The Social Consequences of Expressive Suppression

Emily A. Butler Boris Egloff Stanford University Johannes Gutenberg-Universita¨t Mainz

Frank H. Wilhelm, Nancy C. Smith, Elizabeth A. Erickson, and James J. Gross Stanford University

At times, people keep their emotions from showing during social interactions. The authors’ analysis suggests that such expressive suppression should disrupt communication and increase stress levels. To test this hypothesis, the authors conducted 2 studies in which unacquainted pairs of women discussed an upsetting topic. In Study 1, one member of each pair was randomly assigned to (a) suppress her emotional behavior, (b) respond naturally, or (c) cognitively reappraise in a way that reduced emotional responding. Suppression alone disrupted communication and magnified blood pressure responses in the suppressors’ partners. In Study 2, suppression had a negative impact on the regulators’ emotional experience and increased blood pressure in both regulators and their partners. Suppression also reduced rapport and inhibited relationship formation.

It occasionally seems desirable to reduce or even suppression. It has been systematically studied both in entirely suppress emotional expressions that arise dur- solitary situations (Gross, 1998a; Gross & Levenson, ing our interactions with others. For example, when a 1993, 1997; Richards & Gross, 1999, 2000) and in friend angers us, we may try to prevent further con- noninteractive social situations such as being ob- flict by suppressing our outward signs of anger until served by strangers (Harris, 2001). Surprisingly little we can figure out how best to resolve the issue, or we is known, however, about either the personal or social may hide our distress about witnessing a disturbing outcomes that accompany suppression during social event because we expect that talking about the event exchanges. This is a particularly unfortunate gap in will only increase our feelings of upset. our understanding, given that expressive suppression In the emotion regulation literature, the process of is quite common during social interaction (Gross & consciously inhibiting emotional expressions while John, 2002) and that diverse lines of evidence suggest emotionally aroused has been referred to as expressive important links between emotion expression, social relationships, and health (House, Landis, & Umber- son, 1988; Schwarzer & Leppin, 1991; Seeman, 2001; Uchino, Cacioppo, & Kiekolt-Glaser, 1996). Emily A. Butler, Frank H. Wilhelm, Nancy C. Smith, To generate hypotheses regarding the consequences Elizabeth A. Erickson, and James J. Gross, Department of of expressive suppression during social interaction, Psychology, Stanford University; Boris Egloff, Department we turn to the literatures on emotion regulation of Psychology, Johannes Gutenberg-Universita¨t, Mainz, (Gross, 1998b, 1999a, 1999b) and interpersonal com- Germany. munication (H. H. Clark, 1996; Kappas & Descote- We thank Tina Huang for her help with data collection aux, in press; Keltner & Kring, 1998). As we describe and data reduction for Study 1. We also thank Meg Given, Leonard Horowitz, Oliver John, Jessica Letchamanan, Ni- here, the picture that emerges from the various litera- chole Schectman, and members of the Stanford Psycho- ture is that expressive suppression may be a particu- physiology Laboratory for their help with this research pro- larly costly form of emotion regulation that disrupts gram. multiple aspects of social exchange, creating stress for Correspondence concerning this article should be ad- both the regulator and the interaction partner alike.1 dressed to James J. Gross, Department of Psychology, Stan- ford University, Building 420, Jordan Hall, Stanford, Cali- fornia 94305-2130. E-mail: [email protected] 1Peter Salovey was the Action Editor for this article.

48 SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 49

Personal Consequences of what leads to this discrepancy between suppressing Expressive Suppression positive and negative emotion in noninteractive set- tings (Gross & Levenson, 1997). Gross’s (1998b, 2002) process model of emotion Findings regarding the physiological consequences regulation provides a general framework for studying of expressive suppression are more mixed. Although how expressive suppression affects the regulator. This occasional studies have found either no differences model emphasizes that emotions unfold over time and between expressive suppression and spontaneous re- distinguishes between different forms of emotion sponding (Bush, Barr, McHugo, & Lanzetta, 1989) or regulation on the basis of when they have their pri- even decreases in physiological activation associated mary impact on the emotion-generative process. Ex- with suppression (Zuckerman, Klorman, Larrance, & pressive suppression is seen as a type of response- Spiegel, 1981), the more typical finding is that sup- focused emotion regulation because it selectively pression is accompanied by increased sympathetic down-regulates the behavioral component of an emo- and cardiovascular responding (Gross, 1998a; Gross tional response after the emotional response is already & Levenson, 1993, 1997; Harris, 2001; Richards & underway (Gross, 1998b, 1999a, 1999b). Gross, 1999, 2000). Thus, even though the bulk of the In this model, although we would expect suppres- evidence points toward a physiological cost of sup- sion to decrease emotion-expressive behavior, we pression, the magnitude and boundary conditions of would not expect decreases in the subjective experi- these effects are by no means clear. ence of emotion, which is not the target of regulation. In terms of the cognitive consequences of suppres- Given the cognitive and physical work required to sion, several slide- and film-viewing studies have monitor and suppress ongoing emotion-expressive be- shown that suppression reduces memory for social havior, we would expect increases in physiological information (e.g., individuals’ names and occupa- measures that are sensitive to effort. We also would tions) presented at the time of the regulatory effort, expect decreases in measures of cognitive functioning which suggests increases in cognitive load (Richards that are sensitive to the multitasking required to sup- & Gross, 1999, 2000). This finding is particularly press emotional behavior. Several experiments have relevant when we consider the social consequences of examined these predictions regarding behavioral, ex- suppression. If suppression is cognitively costly, then periential, physiological, and cognitive consequences it might be expected to distract the regulator from his of suppression. or her partner. This should sorely tax the regulator’s These studies show that by early adulthood, indi- ability to carry out the basic processes required for viduals are quite adept at inhibiting the outward signs conversation maintenance, including the generation of of emotion. In both solitary slide- and film-viewing appropriately contingent responses, and this in turn paradigms (Gross, 1998a; Gross & Levenson, 1993, could be quite disruptive to the interaction (Lieber- 1997; Richards & Gross, 1999, 2000), and in non- man & Rosenthal, 2001). interactive social contexts (Friedman & Miller- Herringer, 1991; Harris, 2001), participants are able Social Consequences of Expressive Suppression to substantially reduce positive and negative expres- Theories of interpersonal communication posit that sive behavior. With respect to changes in emotion face-to-face interaction involves not only information experience, several studies have found that suppress- exchange but also (a) interpersonal coordination in the ing negative emotion expressions does not decrease service of accomplishing various joint goals (e.g., ex- negative emotion experience. For example, suppress- changing goods, complying with an experimenter to ing emotion expressions in response to either a dis- have a conversation about a specific topic); and (b) gusting or a sad film (Gross, 1998a; Gross & Leven- the ongoing negotiation of the relative status and in- son, 1997) and suppressing signs of embarrassment in timacy of the interaction partners (H. H. Clark, 1996; public (Harris, 2001) did not alter the experiences of Fridlund, 1994; Harker & Keltner, 2001; Kappas & those negative emotions. Interestingly, suppression Descoteaux, in press; Keltner & Kring, 1998). These does seem to dampen positive experience. Amuse- functions are both intentionally signaled and uninten- ment was decreased both when participants sup- tionally indexed through multiple communicative pressed their expressive behavior while watching an channels including verbal content, facial expression, amusing film (Gross & Levenson, 1997) and while voice intonation, and gesture. If expressive suppres- reading cartoons (McCanne & Anderson, 1987; sion results in reduced positive and negative emotion Strack, Martin, & Stepper, 1988). It is not yet clear expression, as well as distraction, what consequences 50 BUTLER ET AL. would we expect for the regulator’s partner and for vestigated stonewalling, characterized by low levels the relationship between them? of emotion expression, and have linked this process Consider the first function, interpersonal coordina- with increased physiological responding and negative tion. Many authors have emphasized that responsive- emotion experience in the regulator, along with re- ness—the provision of appropriately contingent reduced marital satisfaction in both partners (Gottman sponses—is the minimal requirement for coordination & Levenson, 1988; Levenson, 1994). Clinically ori- among individuals (Berg & Derlega, 1987; Cappella, ented researchers have examined links between pat- 1997; Davis, 1982; Davis & Perkowitz, 1979; terns of emotional expression and the social difficul- Laurenceau, Barrett, & Pietromonaco, 1998; Reis & ties associated with psychopathology (Keltner & Shaver, 1988). It does not matter whether you are Kring, 1998); for example, part of the aversive social discussing a film or asking about the price of bread, if quality of depressed individuals may be their de- the other person does not respond, or responds with creased levels of positive expressivity (Gotlib, 1992). irrelevant content, then the interaction breaks down. There is also evidence that appropriate emotion ex- Thus, if expressive suppression distracts the regulator pression has positive consequences. Expressivity is from the conversation and decreases responsive be- considered to be a component of social skill (Riggio, havior, it should disrupt interpersonal coordination. 1986; Riggio & Friedman, 1982), positive emotion Now consider the second function, negotiating re- expression has been linked with the development of lationship parameters. Emotion expressions provide affiliation and rapport (Bernieri, Gillis, Davis, & information to observers about a person’s social in- Grahe, 1996; Harker & Keltner, 2001; Tickle-Degnan tentions and his or her feelings regarding the current & Rosenthal, 1990), and the self-disclosure of emo- relationship. For example, a smile communicates a tions appears to be central to both initial attraction and willingness to affiliate, but a frown suggests that con- the development of intimacy (Berg & Derlega, 1987; flict is likely. As such, emotion expressions are cen- Collins & Miller, 1994; Hornstein & Truesdell, 1988; tral to establishing interpersonal distance and relative Laurenceau et al., 1998). Along similar lines, respon- status (Fridlund, 1994; Harker & Keltner, 2001; Kap- siveness has also been shown to predict attraction, pas & Descoteaux, in press; Keltner & Kring, 1998). affiliation, and rapport (Davis & Perkowitz, 1979; Given that expressive suppression explicitly involves Laurenceau et al., 1998; Reis & Shaver, 1988). Fi- inhibiting emotional content across communicative nally, a series of experimental studies of the stress- channels, we would expect suppression to dramati- reducing capacity of social support (Christenfeld cally disrupt the relationship negotiation aspect of so- et al., 1997; Glynn, Christenfeld, & Gerin, 1999; cial exchange. Lepore, 1995; Lepore, Allen, & Evans, 1993) has Disrupting both interpersonal coordination and re- demonstrated that supportive behavior, characterized lationship negotiation within a conversation would by high levels of positive emotion expression and make it more difficult for the participants to accom- responsiveness, reduces cardiovascular stress-re- plish even the simplest joint tasks and would create an sponding in the recipient, and nonsupportive be- ambiguous situation with respect to their relationship. havior, characterized by a lack of expressive behavior Taken together, this should produce a fairly stressful and responsiveness, increases cardiovascular stress- encounter (Tomaka, Blascovich, Kelsey, & Leitten, responding. 1993). We thus may expect expressive suppression to result in physiological signs of stress in both the regu- The Present Research lator and his or her partner. Similarly, the disruption On the basis of both theory and prior findings, we of implicit relationship negotiation should inhibit a expected expressive suppression to result in the fol- sense of connection, and so we thus may expect in- lowing outcomes for the regulator: (a) reduced nega- dividuals who interact with someone suppressing his tive and positive emotion expression, (b) reduced or her emotions to experience lower levels of rapport positive emotion experience, (c) increased distraction, (feeling close and connected to the other person) and (d) decreased responsiveness, and (e) increased physi- to be less willing to establish or maintain an intimate ological responding. In terms of social consequences, relationship. we expected expressive suppression to lead to (f) de- Indirect support for these hypotheses can be found creased rapport, (g) decreased willingness to affiliate, across research areas, including investigations of and (h) increased physiological responding for the marital interaction, emotion expression, responsive- partners of the suppression regulators. ness, and social support. Marital researchers have in- A strong test of these hypotheses requires experi- SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 51 mental manipulation of expressive suppression within Similar to expressive suppression, survey data show an ecologically valid social context. To provide such that cognitive reappraisal is common in daily life a test, we conducted two studies that involved un- (Gross & John, 2002), and experimental data show scripted two-person interactions. One person in each that it can be manipulated in a lab setting (Gross, of the experimental dyads, unbeknownst to her part- 1998a; Richards & Gross, 2000). On the basis of our ner, was asked to regulate her emotions during the prior experience with these two forms of emotion conversation. We chose this methodology, rather than regulation, we expected that participants would find using confederates, because it allowed us to ask the two tasks equally difficult to perform. whether engaging in expressive suppression provides Despite these similarities, we expected suppression a sufficient condition for producing significant inter- alone to disrupt communication and hence to be personal outcomes. We used single-sex dyads to sim- uniquely associated with social consequences. As pre- plify our design and chose to focus on women both viously outlined, expressive suppression should re- because they tend to be more emotionally expressive duce both positive and negative expressivity. In con- than men (Kring & Gordon, 1998) and because they trast, cognitive reappraisal should reduce negative show more consistent physiological stress responses emotion experience, and hence negative expressive to supportive behavior (Glynn et al., 1999). We fo- behavior, but leave positive experience and expres- cused on interactions between previously unac- sion unscathed (Gross, 1998a; Richards & Gross, quainted women because we thought that if partici- 2000). In addition, previous research in nonsocial set- pants were familiar with each others’ typical tings illustrates that cognitive reappraisal does not en- communication styles, they might be suspicious when tail the decrements in memory that are associated with a normally expressive individual began to suppress suppression (Richards & Gross, 2000). As a result, we her emotion expressions. An additional motivation for expected cognitive reappraisal to interfere less with using unacquainted individuals was to better under- responsiveness than would expressive suppression. stand relationship formation. The formation of new These expected differences in expressivity and re- friendships is an important process in our increasingly sponsiveness led us to predict that only the partners of mobile social world, and there is evidence that the participants engaging in expressive suppression, and quality of an initial social interaction between two not the partners of participants engaging in cognitive partners can play an important role in determining reappraisal, would experience lower levels of rapport whether the interactants go on to become friends and increased blood pressure during a conversation. (Berg & Clark, 1986). Method Study 1: Comparing Expressive Suppression Participants With Cognitive Reappraisal Seventy-two women from the Stanford University When assessing the personal and social conse- community participated in this study. Participants quences of expressive suppression, one crucial issue were paid $15. The mean age of participants was 20.3 is what comparison standard to use. One natural com- years). Of the participants, 51% 5.2 ס years (SD parison condition is an uninstructed condition in described themselves as Caucasian, 33% as Asian which both members of the conversation dyad are free American, 10% as Latin American, 3.5% as African to respond as they typically do. This comparison stan- American, and 2.5% as other. dard makes it possible to examine the personal and social consequences of emotion suppression versus Procedure free expression. However, an uninstructed compari- Unacquainted pairs of women were briefly intro- son condition alone does not address the possibility duced to each other, and it was verified that the that any observed effects of suppression simply reflect women had never spoken to each other before. They the imposition of a second task. were then seated 2 m apart on either side of an opaque To address this issue, we also used a second com- partition. The experimenter explained that the purpose parison group in Study 1. In this comparison group of of the study was to better understand conversational dyads, one member of each dyad had been randomly processes. The participants were informed that blood assigned to engage in a second common form of emo- pressure would be measured and that their conversation down-regulation, namely cognitive reappraisal, tion would be videotaped. which entails altering how a situation is construed so After blood pressure cuffs were attached, a televi- as to change its emotional impact (Gross, 1998b). sion monitor was positioned so that each participant 52 BUTLER ET AL. could view the monitor but not each other. Partici- (SPAFF; Gottman & Levenson, 1992). The SPAFF pants first watched a 3-min nature film, which pro- uses a “cultural informant” approach to coding in vided our blood pressure reference baseline. To create which the gestalt of all simultaneously occurring com- a shared negative emotion experience, participants municative signals, both verbal and nonverbal, is were shown an upsetting 16-min documentary war taken into account when assigning a behavioral seg- film. This film shows graphic footage of the aftermath ment to one of the coding categories. Some modifi- of the nuclear bomb being dropped on Hiroshima and cations were necessary for our purposes because the Nagasaki during World War II. Pilot testing showed SPAFF tracks specific emotions such as anger, that this film elicits high levels of negative emotions whereas our hypotheses were framed at a broader such as disgust, anger, and sadness, as well as strong level of analysis and includes categories appropriate political and religious opinions. for marital interaction that did not appear in our data Random assignment to conditions took place im- set, such as “whining” or “affection.” Therefore, ag- mediately after viewing the war film. The first step gregates of only the relevant categories were used and was to assign a dyad to either the suppression, reap- each turn of speech was classified as emotionally neu- praisal, or control group. This process resulted in 12 tral, positive expression, or negative expression. suppression dyads, 12 reappraisal dyads, and 12 un- Some turns contained both positive and negative ex- instructed control dyads. Next, the women in the regu- pression, and so a turn could be double-coded. Fi- lation dyads were assigned to be either the regulator nally, the SPAFF does not include an equivalent cat- or the uninstructed partner. The suppression regulator egory to responsiveness. In developing such a received tape-recorded suppression instructions via category, we followed Davis’ (1979, 1982) approach headphones (“During the conversation, behave in and focused on nonresponsiveness because the lack of such a way that your partner does not know you are response is more salient than its presence. A nonre- feeling any emotions at all”). The reappraisal regula- sponse was thus defined as a sequence in which one tor received tape-recorded reappraisal instructions person finished an utterance and the other person ei- (“During the conversation, think about your situation ther did not respond within 2 s, or they responded in such a way that you remain calm and dispassion- with an utterance of less than three words followed by ate”). Their 24 partners, plus the 24 individuals in the silence, or they responded with content that was un- control condition, simply heard a bland musical seg- related to the previous person’s contribution. ment. Because conversations differed in length, propor- The experimenter then removed the partition and tions were used for our analyses. For expressive be- asked participants to discuss their thoughts and feel- havior, the number of either positive or negative ex- ings, the implications of the film for human nature, pressive utterances was divided by the total number of and its relevance to their religious and political be- utterances. For nonresponse, one woman’s number of liefs. Participants were free to signal the end of the nonresponses was divided by her partners’ total num- conversation when they so chose. After the conversa- ber of utterances (i.e., the total number of conversation, the opaque partition was replaced, and partici- tional turns that could be responded to). For ease of pants privately responded to the self-report measures interpretability, this was then translated into a respon- (see the Measures of Rapport, Emotion Experience, siveness score using the conversion of “1 − % non- Task Difficulty, and Distraction section). Finally, par- response.” ticipants were fully debriefed and given a chance to Coders were blind to the participants’ experimental converse with each other about the experiment. condition. One person coded all videotapes, and two other coders provided reliability ratings on 25 of the Measures of Expressive Behavior 36 tapes. For tapes that were coded by multiple raters, Participants were videotaped during the conversa- the mean of the ratings was used for final analyses. ס tion with two cameras hidden behind darkened glass Reliabilities were excellent (positive expression: r responsiveness: r ;92. ס and positioned so that one camera focused on each .90; negative expression: r .(87. ס participant’s face and upper torso. The two camera images were then combined into a single split-screen Two other behavioral measures were obtained: image using a special effects generator. The videos speaking time and general somatic activity. Speaking were used to score the participants’ emotion-expres- time was recorded from the videotapes using a stop- sive behavior and responsiveness using a coding sys- watch. Any utterance of more than two syllables was tem adapted from the Specific Affect Coding System considered as speaking. To measure somatic activity SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 53 levels, piezo-electric sensors were attached to the legs to attempting to suppress emotion in a social situation of the participants’ chairs, which generated electrical (Harris, 2001). signals proportional to the amount of movement in Participants’ blood pressure was measured continu- any direction. ously throughout the baseline film and conversation using an Ohmeda 2300 Finapres (Ohmeda Monitoring Measures of Rapport, Emotion Experience, System, Englewood, CO). The continuous finger ar- Task Difficulty, and Distraction terial pressure waveform was manually edited for ar- All self-report responses used an 11-point Likert tifacts, and beat-by-beat systolic (SBP) and diastolic a great deal). To assess par- blood pressure (DBP) values were computed. SBPs ס none, 10 ס scale (0 and so (95. ס ticipants’ experience of rapport during the conversa- and DBPs were highly correlated (r tion, they were asked to report on the degree to which were combined into mean arterial blood pressure DBP ס they felt the conversation had been warm and smooth (MAP), following standard guidelines (MAP and the extent to which they felt they had “clicked” + 1/3 [SBP − DBP]; Vander, Sherman, & Luciano, with their partner. The alpha for this three-item rap- 1990). Change scores were then computed by sub- port scale was .85. tracting mean neutral-film baseline values from mean To assess emotion experience, after the conversa- conversation values. tion, participants provided a retrospective report of Data Analysis the emotions that they had felt during the conversation. Participants indicated how much they had felt One feature of any unscripted social interaction is 12 negative emotions (anger, anxiety, self-conscious- that the responses of the two individuals within a ness, contempt, disgust, fear, sadness, tension, frus- conversational dyad may be correlated. For example, tration, embarrassment, guilt, shame), and 8 positive if one person acts in a hostile manner, it is very likely emotions (amusement, happiness, joy, love, inter- that her partner will reciprocate. This lack of indepen- est, excitement, pride, pleasantness). The alpha for the dence violates the assumptions of standard analyses of Negative Emotion scale was .87; the alpha for the variance (ANOVAs) and regression models, and if Positive Emotion scale was .78. ignored can render the significance tests inaccurate We expected that our instructions to engage in ex- (Kashy & Kenny, 1997; Kenny, 1988, 1996a, 1996b). pressive suppression and in cognitive reappraisal Therefore, to test our hypotheses, we used Kenny’s would be equally difficult for our participants to carry actor–partner interdependence model, which not only out but that the unique self-monitoring demands im- deals appropriately with dyadic data but actually takes posed by expressive suppression would result in advantage of it to address questions of mutual influ- higher levels of distraction in the suppression regula- ence (Kashy & Kenny, 1997; Kenny, 1996b). This tors. To obtain evidence pertinent to these hypotheses, model entails calculating “actor effects” and “partner our regulation participants were asked, following the effects.” An actor effect represents the influence that conversation, to report on the item “How difficult was an individual’s score on a predictor variable has on it to follow your instructions during the conversa- her own score on a dependent variable, whereas a tion?” and both the regulation and uninstructed par- partner effect represents the influence that an indi- ticipants were asked to report on the item “How dis- vidual’s score on a predictor has on her partner’s tracted were you during the conversation?” score on the dependent variable. For example, if suppressing emotion expressions raised the regulators’ Measures of Physiological Responding blood pressure, then this would be captured by an In the domain of physiological responding, blood actor effect, but if suppressing raised the regulator’s pressure responses were chosen to be the focus for partner’s blood pressure, then this would appear as a two reasons. First, blood pressure is the key-regulated partner effect. Full details of how to implement the variable of the cardiovascular system (Vander, Sher- method are provided in Kashy and Kenny (1997). man, & Luciano, 1990). This makes it a sensitive For each actor or partner effect of interest, the fol- index in situations such as an emotional conversation lowing two hypothesis tests were conducted: (a) the where both the sympathetic and the parasympathetic difference between the uninstructed control partici- autonomic systems are expected to be activated. Sec- pants and the combined mean of both regulation ond, blood pressure appears to be particularly sensi- groups—actor effects reflect the mean of the two tive to social cues of evaluative threat (Cacioppo & types of regulators compared with the controls, Petty, 1986), social support (Uchino et al., 1996), and whereas partner effects reflect the mean of the part- 54 BUTLER ET AL. ners of the two types of regulators compared with the ished in terms of responsiveness. Although the com- controls; and (b) the difference between the two regu- bination of the two regulation groups did not differ lation groups—actor effects reflect the difference be- from the controls, the suppressors were significantly ס tween suppressors and reappraisers, whereas partner less responsive than the reappraisers (actor effect p < .001. Thus, although both ,−3.53 ס (effects reflect the difference between the partners of −0.07), t(67 suppressors and the partners of reappraisers. forms of regulation reduced expressivity, suppression To balance the demands of limiting Type I and alone interfered with responsiveness. Group means Type II error, our hypothesis tests were divided into and standard errors for interpersonal behavior are three families and the Type I error rate was restricted shown in Figure 1. within each family by using a stepwise Bonferroni Personal Consequences: Difficulty procedure (Howell, 1992). The first family of tests and Distraction includes the emotion experience items because these bear on the question of the effects of emotion regu- Because only the regulators received task instruc- lation on subjective experience. This family required tions, an independent samples t test was adequate to a total of 8 hypothesis tests. The second family in- compare their responses on the difficulty item. As cludes all other primary measures and required a total expected, the suppressors’ difficulty in following the did not differ from (1.8 ס SD ,8.4 ס of 15 tests. The third family included the secondary instructions (M With .(2.5 ס SD ,7.3 ס measures and involved five tests. Following Howell that of the reappraisers (M (1992), the obtained p values were ordered for the respect to distraction, we expected suppression to dis- tests within a set in decreasing order of significance tract the regulators from the conversation more than and the critical p value was set for the first test (the reappraisal would. As shown in Figure 1, this was in most significant one) at .05/c, where c equals the total fact the case. Actor–partner analyses showed that the number of tests to be made within the set. Assuming combination of the two types of regulation differed ס this test was significant, the critical p value was set for significantly from the control group (actor effect -p < .004, and that the suppres ,−3.07 ס (the next test (the second most significant one) at .05/ −0.38), t(48 (c − 1), reflecting the total number of tests left to be sors were also significantly more distracted during the made within the set. This process continued until a conversation than were the reappraisers (actor effect .p < .001 ,3.43 ס (t(65 ,(0.48 ס .test was nonsignificant Personal Consequences: Emotion Experience Results Actor–partner analyses on the positive and negative We first present results relevant to the personal emotion experience composites did not show pre- consequences of suppression (actor effects). We next dicted differences among regulation groups. For posi- present results relevant to the social consequences tive emotion, there were no significant differences the ,(0.6 ס SD ,1.4 ס partner effects). Finally, we present secondary analy- between the suppressors (M) and the controls ,(1.4 ס SD ,2.5 ס ses intended to address several alternative accounts of reappraisers (M ,Similarly for negative emotion .(1.3 ס SD ,2.0 ס our findings. (M Personal Consequences: Behavior the means and standard deviations for the suppressors, 2.5 ,(1.9 ס reappraisers, and controls were 2.3 (SD respectively, with no ,(1.9 ס and 2.9 (SD ,(2.0 ס As predicted, the regulators were less expressive of (SD negative emotion than were the uninstructed controls. significant differences. An actor–partner analysis contrasting the mean of the Personal Consequences: Blood Pressure two regulation groups with the controls showed that We also predicted that the suppressors would show ,(−0.06 ס this difference was significant (actor effect -p < .008, whereas the two types of enhanced blood pressure responses during the conver ,−2.76 ס (t(62 regulators did not differ from each other. This pattern sation relative to reappraisal or control participants. was also observed for positive expressivity. The com- Again, this hypothesis was not supported. Consistent bined means of the two regulation groups differed with prior literature on the effects of vocalization ,al’Absi et al., 1997; Egloff, Wilhelm, Neubauer) ,(−0.05 ס significantly from the controls (actor effect -p < .001, but the suppression and Mauss, & Gross, 2002), blood pressure increased dur ,−3.67 ס (t(67 reappraisal regulators again did not differ. In contrast, ing the conversation for all participants. However, the suppression regulators were uniquely impover- there were no significant differences in increases in SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 55

Figure 1. Interpersonal behaviors and distraction by instructional group in Study 1.

SD ,17.1 ס mean blood pressure for suppressors (M and ,(10.0 ס SD ,19.8 ס reappraisers (M ,(10.9 ס .(11.7 ס SD ,18.1 ס controls (M Social Consequences: Rapport We expected the differences in suppressors’ and reappraisers’ interpersonal behavior to result in differences in their partners’ experiences of rapport. Al- though the partners of suppressors did report feeling ס SD ,5.4 ס less rapport during the conversation (M 1.9) than did either the partners of the reappraisers (M ס or the uninstructed controls (M (1.6 ס SD ,6.2 ס our stringent control of Type I error ,(1.6 ס SD ,7.2 prevented this result from reaching significance. Social Consequences: Blood Pressure Increases in mean arterial blood pressure for the partners of the regulators as compared with the uninstructed controls are presented in Figure 2. As predicted, the partners of the suppressors showed the largest increases. An actor–partner analysis confirmed Figure 2. Increases in blood pressure by instructional that although the combination of the two types of group in Study 1. 56 BUTLER ET AL.

,min 2.6 ס min), reappraisal regulators (M 1.4 ס regulation partners did not differ from the controls, SD ,min 3.7 ס min), suppressors’ partners (M 1.6 ס the suppressors’ partners experienced significantly SD ,min 2.7 ס min), reappraisers’ partners (M 2.4 ס larger increases than did either the reappraisers’ part- SD ס min, SD 4.3 ס min), and controls (M 1.5 ס p < .002, SD ,3.24 ס (t(63 ,(4.03 ס ners (partner effect min). To be certain that neither activity levels nor 2.8 ,2.49 ס (t(58 ,(4.67 ס or the controls (partner effect p < .016. speaking time contributed to the blood pressure differences, we reran our actor–partner analyses of blood Social Consequences: Emotion Experience pressure, including those variables as predictors, and We had not predicted any differences in the sub- established that this did not alter the findings. jective experience of emotion for the partners of the regulators. Reports of positive emotion by the sup- Discussion -the reap ,(2.0 ס SD ,1.9 ס pressors’ partners (M -and the un- Study 1 tested the hypothesis that expressive sup ,(1.1 ס SD ,2.4 ס praisers’ partners (M -did not pression would have personal and social conse (1.3 ס SD ,2.0 ס instructed controls (M differ. Similarly, there were no differences in reports quences, and that these consequences would be spe- of negative emotion experience by the suppressors’ cific to suppression. To test this hypothesis, we —the reappraisers’ part- compared two forms of emotion regulation ,(1.1 ס SD ,1.4 ס partners (M —expressive suppression and cognitive reappraisal ,2.9 ס and the controls (M ,(1.0 ס SD ,2.0 ס ners (M with each other and with an uninstructed control .(1.9 ס SD group. We found that suppression, and suppression Secondary Analyses alone, distracted the regulators from the conversation, To ascertain whether the observed differences be- reduced their responsiveness, and led to the regula- tween suppressors’ and reappraisers’ partners’ blood tors’ partners experiencing heightened blood pressure pressure responses could be attributed to general so- increases. Furthermore, the cardiovascular impact on matic activity, we conducted an additional actor– the regulators’ partners could not be accounted for by partner analysis. There were no differences between factors such as task difficulty, general activity levels, any of the groups (means ranged from .17 to .19, with or speaking times. It appears, therefore, that expres- standard deviations from .004 to .008). We similarly sive suppression was particularly disruptive to com- wished to rule out conversation length and speaking munication and produced a uniquely physiologically time as explanations of the blood pressure differences. stressful encounter for the regulators’ partners. Conversations ranged in length from 2.5 min to 22 These findings are intriguing, but several important min). A one-way limitations of Study 1 must be noted. Perhaps the 4.0 ס min, SD 7.4 ס min (M ANOVA, with regulation instructions as a between- most important limitation is the relatively small subjects variable (suppression, reappraisal, control), sample size. On the basis of the observed data, we showed that although suppression conversations (M required a very large effect size to obtain significance min) and reappraisal conver- given our tight control of Type I error. On the one 3.6 ס min, SD 6.8 ס min) tended to be hand, this fact gives us confidence that those results 2.8 ס min, SD 6.4 ס sations (M that did attain significance are substantial. On the 9.1 ס slightly shorter than control conversations (M min), this difference was not signifi- other hand, we did not have the statistical power to 5.0 ס min, SD cant. detect even fairly large group differences. This fact Although regulation instructions did not affect con- may account for several null results. For example, versation length, they nonetheless could have influ- suppressors experienced less positive emotion than enced participants’ speaking time. To test this possi- the controls, whereas the reappraisers experienced bility, we conducted an actor–partner analysis on more positive emotion. Similarly, the partners of sup- individual speaking times. This demonstrated that the pressors felt less rapport than either the reappraisers’ combined speaking length of the two types of regu- partners or the uninstructed controls. Both of these lator was shorter than that of the controls (actor effect results were predicted a priori and have effect sizes in -p < .007, but that the two the range of .80, but the lack of statistical power pre ,−2.77 ס (t(65 ,(−24.94 ס types of regulator did not differ from each other. The vented them from reaching significance. It should be partners of the regulators did not differ from either the noted that this lack of power does not weaken the controls or from each other. The means and standard conclusion that suppression uniquely drove up the -min, regulators’ partners’ blood pressure because the re 2.3 ס deviations were suppression regulators (M SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 57 appraisers’ partners actually experienced smaller ducted Study 2. Because Study 1 showed that expres- blood pressure increases than did the uninstructed sive suppression had consequences not evident in a controls. second common form of emotion down-regulation Another factor that may have contributed to null that matched suppression in terms of task difficulty, results is that our emotion experience measure, de- Study 2 focused on just two conditions to maximize spite having been used successfully in numerous non- cell size: suppression and uninstructed control. We social studies, may not have been optimal in a con- retained most of the procedures from Study 1 but versational setting. Within the context of a social included a more differentiated sampling of emotion interaction, a distinction needs to be made between experience in which we distinguished between emo- emotional experiences arising from the topic of con- tional responses as a result of recalling the film and versation and emotional experiences provoked by those arising in response to one’s partner. In addition, one’s partner. For example, in our context, suppres- we aimed to more directly assess the impact of sup- sion may have resulted in conversations that avoided pression on the potential for relationship develop- the topic of the film, and so when responding to emo- ment, and so we included a self-report measure tap- tion items such as “disgust,” the suppressors might ping the participants’ liking of each other and their have reported reduced feelings because the film willingness to form a friendship. would have been less salient during the conversation. With respect to emotion experience, when referring In contrast, when responding to negative emotion to emotions about their partners, we expected sup- items such as “frustration,” the suppressors may have pressors would be conscious of the disruptions in referred to their experience of disrupted communica- communication brought about by their own purpose- tion and given high ratings to these items. When the ful regulatory efforts and might therefore report more items are combined, they would cancel out and we negative and less positive emotional experiences. As would see no differences between the groups. for the suppressors’ partners, studies of social support The bulk of prior research has concluded that sup- routinely show that participants’ do not report differ- pression is accompanied by increased sympathetic ac- ences in experience because of interacting with non- tivity and increased blood pressure responses in the supportive confederates, although they do show in- regulators (Gross, 1998a; Gross & Levenson, 1993, creased physiological responding and reduced liking 1997; Harris, 2001; Richards & Gross, 1999), but this for the confederate (Christenfeld et al., 1997; Glynn et was not borne out in our results. In this case, even the al., 1999; Lepore et al., 1993). This may be due to direction of the group means does not support the the fact that physiological responding and simple prediction, and so this is not purely a power issue. like–dislike evaluations occur more automatically Although it is possible that the cardiovascular effect than consciously evaluated and articulated emotion found in the prior research was overwhelmed by the experiences. We thus expected the partners of the demands of a conversational setting, it is also possible regulators would not report any changes in emotion that with only 12 suppression regulators, we were experience, although they would show increased observing an anomalous sample. physiological stress responses and more negative One final limitation concerns mediation. In the con- evaluations of the conversation and of their partners text of this study, the only way that one woman’s (e.g., decreased rapport and affiliation). emotion regulation could affect her partner was via her behavior. Although Study 1 provided clear evi- Method dence that expressive suppression was particularly Participants disruptive to interpersonal behavior, the small sample Eighty-four undergraduate women at Stanford Uni- size meant that mediation analyses would not be re- versity participated in this study. They either received liable (Baron & Kenny, 1986). This means that it is partial course credit or were paid $25. The mean age years). Of 1.0 ס not yet clear what behavioral changes in suppression of participants was 18.7 years (SD regulators mediated the observed effects. the participants, 53% described themselves as Cauca- sian, 32% as Asian American, 8.6% as Latin Ameri- Study 2: The Personal and Social Costs of can, 1.2% as African American, and 4.9% as other. Expressive Suppression Procedure To replicate and extend findings from Study 1, and The only differences from Study 1 were that a to address the limitations described above, we con- longer baseline nature film was used (6 min) to in- 58 BUTLER ET AL. crease the reliability of our blood pressure measures and positive emotion expression, and their respon- and the self-report measures were improved and ex- siveness. As shown in Figure 3, this was indeed the tended (see the Measures of Affiliation and Emotion case. Actor–partner analyses demonstrated that com- Experience section). Randomization resulted in 21 pared with the controls, during the conversation sup- suppression dyads and 21 uninstructed control dyads. pressors expressed less negative emotion (actor effect p < .001, less positive ,−3.99 ס (t(58 ,(−0.11 ס Measures of Expressive Behavior > p ,−4.84 ס (t(44 ,(−0.11 ס emotion (actor effect Using the same procedures as in Study 1, one coder .001, and were significantly less responsive (actor ef- .p < .001 ,−4.82 ס (t(62 ,(−0.12 ס rated all videotapes while two others provided reli- fect ability ratings on 28 of the 42 tapes. Reliabilities were Personal Consequences: Distraction -nega ;94. ס again excellent (positive expression: r -As predicted, and as shown in Figure 3, the sup .(85. ס responsiveness: r ;88. ס tive expression: r pressors reported higher levels of distraction during Measures of Affiliation and ס the conversation than did the controls (actor effect Emotion Experience .p < .001 ,5.72 ס (t(68 ,(1.19 In addition to the measures from Study 1, following Personal Consequences: Emotion Experience the conversation, the participants’ mutual liking and willingness to form a friendship were assessed with We predicted that the suppressors would experi- the following five questions: (a) “To what extent do ence reduced positive and increased negative emotion you like your partner?” (b) “How well do you think about their partners. As can be seen in Table 1, these you would get along with your partner?” (c) “To what hypotheses were supported. Suppressors reported sig- extent do you think your partner likes you?” (d) “To nificantly less positive emotions about their partners ס (t(66 ,(−4.90 ס what extent would you be interested in talking to your than did the controls (actor effect partner again?” (e) “To what extent is your partner the −2.61, p < .009, as well as significantly more negative ס (t(60 ,(9.80 ס type of person you could become close friends with?” emotions about them (actor effect The alpha for this affiliation composite was .89. 3.38, p < .001. By contrast, there were no significant Emotion experience was again assessed, following effects for emotion experience in response to the film. the conversation, but this time the referent of our Personal Consequences: Blood Pressure questions was clearly specified. Participants were first asked to indicate “the extent to which your partner As in Study 1, blood pressure increased during the made you feel the following emotions during the con- conversation for all participants. Unlike Study 1, but versation.” Participants were then asked to indicate as originally predicted, the suppressors showed larger “the extent to which recalling the film made you feel increases in blood pressure during the conversation the following emotions during the conversation.” For than did the controls. The means and standard errors each questionnaire (emotions about partner, emotions are provided in Figure 4. An actor–partner analysis of about film), participants indicated how much they had increases in mean arterial blood pressure showed that ,(2.32 ס felt the same 12 negative and 8 positive emotions as this difference was significant (actor effect .p < .032 ,2.18 ס (in Study 1. Alphas ranged from .79 to .93. t(77 Data Analysis Social Consequences: Rapport and Affiliation The same analytic strategy was used as in Study 1. As predicted, and as shown in Figure 5, the partners This resulted in one family of eight comparisons rel- of suppressors reported feeling less rapport during the evant to emotional experience, a second family of conversation than did the uninstructed control partici- eight comparisons to evaluate all other primary out- pants An actor–partner analysis confirmed that this comes, and a third family of five comparisons to difference was significant (suppression partner effect p < .025. In addition, we ,−2.32 ס (t(55 ,(−0.64 ס .evaluate secondary factors predicted that the partners of suppressors would report Results less liking and willingness to form a friendship with their partners (i.e., the suppressors) than would the Personal Consequences: Behavior controls. An actor–partner analysis of the affiliation Compared with the controls, we expected that sup- composite clearly supported these hypotheses. As is pression would decrease both the regulators’ negative also shown in Figure 5, the partners of suppressors SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 59

Figure 3. Interpersonal behaviors and distraction by instructional group in Study 2. liked their partners (i.e., the suppressors) less than did Table 1, and actor–partner analyses confirmed that the controls and were decidedly uninterested in ever there were no significant differences between the speaking with them again or in forming a friendship regulation partners and the controls. .p < .016 ,−2.48 ס (t(63 ,(−2.75 ס partner effect) Social Consequences: Blood Pressure Social Consequences: Emotion Experience Increases in mean arterial blood pressure for the We did not expect any differences in subjective partners of the suppressors and the controls are pre- emotion experience in the partners of the suppressors sented in Figure 4. As in Study 1, the partners of the either with respect to recalling the film or in response suppressors showed the largest increases. Actor– to their partners. Descriptive statistics are provided in partner analyses confirmed that this difference was

Table 1 Mean Emotion Experience Reports (and Standard Deviations) in Study 2 Suppressors’ Variable Controls Suppressors partners Experience about partner Negative experience 1.0 (0.9)a 2.3 (2.2)b 1.7 (1.1)a,b Positive experience 4.0 (1.8)a 2.6 (1.6)b 3.4 (1.7)a,b Experience about film Negative experience 3.3 (1.5)a 3.4 (2.1)a 3.9 (1.8)a Positive experience 2.5 (1.3)a 1.4 (1.2)a 2.8 (1.4)a a great deal). Means ס none, 10 ס Note. Means and standard deviations based on an 11-point scale (0 in a given row with different superscripts differ from one another at p < .01, two-tailed. 60 BUTLER ET AL.

Figure 4. Increases in blood pressure by instructional group in Study 2.

> p ,3.24 ס (t(77 ,(3.45 ס significant (partner effect .002. These results indicate that interacting with someone who is suppressing her emotional expressions can result in heightened cardiovascular responding. Secondary Analyses

As in Study 1, an actor–partner analysis of somatic ס activity revealed that neither the suppressors (M ס SD ,0.3 ס nor their partners (M (0.2 ס SD ,0.4 (0.4 ס SD ,0.5 ס differed from the controls (M (0.2 in terms of general body movement. Turning to conversation length, conversations ranged from 4.0 min min). An 4.7 ס min, SD 9.1 ס to 25.2 min (M independent samples t test showed that although sup- Figure 5. Experiences of rapport and affiliation by instruc- .min) tional group in Study 2 4.9 ס min, SD 8.5 ס pression conversations (M tended to be slightly shorter than control conversa- min), this difference 4.9 ס min, SD 9.8 ס tions (M uted to the blood pressure differences. Nevertheless, was not significant. We also investigated whether ex- we reran our actor–partner analyses of blood pressure, pressive suppression would impact individual speak- including general activity levels and speaking times as ing times. An actor–partner analysis demonstrated predictors. As in Study 1, these secondary analyses 6.0 ס that neither the suppressors’ speaking times (M did not alter our findings. ,min 5.7 ס min) nor their partners’ (M 5.9 ס min, SD min) differed from that of the controls (M Mediation of Partner Outcomes 4.8 ס SD .(min 4.9 ס min, SD 5.8 ס Because neither the suppressors nor their partners We reasoned that the replicated social conse- differed from the controls in terms of activity levels or quences of expressive suppression must be mediated speaking times, these factors should not have contrib- by the regulators’ behavior. In the following sections, SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 61 we present mediation analyses following Baron and ized the suppression dyads yielded no significant as- Kenny (1986). In these analyses, we first predicted sociations either for actors or for partners. In other each partner outcome (rapport, affiliation, and blood words, a woman’s expressivity and responsiveness pressure) from each of the behaviors that differenti- were not associated with either her own or her part- ated between the suppressors and the controls (posi- ner’s blood pressure response. tive and negative expressivity, responsiveness). If a behavior–outcome pairing showed a significant asso- Discussion ciation, we then took the final step in a mediation analysis by predicting the given outcome from an Study 2 replicated and extended the findings from equation, including both regulation condition (sup- Study 1. Expressive suppression distracted the regu- press vs. control) and the behavior. Mediation is dem- lators from the conversation, led to major decrements onstrated when (a) the slope for the regulation condi- in their expressivity and responsiveness, and pro- tion is no longer significant but the one for the duced heightened cardiovascular responding in their mediating behavior is, and (b) the reduction in the partners. Whereas in Study 1 there was a trend for the slope for the regulation condition is significantly dif- suppressors’ partners to feel less rapport than the con- ferent from zero. trols, in Study 2 this difference was clearly apparent Rapport. A sequence of actor–partner analyses and reached statistical significance. In addition, Study predicting rapport from each of the three behaviors 2 extended our investigation by including the impact showed that only responsiveness produced a signifi- of expressive suppression on the potential for the p < .008. women to develop a friendship, finding that when one ,2.73 ס (cant partner effect (2.76), t(77 Thus, individuals who produced high levels of re- woman inhibited her emotion expressions, it dramati- sponsive behavior had partners who felt high levels of cally reduced her partner’s motivation to become fur- rapport. The next steps in the analysis showed that ther acquainted. These findings indicated that expres- responsiveness did in fact mediate the partners’ expe- sive suppression not only directly impacted the rience of rapport. When both regulation condition and regulators’ partners in negative ways but it also un- responsiveness were included in an actor–partner dermined the relationship between them. analysis predicting rapport, the partner effect for con- Turning to the personal consequences of expressive dition, which was originally significant, no longer suppression, Study 2 provided support for the gener- was, but the partner effect for responsiveness contin- alizability of effects that previously have been seen in nonsocial situations. Compared with the uninstructed ס (t(76 ,(2.65 ס ued to be significant (partner effect 2.35, p < .021. Furthermore, the Sobel reduction test control group, women who were asked to suppress demonstrated that the reduction in the partner effect of experienced less positive and more negative emotion about their partners and had larger blood pressure ס regulation condition was significant (Sobel: z -responses. Although we had not observed these ef .(03. ס p ,−2.13 Affiliation. A similar analysis of the participants’ fects in Study 1, this discrepancy may be accounted responses to the affiliation measure showed that both for by the lack of statistical power and less clearly negative expressivity and responsiveness, but not defined emotion experience items in the first study. positive expressivity, predicted the partners’ willing- The increased sample size of Study 2 permitted us ness to form a friendship (negative expressivity part- to examine mediating processes. Our mediational -p < .025; (respon- analyses revealed that the suppressors’ reduced re ,2.28 ס (t(80 ,(9.00 ס ner effect -p < sponsiveness mediated their partners’ reduced rap ,2.40 ס (t(63 ,(10.80 ס siveness partner effect .019. Thus, individuals who expressed high levels of port. This is in accord with theory and research that negative emotion during the conversation, and those posits responsiveness as the minimal requirement for who were responsive, had partners who were willing interpersonal coordination and the experience of rap- to form a friendship. Unlike rapport, however, the port (Bernieri & Rosenthal, 1991; Cappella, 1997; next step in the analysis showed that these behaviors Tickle-Degnan & Rosenthal, 1990). In addition, did not mediate the partners’ affiliation; in both cases, across both the suppression and the control dyads, one neither condition nor behavior remained a significant woman’s responsiveness and expression of strong predictor when both were included in the equation. negative emotions about the film predicted her part- Blood pressure. Unlike rapport and affiliation, a ner’s willingness to establish a friendship. This is not series of actor–partner analyses predicting blood pres- surprising given the extensive literature on the impor- sure responses from the behaviors that had character- tance of both responsiveness and emotional disclosure 62 BUTLER ET AL. for initial attraction and friendship formation (see e.g., lished, expressive suppression may still have the ef- Berg & Derlega, 1987; Collins & Miller, 1994; fect of degrading the quality of longer term relation- Laurenceau et al., 1998; Reis & Shaver, 1988). Un- ships. This speculation receives support from research like rapport, however, these behaviors did not mediate on marriage. Marital researchers have shown that a the group differences in affiliation. Finally, despite conflict-avoidance behavior referred to as stonewall- these clear links between the regulators’ behaviors ing, which appears to be similar if not identical to and their partners’ evaluations of the conversation and what we have called expressive suppression (Gottman the relationship, these behaviors did not predict car- & Levenson, 1988; Levenson, 1994), is associated diovascular responding in either the suppressors or with reduced marital satisfaction in both partners. It their partners. appears that at each stage of relationship development, inhibiting emotion expressions is an effective General Discussion way of avoiding interpersonal connection whether the regulator intends that goal or not. Emotion theorists since Darwin (1872/1998) have emphasized the communicative value of emotion. Implications for Health Emotions are not always allowed free reign during We believe that our findings have implications for social interaction, however. Occasionally, we want to understanding how emotion, emotion regulation, and suppress our emotional expressions. For example, we social support together may promote or inhibit healthy may want to avoid a troubling topic or keep from adaptation. It is an unfortunate fact of life that no one betraying hidden preferences. Research on emotion can entirely escape negative emotional experiences. suppression generally has focused on the conse- Our results suggest, however, that the ways in which quences for the person who is doing the regulating we regulate our emotional reactions may have impor- and has been limited to solitary or noninteractive set- tant psychological and physical health consequences tings. The present research moves the study of emo- not only for ourselves but for the people who interact tion regulation into the social domain and shows that, with us as well. in some contexts at least, suppressing emotion dis- Kennedy-Moore and Watson (2001) recently re- rupts communication, hinders the development of so- viewed extensive evidence that expressing upsetting cial bonds, and is physiologically taxing for both the emotions can help reduce the intensity of those emo- suppressor and her social partner. tions, limit intrusive thoughts, and increase insight Implications for Social Functioning and a sense of meaning, thereby making the negative emotions less aversive. All of these benefits would be Our results suggest that the use of expressive sup- lost if an individual chronically suppressed emotion pression may have implications for the regulator’s expressions. The present studies showed that sup- social functioning, both because suppression may pressing distressing emotions not only does not de- limit access to new relationships and because it may crease aversive emotional experience but can actually hinder the maintenance and growth of existing rela- exacerbate it. Given the importance of excessive tionships. We observed initial interactions between negative emotion in defining many psychological dis- same-sex strangers and found that suppression led the orders such as depression and anxiety, our findings regulators’ partners to be less willing to establish a suggest the possibility that expressive suppression friendship. Initial measures of attraction and affilia- may play a maintaining role in psychopathological tion have been found to predict which relationships go conditions. on to become close and which do not (Berg & Clark, Another way that expressive suppression could im- 1986), implying that these first impressions are indeed pact both psychological and physical health is by lim- important for relationship formation. Furthermore, iting the regulator’s access to supportive relation- both theory and research suggest that intimacy de- ships. Social isolation has been linked with pends on an interactive process involving emotional psychological problems ranging from loneliness to self-disclosure and responsiveness (Berg & Derlega, suicide and has been repeatedly associated with both 1987; Laurenceau et al., 1998; Reis & Shaver, 1988). physical morbidity and mortality (Cohen & Thomas, Thus, during subsequent interactions, the suppressors’ 1985; House et al., 1988; Seeman, 2001; Uchino et reduced expressivity and responsiveness could further al., 1996). Although there is debate as to the precise hinder the development of close relationships. mechanism by which social support fosters health, the Even if some degree of intimacy were to be estab- fact that it does so is uncontested. Thus, if the chronic SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 63 use of suppression increases social isolation and mimicry and contagion (Chartrand & Bargh, 1999; weakens social bonds, we would expect suppression Neumann & Strack, 2000). Given this analysis, we to lead to adverse health consequences. would expect beneficial outcomes of expressive sup- Yet another route by which expressive suppression pression when it (a) served to interrupt negative emo- could impact physical health is via the increased car- tion escalation, including anxiety, hostility, and ag- diovascular responding that we observed in both the gression; and (b) increased interpersonal distance regulators and their partners. Repeated increases in when such distance was desirable. An example of the blood pressure that are in excess of metabolic demand latter that has been empirically studied is the distinc- have been implicated in the development of coronary tion between exchange and communal relationships heart disease (Krantz & Manuck, 1984). Indeed, one (Berg & Clark, 1986; Clark & Taraban, 1991). In of the mechanisms by which social support may foster exchange relationships, individuals engage in a tit- health is by reducing such physiological stress re- for-tat kind of interaction, such as exchanging goods sponses (Christenfeld et al., 1997; Cohen & Thomas, or services. Empirical work has demonstrated that in 1985; Glynn et al., 1999; Lepore, 1995; Lepore et al., this context, people prefer reduced levels of emotion 1993; Uchino et al., 1996). In this regard, expressive expression, and so some degree of suppression would suppression may be the “second-hand smoke” of emo- be normative and should help to preserve smooth in- tion regulation—it not only exacerbates the suppres- teraction (Clark & Taraban, 1991). sors’ own physiological responding in response to an Another important moderator may be whether ex- emotionally taxing conversation but it also increases pressive suppression is used flexibly and sparingly or the cardiovascular responding of their interaction inflexibly and chronically. If a single use of expres- partners as well. If such a pattern were repeated fre- sive suppression serves to bring a halt to a fight, and quently enough in a relationship over the course of the partners later manage to resolve the problem, then decades, suppression may exact a toll on the physical the benefits of suppression may justify its short-term health of both partners (Krantz & Manuck, 1984). costs. If, however, the use of suppression were chronic, then the problem may never be adequately Is Expressive Suppression Ever a Good Thing? discussed, and hence may never be resolved. At the Social commentators have long argued that expres- same time, the suppressor and his or her partner sive suppression is a form of self-restraint that is es- would be repeatedly experiencing the relational and sential for social harmony (Elias, 1978). Our results physiological costs that we observed in our studies. In paint a much more negative picture of suppression. addition, it is possible that in some situations suppres- How can the tension between these two views be sion does not entail the acute costs that we have ob- resolved? served. If suppression were chronic, however, the We believe that the answer to this question requires regulator would not limit him- or herself to the low- us to examine the boundary conditions for the effects cost situations. Thus, the value of expressive suppres- we have identified in these two studies. We focused sion would likely depend on what emotions are being on conversations about a shared negative event be- suppressed, who they are being hidden from, what the cause these are common in everyday life and provide relational and cultural context is, and whether acute or a forum for social bonding and support (Cohen & chronic outcomes are paramount. Thomas, 1985; Luminet, Patrick, Manstead, & Rime, Limitations and Future Directions 2000; Reis & Shaver, 1988). We found that in this context, expressive suppression provided no benefits To the best of our knowledge, the present studies and entailed multiple costs. Nevertheless, not all so- represent the first direct experimental investigation of cial situations are like this one, and the consequences the social consequences of expressive suppression. of expressive suppression may be expected to vary Because we randomly assigned individuals to regulate across contexts. What parameters might determine their emotions in specified ways, we can be sure that when suppression would be unproblematic or perhaps the effects we observed were due to suppression per even beneficial? se rather than to some unmeasured characteristic of As we have seen, expressive suppression should individuals who typically suppress. Our results thus inhibit the development of intimacy. In addition, it give us confidence that inhibiting emotion expres- should impede the interpersonal escalation of both sions is sufficient, at least in the given context, to positive and negative emotions by disrupting the un- generate undesirable personal and social conse- intentional transmission of emotion via behavioral quences. 64 BUTLER ET AL.

One limitation of our approach, however, derives or interchanges that have a quite different affective directly from its core strength: It is possible that the character (e.g., a predominantly positive emotional instructions that we gave, or the social situation that tone) would have similar consequences. we created, do not faithfully represent actual events in A third important limitation of these studies is our everyday life. For example, if our experimental task focus on same-sex female dyads. One important di- was experienced by our participants as artificial, ei- rection for future work will be to examine the per- ther because suppression was instructed or because it sonal and social consequences of expressive suppres- seemed out of place, this artificiality itself might ac- sion in the context of male–male and mixed-sex count for the physiological and relational effects that dyads. Given evidence that men are more likely to use we observed. Although the generalizability of our re- expressive suppression than are women in everyday sults cannot be ascertained without further research, life (Gross & John, 2002), it is possible that the effects several factors suggest that artificiality alone cannot of suppression, versus uninstructed controls, will be account for our findings. First, across both studies, less evident for male–male pairs. As the generaliz- some individuals in our control groups appeared to ability of these findings to dyad types and social con- have spontaneously engaged in suppression, suggest- texts becomes clear, it will be important to further ing that suppression was normative in this situation probe the processes that mediate the personal and so- for at least some of our participants. One corollary of cial effects of expressive suppression. Only the sec- this observation is that our replicated findings may ond of our two studies permitted us to consider issues have been subjected to a particularly stringent test by of mediation, and we were only partially successful in comparing a “suppression” condition with an unin- our efforts to delineate the behavioral changes structed condition in which at least some participants wrought by suppression that were responsible for its suppressed. It is also noteworthy that during debrief- social consequences. Future studies will be needed if ing, our suppression participants reported that they we are to fully understand the processes by which this understood our instructions, tried hard to follow them, and other forms of emotion regulation exact social and generally felt that they had done a good job of costs. doing so. Even more telling, during debriefing not one These limitations notwithstanding, our results sug- of the partners of the suppressors showed any signs of gest the intriguing possibility that attempts to regulate awareness that her partner had been behaving in an negative emotional responses when interacting with artificial manner. All reported amazement that her another person can have personal and social conse- partner had received suppression instructions. Thus, quences spanning multiple domains, including behav- although further research is needed to fully explicate ior, subjective experience, physiological responding, the mechanisms underlying our effects, we do not and relationship outcomes. We regard these findings believe artificiality provides an adequate account. as an initial glimpse of an exciting new domain at the In a related vein, our use of an upsetting film to intersection of social and personality psychology. provide a comparable basis for moderately emotional conversations is also a limitation. Our goal was to References create a conversational context in which emotional al’Absi, M., Bongard, S., Buchanan, T., Pincomb, G. A., expression was appropriate but in which expressive Licinio, J., & Lovallo, W. R. (1997). Cardiovascular and suppression was also a plausible regulatory goal. This neuroendocrine adjustment to public speaking and mental led us to focus on conversations involving shared arithmetic stressors. Psychophysiology, 34, 266–275. emotional content. Such conversations are common in Baron, R. M., & Kenny, D. A. (1986). The moderator– everyday life (Luminet et al., 2000), are central to the mediator variable distinction in social psychological re- development of interpersonal bonds (Laurenceau et search: Conceptual, strategic, and statistical consider- al., 1998; Reis & Shaver, 1988), and are often char- ations. Journal of Personality and Social Psychology, 51, acterized by mutual emotional disclosure and support- 1173–1182. ive responses (Berg & Derlega, 1987; Laurenceau et Berg, J. H., & Clark, M. S. (1986). Differences in social al., 1998). Nevertheless, such expressiveness is not exchange between intimate and other relationships: inevitable, as demonstrated by the spontaneous sup- Gradually evolving or quickly apparent? In V. J. Derlega pression that occurred in our uninstructed control & B. A. Winstead (Eds.), Friendship and social interac- group. It is not known whether emotion regulation tion (pp. 101–128). New York: Springer-Verlag. during more intense emotional interchanges (e.g., Berg, J. H., & Derlega, V. J. (1987). Responsiveness and about more personally relevant conversation topics), self-disclosure. New York: Plenum Press. SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 65

Bernieri, F. J., Gillis, J. S., Davis, J. M., & Grahe, J. E. sponses on interpersonal attraction. Journal of Personal- (1996). Dyad rapport and the accuracy of its judgment ity and Social Psychology, 37, 534–550. across situations: A lens model analysis. Journal of Per- Egloff, B., Wilhelm, F. H., Neubauer, D. H., Mauss, I. B., & sonality and Social Psychology, 71, 110–129. Gross, J. J. (2002). Implicit anxiety measure predicts car- Bernieri, F. J., & Rosenthal, R. F. (1991). Interpersonal co- diovascular reactivity to an evaluated speaking task. ordination: Behavior matching and interactional syn- Emotion, 2, 3–11. chrony. In R. S. Feldman & B. Rime (Eds.), Fundamen- Elias, N. (1978). The civilizing process: The history of man- tals of nonverbal behavior (pp. 401–432). New York: ners. New York: Urizon Books. Cambridge University Press. Fridlund, A. J. (1994). Human facial expression: An evolu- Bush, L. K., Barr, C. L., McHugo, G. J., & Lanzetta, J. T. tionary view. San Diego, CA: Academic Press. (1989). The effects of facial control and facial mimicry Friedman, H. S., & Miller-Herringer, T. (1991). Nonverbal on subjective reactions to comedy routines. Motivation display of emotion in public and in private: Self- and Emotion, 13, 31–52. monitoring, personality, and expressive cues. Journal of Cacioppo, J. T., & Petty, R. E. (1986). Social processes. In Personality and Social Psychology, 61, 766–775. M. G. Coles, E. Donchin, & S. W. Porges (Eds.), Psy- Glynn, L. M., Christenfeld, N., & Gerin, W. (1999). Gen- chophysiology: Systems, processes, and applications (pp. der, social support, and cardiovascular responses to 646–679). New York: Guilford Press. stress. Psychosomatic Medicine, 61, 234–242. Cappella, J. N. (1997). Behavioral and judged coordination Gotlib, I. H. (1992). Interpersonal and cognitive aspects of in adult informal social interactions: Vocal and kinesic depression. Current Directions in Psychological Science, indicators. Journal of Personality and Social Psychology, 1, 149–154. 72, 119–131. Gottman, J. M., & Levenson, R. W. (1988). The social psychophysiology of marriage. In P. Noller & M. A. Fitz- Chartrand, T. L., & Bargh, J. A. (1999). The chameleon patrick (Eds.), Perspectives on marital interaction (pp. effect: The perception–behavior link and social interac- 182–200). Clevedon, England: Multilingual Matters. tion. Journal of Personality and Social Psychology, 76, Gottman, J. M., & Levenson, R. W. (1992). Marital pro- 893–910. cesses predictive of later dissolution: Behavior, physiol- Christenfeld, N., Gerin, W., Linder, W., Sanders, M., ogy, and health. Journal of Personality and Social Psy- Mathur, J., Deich, J. D., et al. (1997). Social support ef- chology, 63, 221–233. fects on cardiovascular reactivity: Is a stranger as effec- Gross, J. J. (1998a). Antecedent- and response-focused tive as a friend? Psychosomatic Medicine, 59, 388–398. emotion regulation: Divergent consequences for experi- Clark, H. H. (1996). Using language. Cambridge, England: ence, expression, and physiology. Journal of Personality Cambridge University Press. and Social Psychology, 74, 224–237. Clark, M. S., & Taraban, C. (1991). Reactions to and will- Gross, J. J. (1998b). The emerging field of emotion regula- ingness to express emotion in communal and exchange tion: An integrative review. Review of General Psychol- relationships. Journal of Experimental Social Psychol- ogy, 2, 271–299. ogy, 27, 324–336. Gross, J. J. (1999a). Emotion and emotion regulation. In Cohen, S., & Thomas, A. W. (1985). Stress, social support, L. A. Pervin & O. P. John (Eds.), Handbook of person- and the buffering hypothesis. Psychological Bulletin, 98, ality: Theory and research (2nd ed., pp. 525–552). New 310–357. York: Guilford Press. Collins, N. L., & Miller, L. C. (1994). Self-disclosure and Gross, J. J. (1999b). Emotion regulation: Past, present, fu- liking: A meta-analytic review. Psychological Bulletin, ture. Cognition and Emotion, 13, 551–573. 116, 457–475. Gross, J. J. (2002). Emotion regulation: Affective, cogni- Darwin, C. (1998). The expression of emotions in man and tive, and social consequences. Psychophysiology, 39, animals. (3rd ed.). New York: Oxford University Press. 281–291. (Original work published 1872) Gross, J. J., & John, O. P. (2002). Individual differences in Davis, D. (1982). Determinants of responsiveness in dyadic two emotion regulation processes: Implications for af- interactions. In W. Ickes & E. G. Knowles (Eds.), Per- fect, relationships, and well-being. Manuscript submitted sonality, roles and social behavior (pp. 85–140). New for publication. York: Springer-Verlag. Gross, J. J., & Levenson, R. W. (1993). Emotional suppres- Davis, D., & Perkowitz, W. T. (1979). Consequences of sion: Physiology, self-report, and expressive behavior. responsiveness in dyadic interaction: Effects of probabil- Journal of Personality and Social Psychology, 64, 970– ity of response and proportion of content-related re- 986. 66 BUTLER ET AL.

Gross, J. J., & Levenson, R. W. (1997). Hiding feelings: Intimacy as an interpersonal process: The importance of The acute effects of inhibiting negative and positive emo- self-disclosure and partner disclosure, and perceived part- tion. Journal of Abnormal Psychology, 106, 95–103. ner responsiveness in interpersonal exchanges. Journal of Harker, L., & Keltner, D. (2001). Expressions of positive Personality and Social Psychology, 74, 1238–1251. emotion in women’s college yearbook pictures and their Lepore, S. J. (1995). Cynicism, social support, and cardio- relationship to personality and life outcomes across adult- vascular reactivity. Health Psychology, 14, 210–216. hood. Journal of Personality and Social Psychology, 80, Lepore, S. J., Allen, K. A. M., & Evans, G. W. (1993). So- 112–124. cial support lowers cardiovascular reactivity to an acute Harris, C. R. (2001). Cardiovascular responses of embar- stressor. Psychosomatic Medicine, 55, 518–524. rassment and effects of emotional suppression in a social Levenson, R. W. (1994). Emotional control: Variation and setting. Journal of Personality and Social Psychology, consequences. In P. Ekman & R. J. Davidson (Eds.), The 81, 886–897. nature of emotion: Fundamental questions (pp. 273– Hornstein, G. A., & Truesdell, S. E. (1988). Development 279). New York: Oxford University Press. of intimate conversation in close relationships. Journal of Lieberman, M. D., & Rosenthal, R. (2001). Why introverts Social and Clinical Psychology, 7, 49–64. can’t always tell who likes them: Multitasking and non- House, J. S., Landis, K. R., & Umberson, D. (1988, July verbal decoding. Journal of Personality and Social Psy- 29). Social relationships and health. Science, 241, 540– chology, 80, 294–310. 545. Luminet, O. B., Patrick, D. F., Manstead, A. S. R., & Rime, Howell, D. C. (1992). Statistical methods for psychology B. (2000). Social sharing of emotion following exposure (3rd ed.). Belmont, CA: Duxbury Press. to a negatively valenced situation. Cognition and Emo- Kappas, A., & Descoteaux, J. (in press). Of butterflies and tion, 14, 661–688. roaring thunder: Nonverbal communication in interaction McCanne, T. R., & Anderson, J. A. (1987). Emotional re- and regulation of emotion. In P. Philippot, E. J. Coats, & sponding following experimental manipulation of facial R. S. Feldman (Eds.), Nonverbal behavior in clinical set- electromyographic activity. Journal of Personality and tings. New York: Oxford University Press. Social Psychology, 52, 759–768. Kashy, D. A., & Kenny, D. A. (1997). The analysis of data Neumann, R., & Strack, F. (2000). “Mood contagion”: The from dyads and groups. In H. Reis & C. M. Judd (Eds.), automatic transfer of mood between persons. Journal of Handbook of research methods in social psychology (pp. Personality and Social Psychology, 79, 211–223. 451–477). New York: Cambridge University Press. Reis, H. T., & Shaver, P. (1988). Intimacy as an interper- Keltner, D., & Kring, A. (1998). Emotion, social function, sonal process. In S. W. Duck (Ed.), Handbook of per- and psychopathology. Review of General Psychology, 2, sonal relationships (pp. 367–389). New York: Wiley. 320–342. Richards, J. M., & Gross, J. J. (1999). Composure at any Kennedy-Moore, E., & Watson, J. C. (2001). How and cost? The cognitive consequences of emotion suppres- when does emotional expression help? Review of General sion. Personality and Social Psychology Bulletin, 25, Psychology, 5, 187–212. 1033–1044. Kenny, D. A. (1988). The analysis of data from two-person Richards, J. M., & Gross, J. J. (2000). Emotion regulation relationships. In S. W. Duck (Ed.), Handbook of personal and memory: The cognitive costs of keeping one’s cool. relationships (pp. 57–77). New York: Wiley. Journal of Personality and Social Psychology, 79, 410– Kenny, D. A. (1996a). The design and analysis of social- 424. interaction research. Annual Review of Psychology, 47, Riggio, R. E. (1986). Assessment of basic social skills. 59–86. Journal of Personality and Social Psychology, 51, 649– Kenny, D. A. (1996b). Models of non-independence in dy- 660. adic research. Journal of Social and Personal Relation- Riggio, R. E., & Friedman, H. S. (1982). Impression forma- ships, 13, 279–294. tion: The role of expressive behavior. Journal of Person- Krantz, D. S., & Manuck, S. B. (1984). Acute psychophysi- ality and Social Psychology, 50, 421–427. ological reactivity and risk of cardiovascular disease: A Schwarzer, R., & Leppin, A. (1991). Social support and review and methodologic critique. Psychological Bulle- health: A theoretical and empirical overview. Journal of tin, 96, 435–464. Social and Personal Relationships, 8, 99–127. Kring, A. M., & Gordon, A. H. (1998). Sex differences in Seeman, T. (2001). How do others get under our skin? In emotion: Expression, experience, and physiology. Jour- C. D. Ryff & B. H. Singer (Eds.), Emotion, social rela- nal of Personality and Social Psychology, 74, 686–703. tionships, and health (pp. 189–210). New York: Oxford Laurenceau, J., Barrett, L. F., & Pietromonaco, P. R. (1998). University Press. SOCIAL CONSEQUENCES OF EXPRESSIVE SUPPRESSION 67

Strack, F., Martin, L. L., & Stepper, S. (1988). Inhibiting mechanisms and implications for health. Psychological and facilitating conditions of the human smile: A nonob- Bulletin, 119, 488–531. trusive test of the facial feedback hypothesis. Journal of Vander, A. J., Sherman, J. H., & Luciano, D. S. (1990). Hu- Personality and Social Psychology, 54, 768–777. man physiology (5th ed.). New York: McGraw-Hill. Tickle-Degnan, L., & Rosenthal, R. (1990). The nature of Zuckerman, M., Klorman, R., Larrance, D. T., & Spiegel, rapport and its nonverbal correlates. Psychological In- N. H. (1981). Facial, autonomic, and subjective compo- quiry, 1, 285–293. nents of emotion: The facial feedback hypothesis versus Tomaka, J., Blascovich, J., Kelsey, R. M., & Leitten, C. L. the externalizer–internalizer distinction. Journal of Per- (1993). Subjective, physiological, and behavioral effects sonality and Social Psychology, 41, 929–944. of threat and challenge appraisal. Journal of Personality and Social Psychology, 65, 248–260. Uchino, B., Cacioppo, J. T., & Kiekolt-Glaser, J. K. (1996). Received January 16, 2001 The relationship between social support and physiologi- Revision received September 27, 2002 cal processes: A review with emphasis in underlying Accepted September 27, 2002 ■