hopeless," "bad," and "unhappy." The total positive feeling was then rescaled so that it had a minimum value of 0 and a maximum value of 1 (a.k.a. min-max normalized), and the values from before the intervention were subtracted from the values afterward.


Statistical analyses were conducted in GraphPad (https://www.graphpad.com/quickcalcs/ttest1/?Format=SD) and Jasp (JASP Team, 2020), except where otherwise specified. Except for Figure Seven and the appendices (most of which were created using a spreadsheet editor), figures were created using Jasp (JASP Team, 2020). Because our outcome data were not normally distributed, Mann-Whitney U-tests were performed, except where otherwise specified. However, due to our large sample sizes, student's t-test results are also reported in the footnotes after each Mann-Whitney U-test result. Please see the Appendices for a note about the implications of conducting multiple comparisons.


Our Study Results


(1) Power posing was followed by tiny but statistically significant increases in overall positive feelings

As shown in Table One and Figure Two below, power posing appeared to increase the total positive feelings that participants reported, relative to both the true controls and the contractive posture group. However, these effects were very small. (And for some people, the effects were non-existent - a total of 17.73% of participants in the power pose groups reported no change in any of the eight mood-related questions after adopting their assigned power pose.) The rescaled (min-max normalized) sum of positive feelings increased by a mean of 0.04 following one minute of power posing, compared to a mean increase of 0.02 in the neutral control groups.


Table OneGroupSample SizeMedianMADMeanSDRank biserial correlation vs neutral postures*Cohen’s d vs neutral posturesPower poses3890.020.040.040.08-0.16-0.25Neutral postures4130.020.040.020.08NANAContractive posture2000.000.04-2.08 ✕ 10-30.120.070.23 MAD = median absolute deviation SD = standard deviation * effect size is given by the rank biserial correlation (from the Mann-Whitney U-test comparing a given group to the pooled neutral postures group).

Increasing someone’s normalized sum of positive feelings by a mean of 0.04 is equivalent to moving their response to the eight mood questions by about 24% of a notch toward more positive feelings. (For example, this could mean moving 24% of the way from Agree to Totally Agree on the positive-mood-related questions, and moving the same amount, but in the opposite direction, on the negative-mood-related questions.)


Unlike most other experiments, the current study also included neutral comparison groups, which enabled us to look at the effect of power posing compared to neutral standing poses (as opposed to only compared to contractive postures). It is worth noting that the control group (i.e., the neutral postures groups pooled together) had an average raw effect size of about 0.02, so if we subtract this effect from that of the power poses, this halves the power posing effect to about 0.02, which is the equivalent of moving the person ~one eighth of a notch in the more positive direction on all of the eight mood questions. Although this difference between the pooled power posing groups and neutral posture groups was small, so of unclear practical importance, it was statistically significant (Mann-Whitney U = 67,533, n1 = 413, n2 = 389, p = 8.47 ✕ 10^-5). [5]


The power posing groups also reported significantly greater improvements in positive feelings compared to the Crumpled group (Mann-Whitney U = 30,652.50, n1 = 389, n2 = 200, p = 2.24 ✕ 10^-5)). [6] Interestingly, however, there was no statistically significant difference in the positive feeling increase reported by the pooled neutral comparison groups compared to the Crumpled group (Mann-Whitney U = 38,223, n1 = 413, n2 = 200, p = 0.13)). [7] So, in this experiment at least, there appear to be small but statistically significant positive effects following power posing, but we did not find significant negative effects following contractive posing.



Figure Two: changes in normalized total positive feelings across the three groups. 2-A: The distributions of all participants’ results. These violin plots show the probability density of data at different values of normalized positive feelings. Power posing appeared to increase the total positive feelings that participants reported, relative to both the true controls and the contractive posture group. However, as noted in the body of the text, the effect sizes here are tiny. These violin plots highlight how subtle the differences were.  2-B: The means and confidence intervals associated with the above results. Dots display the mean value for each group, while the bars show the 95% confidence intervals (CIs). (However, note that the data were non-normally distributed.)
Figure Two: changes in normalized total positive feelings across the three groups. 2-A: The distributions of all participants’ results. These violin plots show the probability density of data at different values of normalized positive feelings. Power posing appeared to increase the total positive feelings that participants reported, relative to both the true controls and the contractive posture group. However, as noted in the body of the text, the effect sizes here are tiny. These violin plots highlight how subtle the differences were. 2-B: The means and confidence intervals associated with the above results. Dots display the mean value for each group, while the bars show the 95% confidence intervals (CIs). (However, note that the data were non-normally distributed.)

The visual differences between 2-A and 2-B highlight the fact that the mean changes for the neutral and power posing groups (shown in 2-B) were larger than the median changes for these groups (discernible from the bulges in the middle of the violin plots in 2-A). When comparing the means between groups (as in 2-B), the differences between the neutral and power posing groups are more obvious. However, even in this case, note that the entire y axis extends from just -0.02 to 0.05, again highlighting that the effects are very small.


(2) In head-to-head comparisons between individual power poses and other poses, both power poses outperformed neutral sitting and contractive postures, but they did not consistently outperform neutral standing postures


Figure Three shows the changes in normalized total positive feelings broken down across the five individual groups. Although the differences are subtle, the rank order of effect sizes of the different groups were aligned with expectations: the highest changes were for the two power posing groups, followed by the noPosingStanding group, the noPosingSeated group, with the Crumpled posture having the smallest observed changes. There was no significant difference between the positive feelings reported following the strongPerson versus the akimboHandsOnHips posture (Mann-Whitney U = 19,437.50, n1 = 195, n2 = 194, p = 0.64), suggesting that neither power pose was superior to the other one. [8]


It may be that standing up is helpful on its own: there was a small trend toward an increase in positive feelings following the neutral standing compared to the neutral seated posture (Mann-Whitney U = 18,858.50, n1 = 207, n2 = 206, p = 0.04). [9] Note that most participants (>90%) were sitting at the start of the experiment, which means that almost everyone in the noPosingStanding was changing from a sitting position to a standing position. Although we lose some power by making comparisons between these subgroups (with smaller sample sizes), we conducted head-to-head comparisons between the two individual power groups and the noPosingStanding group, to see whether the small increase in positive feelings following power posing was significantly greater than the effect of merely standing in a neutral position.


There was a trend toward the akimbo pose producing a greater increase in normalized positive feelings compared to the noPosingStanding posture, but the difference was not statistically significant with either the Mann-Whitney t-test (Mann-Whitney U = 18,247.50, n1 = 207, n2 = 195, p = 0.09) or with student’s t-test. [10] There was a trend toward a small difference between the strongPerson and noPosingStanding groups (Mann-Whitney U = 17,767.50, n1 = 207, n2 = 194, p = 0.04; student’s t-test: t(399) = -1.69, p = 0.09).


These results suggest that there is a trend toward power posing producing more positive feelings than a neutral standing posture, but it is not significant in the case of akimbo and only marginally significant in the case of the strongPerson pose. However, both the power poses were associated with significantly greater positive feelings being reported versus the neutral seated posture (akimbo vs neutral seated: Mann-Whitney U = 15,933.50, n1 = 206, n2 = 195, p = 3.11 ✕ 10^-4; strongPerson vs neutral seated: Mann-Whitney U = 15,584.50, n1 = 206, n2 = 194, p = 1.27 ✕ 10^-4) [11] and versus the standing contractive posture (akimbo vs contractive posture: Mann-Whitney U = 15,507, n1 = 200 , n2 = 195, p = 4.01 ✕ 10^-4; strongPerson vs contractive posture: Mann-Whitney U = 15,145.50, n1 = 200, n2 = 194, p = 1.54 ✕ 10^-4).


Figure Three: average changes in min-max normalized total positive feelings, broken down across the five groups. 3-A: The distributions of all participants’ results. These violin plots show the probability density of data at different values of normalized positive feelings. Power posing appeared to increase the total positive feelings that participants reported, relative to both the true controls and the contractive posture group. However, as noted in the body of the text, the effect sizes here are tiny. These plots highlight how subtle the differences were.  3-B: The means and confidence intervals associated with the above results. Dots display the mean value for each group, while the bars show the 95% confidence intervals (CIs). (However, note that the data were non-normally distributed.)
Figure Three: average changes in min-max normalized total positive feelings, broken down across the five groups. 3-A: The distributions of all participants’ results. These violin plots show the probability density of data at different values of normalized positive feelings. Power posing appeared to increase the total positive feelings that participants reported, relative to both the true controls and the contractive posture group. However, as noted in the body of the text, the effect sizes here are tiny. These plots highlight how subtle the differences were. 3-B: The means and confidence intervals associated with the above results. Dots display the mean value for each group, while the bars show the 95% confidence intervals (CIs). (However, note that the data were non-normally distributed.)

The visual differences between 3-A and 3-B highlight the fact that the mean changes for the neutral and power posing groups (as shown in 3-B) were larger than the median changes for these groups (discernible from the bulges in the middle of the violin plots in 3-A). When comparing the means between groups, the differences between the neutral and power posing groups are more obvious.



(3) Power posing was followed by small but significant changes in felt power


Our focus now turns to the effects of power posing on participants’ subjective feelings of power specifically (as opposed to overall positive feelings), since that was the focus of the psychological findings from the original 2010 paper. As shown in Table Two and Figure Four, the power pose conditions were associated with an increase in participants’ felt power (compared to before the intervention). The average size of change in feelings of power across the (pooled) two power posing groups was 0.64. This represents a move 64% of the way from one point on the Agree-Disagree scale to a higher point on the scale, such as from "Totally disagree" to "Disagree", or from "Neither agree nor disagree" to "Moderately agree."


Table TwoGroupSample SizeMedianMADMeanSDRank biserial correlation vs neutral postures*Cohen’s d vs neutral posturesPower poses389010.641.14-0.24-0.40Neutral postures413000.201.05NANAContractive posture200000.071.180.040.12 MAD = median absolute deviation SD = standard deviation * effect size is given by the rank biserial correlation (from the Mann-Whitney U test comparing a given group to the pooled neutral postures group).

This small change represented a significantly greater increase in felt power than the (pooled) neutral posing groups overall (Mann-Whitney U = 60,990, n1 = 413, n2 = 389, p = 1.35 ✕ 10^-10). [14] We also tried eliminating the neutral sitting participants from the analysis, and found that the felt power following power posing was also significantly greater than the neutral standing group specifically (Mann-Whitney U = 32,495, n1 = 389, n2 = 207, p = 3.13 ✕ 10^-5). [15] However, once again, it is questionable as to whether the effect sizes are of any practical significance. The mean difference between the pooled power posing groups and the neutral standing group was only 0.36, suggesting that power posing only increases power by about 36% of a notch on the scale, when compared to the effects of merely standing up without adopting a specific power pose. [16]


(4) The change in power following contractive postures was less than that following power posing, but was not significantly different from neutral postures


The positive effects of power posing were also evident when the Crumpled condition was used as the comparison group: the power posing groups were associated with significantly greater increases in felt power than the Crumpled condition (Mann-Whitney U = 28,543, n1 = 389, n2 = 200, p = 1.44 ✕ 10^-8) [17], with a mean difference of 0.57 between the two conditions. But was this due to positive effects of power posing or negative effects of contractive postures? People in the Crumpled condition reported an average increase in felt power of 0.075, which was not significantly different from zero [18] and which was also not significantly different to the pooled neutral comparison groups (Mann-Whitney U = 39,580.50, n1 = 413, n2 = 200, p = 0.35). [19] This suggests that, in this study at least, the positive effects of power posing are not likely to be explained away by negative effects of contractive postures (since our data do not appear to support the existence of negative effects of contractive postures).



Figure Four: Changes in felt power following one minute of posing. 4-A: The distributions of all participants’ results. These violin plots show the probability density of data at different values of reported feelings of power. Power posing appeared to increase the subjective power that participants reported, relative to both the true controls and the contractive posture group. However, as noted in the body of the text, the effect sizes here are tiny. These plots highlight how subtle the differences were.  4-B: The means and confidence intervals associated with the above results. Dots display the mean value for each group, while the bars show the 95% confidence intervals (CIs). (However, note that the data were non-normally distributed.)
Figure Four: Changes in felt power following one minute of posing. 4-A: The distributions of all participants’ results. These violin plots show the probability density of data at different values of reported feelings of power. Power posing appeared to increase the subjective power that participants reported, relative to both the true controls and the contractive posture group. However, as noted in the body of the text, the effect sizes here are tiny. These plots highlight how subtle the differences were. 4-B: The means and confidence intervals associated with the above results. Dots display the mean value for each group, while the bars show the 95% confidence intervals (CIs). (However, note that the data were non-normally distributed.)

As in earlier figures, the visual differences between panel A and panel B highlight the fact that the mean changes for the neutral and power posing groups were larger than the median changes for these groups. When comparing the means between groups, the differences between the neutral and power posing groups are more obvious.


Figure Five: Changes in felt power following one minute of posing. 5-A: The distributions of all participants’ results. These violin plots show the probability density of data at different values of reported feelings of power. Power posing appeared to increase the subjective power that participants reported, relative to both the true controls and the contractive posture group. However, as noted in the body of the text, the effect sizes here are tiny. These plots highlight how subtle the differences were.   5-B: The means and confidence intervals associated with the above results. Dots display the mean value for each group, while the bars show the 95% confidence intervals (CIs). (However, note that the data were non-normally distributed.)  Again, the visual differences between 5-A and 5-B highlight the fact that the mean changes for the neutral and power posing groups were larger than the median changes for these groups. When comparing the means between groups, the differences between the neutral and power posing groups are more obvious.
Figure Five: Changes in felt power following one minute of posing. 5-A: The distributions of all participants’ results. These violin plots show the probability density of data at different values of reported feelings of power. Power posing appeared to increase the subjective power that participants reported, relative to both the true controls and the contractive posture group. However, as noted in the body of the text, the effect sizes here are tiny. These plots highlight how subtle the differences were. 5-B: The means and confidence intervals associated with the above results. Dots display the mean value for each group, while the bars show the 95% confidence intervals (CIs). (However, note that the data were non-normally distributed.) Again, the visual differences between 5-A and 5-B highlight the fact that the mean changes for the neutral and power posing groups were larger than the median changes for these groups. When comparing the means between groups, the differences between the neutral and power posing groups are more obvious.

(5) Many respondents experienced no change in felt power following the adoption of power poses, but a minority experienced sizeable changes


As shown in Figure Six, despite the mean increase in subjective power across the power posing groups overall, for many individual participants, there was no change in felt power. This suggests that, if there are real effects of power posing, the effects might only apply to some people, and when they do apply, there can be individual differences in the size of the effects (although it is impossible to know exactly how much of the variation in the histogram below is attributable to this explanation and how much of the variation is due to noise). A total of 174, or 44.7%, of participants in the power posing groups reported no change in subjective feelings of power after adopting their assigned power pose. However, 17.74% of them reported an increase in felt power of two points or more on the 7-point scale, suggesting that - if (or to the extent that) these measurement changes aren’t the result of noise - power posing might have been useful for those people.


Figure Six: distribution of change in feelings of power among participants in either of the power pose groups.
Figure Six: distribution of change in feelings of power among participants in either of the power pose groups.

(6) Power posing also produced small changes in felt power when participants didn’t believe that posture could affect mood


At the end of the experiment, all participants were asked if, before the study, they believed that posture could affect mood. A total of n = 601 people across the whole experiment reported that, before the study, they did not believe posture could affect mood. Among this subgroup (271 of whom were in one of the neutral posing groups and 213 of whom were in one of the power posing groups), the average increase in felt power following power posing was 0.52, compared to an average increase of 0.14 among those in the neutral posing groups; this represented a small but statistically significant difference (Mann-Whitney U = 23,102.50, n1 = 271, n2 = 213, p = 3.06 ✕ 10^-5). [20] There was a trend toward a greater increase in felt power among those in the power posing groups who reported that they did previously believe that posture affects mood compared to those in the power posing groups who reported not previously believing this (Mann-Whitney U = 16,030, n1 = 213, n2 = 176, p = 8.75 ✕ 10^-3). [21] The implications and some possible explanations of these findings are explored in the Discussion section.

Figure Seven: effects of different poses on felt power for participants who reported believing (versus not believing) that postures affect mood.
Figure Seven: effects of different poses on felt power for participants who reported believing (versus not believing) that postures affect mood.

Discussion


Our findings were consistent with Gronau et al.’s (2017) meta-analysis, in that we found a small but statistically significant increase in reported feelings of power after participants adopted a power pose for one minute. Unlike the studies included in that meta-analysis, however, this experiment included two truly neutral posture groups (one sitting and one standing) in addition to a contractive (or low-power) posture group. Interestingly, we found that changes in subjective power in the contractive posture group were not significantly different to those in the pooled neutral posture groups. We also found that the subjective power increased in the (pooled) power posing groups more than it did in the neutral posing groups (both when the neutral standing posture was singled out specifically and when it was combined with the neutral sitting posture group).


The aims of the experiment were relatively transparent to respondents, with 640 people (63.9% of the sample) correctly guessing the aim of the study. This raises the concern of demand characteristics - that is, the effect of the experiment being influenced by participants’ knowledge of the experiment aims. When we limited analyses to only people who reported that before the experiment they did not believe that bodily posture influences one’s mood, there were still significant differences in the subjective power felt following power posing versus neutral postures (on average), but it was not as strong as the result for the group at large.


Why did power posing tend to increase felt power more significantly among those who reported that they already believed (before the experiment) that posture impacts mood? There are multiple possible explanations for this finding, some more obvious than others. The findings in the current experiment might be consistent with any of the explanations below (or with any combination of them):


(1) Firstly, it may be that the mere belief that one’s mood is impacted by one’s posture causes one to truly feel more powerful (and other more positive feelings) following a power pose than one would have otherwise. If this is the case, then this would be an example of a placebo effect. Nevertheless, placebo effects are still real effects (in the sense of producing the intended change on participant feelings). Given that standing up and trying a power pose has ~no cost, is easy to implement, and does not typically come with risks, it seems like a harmless recommendation for people to try it and to see whether they experience any positive effects (especially if the person considering it is open to trying things that might work via placebo effects).


(2) Secondly, the effect could result from reporting bias, wherein a belief that posture can change one’s mood makes someone "look" for, find, or take note of changes more than they otherwise would. Alternatively, some participants may have been reporting what they thought we "wanted to hear" rather than introspecting on their real feelings. On the other hand, though, there were no obvious incentives to do so, as the study participants are anonymous, the format was online (so there was no interaction with the research team), and participants were aware that their pay for participation would not be influenced by what result they reported.


(3) Thirdly, it could be that people whose moods are more affected by posture would be likely to have noticed this prior to the study, and hence would be more likely to have believed in such effects before the study. These same people would also be more likely to report stronger mood effects following power poses during the study. If this explanation is true, power posing effects could be "real" but could vary from person to person, and the more someone experiences them, the more likely they are to believe in them.


If this explanation applies, there are multiple reasons that some people might have developed a posture-mood relationship while others haven’t. For example, it might be that some of the participants had spontaneously adopted power poses when they were feeling powerful or triumphant in the past, and that over time, they may have come to associate such postures with those feelings (either through associative learning/conditioning, social learning, through explicit memory of these past times, or a combination of these). This might cause them to both (1) believe that there is a relationship between posture and mood, and (2) to feel more powerful (either subconsciously or consciously) when they adopt power poses in other settings, such as in the current experiment.


Alternatively (or as well as this), we could speculate that we are genetically predisposed to feel powerful in certain poses, such as when maintaining an expansive posture. If so, it might be that this varies between people, resulting in some people developing a belief in a relationship between posture and mood (along with experiencing an appreciable mood change in the experiment), while others do not develop such a belief (due to being less predisposed to having and noticing posture-related mood effects).


(4) Finally, it could be that people were not properly introspecting on their beliefs from before the study - and instead, when asked about whether they believed that posture affects mood before the study, they may have been influenced by what happened during the study. In other words, if their mood was affected by the posture and they came to believe that posture affects mood, then this new belief may have contaminated their assessment of what they believed before the study. If this occurred, this could explain at least some of the reduction in the effect size following the removal of people who said they previously believed in the effects.


Limitations of this research


This study only focused on the psychological effects of different bodily postures on mood. It did not examine whether the observed mood changes had appreciable effects on outcomes that people would be likely to care about, such as interview performance, cognitive performance, success with professional or personal negotiations, and so on. Even in cases where power posing does what it is supposed to do, this does not necessarily imply that it would improve outcomes in these areas (e.g., see here). The current study also relied on participant self-report, and the aims of the study were relatively transparent to participants, which complicates our interpretation of the reported effects.


Conclusion


Both Gronau and colleagues’ Bayesian meta-analysis and the current study suggest that power posing can have a small, but appreciable, positive impact on felt power. It also seems that the degree to which posture impacts someone’s mood is at least partially dependent on whether they believe in the effect in the first place. Overall, given that adopting power poses is a relatively harmless, low-cost intervention, it seems reasonable for people to test it for themselves and to use it if they think that it works. If you’re planning to try power posing the next time you want to feel a bit more powerful, it’s probably unrealistic to expect anything life-changing - but if you expect at least some effect, this might help to produce that effect.



Acknowledgements


Thank you to Hannah M. Le and Holly Muir for their valuable input on this post! We are also very grateful to all the participants for their time.


Links to the study and anonymized dataset

  • You can experience the study exactly as study participants did by visiting this link (note that there are five groups you may be randomized into, so you’ll need to try it many times to see all five groups): https://www.guidedtrack.com/programs/j0qliha/preview

  • To view the code used to run the study via GuidedTrack go to this link. You can also use this link to easily copy a deployable version of the whole study so you can quickly run your own replication.

  • You can download all the data we collected for the study here.

  • The raw data is an anonymized version of all the data we collected, and the "clean" data is after we removed the incomplete and invalid responses (that is, responses from people who either dropped out in the middle or who said they didn’t do the instructions exactly as written)

  • This study was pre-registered using aspredicted.com, and the pre-registration can be found here: https://aspredicted.org/4jp4y.pdf

  • If you have questions or comments (or to share your own analyses of our data with us), please get in touch!


Footnotes

  1. Pre-registration has been proposed as a promising component of more replicable research, partly because it forces researchers to commit to their methods (including their analyses) in advance.

  2. Regarding the choice of statistical analyses, another of the original study’s authors - the TED talk speaker - reports hiring a statistician, Fosse, and asking him to conduct and publish an independent investigation into the errors of the original study; the resulting report remains on the Harvard dataverse to this day.

  3. One of the first and more prominent examples of the hormonal effects in the paper failing to replicate was a paper by Ranehill and colleagues in 2015, although that paper did reproduce a felt power effect, which is the focus of this post. An example of a prominent paper which did question the psychological effects of power posing is that by Simmons and Simonsohn in 2017.

  4. The meta-analysis was published in a special issue of Comprehensive Results in Social Psychology, dedicated to a series of rigorous pre-registered studies investigating the effects of power posing. Carney and other researchers had expected the Bayesian meta-analysis to settle the issue and to confirm the absence of any real effects, but, as we explain here, the data instead supported the existence of a real (though small) effect.

  5. Student’s t-test results were t(800) = -3.55, p = 4.04 ✕ 10^-4.

  6. Student’s t-test results: t(587) = -4.94, p = 1.01 ✕ 10^-6.

  7. Student’s t-test results: t(611) = -2.68, p = 7.62 ✕ 10^-3.

  8. Student’s t-test result: t(387) = 0.64, p = 0.52.

  9. Student’s t-test result: t(411) = -1.93, p = 0.05. Please also see the note in the Appendix.

  10. Student’s t-test result: t(400) = -1.25, p = 0.21.

  11. Student’s t-test results: akimbo vs neutral seated: t(399) = -3.75, p = 2.00 ✕ 10^-4; strongPerson vs neutral seated: t(398) = -3.83, p = 1.50 ✕ 10^-4.

  12. Student’s t-test results: akimbo vs contractive posture: t(393) = -3.93, p = 9.90 ✕ 10^-5; strongPerson vs contractive posture: t(392) = -4.09, p = 5.21 ✕ 10^-5.

  13. The rank biserial correlation (effect size) for changes in felt power for power posing vs controls was -0.24 (a larger negative value than the effect size for changes in positive feelings, which was -0.16, indicating that the size of the difference between the power posing and control groups was larger for changes in power than for changes in positive feelings).

  14. Student’s t-test result: t(800) = -5.71, p = 1.56 ✕ 10^-8.

  15. Student’s t-test result: t(594) = -3.68, p = 2.53 ✕ 10^-4.

  16. The noPosingStanding had an average effect of increasing felt power by about 0.29 up the scale. It is also worth noting that even noPosingSeated increased felt power by about 0.12 up the scale.

  17. Student’s t-test result: t(587) = -5.66, p = 2.33 ✕ 10^-8.

  18. A Wilcoxon’s signed rank test showed that the change in felt power was not significantly different from zero (W = 2475, p = 0.35).

  19. Student’s t-test results: t(611) = -1.34, p = 0.18.

  20. Student’s t-test results: t(482) = -3.80, p = 1.64 ✕ 10^-4.

  21. Student’s t-test results: t(387) = -2.42, p = 0.02. However, please see the note in the Appendix.


References

Antonakis, J. (2017). On doing better science: From thrill of discovery to policy implications. The Leadership Quarterly, 28(1), 5-21.

Carney, D. R. (2016). My position on "power poses.". Unpublished manuscript. Berkeley: University of California. <http://faculty.haas.berkeley.edu/dana_carney/pdf_My%20position%20on%20power%20poses.pdf>

Carney, D. R., Cuddy, A. J., & Yap, A. J. (2010). Power posing: Brief nonverbal displays affect neuroendocrine levels and risk tolerance. Psychological science, 21(10), 1363-1368.

Cesario, J., & Johnson, D. J. (2018). Power poseur: Bodily expansiveness does not matter in dyadic interactions. Social Psychological and Personality Science, 9(7), 781-789.

Cesario, J., Jonas, K. J., & Carney, D. R. (2017). CRSP special issue on power poses: what was the point and what did we learn?. Comprehensive Results in Social Psychology 2(1):1-5.

Crede, M. (2019). A negative effect of a contractive pose is not evidence for the positive effect of an expansive pose: Comment on Cuddy, Schultz, and Fosse (2018). Meta-Psychology, 3

Cuddy, A. J., 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.

Ingre, M. (2013). Why small low-powered studies are worse than large high-powered studies and how to protect against "trivial" findings in research: Comment on Friston (2012). Neuroimage, 81, 496-498.

JASP Team (2020). JASP (Version 0.14.1) [Computer software].

Lonati, S., Quiroga, B. F., Zehnder, C., & Antonakis, J. (2018). On doing relevant and rigorous experiments: Review and recommendations. Journal of Operations Management, 64, 19-40.

McShane, B. B., Gal, D., Gelman, A., Robert, C., & Tackett, J. L. (2019). Abandon statistical significance. The American Statistician, 73(sup1), 235-245.

Quentin F. Gronau, Sara Van Erp, Daniel W. Heck, Joseph Cesario, Kai J. Jonas & Eric-Jan Wagenmakers (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, DOI: 10.1080/23743603.2017.1326760.

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.

Simmons, J. P., & Simonsohn, U. (2017). Power posing: P-curving the evidence. Psychological science.




Appendices


A note on multiple comparisons

There are a total of 35 statistical tests reported in this piece (including footnotes). Performance of multiple comparisons increases the risk of type I errors (false positive results). One way to correct for this is to use a standard Bonferroni correction (however, this can be overly conservative and can increase the risk of false negative results: see here). Using a Bonferroni-corrected test-wise alpha level of 0.0014 (1.4 ✕ 10^-3), the following findings can be labeled statistically insignificant:

  • There was trend toward a greater increase in positive feelings following the neutral standing compared to the neutral seated posture (Mann-Whitney U = 18,858.50, n1 = 207, n2 = 206, p = 0.04) - this would not be labeled as significant at our Bonferroni-corrected alpha level of 0.0014.

  • There was a trend toward a small difference between the strongPerson and noPosingStanding groups (Mann-Whitney U = 17,767.50, n1 = 207, n2 = 194, p = 0.04 - this would not be labeled as significant at our Bonferroni-corrected alpha level of 0.0014.

  • There was a trend toward a greater increase in felt power among those in the power posing groups who reported that they did previously believe that posture affects mood compared to those in the power posing groups who reported not previously believing this (Mann-Whitney U = 16,030, n1 = 213, n2 = 176, p = 8.75 ✕ 10^-3) - this would not be labeled as significant at our Bonferroni-corrected alpha level of 0.0014.

All of the findings that are labeled as statistically significant in the body of this piece would also be labeled as significant at the Bonferroni-corrected alpha level of 0.0014.

Supplementary Figure One: Distribution of normalized "total good feelings" in the power pose group specifically. The most frequently-reported effect was no change, with a substantial portion of people reporting a change in good feelings between 0 and 0.1. Fewer people were in the range 0.1 to 0.2, and almost nobody had an effect larger than 0.25. This suggests that if there are positive effects of power-posing on overall good feelings, these effects are small, and they vary across people. (It is also possible that some of the fluctuations in the chart represent random fluctuations in mood, something unrelated to the study, or other factors unrelated to the power posing condition.)



Supplementary Figure Two: Participants’ identification of the study goal. A total of 640 people who completed the study and followed directions (63.9% of the final sample) correctly guessed the aim of the study. The next two most commonly-guessed study goals were "How much mood tends to vary randomly when a person is asked to do something pointless" and "The [effects] of distraction on mood." In both these cases, as well, it appears that people were aware of the expectation that their mood would change over the course of the experiment, which may have produced demand characteristics.