Physical Mechanisms of Emotions Evoked by Postural Feedback

ABSTRACT

Adopting physical expressions of emotion has been shown to have feedback effects on individuals' mood and behavior. For example, adopting the expansive and contractive body language of dominance and submission can affect individuals' feelings of power. However, the effects can be subtle and variable; we suggest that this may be due to inter‐individual variability in the physical expression of the experimental posture, including both the magnitude of the posture adopted and the specific muscles used to adopt it. Here, we employed a postural feedback (i.e., “power posing”) paradigm and recorded quantitative measures of body position algorithmically derived from video recordings (N = 101). We demonstrate for the first time that variation in neck flexion mediates the effects of posture on mood. We also investigated several other variables which, based on previous work, could additionally mediate or moderate the postural feedback effects, finding that subjective difficulty also mediates posture effects independently of neck flexion, with effects moderated by body awareness. Finally, we investigated the muscular processes underlying neck flexion in the posture using electromyography, demonstrating that the neck flexion which mediates mood effects is associated with sternocleidomastoid muscle activity. The present work carries implications for our understanding of the adaptive benefits of expansive and contractive postures, and provides important methodological insights into the paradigms typically used for research on postural feedback.

Short abstract

This study provides new evidence for embodied theories of emotion by addressing inconsistencies in the literature on emotions evoked by postural feedback. We demonstrate that individual variation in the muscles and movements involved in adopting power‐related postures mediates how much these poses affect individuals' mood, addressing concerns about alternative mechanisms of postural feedback effects and supporting their original embodied explanation.

1. Introduction

In Darwin's seminal theory of emotion, affective processes prepare individuals for specific adaptive courses of action (Darwin 1872; Frijda 2007). In concert with William James's early theories (James 1884), this embodied perspective suggests that physical expressions of emotion can not only play a key role in social interactions but also have feedback effects on individuals' own mood and behavior. Considerable evidence has supported this view through the use of physiological feedback paradigms, where participants are instructed to adopt specific emotional expressions and mood or behavior are subsequently measured (Coles et al. 2022; Price and Harmon‐Jones 2015); this phenomenon has been mostly investigated with facial expressions, in the context of the Facial Feedback Hypothesis (Tomkins 1962). Although less studied, there is growing evidence that similar effects can be elicited with body expressions of emotion, particularly in the case of dominance‐related displays. Power and dominance are typically expressed using open, expanded postures which take up more space and appear intimidating, while submission is expressed using closed, contractive postures which signal that the individual is not a threat (Burgoon and Dunbar 2006). Despite initial controversy caused by inconsistent results and several replication failures (Jonas et al. 2017; Simmons and Simonsohn 2017), feedback effects from these postures have been demonstrated to have reliable effects on some aspects of behavior and mood (Elkjaer et al. 2022; Körner et al. 2022); changes in self‐report variables such as positive affect, self‐confidence and feelings of power have been particularly robust (Gronau et al. 2017; Körner et al. 2022).

Both the physical and psychological mechanisms driving bodily feedback effects remain largely unknown. Here, we focus on physical mechanisms; that is, the physiological factors which mediate, or at least contribute to, the observed cognitive/affective effects of postural feedback (psychological mechanisms have been discussed elsewhere, notably Körner and Schütz 2020). Indeed, although the initial “power posing” study (Carney et al. 2010) suggested a mechanism of action via cortisol and testosterone, this has since failed to replicate (Elkjaer et al. 2022). Notably, the specific movements and postures in bodily feedback studies can vary from individual to individual: for example, expansive postures include multiple movements such as limb extension, upright posture, shoulder retraction, and chest expansion. This pointis particularly relevant, as findings from the facial feedback literature suggest that the specificity and magnitude of the adopted expression can have important effects on the outcomes. For example, “Duchenne” smiles (which implicate a specific group of muscles in the cheek and periocular regions; Ekman and Friesen 1982) result in more intense feedback effects than non‐Duchenne smiles involving a subset of those muscles (Soussignan 2002); similar results were observed in the large‐scale study by Coles et al. (2022), where participants who rated their smiles as more genuine and more similar to the examples given to them reported greater feedback effects. Similarly, mood can be altered by the inactivation of specific facial muscles such as the corrugator, involved in frowning, using BOTOX (de Costa F et al. 2022; Wollmer et al. 2022). This suggests that feedback effects are linked to the activity of specific muscles, and that the behavioral effects may be stronger or weaker depending of which of the relevant muscles are engaged, and to what extent (Ceolato et al. 2018). In the case of bodily feedback studies, this direct, quantitative characterization of the movements and muscles involved in the posture and its effects on mood and behavior has rarely been applied; to our knowledge, poses have always been treated as categorical variables. This is a notable gap, as some theoretical accounts posit different behavioral outcomes for low‐ and high‐intensity poses, assessed qualitatively from different types of posture manipulations (Körner and Schütz 2020).

Thus, the main aim of the present study was to quantitatively assess the adopted postures in a feedback paradigm and examine their relationship to behavioral outcomes. Specifically, we employed a “power posing” paradigm to investigate the effects of expansive and contractive body postures on participants' mood, while concurrently acquiring video‐based measurements of participants' postures and physiological data from electromyography. We previously investigated quantitative measures of participants' dominance‐related natural postures in Wainio‐Theberge and Armony (2024): there, we observed that neck flexion (obtained from photos using computer vision techniques) was associated with antisocial personality traits; notably, this angle was the largest factor discriminating between explicit dominant and submissive poses made by participants (for details, see Study 3 in Wainio‐Theberge and Armony 2024). Following this, we hypothesized that objectively‐measured neck flexion would mediate the effects of the posture manipulation in this paradigm. Moreover, also following the findings of that study, we hypothesized that expansive and contractive postures would be distinguishable based on the muscular activity in the neck flexors. Finally, we assessed several other potential moderators and mediators of power‐pose effects. Several of these have been previously shown to be relevant for power‐pose effects, such as participants' awareness of the research hypothesis (more aware participants show larger effects; Körner et al. 2022), and the degree of effort and difficulty in adopting the posture (contractive postures are more effortful than expansive; Bombari et al. 2017). We also examined participants' awareness of their own bodies during the posture, as attention to somatosensory processes may enhance the posture's effects; this has not been previously investigated, though (Weineck et al. 2019) did examine this as an outcome of posture manipulations. The aim was to ensure that these variables did not confound our effects of interest and to examine them in their own right as possible factors influencing postural feedback effects, as either suggested by previous studies or by relevant theory (James 1884).

2. Methods

2.1. Participants

A total of 147 participants (mean age = 21.8, SD = 3.6, 109 self‐assigned as female) participated in this study, which was approved by the McGill University Faculty of Medicine Institutional Review Board (IRB #A00‐B62‐21A). A sample size of 124 was determined using a power analysis in G*Power, aiming at 80% power to detect an effect of size 0.45 (Hedges' g for self‐report outcomes reported by Elkjaer et al. (2022)) based on a one‐tailed independent samples t‐test; we recruited an additional 15% of participants to compensate for expected data loss. The power analysis was conducted for the main outcome of posture, which is reported in the results as model 1. Other individual‐level data unrelated to the present study (namely personality traits, natural body posture, and resting‐state EMG data) were previously reported in Wainio‐Theberge and Armony (2024), Study 4. Participants were recruited from the general population using social media advertisements and from the McGill Psychology Department extra‐credit participant pool. Participants were between 18 and 35 years of age, reported no history of diagnosed psychiatric or neurological disorders and were not presently taking psychiatric medication at the time of the study. Participants were compensated CAD $50 for taking part in the study or received equivalent course credit. Testing was conducted in person in Montreal, Québec.

2.2. Procedure

The experiment involved a series of resting‐state electrophysiological recordings, including EEG, EMG, and ECG, with accompanying mood questionnaires, interspersed with blocks of a perceptual task (Chadwick et al. 2019). Results from the task and EEG components will be presented elsewhere. In the present study, we focused on a section of the study involving a mood questionnaire and 3‐min resting‐state EMG recording, which was administered directly before and after participants adopted the experimental posture for 3 min. We thus examined the effects of body posture on mood by comparing the changes in questionnaire scores between expansive and contractive postures. A cover story was also used to prevent participants from discovering the intention of the manipulation: they were told that the study was about the effect of body posture and heart‐rate variability on cognitive performance; body posture was meant to affect heart‐rate variability, and thereby performance in the task. Participants were debriefed afterward, and those who suspected any effects of the posture on mood or emotion were noted for further subgroup analyses (see next section). A schematic of the timeline of the experiment is given in Figure 1.

FIGURE 1.

Timeline of the relevant section of the experiment. Results from the task sessions will be reported in forthcoming publications.

Participants were randomly assigned to an expansive or contractive posture group, and completed only the posture of the group to which they were assigned, resulting in a between‐subjects design for the posture factor. In the expansive condition, participants were instructed to raise their heads, straighten their spines, and bring their shoulders back; in the contractive condition, participants were instructed to lower their heads, curve their spines, and bring their shoulders together in front (for the exact instructions given to participants, see Figure S2). This is most similar to the postures defined in early postural feedback studies (e.g., Riskind and Gotay 1982), but incorporated expansion and contraction of the chest as in Carney et al. (2010). Experimenters did not demonstrate the posture to avoid the confounding influence of viewing the experimenter's posture but checked the participant's understanding of the intended posture. Participants completed the posture once, holding it for 3 min in a seated position without performing any other task. Participants were video‐recorded during the experiment to ensure compliance and, critically, measure their postural angles, as described below.

2.3. Measures, Scales, and Creation of Variables of Interest

Mood questionnaires consisted of the Affect Valuation Index (AVI; Tsai et al. 2006) and the authentic pride scale (AP; Tracy and Robins 2007); the Toronto Hospital Alertness Test (THAT; Shapiro et al. 2006) was also administered to distract participants from the true intention of the questionnaires and support the cognitive cover story. The AVI was chosen as it samples the full affective circumplex, providing a comprehensive characterization of valence and arousal; it has furthermore been shown to be affected by expansive and contractive postures (Körner et al. 2021). Items from the AVI consist of simple emotion labels, such as “Enthusiastic” (High Arousal Positive subscale) or “Dull” (Low Arousal Negative subscale). Pride was chosen following the theoretical model presented by Körner and Schütz (2020), which argues that the vertical dimension of expansive and contractive posture (in which neck flexion plays a key role) is related to prestige; similar to the AVI, items consist of adjectives or short descriptors, such as “Confident” or “Like I am achieving something”. Following the posture, participants also answered 5 questions related to their experience of the posture: (1) how comfortable did you find the posture?; (2) how difficult did you find the posture to maintain?; (3) how natural did the posture feel to you?; (4) how aware were you of your body during the posture?; and (5) how aware were you of your heartbeat during the posture? (this latter question was intended to support the cover story related to heart‐rate variability). All scales and questions were rated on a 10‐point Likert scale, ranging from 1 (“Not at all”) to 10 (“Extremely”).

To reduce the data and to improve the interpretability of the findings, the AVI subscales were combined into two dimensions reflecting the two axes of the affective circumplex (Posner et al. 2005). The composite Arousal dimension was calculated as the difference between the high arousal scales (high arousal (HA), high arousal positive (HAP), and high arousal negative (HAN)) and the low arousal scales (low arousal (LA), low arousal negative (LAN), and low arousal positive (LAP)). The composite Valence dimension was calculated as the difference between the positive scales (positive (P), HAP, and LAP) and the negative scales (negative (N), HAN, and LAN). The Authentic Pride Scale (AP) was analyzed separately. Focusing on the time points immediately preceding and following the posture, we computed the difference between post‐ and pre‐posture for Arousal, Valence, and Pride; these scores were used as dependent variables for the statistical analysis.

Difficulty/discomfort with the posture was also defined as a composite variable, based on the follow‐up questions about participants' experience of the posture. Difficulty was defined based on the first three items about participants' experience of the posture (discomfort, difficulty maintaining the posture, and naturalness of the posture), which were strongly correlated. To create this variable, we summed the z‐transformed scores of each of the three items, with the naturalness question reverse‐coded as it was negatively correlated with the others. Suspicion of the research hypothesis was coded based on participants' responses to verbal questioning by the researchers at the end of the experiment; this was coded as 0 or 1, with 1 indicating any level of suspicion the participants had that the posture was meant to affect their mood. Body awareness was defined based on the single‐item Likert scale administered during the post‐posture questionnaires (“I was more aware of my body during the posture”).

2.4. Physiological Recording and Preprocessing

Objective measures of body posture were obtained throughout the experiment, using photogrammetric and physiological approaches. Participants were video‐recorded at a distance of 0.5 m using a Panasonic Lumix GX‐85 mirrorless camera with a 12 mm Panasonic lens. Bipolar surface EMG was recorded from the sternocleidomastoid (SCM) muscle and the upper trapezius (TRAP) on the participant's dominant side; these muscles were chosen as they are neck flexors and neck extensors, respectively (Brennan et al. 2015), and are easily accessible by surface EMG. EMG electrodes were positioned according to Falla et al. (2002) for the SCM and according to the SENIAM 8 guidelines for the TRAP (Hermens et al. 2000).

Next, we computed a measure of expansive and contractive postures based on the video recordings. As neck flexion is implicated in these postures, and following our findings that neck angle reliably distinguishes between posed dominant and submissive postures (Wainio‐Theberge and Armony 2024), we chose to focus on participants' neck angles as a quantitative measure of their adoption of the posture. Neck angle was computed during the posture following the methods of Wainio‐Theberge and Armony (2024). Video recordings during the posture were downsampled to 1 frame per second, and OpenPOSE (Cao et al. 2019) was applied to each frame to measure keypoints. Neck angle was then calculated as the angle between the shoulder keypoint (#2) and the ear keypoint (#17); these points were defined based on annotations of body parts provided by raters in the Microsoft COCO dataset (Lin et al. 2014). As videos could not be manually inspected due to the number of frames (~180 frames per participant × 100 participants), we automatically excluded any frames where the neck angle was less than −20°, as these reflect physiologically improbable angles even for the most upright and expansive postures. The median neck angle value across frames was taken as a measure of during‐posture neck angle for each participant.

EMG was recorded via the external channels of a BioSemi ActiveTwo amplifier, referenced at the time of recording to the Cz EEG electrode; an integrated CMS/DRL circuit was used to reduce power‐line interference (Merletti and Cerone 2020). Data was recorded at a sampling rate of 2000 Hz, and downsampled to 500 Hz for analysis. EMG preprocessing followed the recommendations of the SENIAM committee (Stegeman and Hermens 2007). First, raw EMG signals were re‐referenced to a bipolar montage for each muscle. Next, bipolar EMG signals were bandpass filtered from 10 to 500 Hz to remove low‐frequency drift. Finally, the filtered signals were rectified and lowpass filtered at 20 Hz to create the EMG power time series. These power time series were averaged within each resting state to obtain a measure of raw EMG power. Nine large outliers determined via Grubbs' test; (Grubbs 1950) were removed, as these likely reflect bad electrode‐to‐skin contact. EMG activity during the posture was log‐transformed and baseline‐corrected relative to EMG activity within the pre‐posture resting‐state recording.

2.5. Statistical Analysis

We used structural equation models implemented in the lavaan package in R for our statistical analyses (Rosseel 2012). To test our hypotheses, we adopted an incremental approach, beginning with simpler models and moving to more complex ones only when we found significant effects. Code and lavaan syntax for each model are provided in the OSF repository associated with this project. Each model is depicted graphically in Figures 2 and 3. All continuous variables were standardized prior to analysis.

FIGURE 2.

Results of SEM models 1 and 2 for each mood variable (Arousal, Valence, and Pride). Boxes indicate exogenous variables (variables not used as a dependent variable in any regression), and circles indicate endogenous variables (variables which are dependent variables in at least one regression). Arrows indicate paths tested in the model. Green paths indicate coefficients which are significant at p < 0.05. Gray paths indicate non‐significant coefficients.

FIGURE 3.

Results of SEM models 3 and 4 for the mood variables where significant mediation effects were observed in model 2 (Valence and Pride). Boxes indicate exogenous variables (variables not used as a dependent variable in any regression), and circles indicate endogenous variables (variables which are dependent variables in at least one regression). Arrows indicate paths tested in the model. Green paths indicate coefficients which are significant at p < 0.05. Gray paths indicate non‐significant coefficients.

In model 1, we aimed to determine whether the posture manipulation indeed affected mood, and if so, on which variables. That is, we tested three models with posture category as the exogenous variable influencing the endogenous variables Arousal, Valence, and Pride change scores, respectively.

In model 2, we assessed whether neck angle mediated the effects of posture on the mood variables obtained in model 1. To do so, we added during‐posture neck angle as a mediator of the posture category's effect on the corresponding mood change. The direct and indirect effects were calculated and used to determine whether mediation was present.

In model 3, we aimed to exclude a possible confounding effect, namely the difficulty and discomfort participants felt while adopting the posture. Indeed, effort and self‐reported ability to take on the posture were previously found to predict outcomes in a postural feedback experiment (Bombari et al. 2017). To do so, we added to model 2 a parallel path between posture category and mood change going through the composite difficulty variable defined earlier. Direct and indirect effects were calculated for both paths. An additional model was constructed to determine whether there was a serial mediation effect of the neck angle and difficulty—this is reported in the Data S1.

Finally, in model 4 we included two other variables that might moderate the association between posture and mood: suspicion of the research hypothesis and body awareness during the posture. The former was coded as a binary variable with 0 indicating no suspicion at all and 1 indicating any level of suspicion (see above), and the latter corresponded to the question “How aware were you of your body during the posture?” (rated from 1 to 10 on a Likert scale in the post‐posture questionnaire). Moderation effects on each path in the parallel mediation model of model 3 were included, and the index of moderated mediation was calculated in each case.

Additionally, we examined the physiological correlates of our posture manipulation and neck angle. To do so, we first separately examined the effects of posture category and neck angle on sternocleidomastoid (SCM) and trapezoid (TRAP) activity using t‐tests. We further correlated SCM and TRAP activity with during‐posture neck angle and mood changes using Pearson correlation.

For mediation models in model 2, we report the proportion mediated as an index of the effect size of mediation. This measure is defined as the indirect effect divided by the direct effect, and can be interpreted as the proportion of the direct effect which operates through the mediator. Note that the proportion mediated can be numerically greater than 100% if the direct effect changes direction (VanderWeele and Vansteelandt 2014); in this case, the proportion mediated should be interpreted as 100%, as the inclusion of the mediator results in an uncovered effect in the opposite direction. All confidence intervals reported in the text correspond to 95% confidence intervals.

2.6. Transparency and Openness

We report how we determined our sample size (Methods section Participants), all data exclusions, all measures in the study (Methods section Data preprocessing and creation of variables of interest), and all manipulations (Methods section Procedure), and the study follows JARS (Appelbaum et al. 2018). Code to run the relevant analyses and preprocessed data from the study are available at https://osf.io/gpbkv/; OpenPOSE keypoints and raw EMG data files are also provided, as raw videos could not be shared due to privacy considerations. Data analysis was conducted in MATLAB R2020a and R 4.2.3, using the package lavaan 0.6–17 (Rosseel 2012). This study's design and its analyses were not pre‐registered.

3. Results

Following the exclusion of participants with missing (1) or outlier EMG data (9, see Methods) and unusable video data due to technical issues (24) or failure of OpenPOSE to determine posture (13), our final sample contained 101 participants (mean age 21.8, SD 3.77, 70 women; 44/56 in the contractive/expansive group). Values for the relevant demographics and experimental variables, with their corresponding statistics, are reported in Table 1.

TABLE 1.

Descriptive statistics for demographics and all variables used in the analysis.

	Expansive	Contractive
Excluded	18	28
Age	21.8, [20.7, 22.8]	21.9, [20.8, 23]
Sex	38 women	32 women
Arousal change	1.68, [−1.4, 4.76]	−3.09, [−5.98, −0.21]
Valence change	3.94, [0.79, 7.1]	−1.65, [−5.19, 1.9]
Pride change	0.52, [−0.94, 1.98]	−1.58, [−2.85, −0.31]
Neck angle	19.1, [16.7, 21.5]	43, [38.7, 47.3]
Difficulty	−0.6, [−1.18, −0.0084]	0.8, [0.11, 1.49]
Suspicion of research hypothesis	11 guessed	9 guessed
Body awareness	7.27, [6.72, 7.82]	6.91, [6.24, 7.58]
SCM power	0.091, [0.0079, 0.17]	0.44, [0.31, 0.58]
TRAP power	0.33, [0.14, 0.51]	0.48, [0.31, 0.65]

Note: Means are presented with 95% confidence intervals in square brackets. Positive values of arousal and valence indicate higher arousal and positive valence, respectively.

3.1. Model 1: Effects of Posture Manipulation on Mood

As a necessary first step, we assessed whether the posture manipulation elicited the previously reported behavioral effects, specifically changes in positive mood, emotional arousal, and feelings of pride. Using three simple models with posture as a categorical experimental variable predicting changes in Arousal, Valence, and Pride, respectively, we found significant effects for all mood variables. Relative to contractive postures, expansive postures resulted in increases in arousal (β = 0.42, z = 2.1, p = 0.03, 95% CI = [0.03, 0.80]), positive‐valenced emotions (β = 0.47, z = 2.3, p = 0.02, 95% CI = [0.09, 0.85]), and pride (β = 0.43, z = 2.2, p = 0.03, 95% CI = [0.04, 0.81]; Figure 2). To determine whether these changes were driven by the expansive condition, the contractive condition, or both, we conducted follow‐up one‐sample t‐tests on the mood change scores in each group. Changes in Arousal and Pride were driven primarily by the contractive condition (Arousal: t(43) = −2.1, p = 0.04, 95% CI = [−6.1, −0.13]; Pride: t(43) = −2.4, p = 0.02, 95% CI = [−2.9, −0.27]), while changes in expansive were non‐significant (Arousal: t(56) = 0.96, p = 0.29, 95% CI = [−1.6, 4.6]; Pride: t(56) = 0.83, p = 0.49, 95% CI = [−0.9, 2.1]). For Valence, the opposite pattern was observed, as changes in the expansive condition were significant (t(56) = 2.6, p = 0.01, 95% CI = [0.9, 7.24]), but not changes in the contractive one (t(43) = −0.91, p = 0.37, 95% CI = [−5.3, 2.0]).

3.2. Model 2: Mediation of Posture Effects by Neck Angle

Once we established the efficacy of the manipulation in eliciting the hypothesized mood changes, we proceeded to test our main hypothesis that these effects were mediated by changes in neck angle; that is, we assessed whether the posture manipulation exerted its effects through an influence on neck angle which, in turn, determined the observed changes in mood.

Examining the path between posture category and neck angle, we found that, as expected, participants in the expansive posture had lower (i.e., more upright) neck angles than participants in the contractive posture (β = −1.44, z = −10.3, p < 0.001, 95% CI = [−1.71, −1.16]), consistent with the instructions for adopting the posture given to each group (see Figure S2). Moreover, during‐posture neck angle significantly predicted both changes in Valence (β = −0.30, z = −2.2, p = 0.03, 95% CI = [−0.56, −0.03]) and changes in Pride (β = −0.37, z = −2.7, p = 0.007, 95% CI = [−0.63, −0.11]). In both cases, the direct effect of posture category (controlling for during‐posture neck angle) was no longer significant (Valence: β = 0.04, z = 0.14, p = 0.88, 95% CI = [−0.49, 0.57]; Pride: β = −0.11, z = −0.41, p = 0.68, 95% CI = [−0.63, 0.42]), while the indirect effect was (Valence: β = 0.43, z = 2.1, p = 0.03, 95% CI = [0.04, 0.81]; Pride: β = 0.54, z = 2.7, p = 0.007, 95% CI = [0.14, 0.92]). Moreover, in both cases, the indirect effect through neck angle explained a large majority of the total effect (Valence: Proportion mediated = 91%; Pride: Proportion mediated = 125%; note that the proportion mediated is greater than 100% due to the reversed direction of the direct effect of posture on pride). The indirect effect for Arousal changes was in the expected direction but not significant (see Table 2). Values and corresponding statistics for individual paths are presented in Figure 2, while derived quantities such as indirect effects are presented in Table 2. Taken together, these results indicate that the effect of body posture on Valence and Pride was indeed mediated by the posture's effect on neck angle. In other words, neck angle significantly predicted changes in mood above and beyond the categorical effects of the manipulation. This suggests that the effect of the manipulation on mood, obtained in model 1, was due to the fact that the former group had, on average, a lower (more vertical) neck angle compared to the latter, as shown in Figure 4.

TABLE 2.

Derived statistics for model 2, including estimates of indirect effects.

Model 2
Formula: Neck angle ~  β pn *Posture Mood ~  β pm *Posture +  β nm *Neck angle

Dependent variable	Parameter	β	SE	CI	p
Arousal	β pn * β nm (Indirect effect of posture on DV)	0.091	0.2	[−0.30, 0.48]	0.65
Valence	β pn * β nm (Indirect effect of posture on DV)	0.43	0.2	[0.038, 0.81]	0.032
Pride	β pn * β nm (Indirect effect of posture on DV)	0.53	0.2	[0.14, 0.92]	0.007

Note: Standardized estimates, standard error, 95% confidence intervals, and p‐values are provided. Coefficient subscripts are named for the independent and dependent variables they relate (e.g., β _pn is the effect of Posture on Neck angle). Estimated parameters are indicated by their formula (i.e., the product of two regression coefficients), with a descriptive title given in parentheses.

FIGURE 4.

Scatterplot showing the correlation of neck angle with mood change for (a) Arousal, (b) Valence, and (c) Pride. Participants in the expansive condition are plotted in orange, while participants in the contractive condition are plotted in blue. An overall line of best fit is shown, as well as Pearson's correlation coefficient and p‐value. Plus corresponds to the data mean, while the bounds of the ellipse correspond to the 95% confidence interval of the mean.

3.3. Model 3: Parallel Mediation Effects of Posture Difficulty

In models 3 and 4, we sought to account for potential confounding variables that could impact the posture‐personality relationship. One of these is the difficulty, discomfort, or effort involved in maintaining the posture. Previous work by Bombari et al. (2017) showed that participants found the contractive posture more effortful than the expansive one, and that effort was correlated with the behavioral effects of the manipulation. This implies a mediation effect similar to the one described above, whereby posture exerts effects on mood through a third variable, namely difficulty/effort. Thus, in model 3 we examined whether there was a parallel mediation effect through difficulty, and whether the indirect effect through neck angle was independent of this effect.

Consistent with Bombari et al. (2017), we observed that self‐reported difficulty was significantly different across posture categories; people generally found the contractive posture less comfortable and more difficult to maintain (β = −0.58, z = −3.0, p = 0.002, 95% CI = [−0.95, −0.20]; Figure 3). Moreover, we found that difficulty maintaining the posture negatively predicted both Valence change (β = −0.32, z = −3.4, p < 0.001, 95% CI = [−0.50, −0.13]) and Pride change (β = −0.26, z = −2.7, p = 0.006, 95% CI = [−0.44, −0.07]); the indirect effect of posture via difficulty was also significant (Valence: β = 0.18, z = 2.3, p = 0.02, 95% CI = [0.02, 0.34]; Pride: β = 0.15, z = 2.0, p = 0.04, 95% CI = [0.005, 0.29]; Table 3). However, the parallel mediation path involving neck angle remained largely the same as in model 2 (Valence: β = 0.38, z = 2.0, p = 0.04, 95% CI = [0.009, 0.74]; Pride: β = 0.49, z = 2.6, p = 0.01, 95% CI = [0.12, 0.87]). As before, the direct effect was not significant (Valence: β = −0.09, z = −0.35, p = 0.73, 95% CI = [−0.61, 0.42]; Pride: β = −0.22, z = −0.81, p = 0.42, 95% CI = [−0.73, 0.30]). We tested additional models in which difficulty maintaining the posture was on the mediation path between neck angle and mood change; the serial mediation effect in these models were not significant (see Data S1). Importantly, we note that difficulty maintaining the posture had a generally negative effect on mood changes—thus, it cannot explain why expansive postures result in a more positive mood relative to the pre‐posture time point, as obtained with model 1.

TABLE 3.

Derived statistics for model 3, including estimates of indirect effects.

Model 3
Formula: Neck angle ~  β pn *Posture Difficulty ~  β pd *Posture Mood ~  β pm *Posture +  β nm *Neck angle +  β dm *Difficulty

Dependent variable	Parameter	β	SE	CI	p
Valence	β pn * β nm (Indirect effect of posture on DV)	0.38	0.19	[0.0086, 0.74]	0.045
Valence	β pd * β dm (Indirect effect of difficulty on DV)	0.18	0.08	[0.024, 0.34]	0.024
Pride	β pn * β nm (Indirect effect of posture on DV)	0.49	0.19	[0.12, 0.87]	0.01
Pride	β pd * β dm (Indirect effect of difficulty on DV)	0.15	0.07	[0.0052, 0.29]	0.042

Note: Standardized estimates, standard error, 95% confidence intervals, and p‐values are provided. Coefficient subscripts are named for the independent and dependent variables they relate (e.g., β _pn is the effect of Posture on Neck angle). Estimated parameters are indicated by their formula (i.e., the product of two regression coefficients), with a descriptive title given in parentheses.

3.4. Model 4: Moderators of Postural Feedback Effects on Mood

Finally, in model 4, we examined two potential moderators of the effects observed in our studies. First, as cover stories have been shown to reduce the effect size of postural feedback manipulations (Körner et al. 2022), we examined whether participants who were suspicious of the true intentions of the research would show greater effects than those who were not. Secondly, and to our knowledge not previously tested by any study, we investigated whether body awareness had a moderation effect on postural feedback. This was based on the hypothesis that participants who are more aware of their bodies will feel somatosensory feedback from the posture to a greater extent and should therefore have larger feedback effects according to the theoretical model of bodily feedback articulated in the introduction (James 1884; Price and Harmon‐Jones 2015). We assessed the effects of these two moderators on Model 3.

As before, the indirect effects of posture on mood were significant for neck angle (Valence: β = 0.41, z = 2.2, p = 0.03, 95% CI = [0.05, 0.77]; Pride: β = 0.52, z = 2.8, p = 0.006, 95% CI = [0.15, 0.89]); perceived difficulty was significant for Valence, but only a trend was observed for Pride (Valence: β = 0.14, z = 2.0, p = 0.05, 95% CI = [0.004, 0.28]; Pride: β = 0.12, z = 1.8, p = 0.07, 95% CI = [−0.009, 0.25]). The direct effect of the posture category was not significant in either case (see Table 2). For suspicion of the research hypothesis, indices of moderated mediation were not significant for any path for either Valence or Pride, indicating no significant moderation (all p > 0.2—see Table 4). In contrast, there was a trend for a moderation effect of body awareness on both of the difficulty‐related paths for both Valence and Pride: participants who rated themselves as more aware of their bodies during the posture reported greater difficulty and discomfort with the posture, and also exhibited greater effects of perceived difficulty on their mood (all 0.05 < p < 0.1; see Table 4). No moderation effects were found on the neck angle mediation paths (all p > 0.2).

TABLE 4.

Derived statistics for model 4, including estimates of indirect effects and indices of moderated mediation.

Model 4
Formula: Neck angle ~  β pn *Posture +  β an *Body Awareness*Posture +  β sn *Suspicion Posture Difficulty ~  β pd *Posture +  β ad *Body Awareness*Posture +  β sd *Suspicion*Posture Mood ~  β pm *Posture +  β nm *Neck angle +  β dm *Difficulty +  β am *Body Awareness*Posture +  β sm *Suspicion*Posture

Dependent variable	Parameter	β	SE	CI	p
Valence	β pn * β nm (Indirect effect of posture on DV)	0.41	0.18	[0.048, 0.77]	0.027
Valence	β pd * β dm (Indirect effect of difficulty on DV)	0.14	0.07	[0.0039, 0.28]	0.044
Valence	β an * β nm (Moderated mediation effect of body awareness via posture's effect on neck angle)	0.053	0.05	[−0.037, 0.14]	0.25
Valence	β pn * β am (Moderated mediation effect of body awareness via posture's effect on neck angle)	0.097	0.14	[−0.17, 0.36]	0.48
Valence	β sn * β nm (Moderated mediation effect of suspicion via posture's effect on neck angle)	0.067	0.32	[−0.55, 0.68]	0.83
Valence	β pn * β sm (Moderated mediation effect of suspicion via posture's effect on neck angle)	0.1	0.11	[−0.11, 0.31]	0.35
Valence	β ad * β dm (Moderated mediation effect of body awareness via difficulty's effect on neck angle)	0.12	0.07	[−0.018, 0.25]	0.09
Valence	β pd * β am (Moderated mediation effect of body awareness via difficulty's effect on neck angle)	0.13	0.07	[−0.0033, 0.25]	0.056
Valence	β sd * β dm (Moderated mediation effect of suspicion via difficulty's effect on neck angle)	−0.011	0.13	[−0.27, 0.25]	0.94
Valence	β pd * β sm (Moderated mediation effect of suspicion via difficulty's effect on neck angle)	0.017	0.11	[−0.19, 0.23]	0.87
Pride	β pn * β nm (Indirect effect of posture on DV)	0.52	0.19	[0.15, 0.89]	0.006
Pride	β pd * β dm (Indirect effect of difficulty on DV)	0.12	0.07	[−0.009, 0.25]	0.068
Pride	β an * β nm (Moderated mediation effect of body awareness via posture's effect on neck angle)	0.07	0.06	[−0.039, 0.18]	0.21
Pride	β pn * β am (Moderated mediation effect of body awareness via posture's effect on neck angle)	−0.073	0.14	[−0.34, 0.19]	0.59
Pride	β sn * β nm (Moderated mediation effect of suspicion via posture's effect on neck angle)	−0.29	0.32	[−0.91, 0.34]	0.37
Pride	β pn * β sm (Moderated mediation effect of suspicion via posture's effect on neck angle)	0.14	0.13	[−0.12, 0.41]	0.29
Pride	β ad * β dm (Moderated mediation effect of body awareness via difficulty's effect on neck angle)	0.14	0.07	[−0.0045, 0.28]	0.058
Pride	β pd * β am (Moderated mediation effect of body awareness via difficulty's effect on neck angle)	0.11	0.06	[−0.011, 0.24]	0.074
Pride	β sd * β dm (Moderated mediation effect of suspicion via difficulty's effect on neck angle)	−0.15	0.14	[−0.42, 0.12]	0.28
Pride	β pd * β sm (Moderated mediation effect of suspicion via difficulty's effect on neck angle)	0.029	0.1	[−0.16, 0.22]	0.76

Note: Standardized estimates, standard error, 95% confidence intervals, and p‐values are provided. Coefficient subscripts are named for the independent and dependent variables they relate (e.g., β _pn is the effect of Posture on Neck angle). Estimated parameters are indicated by their formula (i.e., the product of two regression coefficients), with a descriptive title given in parentheses.

3.5. Physiological Correlates

As it was previously found that neck angle is related to muscle activity (Wainio‐Theberge and Armony 2024), we also examined the relationship between our neck angle measure and EMG activity from neck flexors and extensors. We first examined changes in muscle activity in each posture: using one‐sample t‐tests, we found that SCM activity increased in both the expansive and contractive postures (expansive: t(56) = 2.1, p = 0.04, 95% CI = [0.004, 0.17]; contractive: t(43) = 6.6, p < 0.001, 95% CI = [0.31, 0.58]), though significantly more so in the latter (t(99) = 4.7, p < 0.001, 95% CI = [0.21, 0.51]). TRAP power also increased in both postures, but did not differ between conditions (expansive: t(56) = 3.6, p < 0.001, 95% CI = [0.15, 0.53]; contractive: t(43) = 5.6, p < 0.001, 95% CI = [0.31, 0.66]; difference: t(99) = −1.1, p = 0.29, 95% CI = [−0.41, 0.12]). Importantly, we found that SCM and TRAP activity were positively correlated with during‐posture neck angle (r = 0.42, p < 0.001, 95% CI = [0.24, 0.57] and r = 0.20, p = 0.04, 95% CI = [0.006, 0.38], respectively; Figure 5); partial correlations revealed that SCM power remained significant when controlling for TRAP (r = 0.39, p < 0.001, 95% CI = [0.21, 0.55]), but the reverse was not true (r = 0.13, p = 0.19, 95% CI = [−0.07, 0.32]), indicating that TRAP power reflected a general increase of muscle tension in more contractive postures, rather than a component specific to neck angle. Moreover, only SCM power was also correlated with changes in mood, specifically Valence (r = −0.26, p = 0.009, 95% CI = [−0.43, −0.07]) and Pride (r = −0.20, p = 0.05, 95% CI = [−0.38, −0.005]). Thus, the changes in neck flexion observed in our experiment (and their associated mood effects) appear to be partly, though not wholly, attributable to activity in the sternocleidomastoid muscle.

FIGURE 5.

Results of the physiological validation. (a) Scatterplots as in Figure 4 showing the correlation of neck angle with sternocleidomastoid (SCM) EMG power, (b) Scatterplots as in Figure 4 showing the correlation of SCM power with mood change for Valence (left) and Pride (right). Plus corresponds to the data mean, while the bounds of the ellipse correspond to the 95% confidence interval of the mean.

3.6. Additional Validation Analyses

Following recommendations by Creedon and Hayes (2015) for small sample size research using mediation models, we repeated our analyses using bootstrap estimated standard errors. These results are reported in the Data S1, and closely mirror those presented here. Furthermore, to further validate the mediation effect, we employed a permutation approach to test the mediation effect of model 2. Within each posture group (expansive and contractive), we shuffled participants' neck angle values 1000 times, and recomputed the mediation effect. We then compared our empirically observed mediation effect to this null distribution. The procedure revealed that the mediation effect observed in our data was significantly greater than would be expected by chance (Valence: z = 2.0, p = 0.02; Pride: z = 2.7, p = 0.001). This suggests that the mediation effect of neck angle on mood changes was not merely a consequence of the average group difference in this variable but, rather, reflected a direct, within‐subject significant relation between angle and mood change.

In addition to neck angle as a quantitative measure of contractive postures, we also measured two other relevant angles from the same video data which indexed participants' expansive and contractive postures. First, we measured the participant's up or down tilt of the head; this angle has similarly been implicated in dominance research, particularly for faces (Witkower and Tracy 2019), and should likewise index postural expansion or contraction (i.e., participants in the contractive pose look down more, while participants in the expansive pose look up more). Second, we computed the forward and backward lean of the trunk, which can be thought of as a “bowing” motion, indicating deference (Wierzbicka 1995), and thus also related to dominance. We conducted the analyses of model 2 using these angles instead of the neck angle. As could be expected, both angles were correlated with neck angle and differed between posture categories. However, although they correlated to some extent with mood questionnaires, we found no mediation effects for either of them (see Data S1); that is, the indirect effects of head tilt angle and trunk angle were not significant. Thus, it appears that the mediation effects we observed for neck angle did not simply reflect a general effect present for any quantitative measure associated with expansive and contractive postures, but were fairly specific to that particular postural angle; this provides tentative evidence that neck flexion may play a key role in dominance signaling, rather than serving simply as an index of a more global postural expansion or contraction.

4. Discussion

Embodied theories of emotion suggest that adopting physical expressions of emotion can have feedback effects on an individual's mood and behavior; however, the influence of these postures appear heterogeneous and well‐known paradigms struggle to achieve consistent effects. This inconsistency may stem from variation in physical aspects of the pose participants adopt: for example, how much they adopt the expression, using which muscles, or how much tension is used to maintain the posture. In the present work, we examined these factors for the first time, hypothesizing that the flexion angle of their neck would mediate the effects of expansive and contractive postures on mood in a postural feedback paradigm. In agreement with our hypothesis, we found a mediation effect on Valence and Pride changes: the effects of the categorical posture manipulation were primarily mediated by the manipulation's effect on neck flexion, while the residual direct influence of the manipulation was small and non‐significant. Notably, participants with similar neck angles, even in different posture categories, experienced similar changes in mood (Figure 4). These findings constitute the first experimental evidence for the proposal that postural feedback paradigms using expansive and contractive postures operate via inducing variability in an underlying continuous physiological variable, and thus provide new strong support for embodied theories of emotion (Frijda 2007; Varela et al. 2017). Our results can also help improve the measurement and design of bodily feedback experiments, as well as provide new options for their application to clinical or social interventions. They also accord with previous theoretical work predicting that different types of posture manipulation (expansive/contractive vs. upright/stooped; Körner and Schütz 2020), which differ both in the movements involved and their magnitude or intensity, may have different psychological effects.

Our work sheds additional light on the physiological correlates of dominant and submissive body language by demonstrating the role of specific muscles in these displays. In the present work, we found that sternocleidomastoid activity reliably distinguished between the expansive and contractive postures, with greater SCM power in more flexed, contractive positions; moreover, we found that SCM power functioned like neck angle in mediating the effects of the posture manipulation. In contrast, while TRAP power also correlated with neck angle, it did not correlate with mood change and was no longer significant when controlling for SCM power, suggesting that the effect was likely due to a general increase in muscular effort associated with the posture. While this is in line with the muscle's role as a neck flexor, it is in apparent contrast with the results of our previous study (Wainio‐Theberge and Armony 2024), where we observed that larger neck angles in participants' natural body posture were associated with lower values of resting SCM power; we suggested that this effect was due to passive lowering of the head in naturally submissive postures. In contrast, in the present study, individuals were encouraged to actively take on the postures to an extreme degree and to maintain them for several minutes, making the movement much more likely to elicit SCM activity. Thus, while the SCM appears to be consistently relevant for dominance‐related body language, its involvement is complex and depends on the actions involved; active expansion or contraction may both increase SCM activity, due to the influence of stretch‐reflex firing (Young 1994). In addition to their theoretical relevance in terms of the physiological correlates of the behavioral effects of dominance‐related body postures, our findings may help resolve some of the controversy in the “power posing” literature generated by failed replications and inconsistent findings, as well as propose some recommendations for improving the reliability of the manipulation. First, we note that simple correlations between during‐posture head angle and mood were 50% larger than group comparisons between the expansive and contractive posture categories. This is statistically expected when comparing an arbitrary categorical split of an underlying continuous variable (Irwin and McClelland 2003). Our results suggest that the expansive and contractive posture conditions applied in dominance‐related body feedback studies are a means of achieving variability in neck angle, rather than having effects by themselves. Thus, we suggest that future studies of power posing measure neck angle during the posture using the methods outlined here and use it as a continuous independent variable, rather than relying solely on the posture category. Moreover, such manipulations may benefit from ensuring that expansive and contractive postures are exaggerated to maximize variability. However, this recommendation must be balanced with its potential impact on the degree of difficulty; participants may find exaggerated postures uncomfortable, creating spurious effects. Consistent with previous reports (Bombari et al. 2017), in our study, participants found contractive postures more effortful and difficult than expansive ones, which produced spurious effects in the same direction as the theoretical effects of power posing (i.e., greater negative mood for contractive postures). To mitigate this, we suggest that researchers ask participants in the expansive group to exaggerate their posture; this aims to maximize variability in neck angle, while balancing muscular effort with the contractive posture to minimize potentially confounding effects.

Additionally, our analysis of moderators suggests that body awareness had an impact on the magnitude of the effect, such that participants who were more aware of their bodies during the posture showed larger effects in mood change. Thus, it may be beneficial to direct participants' attention to their bodies during the manipulation. However, as shown in model 4, body awareness moderated the mediating effect of difficulty on the pose‐mood path and therefore could introduce further confounding effects on the results. Indeed, it may be that increasing participants' conscious awareness of their bodies merely increases focus on irrelevant physical sensations which co‐occur with expansive and contractive postures.

In contrast with previous observations of a weaker effect of posture on mood in studies where cover stories were employed to reduce participants' awareness of the research hypothesis, we observed no significant influence of suspicion of the real aim of the study on the posture effects we observed. However, given the relatively small number (20%) of individuals who were suspicious of the hypothesis in our study, we cannot rule out such an effect, as we may have lacked the power to detect it. Moreover, between‐study findings that cover stories reduce effect sizes may not translate directly to within‐study effects, as studies with cover stories may vary in other parameters besides participants' awareness of the research hypothesis. Taken together, our findings suggest that the effects of induced poses on mood can be optimized by having participants maximize their neck angle and be as aware as possible of their body posture, but doing so in a natural way that minimizes effort and discomfort. Further, our results suggest that measuring the actual body posture participants adopt, as well as the key variables of difficulty and body awareness, may be useful in increasing the robustness of power posing designs. While the measurement of neck angle is relatively simple and can be integrated into most experimental paradigms, we encourage researchers to also examine other variables from photogrammetric assessments (such as head tilt and trunk angle, examined here) and to improve measurement procedures in order to assess other dimensions of postural expansion and contraction (see below); different experimental contexts (such as dyadic interaction or implicit posture manipulations where expansion and contraction are encouraged by the experimental environment) may result in different variables being relevant.

4.1. Limitations and Future Directions

A primary limitation of our study is the sample size, which while large for physiological studies, is small for conducting mediation analyses (Fritz and MacKinnon 2007). We thus urge caution in interpreting our findings, as our study was not powered for mediation analysis, but rather for the main effects of posture on mood. Accordingly, while we observed a full mediation effect, in the statistical sense, of neck angle on the posture‐mood relation, this should not be interpreted as definitive evidence that no categorical effects of posture manipulations exist; we would need substantially higher power to rule out the possible presence of a small direct effect of expansive or contractive posture categories (Schoemann et al. 2017).

The present study contributes to a growing literature emphasizing the role of the neck angle or cervical angle in displays of dominance and submission (Mignault and Chaudhuri 2003); however, multiple component actions and movements have been implicated in dominant body language, including upright posture, an expanded chest/shoulders, leaning back while seated, and limb extension (Hall et al. 2005). In particular, expansive and contractive postures typically used in power posing can be decomposed into a vertical component (i.e., only neck flexion) and a horizontal component (i.e., only chest compression and limb contraction), the distinction between which has been theorized to underlie different dimensions of power‐related affect (Körner and Schütz 2020). However, due to our experimental setup and the photogrammetric method we used, we were only able to accurately measure the former. Future studies designed to isolate the different components of expansive and contractive postures should reveal which specific actions have relevant feedback effects.

Finally, future work should examine other muscular correlates of dominant and submissive body language. Our study focused only on two muscles (sternocleidomastoid and upper trapezius), selected for their relevance for neck flexion and extension; however, additional muscles may be implicated, especially for expansive postures, where we found no significant increase in trapezius power compared with the contractive posture.

4.2. Constraints on Generality

The present study, as with many psychological studies, is limited in that its sample primarily consisted of young (ages 18–35) females in psychology and social science undergraduate programs at a Canadian university. These samples have been shown to be outliers in the global distribution of cognitive and personality traits relating to social norms, education, individualism, and even some basic perceptual functions (e.g., perception of optical illusions); therefore, findings obtained in these samples may not generalize well to other populations (Henrich et al. 2010). In particular, although we found no evidence of sex as a moderator of the effects observed here (see Data S1), sex is of particular theoretical importance in research on social dominance, as dominance is frequently employed in inter‐male competition (Andersson 1994). However, given the relatively small overall sample size and the imbalance in sex distribution, our study was underpowered to reveal any potential sex differences. Future work should therefore focus on improving the size and diversity of samples and replicating our results in different societies and cultures.

5. Conclusion

Bodily feedback from emotions is a well‐studied and theoretically grounded paradigm, but experimental effects can be small and variable. In particular, feedback from expansive and contractive postures which signal power and dominance has been well studied, but its mechanism remains largely unknown. Here, we demonstrated that physiological and objective measures of posture, including photogrammetric measurements of neck flexion and muscle activity in neck flexors, mediate the feedback effects of expansive and contractive postures. These results suggest that postural feedback paradigms, which encourage individuals to adopt stereotypical emotional expressions, operate through underlying muscle systems. In addition to their theoretical relevance for bodily feedback, our results provide guidance to researchers using paradigms like “power posing” for how to make these experiments more reliable and robust.

Author Contributions

Soren Wainio‐Theberge: conceptualization, data curation, formal analysis, investigation, methodology, project administration, validation, visualization, writing – original draft, writing – review and editing. Ignacio Spiousas: methodology, validation, writing – review and editing. Jorge L. Armony: conceptualization, project administration, supervision, funding acquisition, resources, writing – review and editing.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Data S1.

Wainio‐Theberge, S. , Spiousas I., and Armony J. L.. 2025. “Physical Mechanisms of Emotions Evoked by Postural Feedback.” Psychophysiology 62, no. 7: e70095. 10.1111/psyp.70095.

Data Availability Statement

Data and code used for this manuscript are available at the following OSF repository: https://osf.io/gpbkv/.