The role of bodily posture is centrally featured in classic writings of emotion (see Darwin, 1872; James, 1884; Dewey, 1894). Bodily posture conveys action tendencies associated with corresponding emotions (Frijda, 1986). For example, a fear posture typically involves an avoidant and protective physical response to an external referent, whereas an anger posture is associated with the individual extending toward a referent and becoming larger. More recent empirical research demonstrates that even relatively small differences in body posture and movement can influence emotion perception and categorization (Dael et al., 2012).

Of particular interest is work investigating the effect of body posture on the perception of affective expressions (Meeren et al., 2005; Van den Stock et al., 2007; Aviezer et al., 2008b, 2011; de Gelder, 2009). For example, a disgust face transposed onto a posture demonstrating a punching motion is perceived as a facial expression of anger (Aviezer et al., 2008b). Such research makes central the need to more carefully consider the perception of emotion as communicated through bodily posture. However, although recent studies have manipulated the postures on which facial expressions were placed (e.g., Aviezer et al., 2008b), these postures often included additional non-postural elements (e.g., a headstone in the background; a soiled undergarment) that may have influenced emotion perception independently of the accompanying posture. Moreover, studies specifically examining the perception of emotion postures frequently display the body in conjunction with corresponding facial expressions (e.g., Bänziger et al., 2012), creating an obvious confound with regards to the perception of the posture in and of itself. More careful study of postural communication of emotion is critical given the influential role of bodily posture in emotion perception.

This study examined adult ratings of postures expressing five discrete emotions: joy, sadness, fear, anger, and disgust. The study was guided by three primary aims to address the above limitations. First, we attempted to validate a static postural display of disgust. Second, we examined participant emotion categorization of distinct bodily postures conveying different action tendencies (e.g., anger expressed as striking out vs. reaching upward in exasperation). Finally, we assessed participant ratings of valence and arousal for each posture to corroborate categorical and dimensional properties of postural communications of emotion.

Two actors (one male, one female) posed postural expressions of five discrete emotions: joy, sadness, fear, anger, and disgust. Actors wore gender-neutral clothing (collared shirt or sweater with jeans) that covered their arms and legs. Pictures were taken with an 8 megapixels digital camera in front of a white background under controlled lighting. Images were digitally altered using Adobe Photoshop to remove the head and neck of each actor, resulting in each image displaying the body from the collar down. Additionally, all images were converted to gray scale to minimize any effect of clothing color and actor skin tone (visible on the hands) on participants’ ratings.

Each discrete emotion category included a bodily pose similar to previously validated stimuli (Aviezer et al., 2011; de Gelder and Van den Stock, 2011b), as well as iterations of related, but structurally novel, postural expressions. The resulting set of 30 stimuli consisted of distinct poses for joy (3), sadness (4), fear (2), anger (3), and disgust (3), each including a male and female version (see Figure 1, for examples¹). The number of poses per emotion varied as a function of the flexibility with which particular emotions are likely to be expressed. For example, it is theorized that there are many ways to express sadness and anger, but likely fewer ways with which to express fear (see Frijda, 1986). All stimuli subtended a vertical visual angle of 11.7° and an average horizontal visual angle of the torso of 2.27° (SD = 0.15°) when viewed from a distance of 60 cm, and exuded an average luminance of 63.78 cd/m² (SD = 19.21). Stimuli did not differ significantly on these low-level features across emotions [Subtension: F(4,15) = 2.29, p = 0.11; Luminance: F(4,15) = 1.55, p = 0.24] or within emotions [Subtension: F(7,15) = 1.16, p = 0.38; Luminance: F(7,15) = 0.28, p = 0.95].

Separate samples were recruited to complete the categorization and dimensional portions of the study to prevent potential carryover effects across ratings. The UC Merced Institutional Review Board approved all procedures and all participants provided written informed consent.

The overall percent agreement for the posture images was high (79%) and the corresponding Cohen’s kappa value indicated substantial inter-rater agreement (κ = 0.64; Landis and Koch, 1977). The majority of images were categorized as the target emotion significantly above chance levels (one-tailed binomial tests, ps < 0.003), with the exception of the female Joy 3 posture (p = 0.23) and the male Sadness 4 posture (p = 0.06).

Bonferroni-corrected paired t-tests (α = 0.0125) revealed that 14 of the 15 posture versions were identified as the target emotion significantly more frequently than any other emotion category (all ps < 0.003). The exception was the Joy 3 posture, which was rated as joy in the majority of instances (M = 52%), but not significantly more than it was miscategorized as anger, (M = 33%), t(25) = 1.31, p = 0.20, CI [-0.20, 0.59]. The confusion matrix presented in Table 1 provides the proportions of participant emotion classifications for the emotion postures.

Participant ratings of valence and arousal were analyzed using Bonferroni-corrected paired t-tests to examine differences in each dimension between emotion categories and between poses within the same emotion category. The mean ratings of valence and arousal dimensions for each pose are presented in Table 1. Additionally, the spatial coordinates of the valence and arousal ratings of each posture are plotted in Figure 2.

Participants accurately categorized postural expressions of five discrete emotions (i.e., joy, sadness, fear, anger, and disgust). The identification of a postural expression of disgust represents a novel contribution to the emotion literature, and to our knowledge is the first study to validate a static disgust posture in which no facial or contextual cues were present. Interestingly, disgust postures were identified at comparable, if not better, rates as other emotion postures.

Further examination of participant miscategorizations of emotion postures indicated that joy postures were most commonly miscategorized as anger, and sadness postures were sometimes confused with fear, particularly when the body was angled to one side (i.e., Sadness 2). The observed miscategorizations provide unique insight into the potential confusability of discrete bodily postures akin to structural similarities documented in facial expressions of emotion (Susskind et al., 2007). While such a bottom-up processing of emotional postures is possible, equally plausible is a top-down explanation centering on likely actions associated with the posture. For example, when disgust was miscategorized it was typically labeled as fear, specifically when the arms were outstretched in a downward motion (i.e., Disgust 2). This pattern of miscategorization may be due to the functionally similar action tendencies of these emotions to avoid a stimulus or source of threat (Fridja et al., 1989).

Postures for each emotion were successfully categorized using multiple physically distinct poses. Notable exemplars included poses varying in the communicated action tendency, such as expressing “anger in” and “anger out” (see Frijda, 2007). While prior studies have included multiple actors expressing the same emotion (e.g., de Gelder and Van den Stock, 2011b), the pose and action tendency expressed for each emotion were typically invariant in such studies. Furthermore, although prior studies including multiple actors could be said to have inadvertently included distinctions in poses across actors, lack of statistical comparisons of such differences prevent any definitive conclusions on this point. Also, while the present study included a range of poses within each emotion category, there are likely numerous other poses of these emotions that are distinct from those tested. Emotion poses in the present study that resembled postures used in prior research (i.e., Joy 1, Fear 2, and Anger 3) were rated with similar or better accuracy than reported previously (Atkinson et al., 2004; de Gelder and Van den Stock, 2011b). It is also worth noting that although all postures in the current study were correctly categorized as the target emotion, some poses within emotion categories were more accurately recognized than others. For example, the anger pose featuring hands clawing upward in the air (Anger 2) was more accurately identified than the anger pose with clenched fists in the air (Anger 1). Interestingly, the latter was often confused as joy, which may be due to its perceptual similarity with bodily expressions of pride (e.g., Aviezer et al., 2008b; Tracy and Matsumoto, 2008).

Participants’ successful identification of perceptually distinct, yet categorically similar, emotion postures also substantiates a broader theoretical point: the equifinality of emotional expression. Researchers seeking a singular archetype expression of discrete emotions risk ignoring the underlying behavioral functions of different emotional responses (Campos et al., 1989). The present findings support the view that different action tendencies can be associated with the same emotion (Frijda, 2007). Thus, rather than utilizing a 1:1 mapping of expression and emotion, the processing of emotional communication likely relies on an appreciation for the nature of a social partner’s relation with the environment (see Barrett and Campos, 1987).

Furthermore, the inclusion of a variety of postural expressions of the same discrete emotion expands the repertoire of postures available to researchers and increases flexibility for conducting research on emotion perception. This is likely of particular relevance for studies integrating emotion postures with other emotional stimuli, such as faces or contextual scenes (e.g., Meeren et al., 2005; Aviezer et al., 2008b; Righart and de Gelder, 2008; Kret and de Gelder, 2010; Mondloch, 2012). For example, an anger posture depicting a raised fist toward a rival may be more contextually appropriate than the same posture directed toward a broken computer, for which the upward hands in exasperation is likely more ecologically valid. The present set of stimuli provides additional avenues for combining emotion postures with other emotion-related elements.

In addition to categorizing each postural pose of emotion, a separate group of participants rated the stimuli on dimensions of valence and arousal using the Affect Grid (Russell et al., 1989). The majority of postural expressions of discrete emotion categories were distinct in valence and arousal, though there were some exceptions. Fear and anger postures were similarly negative in valence and high in arousal despite typically being associated with opposing action tendencies. Additionally, valence ratings revealed that sadness and disgust postures were similarly negative, and that ratings of arousal did not differ for joy and disgust postures.

In general, the observed pattern of results is in line with previous research examining dimensional ratings of emotion facial expressions. However, it is notable that joy postures were rated less positively in valence than facial expressions of joy included in prior research (Russell and Bullock, 1985). This may help account for joy postures being consistently more difficult to validate in comparison with other bodily expressions of emotion (de Gelder and Van den Stock, 2011b). It is also interesting that valence and arousal varied between some poses belonging to the same emotion category. For example, Anger 2 (punching anger) and Anger 3 (inner anger) were rated as significantly more negative in valence than Anger 1 (both fists in the air). As stated earlier, these differences may suggest that some raters may have perceived Anger 1 as communicating pride, resulting in Anger 1 being rated less negatively valenced.

More generally, the present findings demonstrate the value of including both categorical and dimensional data to better understand differences in the perception of emotional communication. The dimensional data can help inform the selection of specific emotion postures for use with contextual cues. For example, high-arousal postures may exert increased influence on the perception of other modalities of emotional communication, such as the face (see Aviezer et al., 2012). Additionally, ratings of valence and arousal can allow investigators to compare emotion postures beyond their perceptual features (e.g., arms being up or arms being down) and consider similarities and distinctions in their communication of these dimensions.

Several considerations relating to the limitations of the present study and opportunities for further research warrant mentioning. First, the procedures and design of the study may have limited participants’ responding. Our procedure utilized a forced-choice method using five basic emotion categories. Although this methodology is consistent with previous posture studies (e.g., Atkinson et al., 2004; de Gelder and Van den Stock, 2011b), a less constrained response set (e.g., including an ‘other’ or ‘fill-in-the-blank’ option) may have yielded greater variation in participant ratings. Thus, while our inclusion of five distinct responses may have been more liberal than other emotion classification studies (e.g., de Gelder and Van den Stock, 2011b), this design may have restricted the richness of participant differences in perceiving the emotion poses. This concern may be particularly relevant for the parsing of emotions within an emotion category (e.g., ‘solemn’ vs. ‘sad’ vs. ‘depressed’) or identifying other discrete emotion categories (e.g., pride, shame, awe). The absence of a debriefing session limits our knowledge of how appropriately the provided emotion labels accurately captured participant perceptions of the poses. Additionally, the lack of a neutral posture precludes the ability to compare the present postures with a “control” posture (though whether a neutral expression would provide such a control is debatable; see Donegan et al., 2003; Somerville et al., 2004). Although, previous studies have similarly omitted inclusion of a neutral posture or a neutral option in categorization procedures (Atkinson et al., 2004; de Gelder and Van den Stock, 2011b), the inclusion of such a posture or option would have provided greater distinction in assessing the categorical and dimensional ratings for the emotions of interest.

Second, it is important to consider aspects relating to the construction of the stimuli. The decision to exclude the head of the actor in our posture images may have eliminated a relevant element for emotion perception (Atkinson et al., 2004; de Gelder and Van den Stock, 2011b). Inclusion of the head with a blurred face may have increased the validation scores in the present study. For example, including the head being pulled back so as to avoid sensory contamination could further increase categorization for disgust poses (Rozin and Fallon, 1987). While the positioning of the head can certainly provide information regarding the communicated emotion (Dael et al., 2012), the decision to omit the head from the present set of stimuli was made to accentuate the role of the body. Importantly, it seems unlikely that our decision to exclude the head from our images artificially inflated participants’ ratings – in fact, the obtained ratings may have been more robust had head orientation been included. Additionally, each of the included emotion poses was artificially staged. Research examining facial expressions of emotion has noted that classic displays (e.g., Ekman and Friesen, 1976) are often different from those deemed naturalistic (e.g., Gosselin et al., 1995; Carroll and Russell, 1997; Scherer and Ellgring, 2007) or observed spontaneously (for reviews, see Fernández-Dols and Crivelli, 2013; Reisenzein et al., 2013). Whether such discrepancies exist in postural expressions of emotion remains to be studied. Comparison of the present stimuli with those naturally observed would help to account for any perceived artificiality of the stimuli.

Finally, although the present study isolated postural communication of emotion, it is crucial that research on emotion perception also consider the gestalt of emotion contexts. We encourage future studies to explore how emotion perception is affected by different elements of relational contexts, and particularly how such elements may interact with one another. For example, a disgust face superimposed onto an angry posture with both fists raised (i.e., Anger 1 pose) in the context of a victorious sporting event may be judged as pride despite neither the face nor the posture being categorized as pride when viewed in isolation (Aviezer et al., 2012). Examining how specific emotion-related elements (e.g., face, posture, voice, physical environment, cultural context, personal history of the protagonist) differentially influence emotion perception is a topic on which researchers have only begun to scratch the surface (for an excellent example of such research, see Van den Stock et al., 2007).