Emotional Mimicry Beyond the Face?

Method

Stimuli

Face stimuli were 40 color photographs of models (five males, five females; Figure 1) posing to convey the emotions of anger and fear. Angry facial displays were characterized by knitted brows and pursed lips or bared teeth. Fearful facial expressions were characterized by raised brows and an open mouth. Each model posed for each emotion twice. These images were pilot tested to ensure they accurately represented these emotions and to select the photograph for each model that conveyed the most anger and fear. Participants viewed all images and rated the extent to which the image conveyed anger and the extent to which it conveyed fear on a scale of 1 (not at all) to 7 (a great deal). Those images with the highest average differences between anger and fear were selected for use in the current study. All selected images had an average target emotion rating above 4.27 and an average nontarget emotion rating below 2.04.

OPEN IN VIEWER

Figure 1.

Example fear and anger facial emotion stimuli.

EMG Measures

For facial muscle movement, EMG was used to record levels of muscle activity over the corrugator supercilli (knits brow) and the medial frontalis (raises inner eyebrow). Corrugator activity has been shown to be a marker of negative emotions such as anger (Cacioppo, Petty, Losch, & Kim, 1986). Activity over the medial portion of the frontalis has been associated with fear because often the brow raises when one is afraid (Darwin, 1872/1998; Ekman & Friesen, 1976; Frois-Wittman, 1930; Moody et al., 2007; Smith, 1989). For arm muscle activity, EMG electrodes were placed over the forearm extensor and flexor forearm groups. Forearm extensor muscle activity is associated with lifting of the fingers and hands to ward off danger, whereas flexor and extensor activity would be consistent with making a fist as in anger (see Figure 3 for electrode placements).

OPEN IN VIEWER

Figure 2.

Schematic of stimulus presentation.

OPEN IN VIEWER

Figure 3.

Sensor placement and associated actions.

Standard EMG site preparation and electrode placement procedures were followed (Tassinary, Cacioppo, & Geen, 1989). Before electrode placement, skin over the muscle group was cleansed with rubbing alcohol and gently abraded with NuPrep Gel®. Electrodes were 4 mm Ag-AgCl, cup-style electrodes. All impedances were within tolerable ranges per standard laboratory procedure (≤20 kΩ). Sites were recleaned and abraded as needed until reaching this threshold. Muscle activity was recorded using a NeuroScan Labs, SynAmps® model 5083 electroencephalograph amplifier. Activity over each muscle group was recorded using two electrodes placed approximately 1.25 cm apart from center to center, roughly parallel to the length of the muscle. Activity over each muscle was continuously recorded at a sampling rate of 2,000 Hz with a 10–500 Hz band pass filter and a 60 Hz notch filter. The EMG signals were immediately amplified at the head-box by a factor of 150 and again by the main amplifier by a factor of 500.

To analyze EMG, each continuous file was inspected visually for noise and artifacts (Dimberg, Hansson, & Thunberg, 1998; Winkielman & Cacioppo, 2001). Next, the waveform around each stimulus presentation was visually inspected to look for artifacts and anomalous waveforms. Sweeps in which the baseline was not intact or that contained clearly anomalous waveforms were dropped from the analyses. No more than 10% of the total number of sweeps for each individual was dropped.

Following visual inspection, the EMG waveforms were divided into 500 ms chunks. These chunks were smoothed and rectified, and the integral under the curve for each time window was calculated using CNS Analysis Suite, Version 5.51 (The CNS Analysis Suite [Computer software], 1999). Consistent with other research (e.g., Moody et al., 2007; Winkielman & Cacioppo, 2001), the integral values were log10 transformed to reduce the impact of extreme values. These values were then standardized within participant and within muscle so meaningful comparisons could be made across muscles and participants. Next, the prestimulus level of activity was subtracted from each poststimulus chunk to measure the change in activity caused by viewing each stimulus (i.e., to calculate the change from baseline). The prestimulus baseline window was the 500 ms before stimulus onset in which the participants saw a blank screen.

To analyze these data, we first plotted representative line graphs of all muscle activity to anger and fear faces over the 3 s to visualize the responses (see Figures 4). To facilitate interpretation of subsequent statistical models and consistent with previous research, we divided the time window of interest into two roughly equal segments and averaged the muscle activity for each muscle (corrugator, frontalis, forearm extensor group, and forearm flexor group) within those segments (500–1,800 ms and 1,800–3000 ms poststimulus onset). This creates two time periods of interest and is a common method to examine EMG mimicry data. These periods were chosen because facial mimicry typically emerges by 500 ms after stimulus onset (Dimberg, 1982; Dimberg & Petterson, 2000; Moody et al., 2007), while arm muscles likely require a longer window after stimulus onset to reach maximal reaction due to their physical distance from the brain.

OPEN IN VIEWER

Figure 4.

Face and body muscle responses in z scores for fear and anger facial stimuli over first 3 s.

Results

To evaluate the effect of the emotional facial expressions on specific muscle group responses, a 2 (emotion: anger, fear) × 4 (muscle group: frontalis, corrugator, extensor, flexor) × 2 (time window: 500–1,800 ms, 1,800–3,000 ms) repeated-measures analysis of variance (ANOVA) was conducted combining face and body muscle group responses. Note that, we use ω p 2 as our measure of effect size, which is the amount of variance in the dependent variable that is attributable to a given factor, accounting for all other factors in the model (Olejnik & Algina, 2003). This measure of association is preferable because it is less likely to overestimate the effect size than η p 2 and can be generalized to numerous research designs (Lakens, 2013). Evidence of mimicry would be observed from interactions of emotion with muscle groups and potentially time window. The ANOVA indicated a three-way interaction, F(3, 42) = 5.68, p = .002, ω p 2 = .23. To understand this three-way interaction, separate ANOVAs were conducted for face and body muscles: 2 (emotion: anger, fear) × 2 (muscle group: frontalis, corrugator or extensor, flexor) × 2 (time window: 500–1,800 ms, 1,800–3,000 ms) ANOVA.

Our findings for facial EMG responses are consistent with previous findings for facial mimicry in response to viewing angry and fear facial expressions. For face muscles, there was a significant interaction for emotion and muscle group, F(1, 44) = 9.20, p = .004, ω p 2 = .15. Simple effects analyses confirmed significantly greater frontalis activity in response to fear faces in response to fear faces (μ = .07, SE = .03) versus anger faces (μ = −.05, SE = .03), F(1, 44) = 9.17, p = .004, ω p 2 = .15; overall corrugator activity was similar for both anger faces (μ = −.02, SE = .03) and fear faces (μ = −.29, SE = .03), F(1, 44) < 1, p = .71, ω p 2 = 0. However, emotion-specific activation of the corrugator and frontalis muscles changed over time (Figure 5). The significant Emotion × Muscle Group × Time Window interaction indicated that for anger face stimuli, differentiation of corrugator over the frontalis activation emerged only in the later time window, but for fear face stimuli, differentiation of the frontalis emerged in the early time window and was maintained into the later time window, F(1, 44) = 12.77, p = .001, ω p 2 = .20. Simple effect analyses confirm that for anger face, there was no corrugator/frontalis (c/f) difference early in time (μ c f dif 500–1,800 ms = −.01, SE = .03; F(1, 44) = 1, p = .73, ω p 2 = 0), but there was later in time (μ c f dif 1,800–3,000 ms = −.08, SE = .04; F(1, 44) = 5.56, p = .02, ω p 2 = .09). For fear face stimuli, frontalis activation was emerged in the early time window μ c f dif 500–1,800 ms = −.08, SE = .03; F(1, 44) = 4.82, p = .03, ω p 2 = .08 and was maintained through the later time window (μ c f dif 1,800–3,000 ms = −.11, SE = .04, F(1, 44) = 7.48, p = .01, ω p 2 = .12). These results replicate our previous work and provide evidence of emotional mimicry of angry and fearful expressions using facial muscles (Moody & McIntosh, 2011; Moody et al., 2007).

OPEN IN VIEWER

Figure 5.

Face muscle responses in z scores for fear and anger facial stimuli.

Of interest were the responses from arm muscles to emotional face stimuli. For arm muscles, the ANOVA showed a main effect of time window, F(1, 44) = 6.90, p = .01, ω p 2 = .11. Arm muscle activation generally increased over time (μ 500–1,800 ms = .03, SE = .01 vs. μ 1,800–3,000 ms = .05, SE = .01), perhaps because larger muscles take longer to activate. Importantly, this main effect was modified by the Emotion × Muscle Group × Time Window interaction, F(1, 44) = 5.49, p = .02, ω p 2 = .09]. Consistent with our hypothesis, participants’ arm muscles generally demonstrated fist-making, flexor and extensor activity, when observing angry faces and hand-raising, extensor activity, when observing fearful faces, although stronger emotion-congruent arm muscle responses were found for the later time window (Figure 6). There was small effect for fist-making activity that did not reach the .05 level of significance (µ flexor 500–1,800 ms vs. 1,800–3,000 dif = −.03 ms, SE = .02; F(1, 44) = 2.72, p = .10, ω p 2 = .04) increased across time periods, but hand-raising activity remained constant (µ extensor 500–1,800 ms vs. 1,800–3,000 ms dif = −.01 ms, SE = .02, F(1, 44) < 1 ms, p = .53, ω p 2 = 0). In contrast, there was stronger evidence that for fear faces, the hand-raising activity (µ extensor 500–1,800 ms vs. 1,800–3,000 dif = −.05 ms, SE = .03, F(1, 44) = 9.42, p = .004, ω p 2 = .15) increased over time, but fist-making activity remained constant (µ flexor 500–1,800 ms vs. 1,800–3,000 ms dif = −.01 ms, SE = .03; F(1, 44) = 1, p = .65, ω p 2 = 0). In sum, emotional face stimuli evoked congruent muscular responses in both the face and the arm; however, the time course of this activation varied by emotion.

OPEN IN VIEWER

Figure 6.

Arm muscles (extensor and flexor) responses in z scores to anger and fear facial stimuli.

Discussion

Several theories (Winkielman et al., 2015; Wood et al., 2016) have posited a whole-body response to facial expressions of emotion assuming that we neurally simulate observed emotions. The data presented here are the first evidence, to our knowledge, demonstrating that participants showed the typical pattern of facial mimicry to emotional faces, while at the same time demonstrated reactions over arm muscles. This occurs despite there being no other emotional information presented (e.g., body postures or vocal information). This expands other research that has found cross-modal mimicry (e.g., Hawk et al., 2012). Here, we replicated the finding that participants show greater activity over the corrugator muscle in response to angry faces and greater activity over the frontalis muscle in response to fear expressions. Additionally, we also found, for the first time, that participants responded with muscle reactions over arm muscles in response to facial expressions of emotions. Participants’ arm muscles demonstrated activity consistent with fist-making when observing angry faces and hand-raising when observing fearful faces. Although the pattern of facial response is consistent with a long line of research showing facial mimicry of emotional expressions (e.g., Dimberg, 1982), the reactions of arm muscles is novel and expands our understanding of the nature of rapid matching of emotional expressions. It appears that observed facial emotions may cause sufficient activity in nonfacial motor areas such that subtle muscle reactions in the body can be detected using EMG.

We also found that muscle responses in face and body muscles unfold over the course of several seconds. Specifically, the results presented here found significant changes in much later time windows than some other studies (e.g., Moody et al., 2007). While it is unclear whether the responses at these later time windows are the result of the same underlying mechanism as mimicry found within the first second following stimulus onset, this still demonstrates that EMG can be successfully used to study mimicry and other subtle muscle responses over several seconds. A number of possibilities exist for why we find these later responses. For instance, body muscles may take longer to activate given that they are further away from the brain than face muscles, or neural simulation may be longer lived process than earlier mimicry research recognized. In any case, additional research is needed to further explore the temporal emergence of mimicry and how body responses differ compared to face muscles.

This finding is important as it is consistent with simulation and embodiment theories of emotions (Niedenthal, 2007; Wood et al., 2016). Here, participants may have viewed emotional information from faces, simulated those emotional states neurally, and that activity led to both facial and body mimicry. If true, this suggests that whole-body emotional simulation occurs, even if only information limited to the face is provided. This is noteworthy because these responses seem to have occurred without the observers processing the images deeply, suggesting that the process of mimicry engages multiple neural systems. What is not clear from these data is which mechanisms are being activated to lead to this response. For instance, it is plausible that participants’ own emotional systems were engaged to process these stimuli or that more conceptual understanding of the emotion led to the response. If true, this phenomenon might be better characterized as rapid emotional microexpression rather than mimicry. Exploring how this process relates to theorized functions of mimicry might provide a framework through which to better understand the underlying mechanisms (Hawk & Fischer, 2016). In any event, additional research will be needed to determine which neural systems are being activated and to better understand how these processes interact to lead to mimicry.

This finding is also important because there have been inconsistent findings regarding the matching of body responses. Berger and Hadley (1975) demonstrated bodily mimicry to arm wrestling; however, Moody and McIntosh (2011) were not able to replicate that finding when presenting videos of only the arms. Importantly, the stimuli in these two studies may have differed in how much information was presented and the degree to which the stimuli were simulated by the observer. Berger & Hadley (1975) seem to have created a more elaborate situation in which the stimuli were of “arm wrestling bouts” (p. 265, Berger & Hadley, 1975), whereas Moody & McIntosh (2011) showed short videos of arm wrestling, in which the movements were overaccentuated, with no faces shown and no context for the actions. In the present study, we found that bodily responses occur when only facial information is presented. This suggests that for mimicry to occur, the viewer may need to be able to have sufficient information to simulate the observed action, whether it is emotional or not. That is, Berger and Hadley may have successfully generated body mimicry because they set up a situation in which the participants were invested in the “bouts.” This may have led to motivation to closely observe and simulate what was being observed, perhaps to understand what was happening or to empathize with the actors. Merely showing actions without context as to why they are occurring may not be sufficient to lead to neural simulation. In the present study, emotions may be motivating, in and of themselves, and led to neural simulation of what was observed. Again, additional research is needed to determine whether this is the case.

These results also have implications for our understanding mimicry between two or more individuals in a more general sense. For example, emotional contagion is a well-known phenomenon (Hatfield, Cacioppo, & Rapson, 1993) and some have speculated that it emerges through afferent feedback (McIntosh, 1996). While this mechanism may exist, it is also clear that merely observing a facial expression of emotion engages more mechanisms than we were previously aware. Indeed, recent research (Hess & Fischer, 2013) has suggested that mimicry is likely more than a mere behavioral matching phenomenon. Our data also suggest that more diverse mechanisms seem to be involved in these rapid responses than just a behavioral correspondence. Therefore, researchers should to be cautious when making claims about what mimicry is and does (e.g., that it causes shared emotions) before considering the underlying processes that produce it.

Another important consideration when studying underlying mechanisms is their timing relative to the onset of the stimulus. Previous research has often found effects well before the end of the first second poststimulus onset (Moody et al., 2007), while others focus on effects that occur much later and that are more obvious (Chartrand & Bargh, 1999). Here, we chose to explore reactions that occur after a relatively long time period given that face and body muscles may have differences in their physiology. For example, signals may simply take longer to reach the arms and arm muscles may take longer for EMG signals to reach their maximum strength. It is possible that we are capturing distinct mechanisms from those that occur earlier. Indeed, previous research has found differences in more rapid, automatic levels of cognitive processing relative to more controlled levels (Sonnby-Borgstrom & Jonsson, 2004). Relatedly, exploring the degree to which face and body reactions occur as unique phenomena or are part of a more generalized reaction could help understanding of mimicry. We think it is likely that they are both the result of an underlying simulation process; however, ultimately, a better understanding of the mechanisms that produce mimicry and how they interact over time will provide insight into typical and atypical social functioning, particularly in these rapid social interactions.

Finally, this finding may provide for a better understanding of why mimicry is not observed in some special populations. For example, mimicry is not found as reliably in those with ASD (Beall et al., 2008; McIntosh, Reichmann-Decker, Winkielman, & Wilbarger, 2006; Oberman et al., 2009). Weak central coherence, the tendency to focus on details and parts of stimuli rather than global form and meaning, has been proposed to be why those with ASD struggle with some social tasks (Happé, 1997; Happé & Frith, 2006). If the mimicry and body responses found here are the results of simulated global meaning of observed emotion, it is possible that mimicry deficits in those with ASD emerge from limited neural simulation. If so, we would expect that body reactions would be absent in those with ASD. Further, to the extent that ASD interventions improve global processing, simulation, or central coherence, mimicry could provide a marker for improved social functioning and an indication that central coherence is improved. However, additional research would be needed to better understand the nature of the mimicry deficit observed in those with ASD and to determine whether it is sensitive to intervention.

Conclusion

Mimicry to emotional expressions appears to spread to other nonface regions of the body. This suggests that mimicry may result from neural simulation of those emotions. Future research is needed to further understand how simulation leads to phenomena like mimicry and how it relates to other phenomena such as emotional contagion.