Broadening the Perspective on Emotional Contagion and Emotional Mimicry: The Correction Hypothesis

The Automaticity of Emotional Contagion

Over the years, the term automatic has aroused some controversy because there is no consensus on the characteristics that define automaticity (Bargh, 1994; Moors & De Houwer, 2006). However, the majority of scientists agree that there are four criteria for automatic processing: (a) the process is elicited unintentionally, (b) it requires only minimal cognitive resources, (c) it cannot be stopped voluntarily, and (d) it occurs without conscious awareness. In contrast, controlled processing (a) is initiated intentionally, (b) requires considerable cognitive resources, (3) can be stopped voluntarily at any moment, and (4) occurs with conscious awareness (Moors & De Houwer, 2006).

Several studies confirm that emotional contagion, conceptualized as a two-stage process, is largely automatic because both emotional mimicry and afferent feedback fulfill the criteria of automaticity. For instance, emotional mimicry occurs even when emotional displays cannot be perceived consciously—for example, when the sender’s face is presented subliminally (Dimberg, Thunberg, & Elmehed, 2000) or presented to the blind visual field of patients who lost their vision because of cortical blindness (Tamietto, Castelli, Vighetti, Perozzo, Geminiani, Weiskrantz, & de Gelder, 2009). Moreover, people are unable to refrain from mimicking emotional displays of others even when instructed to do so (Dimberg, Thunberg, & Grunedal, 2002). These findings are often explained by the perception-action link—the fact that mimicry is based on the mirror-neuron system that provides a direct automatic link between perception (i.e., the sender’s emotional display) and action (i.e., the receiver’s imitation of this display; Catmur, Walsh, & Heyes, 2009). Much evidence supports the notion that the system fires independently of the receiver’s intentions because its capacity to match observed and executed actions depends on automatically triggered learned associations (for a review, see Heyes, 2011). In a similar vein, studies on afferent feedback indicate that people do not control the influence of their expressive behavior on emotional experience. For instance, they do not realize that they adopt various emotional expressions by simply activating their facial muscles (e.g., holding a pen in their mouths or moving their brow region so that two golf tees attached to that region could touch each other) and do not notice that these adopted expressions (e.g., smiles or frowns, respectively) affect their feelings (Larsen, Kasimatis, & Frey, 1992; Strack, Martin, & Stepper, 1988).

In contrast, social appraisal may occur at a less automatic level of processing (Bruder et al., 2014). For instance, in some unusual or uncertain situations, the receiver may intentionally seek emotional cues from others to grasp the emotional meaning of these situations (Bruder et al., 2014; Manstead & Fischer, 2001). This suggests that under certain conditions, social appraisal may act as a deliberate process in which the receiver draws conclusions about the evaluative implications of the sender’s emotional expression (Parkinson, 2011). Yet when demand for detailed processing is absent, social appraisal may be carried out with much less cognitive effort (Bruder et al., 2014; Parkinson, 2011). This is supported by findings indicating that people are often not aware that someone else’s emotional expressions affect their own evaluations of objects or situations (for a review, see Parkinson, 2011). For instance, Bayliss, Frischen, Fenske, and Tipper (2007) demonstrated that objects paired with disgusted faces were evaluated as less likeable than objects paired with smiling faces, but participants did not realize that their ratings were in any way related to the facial expressions to which they had been exposed. Even initially controlled social-appraisal processes, similar to other psychological actions that are performed repeatedly, may become automated and thus may be performed with little or no effort (Mussweiler, 2007).

Overall, these findings suggest that although some mechanisms leading to emotional convergence between the sender and the receiver may occur at a fairly controlled level of processing, this is an exception rather than a rule. However, in the light of the latest research, this does not mean that emotional contagion or its mechanisms are purely reflexlike or inflexible. Evidence is now accumulating that these processes are highly selective; that is, the receiver’s tendency to react with a congruent emotional expression or feeling to the sender’s emotional display may be moderated by various contextual factors such as the sender’s similarity or dissimilarity, familiarity or unfamiliarity, cooperative or competitive intent, or in-group or out-group membership (Epstude & Mussweiler, 2009; Van der Schalk et al., 2011; Weisbuch & Ambady, 2008; Weyers, Mühlberger, Kund, Hess, & Pauli, 2009; Wróbel & Królewiak, 2017). For instance, a smiling face of a likable sender may evoke a congruent emotional expression or feeling in the receiver, whereas a smiling face of an unlikeable sender may not (Likowski, Mühlberger, Seibt, Pauli, & Weyers, 2008; Wróbel & Królewiak, 2017). Taken together, these findings demonstrate that in affiliative social contexts, the probability of congruent emotional reactions increases, whereas in nonaffiliative social contexts, the process may be attenuated, blocked, or may even lead to an incongruent emotional reaction. This, according to Hess and Fischer (2013; Fischer & Hess, 2017), indicates that people do not mimic emotions if they do not have at least a minimal reason to affiliate with the senders. Apparently, emotions are social signals, and the reactions they evoke—in contrast to infections caused by germs—depend crucially on the social meaning of these emotions (Fischer & Hess, 2017).

So far, the discussion on the modulating role of factors that signal the affiliative versus nonaffiliative intent of the sender has been focused on emotional mimicry and has neglected the role of these factors in social appraisal. This may be attributed to the fact that the main goal of social appraisal is to correctly interpret the emotional meaning of the situation rather than affiliate with the sender. At the same time, however, such accuracy goals—like affiliative ones—may motivate the receiver to pay attention to the sender’s affiliative and nonaffiliative characteristics because these characteristics indicate whether the sender can be trusted. For instance, Bruder et al. (2014) argue that, when appraising the situation, receivers seek for emotional cues that are reliable (i.e., provided by not only competent senders but also trustworthy senders). Although competence is manifested through social status, trustworthiness (often referred to as warmth) is determined by social characteristics that communicate affiliative intent (Abele & Wojciszke, 2014).

From a common-sense perspective, the moderating role of social factors seems quite obvious when emotional contagion involves social appraisal, especially when the process occurs at a controlled level. In this case, the receiver may deliberately analyze the sender’s social characteristics. Yet when social appraisal operates automatically, the receiver may also take control over it once he or she realizes that the sender’s expression affects his or her interpretation of the situation and thus is important to his or her accuracy goals. As a result, the receiver may start paying conscious attention to the sender’s social characteristics to decide whether the sender’s emotional display is reliable. Driving a car is a good analogy: Although an experienced driver does not consciously monitor the process, he or she may start controlling it when traffic becomes difficult (Wheatley & Wegner, 2001). What seems more counterintuitive is how the processes operating completely outside of awareness (such as emotional mimicry) can be controlled by the receiver’s goals or attitudes toward the sender. Here, the receiver cannot consciously realize that he or she imitates the sender’s expression, which eliminates the possibility of conscious control. We think that such circumstances leave room for a process called automatic correction. This idea, taken from priming literature (Glaser & Banaji, 1999), may explain how emotional mimicry may be controlled with no awareness or intent. We also propose that similar yet more controlled corrective processes (Wegener & Petty, 1995, 1997) may be involved in emotional contagion triggered by social appraisal.

The Top-Down Control of Emotional Contagion and Mimicry by Social Signals: The Correction Hypothesis

Although the idea that people mimic others’ emotional expressions or “catch” their feelings only in affiliative social contexts has received strong empirical support (for a review, see Hess & Fischer, 2013), some research seems to contradict it. Specifically, studies show that people typically react to pictures or videos of emotional faces with congruent emotional expressions or feelings even when not given any cues about the sender’s intent (e.g., no information about the photographed or video-taped people is provided; Hühnel, Fölster, Werheid, & Hess, 2014; Lundqvist & Dimberg, 1995; Rymarczyk, Żurawski, Jankowiak-Siuda, & Szatkowska, 2016; Sato, Fujimura, & Suzuki, 2008). Of course, it is possible that receivers may assume that such acontextual settings are affiliative by default. This idea is in line with results from studies demonstrating that when people are not provided with any explicit information, they trust complete strangers or naturally presume that others think the same way they do (Dunning, Anderson, Schlösser, Ehlebracht, & Fetchenhauer, 2014; Ross, Greene, & House, 1977). However, this explanation seems unlikely when a photographed or video-recorded sender expresses an intrinsically nonaffiliative emotion (e.g., anger or contempt; Miceli & Castelfranchi, 2017). Here, the lack of accompanying explicit information may suggest that the sender’s intent is negative (e.g., the receiver may think that a frown or a sneer is directed at him or her), and thus no congruent reaction should occur.

Hess and Fischer (2013, 2014) explain these seemingly paradoxical findings by suggesting that reactions to acontextual emotional displays may represent not emotional mimicry or contagion but rather reactive responses to the content of the images (similar to the reactions that people may demonstrate in response to, for instance, smiling babies, cute puppies, or intimidating objects). On the basis of this differentiation, a congruent emotional expression to an acontextual frown or sneer may be qualified as a reactive response (e.g., reacting to the sender’s anger or contempt) rather than emotional mimicry (i.e., imitating the sender’s anger or contempt). The situation may change when the receiver is exposed to an intrinsically affiliative emotion. For instance, smiles typically signal positive intent (Hess, Adams, & Kleck, 2005) and are thus frequently imitated even when no contextual information is provided. Hence, congruent emotional reactions to smiles usually represent emotional mimicry (Hess & Fischer, 2013). It should be noted, though, that reactive responses and emotional mimicry are often very difficult to differentiate in acontextual settings because the receiver may rely on additional implicit cues to interpret the sender’s intent (e.g., gender, ethnicity, or age may be indicative of the sender’s group membership).

Moreover, contextual cues may change the social meaning of intrinsically affiliative and nonaffiliative expressions (Hess & Fischer, 2013). For instance, although a smile in itself is an affiliative social signal, it takes on a nonaffiliative meaning when it is shown on the face of a rival or a unlikable person (e.g., Likowski et al., 2008; Weyers et al., 2009; Wróbel & Królewiak, 2017). In a similar vein, displays of anger or contempt cease to be nonaffiliative social signals once the receiver realizes that the sender is directing these emotions at a person the receiver does not like. This explains why in some contexts (e.g., when the sender’s eyes are directed at an unlikeable person), displays of intrinsically nonaffiliative emotions are mimicked (Bourgeois & Hess, 2008). Taken together, these assumptions, made within the contextual model of emotional mimicry (Hess & Fischer, 2013; Fischer & Hess, 2017), indicate that emotional mimicry is triggered not by the observed expression itself but by the overall interpretation of this expression in a specific social context. Note that this interpretation is based on prior knowledge of social interactions, which suggests that emotional mimicry is not a blind, bottom-up mechanism but a phenomenon sophisticatedly controlled by top-down processes (Fischer & Hess, 2017). Such processes, according to Hess and Fischer (2014), probably depend on the activation of affiliation goals—when the goal to affiliate is activated (which happens when the relationship is perceived as at least neutral), the receiver mimics the sender’s expression, whereas when it is not activated (when the relationship is perceived as negative), no emotional mimicry occurs.

A similar idea can be found in the social top-down response-modulation (STORM) theory (Wang & Hamilton, 2012),which holds that the top-down control of emotional mimicry by social signals is implemented by a mentalizing system located in the medial prefrontal cortex. The system constantly supervises the activity of mirror neurons and, after evaluating the social context, inhibits this activity any time mimicry is socially inappropriate (e.g., in nonaffiliative social contexts). This supervising mechanism thus enables the receiver to strategically control mimicry for his or her social advantage. This assumption is supported by studies indicating that mentalizing abilities (i.e., the abilities to understand mental processes of oneself and others) are strongly related to the inhibition of mimicry (for a review, see Wang & Hamilton, 2012).

The STORM account focuses on mirror neurons because it refers not only to emotional but also to behavioral mimicry (i.e., the imitation of movements, gestures, or postures). Wang and Hamilton (2012) assume that mimicry of emotionally meaningful actions (i.e., emotional expressions) and mimicry of nonemotional actions are controlled by social factors in a similar way. By highlighting this similarity, the STORM account emphasizes that bottom-up mechanisms responsible for mimicry are activated regardless of the meaning of the observed action. This activation is possible because the mere perception of an action (e.g., the sender’s expression) automatically triggers the tendency to mimic this action (perception-action link). The tendency, if not inhibited by top-down processes, leads to a congruent response. However, each time the receiver realizes that mimicry does not serve his or her social goals, bottom-up mechanisms linking perception and action are inhibited. In other words, a mentalizing system corrects the activity of mirror neurons by blocking their automatic response.

It is noteworthy that the STORM model, by treating emotional mimicry as a form of behavioral mimicry, does not assume that the intrinsic social meaning of emotional expressions is irrelevant to the receiver’s evaluation of the sender. In line with the contextual model of emotional mimicry, the STORM theory holds that receivers use a variety of explicit and implicit social clues (including the intrinsic meaning of the expressed emotion) to evaluate the sender’s intent. The focus on the perception-action link, however, illuminates the fact that social cues do not influence the tendency to mimic per se (because this tendency is activated by the mere perceptual input); rather, they influence top-down processes controlling this tendency (because the mentalizing system is sensitive to social cues).

Overall, both the contextual model of emotional mimicry and the STORM theory highlight the idea of the top-down control of emotional mimicry by integrative evaluation of the current social context. Both models share the basic assumption that mimicry serves the receiver’s affiliative goals and is thus corrected (i.e., blocked or attenuated) each time the receiver discovers that these goals cannot be achieved. We propose that, consistent with this assumption, social modulation of emotional mimicry (and emotional contagion based on emotional mimicry) could be conceptualized as correction of an automatic action: The receiver has an automatic tendency to imitate the sender’s emotional display activated by perceptual input but, when the social context suggests that imitation is inappropriate or undesirable (i.e., does not serve the receiver’s affiliative goals), he or she attempts to counteract the influence of this display on his or her emotional expression or feelings. In this respect, social modulation of emotional mimicry resembles correction processes observed in priming research (Chien, Wegener, Petty, & Hsiao, 2014; Glaser & Banaji, 1999; Wegener & Petty, 1995, 1997).

Can similar goal-driven corrections be involved in less automatic mechanisms of emotional contagion, such as social appraisal? Here, no mirror neurons are involved and thus no initial automatic tendency to converge with the sender occurs. Instead, the receiver’s congruent or incongruent reactions (potentially requiring correction) depend on the initial appraisal. For instance, to expand on the aforementioned example of airplane turbulence (Bruder et al., 2014), consider a receiver who sits next to an anxious passenger. The receiver may also appraise the situation as dangerous and start feeling anxious, but once he or she realizes that this passenger is an inexperienced flyer, this initial tendency to react congruently may be blocked, and an incongruent reaction such as relief may occur. The corrective process may also operate in an opposite direction. For instance, if the receiver’s initial appraisal suggests that the anxious passenger is an inexperienced flyer (e.g., because the passenger looks relatively young), the receiver will not display the tendency to “catch” the passenger’s anxiety. This response, however, may be corrected once the receiver discovers that the sender has been flying regularly for the past few years, and thus information gained from his or her emotional expression is reliable (i.e., it serves the receiver’s accuracy goals).

Both examples suggest that, in contrast to emotional mimicry, emotional contagion involving social appraisal may be modulated by social factors at two different levels. First, social factors may shape an initial appraisal (and thus the direction of the receivers’ initial emotional reaction); second, social factors may trigger a corrective process that may reverse the direction of the initial reaction (which happens when the receiver realizes that the initial tendency to converge or diverge with the sender’s emotion is incompatible with his or her accuracy goals). We think that the mechanisms activated at both levels also resemble processes observed in priming research but—in contrast to top-down social control of emotional mimicry—may be less automatic.

Priming Effects as Analogues of Congruent Versus Incongruent Reactions to the Sender’s Emotional Expressions

Decades ago, priming researchers established that exposure to one stimulus (a prime) affects processing of a subsequently presented stimulus (a target). They also demonstrated that this influence can take two forms: assimilation or contrast (see, Bless & Burger, 2016; DeCoster & Claypool, 2004, for reviews). Assimilation effects occur when judgments of the target (or reactions to it) are biased in the direction of the prime (e.g., people primed with concepts associated with hostility perceive others’ behavior as more hostile; Srull & Wyer, 1979), whereas contrast effects occur when judgments of the target (or reactions to it) are biased away from the prime (e.g., people primed with images of highly attractive individuals rate others as less attractive; Kenrick & Gutierres, 1980). These two biasing effects have been observed in a variety of judgmental domains, including not only the assessments of others but also self-evaluations (Brown, Novick, Lord, & Richards, 1992; Mussweiler, Rüter, & Epstude, 2004).

Over the years, it has become clear that the direction of priming depends on specific conditions, and psychologists became interested in studying variables determining assimilation versus contrast as well as the mechanisms producing these effects (for a review, see Bless & Burger, 2016). Initially, it was thought that assimilation is likely to occur when participants are not aware of the influence of a prime, whereas contrast is observed when this influence is apparent. This was supported by findings showing that (a) primes presented subliminally (i.e., below the threshold of consciousness) are more likely to elicit assimilation, whereas primes presented supraliminally (i.e., above this threshold) are more likely to produce contrast (e.g., Bargh & Pietromonaco, 1982); (b) contrast effects are less likely to occur when participants are cognitively distracted and thus they cannot pay enough attention to the prime (Martin, Seta, & Crelia, 1990); and (c) contrast effects are more probable when participants are reminded of the priming episode (Strack, Schwarz, Bless, Kübler, & Wänke, 1993). Subsequent studies, however, challenged this view by demonstrating that participants do not have to be aware of the priming manipulation for contrast effects to occur (e.g., Glaser & Banaji, 1999; Mussweiler et al., 2004). These observations shifted researchers’ attention from participants’ awareness to other variables such as the features of the prime or participants’ processing style. Not only did this shift result in the identification of a wide array of factors for determining assimilation versus contrast, but it also led to systematizing these factors within broader theoretical frameworks (for a review, see Biernat & Eidelman, 2007).

In general, these frameworks differ considerably in their basic assumptions, but many theorists agree that assimilation is more likely to occur when the prime shares some features with the target, whereas contrast is more likely to occur when there is no overlap between these features (Chien, Wegener, Hsiao, & Petty, 2010). This may be explained by the fact that the overlap or the lack thereof may prompt different processing styles (Chien et al., 2010). For instance, Mussweiler (2003) proposes that the emergence of assimilation versus contrasts depends on whether individuals focus on similarities or dissimilarities between the prime and the target (a process called similarity vs. dissimilarity testing). Specifically, when making a judgment, an individual initially engages in a brief general assessment of similarity between the target and the prime. If this initial assessment suggests that the prime shares some features with the target, an individual further tests for similarities and assimilation occurs. However, if the initial assessment suggests that the prime does not share any features with the target, an individual searches for differences, which causes contrast.

Strong support for the role of shared features comes from studies examining how the characteristics of other people affect self-evaluations. For instance, Brown et al. (1992) demonstrated that participants’ evaluations of their own physical attractiveness were assimilated toward or contrasted away from the physical attractiveness of another person depending on whether participants believed that this person shared their attitudes or birthday. Given that similarity is a strong correlate of psychological proximity (Wrzus, 2008), these findings suggest that, when self-evaluative judgments are made, the direction of priming may depend on a sense of communality or psychological closeness between the prime and the target (Brown et al., 1992; Colpaert, Muller, Fayant, & Butera, 2015). Supporting this notion, other studies demonstrate that psychologically proximal primes (e.g., members of an in-group or collaborators) are more likely to cause assimilation, whereas psychologically distant primes (e.g., members of an out-group or competitors) are more likely to elicit contrast (Colpaert et al., 2015; Ledgerwood & Chaiken, 2007; Mussweiler & Bodenhausen, 2002). These effects also extend to physical proximity. For instance, walking toward another person leads to self-evaluative assimilation (i.e., assimilating one’s own physical attractiveness toward physical attractiveness of the approached person), whereas moving away from this person leads to self-evaluative contrast (Fayant, Muller, Nurra, Alexopoulos, & Palluel-Germain, 2011). This may be explained by the fact that approach behavior is associated with psychological closeness, whereas avoidance behavior is associated with psychological distance (Nussinson, Häfner, Seibt, Strack, & Trope, 2012). Taken together, these results indicate that psychological closeness may determine the direction in which other people’s characteristics influence self-evaluations.

Because psychological proximity is indicative of affiliative intentions (Hammock & Young, 2006; Marinović, Wahl, & Träuble, 2017), the aforementioned findings highlight the analogy between priming effects and congruent or incongruent affective reactions to the sender’s emotional displays. The variables that promote self-evaluative assimilation (e.g., similarity, in-group membership, or cooperation) have also been found to foster congruent reactions to the sender’s expressions, whereas the variables that promote self-evaluative contrast (e.g., dissimilarity, out-group membership, or competition) have also been found to hinder congruent reactions or foster incongruent ones (Weisbuch & Ambady, 2008; Weyers et al., 2009; Wróbel & Królewiak, 2017). This suggests that emotional mimicry or contagion might be treated as a special type of priming in which the receiver’s expression or feeling is either assimilated with or contrasted away from the sender’s emotional expression. If this is indeed the case, the mechanisms underlying assimilation versus contrast (e.g., similarity vs. dissimilarity testing) should also trigger congruent versus incongruent affective reactions to the sender’s emotional displays.

To our knowledge, so far only one line of research has addressed this possibility. In a series of three studies, Epstude and Mussweiler (2009) investigated whether the two types of comparison processes (i.e., similarity and dissimilarity testing) contribute to the direction of affect induction in response to facial and vocal expressions of happiness and sadness. Results showed that participants in the similarity-testing condition assimilated their affect to that of the senders, whereas participants in the dissimilarity-testing condition contrasted their affect away from that of the senders. These findings demonstrate that cross-talk between research on priming and emotional contagion or mimicry is possible. We believe that this cross-talk is essential because it may help explain the role of social factors in emotional contagion/mimicry. What may advance this explanation is the concept of correction originating from priming research. In what follows, we briefly describe this concept and elucidate how it might be applied to emotional contagion mimicry settings.

Correction in Priming and Emotional Contagion/Mimicry

Although people tend to believe that how they see the world reflects objective reality, the research reviewed above demonstrates that this is often not the case; people’s judgments about objects, other people, or even themselves are frequently biased. At the same time, they want to be accurate and thus avoid having their evaluations distorted by inappropriate or undesirable influence. Therefore, although assimilation and contrast are fairly automatic in nature, people sometimes suspect the potential bias and try to correct for its influence. As a result, their initial judgments may be corrected or overcorrected, which may result in the transition from assimilation to contrast or vice versa (Wegener & Petty, 1995, 1997).

The direction of correction has been a matter of discussion for many years. Some theorists proposed that correction should lead to contrast because assimilation occurs spontaneously (by default), whereas contrast requires active attempts to counteract the unwanted biasing influence of the prime (Martin et al., 1990; Schwarz & Bless, 1992). Yet, as noted earlier, when the prime does not share any features with the target, contrast may also occur by default (Chien et al., 2010; DeCoster & Claypool, 2004). In this case, the attempts to avoid the bias should lead to assimilation. This possibility has been widely discussed by the creators of the flexible correction model (FCM; Chien et al., 2014; Wegener & Petty, 1995, 1997), who proposed that assimilation and contrast may occur either by default (uncorrected effects) or when people attempt to avoid the biasing influence of the prime (corrected effects). Put differently, the direction of correction is the opposite of the initial (default) influence. For instance, in Wegener and Petty’s (1995, 1997) studies, participants who believed that the prime would lead to assimilation corrected away from it, whereas those who expected contrast corrected toward it. Note that many theories of correction, including the FCM, assume that default (uncorrected) effects occur quite automatically and thus require little conscious attention, whereas correction results from controlled processing and thus takes place when people become aware of the potential biasing influence of the prime (Chien et al., 2014; DeCoster & Claypool, 2004; Strack, 1992; Wegener & Petty, 1995, 1997). As a result, if an individual does not have enough cognitive capacity or motivation to search for potential sources of bias, his or her judgments should reflect a default reaction to the target (Wegener & Petty, 1997).

We believe that the idea of such controlled correction is applicable to emotional contagion triggered by social appraisal. Getting back to the example of airplane turbulence, the initial appraisal of the anxious passenger as an experienced or inexperienced flyer may result in a tendency to converge with or diverge from the passenger’s affective state (uncorrected assimilation or contrast, respectively), but this tendency may be corrected (i.e., attenuated or reversed) once the receiver discovers that the initial tendency is incompatible with his or her accuracy goals. As mentioned above, even when the default reaction occurs automatically, the receiver may become aware that the sender’s expression is the source of his or her reaction. This may result in controlled corrective attempts to counteract the biasing influence of the sender’s expression on the receiver’s interpretation of the situation and, consequently, on his or her feelings about this situation. Of course, as with social appraisal itself (Bruder et al., 2014) or corrective processes in priming (Chien et al., 2014), corrective attempts in social appraisal might also become more automated with time and practice.

When it comes to emotional mimicry, the concept of controlled correction loses its applicability. Here, deliberate corrective processes cannot be involved because people are not only unaware that they imitate the sender’s expression but also that they cannot consciously realize that they are under this biasing influence. Studies on automatic evaluations, however, demonstrate that—under conditions in which controlled processing is precluded (e.g., the primes are presented subliminally or people are not aware that they evaluate the targets)—people may engage in automatic correction; that is, a process whereby an individual corrects for assimilative or contrastive influences with no engagement of control (Espinoza, 2012; Glaser & Banaji, 1999; Maddux, Barden, Brewer, & Petty, 2005). For instance, in Glaser and Banaji’s (1999) studies, participants were not aware that they made evaluations because the authors did not ask them explicitly to rate the targets. Instead, they used an implicit procedure based on how quickly participants pronounced the target words that were evaluatively congruent or incongruent ² with the primes. The results showed that unconscious vigilance for the biasing influence of the prime led to corrective processes (i.e., faster responses when the prime and the target were evaluatively incongruent) that operated without conscious awareness or intention. This, according to Glaser and Banaji (1999), suggests that even under conditions in which the biasing influence cannot be perceived consciously, individuals may be motivated to judge the targets accurately and thus actively yet unconsciously correct their initial biased judgments. This provides clear evidence for compensatory automaticity, defined as “strategic yet nonconscious compensations for unwanted thoughts, feelings, or behaviors” (Glaser & Kihlstrom, 2005, p. 171).

Although at first this idea seems counterintuitive because automaticity is often equated with the lack of intent, some intentions (e.g., the aforementioned goal to judge the target accurately) may operate preconsciously and may thus be protected and promoted by automatic compensatory processes (Glaser & Kihlstrom, 2005; Moskowitz, 2001). Given that correction may be driven by goals other than accuracy (Chien et al., 2014), we believe that affiliation goals may also modulate emotional mimicry (and emotional contagion involving this mechanism) by automatic correction. Such goals, once automatically activated by integrative evaluation of social context, may exert top-down control on the receiver’s reactions to the sender’s emotional displays by correcting the activity of mirror neurons. In this case, the initial reaction to the sender’s emotional display is always assimilative (because the mere perception of this display, if not inhibited, always leads to congruent reactions); consequently, we think that corrective mechanisms involved in emotional mimicry may result in the transition from assimilation to contrast but not vice versa.

In summary, regardless of whether emotional contagion is triggered by social appraisal or emotional mimicry, the process—similar to priming—starts with a default reaction to the sender’s expression that may be either assimilative (in social appraisal and emotional mimicry) or contrastive (only in social appraisal). When the receiver realizes that this initial, typically automatic reaction is incompatible with his or her goals, he or she may engage in automatic or controlled corrective processes that may block, attenuate, or reverse this reaction. The notion that emotional contagion and its mechanisms may be corrected on either an automatic or a controlled level illuminates the role of impulsive and reflective processing in the phenomena responsible for emotional convergence between people, thereby providing a fresh view of these processes. This view implies that the modulating role of social factors in emotional mimicry and contagion may be analyzed from the perspective of dual-process theories.

A Dual-Process Perspective on the Correction Hypothesis

The correction of an automatic action caused by top-down mental processes has been postulated in a number of dual-process theories (Epstein, 2003; Kahneman, 2003; Petty & Brinol, 2014; Smith & DeCoster, 2000). What these theories have in common is the explanation of mental processes underlying social judgments and behavior as the interplay of two modes: impulsive (associative) and reflective (rule-based; Smith & DeCoster, 2000; Strack & Deutsch, 2014). Impulsive processing is based on associations that are built through previous experiences and structured by similarity and contiguity (i.e., an associative network). For instance, the sight of a delicious-looking apple pie activates one node in the network (e.g., tasty) and, as a result, activation spreads to the related, evaluatively congruent nodes (e.g., sweet, luscious, pleasant). The activated nodes may further elicit associated behavioral schemata (e.g., the willingness to eat the pie). Although the results of such processing are subjects of awareness, the processing itself requires little cognitive capacity and occurs rapidly, automatically, and preconsciously. In contrast, reflective processing operates in a controlled fashion and is based on rules that are structured by language and logic (i.e., propositional knowledge). An individual is aware not only of the outcomes but also of the steps of such processing. Thus, reflective processes are slow, require more attentional resources, and occur only when an individual has sufficient cognitive capacity and motivation.

Most theorists state that the two modes interact at various stages of processing because the majority of real-life situations involve both impulsive and reflective actions (Imbir, 2016; Kahneman, 2011; Strack & Deutsch, 2014). For instance, the reflective-impulsive model (RIM; Strack & Deutsch, 2004, 2014) holds that both types of processing may either cooperate or compete when influencing social behavior or judgments. Cooperation occurs when both modes activate compatible schemata (e.g., convergent goals), whereas competition takes place when the schemata activated by the modes are incompatible. For instance, the sight of a tasty yet unhealthy pie may activate an urge to eat it (as the impulsive mode reacts to the visible features of an object), but the reflective mode may inhibit the urge by activating the knowledge about the negative consequences of unhealthy diet. Put differently, reflective processes may exert top-down control on impulsive processes and correct impulsively activated actions (for a similar proposition, see Smith & DeCoster, 2000).

The compound outcomes of processing specific to each mode should be observed in most of the phenomena investigated in psychological research (Imbir, 2016; Strack & Deutsch, 2014). Priming is a good example here; as already mentioned, in most theories, default effects are conceptualized as fairly automatic, whereas correction is treated as a result of controlled processing (Chien et al., 2014; Strack, 1992; Wegener & Petty, 1995, 1997). For instance, DeCoster and Claypool (2004) argue that the biasing information may influence judgments of the target through both impulsive and reflective modes: First, the prime influences judgments through impulsive processing, which results in the initial (default) biasing effect; second, if an individual has enough capacity and motivation, reflective mode is activated, which triggers the corrective process. These two steps, as already stated, are similar to corrected and uncorrected reactions observed in emotional contagion based on social appraisal. Importantly, the fact that the initial reaction to the sender’s expression may by counteracted by decisions generated by the reflective mode suggests that controlled correction in social appraisal occurs when the receiver has sufficient capacity and motivation to take these decisions.

The remaining question is whether automatic correction also requires the interplay of both modes and, if so, whether cognitive resources and motivation are needed for this process. In other words, is automatic correction possible from the dual-process point of view? On the one hand, without the activation of reflective processing, no correction of automatic actions should be expected. Rule-based reflective operations require awareness, and therefore the top-down control they exert should be associated with effort and subjectively conscious decision making. In this respect, correction processes should be controlled, effortful, and conscious (Smith & DeCoster, 2000). Not surprisingly, the idea of controlled correction of emotional mimicry can be found in the literature. For instance, in their neurocognitive model of emotional contagion, Prochazkova and Kret (2017) argue that mimicry may be “consciously inhibited and controlled” (p. 104). We believe, however, that conscious control of actions that are not consciously accessible is unlikely and, on the basis of the aforementioned priming studies, we argue that in such circumstances people will rather engage in automatic correction. At the same time, the assumption that the interpretation of the sender’s intent determines the direction of emotional mimicry without conscious awareness or intent does not mean that automatic correction is generated solely by the impulsive mode. The fact that this interpretation is based on prior knowledge of social interactions (Fischer & Hess, 2017; Hess & Fischer, 2013) suggests that reflective mode must be somehow involved in the corrective process. From this perspective, automatic correction may be treated as an automatic outcome of the interaction between reflective and impulsive processes (rather than processing itself; Glaser & Kihlstrom, 2005). For instance, the RIM (Strack & Deutsch, 2004) holds that the reflective mode may influence the impulsive mode not only through deliberate decisions (which, as stated above, are responsible for controlled correction) but also through a processes called propositional categorization, whereby a perceived stimulus may be assigned to a semantic category.

Turning back to the apple-pie example, the pie might be categorized as representing industrial pastries (i.e., food often packed with preservatives, artificial colors, and other unhealthy ingredients). This categorization, performed in the reflective mode, may activate related contents in the associative network (e.g., fattening, toxic, harmful) that are evaluatively incongruent with the aforementioned impulsively activated nodes and thus incompatible with initially elicited behavioral schemata. As a result, the willingness to eat the pie might be automatically blocked. Put briefly, if some information is activated in the reflective mode, it further activates related contents in the associative network, thereby influencing automatic actions without a person’s conscious awareness of it (see also Gawronski & Bodenhausen, 2014).

Data from cognitive psychology support this reasoning by showing that the activation of some ideas in mind (e.g., the sin) may result in behavioral tendencies (e.g., washing hands) without conscious awareness of this relation (Zhong & Liljenquist, 2006). In other words, experimentally activated reflection over morality (i.e., recalling one’s own unethical behavior) can cause changes in the associative network that are automatic in nature. In a similar vein, research on mutual influences between implicit and explicit self-esteem (automatic and controlled self-evaluations, respectively) shows that the activation of information about the self results in changes in explicit self-esteem (measured with self-report) that, in turn, cause changes in implicit self-esteem (assessed with the Implicit Association Test, a measure of automatic associations; Peters & Gawronski, 2011). Because the former is based on propositional knowledge and the latter is based on the associative network, this provides clear evidence for the automatic top-down influence of propositional categorization outcomes on the associative network (see also Gawronski & Bodenhausen, 2014).

Given that the activity of mirror neurons is based on associations (Heyes, 2011), we believe that propositional categorization processes may be responsible for automatic correction observed in emotional mimicry (and emotional contagion involving this mechanism). On the basis of the RIM, we think that exposure to the sender’s expression evokes an automatic tendency to imitate this expression by activating the perception-action link in the associative network (i.e., a link compatible with an affiliation goal), but this tendency may be inhibited by providing the receiver with information that is incompatible with this goal and activates the reflective mode. For instance, propositional knowledge about the sender’s nonaffiliative intent (e.g., the sender is immoral, belongs to an outgroup, has competitive intent, or is angry with the receiver) activates corresponding nodes in the associative network (e.g., wickedness, disliking, rivalry), which in turn block the automatic perception-action link. As a consequence, the receiver’s congruent reactions may be inhibited.

Summing up, the reflective mode may influence the impulsive mode via deliberate decisions that are involved in controlled correction and propositional categorization processes that are engaged in automatic correction. This suggests that both types of correction should occur only when the reflective mode is activated. One of the implications of this explanation is that even automatic correction requires at least minimal capacity or effort. Accordingly, if insufficient attention is given to the information about the sender, the influence of propositional knowledge on emotional mimicry or contagion may disappear. Two recent studies provide useful examples here.

The first of these studies investigated the role of the sender’s characteristics in emotional contagion that goes beyond dyads. The study demonstrated that people who watched a video recording of a happy or sad man were able to “catch” his emotional state and then pass it along to other people sitting in a separate room, but only if these people were their friends; when they were strangers, this “second-hand” contagion was blocked (Wróbel, 2018). At first sight, these findings may seem paradoxical because participants caught emotions from a complete stranger (i.e., the man in the video recording) but then passed these emotions only to their friends. A closer examination of these findings, however, reveals that these two types of contagion differed in terms of the explicitness of the social context. Specifically, participants who were exposed to the video-recorded sender were provided with no information about his social characteristics. Thus, no propositional knowledge about the social context was activated (i.e., the relationship with the video-recorded man was relatively unimportant to participants). The opposite was true for participants observing their friends or strangers in the adjacent room. Here, relationship closeness was explicitly manipulated by making participants aware that the sender they observed on the computer screen was a “real” person who was either their friend or a stranger they had briefly met before the experiment. This manipulation made the social context more salient, which probably activated propositional knowledge about the sender’s social characteristics. As a result, participants might have felt that the stranger who had been briefly met in the hall had a suspicious expression on his or her face, and thus they assessed his or her intent as nonaffiliative. Thus, their affective reactions to strangers were automatically blocked.

A second example is a study by Sachisthal, Sauter, and Fischer (2016). The authors aimed to replicate the effect of group membership on emotional mimicry previously reported by Van der Schalk et al. (2011). Specifically, drawing on the contextual model of emotional mimicry, they expected that facial expressions of negative emotions would be mimicked to a greater extent when displayed by in-group members than when displayed by out-group members. Their findings, however, did not support this effect, which may be attributed to the differences between the original study and the replication study. Specifically, the former used a between-subjects design, whereas the latter used a within-subjects design. Although Sachisthal et al. (2016) assumed that a within-subjects design should have made group membership more salient and thus the effects of the social context stronger, the opposite turned out to be true. Thus, another explanation given by the authors seems more valid. Note that this explanation is in line with our reasoning. According to it, a within-subjects design made the replication study much longer than the original one, which might have reduced participants’ involvement in the task and made them less attentive to the sender’s social characteristics (i.e., in-group vs. out-group membership). In consequence, reduced cognitive resources might have cushioned the modulating effect of the sender’s group membership (i.e., propositional categorization) on the perception-action link (i.e., associative network).

Both examples suggest that the social context plays its modulating role only when the reflective mode is activated. The activation is possible when information provided by social cues enables the receiver to recognize the sender’s intent. This implies that these cues should be clear enough and the receiver should have enough capacity and motivation to process them. Of course, it does not mean that these cues must be explicitly given because, as mentioned previously, the receiver may also use implicit information to understand the sender’s intent. Yet, if the receiver does not have enough cognitive capacity and motivation, implicit cues may be insufficient for the receiver to decide whether the sender’s intent is affiliative or not. For instance, in a typical “acontextual” experimental paradigm, correct interpretation of the sender’s intent is almost impossible, even if the receiver takes the intrinsic social meaning of the expressed emotion into consideration. Without knowing at whom a particular emotion is expressed, and why, the overall meaning of the social context remains unclear, and the receiver may not be able or motivated to untangle it. As a result, reflective mode may not be activated, and the receiver will react to the mere perception of the sender’s display, even if the expressed emotion is intrinsically nonaffiliative. From this perspective, “acontextual” congruent reactions to the sender’s frowns or sneers may represent not only reactive responses but also “unreflective,” uncorrected emotional mimicry or contagion. Put briefly, when the conditions necessary for the social context to “work” are not met, reflective mode is not activated, and the receiver may react with a congruent expression or feeling even in “objectively” nonaffiliative contexts (e.g., when the sender is an out-group member or he or she expresses an intrinsically nonaffiliative emotion). We believe that this idea provides an important extension of existing theories.

We should also note that although we discussed emotional mimicry and social appraisal separately for the sake of clarity, in reality these two mechanisms may operate simultaneously (Bruder et al., 2014; Parkinson, 2011) and thus automatic and corrective processes may also cooccur. As a result, the receiver’s feelings may be triggered by compound effects of all these complex phenomena. Future studies could investigate these effects from the dual-process perspective. For instance, as mentioned previously, the results observed for emotional mimicry and emotional contagion (i.e., self-reported feelings) do not always mirror each other (McIntosh, 2006; Van der Schalk et al., 2011). It seems likely that these inconsistencies may result from the fact that self-reports used to assess emotional contagion are more strongly affected by the reflective mode (and thus more strongly controlled) than methods used to measure emotional mimicry (i.e., electromyography or facial action coding).

Conclusions

In the current review, we propose that the social top-down control of emotional contagion and emotional mimicry may be explained by the correction hypothesis. We also argue that the analogy between assimilation or contrast effects and the receiver’s congruent or incongruent reactions to the sender’s expressions suggests that emotional contagion and mimicry might be perceived as a special type of priming in which the receiver’s affective responses are assimilated with or contrasted away from the sender’s emotional displays. Building on this analogy, we argue that correction processes observed in emotional mimicry or contagion, as with correction processes observed in priming, may result from the interplay between reflective and impulsive processes. Put differently, we extend the assumptions of dual-process theories to a “new” class of phenomena: emotional contagion and its basic mechanisms (i.e., emotional mimicry and social appraisal). The aforementioned two experiments (Sachisthal et al., 2016; Wróbel, 2018) demonstrate that this extension provides fresh insight into the results of previous studies on the effects of social factors on emotional contagion or mimicry. Thus, we believe that the integration of information from three areas of research (i.e., emotional contagion, priming, and dual-process-theories) is a promising avenue for future research addressing the role of social context in processes responsible for emotional convergence between people.