هراس اجتماعی حالت چهره احساسات را اشتباه تفسیر نمی کنند

Abstract Attentional biases in the processing of threatening facial expressions in social anxiety are well documented. It is generally assumed that these attentional biases originate in an evaluative bias: socially threatening information would be evaluated more negatively by socially anxious individuals. However, three studies have failed to evidence a negative evaluative bias in the processing of emotional facial expression (EFE) in socially anxious individuals. These studies however suffer from several methodological limitations that the present study has attempted to overcome. Twenty-one out-patients diagnosed with generalized social phobia have been compared to 20 out-patients diagnosed with another anxiety disorder and with 39 normal controls matched for gender, age and level of education. They had to decode on seven emotion intensity scales a set of 40 EFE whose intensity and emotional nature were manipulated. Although sufficient statistical power was ensured, no differences among groups could be found in terms of decoding accuracy, attributed emotion intensity, or reported difficulty of the task. Based on these findings as well as on other evidences, we propose that, if they exist, evaluative biases in social anxiety should be implicit and automatic and that they might be determined by the relevance of the stimulus to the person’s concern rather than by the stimulus valence. The implications of these findings for the interpersonal processes involved in social phobia are discussed.

. Introduction Etiological models (Beck, Emery, & Greenberg, 1985; Clark & Wells, 1995; Rapee & Heimberg, 1997) as well as empirical evidences (see Clark & McManus, 2002; Heinrichs & Hofmann, 2001; Musa & Lépine, 2000, for reviews) support the idea that cognitive processes play a key role in social phobia. The interpretation of social information seems to be specifically at stake (Amir, Foa, & Coles, 1998; Stopa & Clark, 2000), specially the information related to negative social evaluation and its negative consequences: social humiliation and embarrassment. In this perspective, the processing of non-verbal information is likely to be particularly important, as social disapproval is rarely verbally and directly expressed due to social convention. Given that emotional facial expressions (EFE) are the most used channel to convey inner states such as emotion and attitude (Patterson, 1999), the ability to decode them might constitute a cognitive skill particularly relevant for the explanation of social anxiety. Several lines of research have examined possible biases in the way socially anxious individuals process EFE. The most fecund line has addressed whether socially anxious individuals present an attentional bias for threatening EFE (Bradley, Mogg, & Millar, 2000; Chen, Ehlers, Clark, & Mansell, 2002; Gilboa-Schechtman, Foa, & Amir, 1999; Mansell, Clark, Ehlers, & Chen,1999; Mogg & Bradley, 2002). Articulating this literature, several authors (e.g. Mogg & Bradley, 2002) have proposed a two-stage model in which anxious individuals show an initial hypervigilance for threatening EFE. This hypervigilance would be the consequence of automatic processes and it could be observed without conscious perception of threat-relevant information (Mogg & Bradley, 2002). However, at further and less automatic stages of information processing, anxious people would actively turn away from threatening information. Thus, this model postulates a dynamic shift of attention allocation from initial threat hypervigilance to later threat avoidance. Other lines of research have addressed the issues of memory and evaluative biases for EFE in social anxiety. There have been surprisingly few studies conducted on these topics. Concerning memory biases for threatening EFE, three studies have provided inconclusive results: Lundh and Öst (1996) as well as Foa, Gilboa-Schechtman, Amir, and Freshman (2000) have observed that socially anxious individuals exhibit a negative memory bias towards threatening EFE. In contrast, Perez-Lopez and Woody (2001) failed to observe such a bias. Concerning evaluative biases in EFE decoding, to our knowledge, only three studies on adults have been published. Merckelbach, Van Hout, Van den Hout, and Mersch (1989) observed no differences between nine social phobics and nine controls in their evaluation of angry, neutral, and joyful faces with respect to pleasantness. We recently replicated this intriguing result (Douilliez & Philippot, 2003): socially anxious and non-anxious students were asked to evaluate the threatening value of angry-threatening, joyful, and neutral faces. No differences between anxious individuals and controls were observed for the evaluation of faces. However, some differences between socially anxious and non-anxious were found in a study using a different procedure: Winton, Clark, and Edelmann (1995) very briefly (60 ms) exposed socially anxious and non-anxious students to negative and neutral facial expressions. Following this subliminal presentation, participants had to guess whether the EFE was negative or neutral. The results showed that students high in social anxiety identified negative EFE better than non-anxious students while the reverse was true for neutral expressions. However, further analyses demonstrated that these effects were due to an overall negativity bias in the anxious students and the authors conclude to an “absence of an enhanced ability to discriminate between different emotional states in others” (p. 193) in their socially anxious participants. Some studies have also attempted to relate EFE decoding skills and social anxiety in children. Mc Clure and Nowicki (2001) did not find relationship between self-reported social anxiety and the ability to correctly label EFE presented for 1 s to 8–10 years old. In a study similar to Winton et al. (1995), Melfsen and Florin (2002) exposed 8–12 years old to EFE for 60 ms. Children had to guess whether the EFE presented was positive, neutral or negative. Socially anxious children reported more often that they saw emotion when a neutral face was presented. However, there was neither an indication of an enhanced ability to decode negative facial expressions in socially anxious children, nor was there a specific tendency to interpret neutral or positive faces as negative. In sum, the few existing studies have failed to evidence any consistent evaluative bias in EFE decoding in socially anxious adults or children. This fact is surprising for several reasons. First, most cognitive models of social phobia (e.g. Clark & Wells, 1995; Rapee & Heimberg, 1997) postulate that socially anxious individuals over-attribute a meaning of social threat to social signals. This implicates that they should decode more threat in EFE, and thus show a negativity bias. Second, there is a strong belief in the existence of such a bias among practitioners (Beck, 1976; Beck, Emery, & Greenberg, 1985). A third reason is that most models of attentional biases postulate the existence of an evaluative bias prior to the attentional bias. For instance, the cognitivo-motivational model of anxiety of Mogg and Bradley (1998) relies on two different systems: The Valence Evaluation System and the Goal Engagement System. The Valence Evaluation System automatically assesses the stimulus threat value according to the nature of the stimulus, the situational context, the person’s state anxiety and prior learning experiences. The Goal Engagement System orients allocation of attention as a function of the output of the former system. If a stimulus in the environment is evaluated as threatening, the Goal Engagement System interrupts ongoing activities and orients attention toward the threat stimulus. This model thus postulates that attentional biases in anxious individuals result from a negative and automatic appraisal of social situations (Mogg & Bradley, 2002). One critical element might be the explicit versus implicit/automatic nature of hypothetical evaluative biases in social anxiety. The cognitivo-motivational model of anxiety of Mogg and Bradley (1998) is unequivocal on the implicit/automatic nature of their Valence Evaluation System. However, all five studies reviewed above requested participants to explicitly evaluate EFE. It is to be noted that two of the five studies used subliminal presentations and might thus have been more sensitive to implicit bias. Interestingly, they were the only ones to evidence some differences between socially anxious and non-anxious participants, although not in the sense of the expected bias for negative EFE, but in the sense of an overall negativity (Winton et al., 1995) or overall emotionality (Melfsen & Florin, 2002) bias. Still, no firm conclusion can be drawn from the very few existing studies that have examined explicit evaluative bias in social anxiety. Yet, establishing the presence or the absence of a significant explicit evaluative bias bares important implications for social anxiety at the etiological as well as clinical levels. Indeed, intervention strategies are markedly different if they address an explicit evaluative (appraisal) dysfunction versus an implicit and automatic process dysfunction (McNally, 1995). In the former case, cognitive restructuring might be the preferred intervention, while in the latter case exposure might be favored. That no firm conclusion can be drawn from existing studies is largely due to the fact that their interpretation is limited by a number of methodological factors related to participant sampling, and to the choice of appropriate control groups, experimental stimuli and dependent measures. Regarding sampling, Merckelbach et al. (1989) tested only 9 social phobic patients and 9 matched controls. This study might thus have lacked statistical power to evidence differences between phobic patients and controls. Douilliez and Philippot’s (2003) sample was larger, but it consisted in a non-clinical population of socially anxious and non-anxious students. In this case, it might be that the difference in social anxiety between the two groups was not sufficient to establish statistically significant differences. Similarly, Winton et al. (1995) used small non-clinical samples of socially anxious (n=13) and non-anxious (n=11) students. All these studies used normal, non-socially anxious individuals as control group. However, to ascertain that potential differences are proper to social anxiety, an additional control group should be considered: anxious patients not suffering from social anxiety. Regarding experimental stimuli, all studies have used extreme prototypical EFE expressing anger-threat, neutral state, or joy. Not only do they have little ecological validity, but they are also easy to decode and the use of such a material is likely to produce ceiling effects (Hess, Blairy, & Kleck, 1997). Further, such extreme stimuli might have acted as UCs (Öhman, 1996; Öhman & Soares, 1993) and have left little room for individual variance. None of the studies used EFE expressing disapproval or rejection, such as disgust or contempt, which would have been particularly appropriate for the present question. Finally, Merckelbach et al. (1989) and Douilliez and Philippot (2003) used an unidimensional dependent measure (pleasantness or threat). A wider array of dependent measures would allow more opportunity of evidencing a potential evaluative bias in the decoding of EFE in social phobia. In sum, despite the theoretical as well as clinical importance of this issue, we are still lacking conclusive evidences regarding whether social phobics present an explicit evaluative bias for social threat and reject in the decoding of EFE. To attempt to answer satisfactorily this question, we have presented a large set of photographs of EFE to patients diagnosed with generalized social phobia, patients suffering from an anxiety disorder without presenting social phobia, and normal controls who were matched for sex, age, and level of education. EFE stimuli were portraying joy, anger, fear, sadness and disgust. The latter emotion was used because pre-tests had shown that it was interpreted as a strong signal of rejection. To avoid ceiling effects and to use a material reflecting real life expressions, stimuli varied in the level of emotional intensity of the expression. To increase the sensitivity of our measures, participants had to rate each EFE on a large emotion profile, including items such as shame or disgust. We had used the exact same procedure (stimuli and dependent measures) to successfully evidence evaluative biases in alcoholics (e.g. Philippot, Kornreich, Blairy, Baert, & Den Dulk, 1999). Finally, mood and individual characteristics were statistically controlled with measures of depression, anxiety, and social anxiety.

3. Results The data analysis proceeded in three steps, each addressing a specific question. It was examined whether the clinical status of the participants altered, first, the emotional profile attributed to facial expression on the basis of the seven emotion scales, second, the level of accuracy in the decoding of EFE, third, the level of difficulty attributed to the decoding task. 3.1. Emotional profiles attributed to EFEs For each of the seven emotion scale, intensity ratings attributed to a given facial expression were averaged across the two actors portraying the same facial expression. On these scores, a MANOVA2 was computed with emotion portrayed by the stimulus (joy, anger, fear, sadness, disgust), intensity of the emotion portrayed (0% or neutral, 30%, 70%, and 100%), and emotion rating scale (joy, anger, fear, sadness, disgust, shame and surprise) as within-subject factors and experimental group (social phobics, anxious controls, and normal controls) as between-subjects factor. No effect or interaction involving group approached even a statistical tendency (all p’s>.12). This cannot be attributed to a lack of statistical power, as the observed power for the Stimulus emotion X Emotion scale X Group interaction, which is the most relevant effect for the present question, is very high: .97, and its η2 very low: .028; F(48, 110)=1.22, p>.19. Other effects were clearly significant, all p’s<.001: main effects for stimulus emotion : F(4, 74)=29.86, for stimulus intensity : F(3, 75)=64.29, for emotion scale: F(6, 72)=29.48; stimulus emotion X Emotion scale interaction: F(24, 54)=61.32, and Stimulus emotion X Stimulus Intensity: F(12, 66)=19.18. In sum, the profiles of emotional state attributed to the facial stimuli by the participants were modulated, as expected, by the type of emotion portrayed as well as by their intensity. However, it was not affected by the clinical status of the participants. An additional analysis was computed on the emotion scale intensity ratings attributed to the neutral expressions. As neutral facial expressions can been seen as ambiguous (Wallbott, 1988), this specific part of the task is the most likely to reveal an evaluative bias. Indeed, in such condition, people tend to rely on a priori representations rather than on stimulus information (Petty & Cacioppo, 1986). Emotion intensity ratings were aggregated by averaging, for each emotion scale, the ratings attributed to each of the 10 neutral facial expressions. On these scores, a MANOVA was computed with the seven emotion scales as within-subject factor and experimental group (social phobics, anxious controls, and normal controls) as between-subjects factor. Again, no effect or interaction including group was observed; for the group main effect, F(2, 80)=.12, p >.88, the observed power was .068 (η2=.003η2=.003), and for the interaction with emotion scales, F(12, 152)=.89, p >.55, power=.51 (η2=.025η2=.025). Finally, correlations were computed within the control group and within the social phobic group between emotion intensity ratings for neutral facial expression and scores of depression (BDI), anxiety (STAI) and social phobia (FNE). None of these correlations reached statistical significance. 3.2. Accuracy in the decoding of facial expression Accuracy scores for each facial stimulus were averaged across the two actors portraying the same facial expression. On these scores, a MANOVA was computed with emotion portrayed by the stimulus (joy, anger, fear, sadness, disgust), and intensity of the emotion portrayed (30%, 70%, and 100%), as within-subject factors and experimental group (social phobics, anxious controls, and normal controls) as between-subjects factor. No effect or interaction involving group approached statistical tendency (all p ’s>.28). The statistical power of the main effect of group is: .28 (η2=.033η2=.033), for F(2, 77)=1.31, p >.27, and the power of the interactions involving group, all relevant for the present question, varies between .36 and .55 (η2η2 between .019 and .033). Other effects were clearly significant, all p’s<.001: main effects for stimulus emotion : F(4, 74)=65.83, for stimulus intensity : F(2, 76)=307.01; Stimulus emotion X Stimulus Intensity interaction: F(8, 70)=24.74. Table 2 illustrates the pattern of these findings. As expected, more intense stimuli were more accurately decoded. Joy expression yielded the most accurate decoding, followed by disgust and sadness, while fear and anger were the least accurately decoded facial expression. Intensity of the stimulus had more impact for anger and fear than for other emotions. Table 2. Decoding accuracy (mean percentage of correctly identified facial expression; standard error into parentheses) as a function of the emotion portrayed by the facial expression and its intensity Emotion portrayed Facial expression intensity Mean 30% 70% 100% Joy .72a (.04) .93a (.02) .88a (.03) .84a (.02) Anger .14d (.03) .55c (.04) .85a (.03) .54c (.02) Fear .08d (.03) .75b (.04) .77a (.04) .56c (.02) Sadness .53b (.04) .76b (.03) .82a (.04) .70b (.03) Disgust .35c (.04) .90a (.03) .84a (.03) .70b (.02) Mean .37 (.02) .78 (.02) .83 (.02) Note: Superscripts correspond to comparisons among emotions portrayed by the facial expression stimuli. Means with different superscripts differ at least at the p<.05 level. Table options Accuracy scores for each emotion were computed by averaging scores across intensities. A general accuracy score was also computed by averaging all accuracy scores. These scores were correlated within the control group and within the social phobic group with the scores of depression (BDI), anxiety (STAI) and social anxiety (FNE). No correlation reached statistical significance. 3.3. Difficulty judgment of the task Difficulty ratings for each facial stimulus were averaged across the two actors portraying the same facial expression. On these scores, a MANOVA was computed with emotion portrayed by the stimulus (joy, anger, fear, sadness, disgust), and intensity of the emotion portrayed (0% or neutral, 30%, 70%, and 100%), as within-subject factors and experimental group (social phobics, anxious controls, and normal controls) as between-subjects factor. No effect or interaction involving group approached even a statistical tendency (all p ’s>.43). The statistical power of the main effect of group is: .19 (η2=.022η2=.022), for F(2, 77)=.86, p >.43, and the power of the interactions involving group, all relevant for the present question varies between .22 and .72, (η2η2 between .014 and .033). Other effects were clearly significant, all p’s<.001: main effects for stimulus emotion : F(4, 74)=9.36, for stimulus intensity: F(3, 75)=20.75; Stimulus emotion X Stimulus Intensity interaction: F(12, 66)=7.02. As can be seen in Table 3, the pattern of the difficulty ratings is parallel to the pattern of accuracy scores. Table 3. Mean difficulty ratings as a function of the emotion portrayed by the facial expression and its intensity (standard error into parentheses) Emotion portrayed Facial expression intensity Mean 0% 30% 70% 100% Joy 3.59a (.16) 3.13b (.20) 2.26b (.17) 2.30b (.17) 2.82b (.14) Anger 2.95b (.16) 3.87a (.19) 3.29a (.19) 2.81ab (.17) 3.23a (.14) Fear 3.56a (.18) 3.63ab (.19) 3.14a (.15) 2.92a (.15) 3.31a (.14) Sadness 3.54a (.19) 3.37b (.17) 3.02a (.16) 2.87a (.16) 3.20b (.14) Disgust 3.41a (.16) 3.41b (.18) 2.52b (.17) 2.54b (.14) 2.97ab (.13) Mean 3.41 (.14) 3.48 (.14) 2.84 (.13) 2.69 (.13) Note: Superscripts correspond to comparisons among emotions portrayed by the facial expression stimuli. Means with different superscripts differ at least at the p<.05 level. Table options Difficulty scores for each emotion were computed by averaging scores across intensities. A general difficulty score was also computed by averaging all difficulty scores. These scores were correlated within the control group and within the social phobic group, with the scores of depression (BDI), anxiety (STAI) and social phobia (FNE). No correlation reached significance in the control group. The pattern of correlations for the social phobic group is displayed in Table 4. It can be seen that the more anxious as a trait participants are, the greater difficulty they experienced during the decoding task. This is particularly true for FNE and for disgust expression. The patterns of correlations for the STAI-trait and the FNE are very similar. As the STAI-trait scale is sensitive to many forms of anxiety, including social anxiety, one could wonder whether the correlations between difficulty scores and the STAI-trait are due to FNE uniquely or to other form of anxiety as well. To address this question, partial correlations were computed between difficulty ratings and STAI-trait, while controlling for FNE. None of these correlations approached a level of statistical tendency (all p’s>.20, one-tailed). However, partial correlations between the difficulty ratings and the FNE, while controlling for STAI-trait, remained significant (r=.43 with total difficulty, r=.45 with difficulty for anger, r=.38 with difficulty for disgust, and r=.43 with difficulty for fear) but for joy and sadness. Thus, it can be concluded that it is social anxiety, and more specifically FNE, that is predictive of difficulty ratings in the decoding task. Interestingly, anxious mood during the task (Stai-State score) is not related to difficulty ratings, so is the score of depression. It is thus unlikely that observed significant correlations are mediated by mood. Table 4. Correlations between difficulty ratings of the decoding task and depression (BDI), state and trait anxiety (STAI-S, STAI-T) and social anxiety (FNE) in the social phobics group BDI STAI State STAI Trait FNE Overall difficulty .29 .13 .47* .52* Difficulty for anger .11 .11 .32 .43* Difficulty for joy .10 −.12 .34 .31 Difficulty for disgust .42 .18 .57* .59* Difficulty for sadness .36 .34 .45* .38 Difficulty for fear .19 .03 .23 .32 Note: *Correlation is significant at the .05 level (1-tailed).