رابطه پاسخ موثر به مقایسه اجتماعی و عملکرد تحصیلی در دبیرستان

Abstract The goal of the present study was to study the relationship between affective responses to social comparison and test scores among high school students. Our analyses showed that three types of responses to social comparison could be distinguished: an empathic, constructive, and destructive response. Whereas girls scored higher on empathic response, boys scored higher on destructive response. In addition, students who had a high social comparison orientation (SCO) scored higher on all three types of responses than students who expressed a low SCO. Multilevel regression analyses indicated that, after controlling for previous performance, a destructive response was negatively related to performance on tests for reading comprehension and mathematics. An empathic response was positively related to performance on reading comprehension only whereas a constructive response compensated the negative relationship between destructive response and reading comprehension. Theoretical and practical implications are discussed.

1. Introduction At school, the typical classroom is characterized by a strongly evaluative atmosphere, because both teachers and parents are concerned with academic accomplishments. The evaluation of a student’s performance at school partly depends on how well his or her classmates are doing. That is, a high grade is not outstanding anymore when many classmates receive a high grade or an even higher one. Because of the evaluative atmosphere and the dependence of evaluation outcomes on the performance of classmates, comparisons of grades between students are very common (Dijkstra et al., 2008, Levine, 1983 and Pepitone, 1972). Students know fairly well how their performance relates to that of classmates since grades are usually made known in class. Both children at primary school (e.g., Dumas, Huguet, Monteil, Rastoul, & Nezlek, 2005) and secondary school (e.g., Blanton et al., 1999, Huguet et al., 2001, Huguet et al., 2009 and Wehrens et al., in press) have been shown to compare their grades and performances to those of classmates. Social comparisons such as these, help students make appraisals of their abilities and opinions and, as a result, may profoundly affect students’ self-concepts and self-esteem, in both positive and negative ways (e.g., Huguet et al., 2009 and Marsh and Parker, 1984). Social comparisons are not merely a matter of choice (Huguet et al., 2009 and Levine, 1983). Due to the structure of the classroom and its strong evaluative nature, many comparisons are ‘forced’ upon children, as are its consequences. Even in individualized classrooms explicitly designed to minimize grade competition by allowing children to move at their own pace and to work at tasks within their own level of competence, a good deal of these social comparisons occur (Crockenberg and Bryant, 1978 and Levine, 1983). 1.1. The development of social comparison in the classroom Evidence has been found that children as young as preschoolers are capable of comparing themselves with classmates (e.g., Chafel, 1984 and Masters, 1969). In general, however, studies show that children’s ability-related self-evaluations are little affected by social comparisons until age 7 or 8 (e.g., Cremeens et al., 2007, Ruble et al., 1980, Schoeneman et al., 1984 and Veroff, 1969). From that age on, children become increasingly receptive to social comparisons. More specifically, Keil, McClintock, Kramer, and Platow (1990) reported that 42% of second graders used social comparisons to make self and other evaluations, whereas 56% of fourth graders, 68% of sixth graders, and 76% of eighth graders did. Similar findings have been reported by Aboud, 1976, Bear et al., 1998, Boggiano and Ruble, 1979, Feld et al., 1979, Ruble et al., 1994, Spear and Armstrong, 1978, Stipek and Tannatt, 1984 and Xiang et al., 2001. Pupils’ attention to social comparisons as a means of self-evaluation increases even more upon entry to secondary school. Following a school transition, students usually re-evaluate their scholastic competence, given their new reference groups. In addition, the shift from elementary to (junior) high school is usually associated with an increase in whole class task organization, between-classroom ability grouping, and an external emphasis on academic performance; practices that increase the salience of social comparison information (Feldlaufer et al., 1988 and Harter et al., 1992). In general, therefore, secondary schools are more performance-focused (i.e., oriented at achieving well in comparison with others) and less mastery-focused (i.e., oriented at achieving well for the sake of competence only; Kumar, 2006). The present study examined social comparisons in the classroom of high school students in grade 9 (age about 15 years). As pointed out above, at this age, the need for social comparison is high. In addition, because, at this age, adolescents develop their identity (Krayer, Ingledew, & Iphofen, 2008), social comparisons at this age may have relatively profound effects on students’ self-evaluations, self-concept, motivation and performances (e.g., Ryant & Pintrich, 1998). Therefore, studying social comparisons and their consequences among secondary school students seems a highly relevant undertaking. 1.2. Affective consequences of social comparison As noted before, social comparisons in the classroom may have both negative and positive consequences. In general, social comparisons with classmates make students feel and think more negative about themselves (e.g., Levine, 1983). Illustrative is the widely observed Big-Fish–Little-Pond-Effect (Marsh, 1991), i.e., the finding that pupils in higher ability schools show substantially lower academic self-concept than equally intelligent and performing pupils in lower ability schools, and that negatively influences academic performance. Likewise, social comparison with classmates may evoke evaluative anxiety, especially when the academic standard in the classroom is high (Bossong, 1985, Butler, 1989, Gröbel and Schwarzer, 1982 and Zeidner and Schleyer, 1999). Research, however, also shows that, although students may feel and think worse about themselves, comparisons with better performing others, may lead to better performances later in the school year (e.g., Blanton et al., 1999, Huguet et al., 2001, Huguet et al., 2009 and Wehrens et al., in press). In their review on social comparisons in the classroom, Dijkstra et al. (2008) conclude that especially social comparison (with those who perform better) seems to be a two-edged sword: although they may lead students to do better, it makes them feel and think worse about themselves. Recently, studies have confirmed the co-existence of these two contradictory effects of social comparisons ( Huguet et al., 2009 and Seaton et al., 2008). One type of consequences of social comparisons are students’ ‘affective responses’. In general, students’ emotions may have far reaching influences on students’ attention, the way they organize information, their efforts to learn, their motivation, their interests, their interactions with peers and teachers, their achievements and their self-concepts (e.g., Pekrun et al., 2009 and Schutz and Pekrun, 2007). It has, for instance, been found that whereas negative emotions, such as anxiety and boredom, are negatively related to motivation and achievement, positive emotions, such as enjoyment and pride, are positively related to motivation and achievement (e.g., Ahmed, in preparation and Pekrun et al., 2009). Despite the importance of emotions in the classroom, students’ affective responses to social comparisons in the classroom have only been scarcely the topic of study (for an overview, see Dijkstra et al., 2008). Both Smith, 2000 and Buunk et al., 2005 propose a model of affective responses to social comparisons in the classroom in which they distinguish between twelve (Smith, 2000) respectively eight (Buunk et al., 2005) affective responses. These models are both based on three underlying dimensions of the social comparison process. First, the specific affective response that is evoked by social comparisons in the classroom depends on the direction of the comparison. In general, it is assumed that downward comparison (i.e., comparison with a worse performing target) will make individuals realize that they are better off than the target and, as a result, evoke a positive response (e.g., Gibbons, 1986 and Wills, 1981). In contrast, individuals will usually interpret upward comparison (i.e., comparison with a better performing target) as evidence that they are worse off than the target and, as a result, evoke a negative response (e.g., Salovey and Rodin, 1984 and Tesser et al., 1988). However, according to Buunk, Collins, Taylor, Van Yperen, and Dakof (1990), the affective reactions people show to social comparisons are not merely intrinsic to its direction: a comparison can be interpreted in different ways, resulting in different responses. Therefore, the second dimension underlying the models of Buunk et al., 2005 and Smith, 2000 is the interpretation of social comparison information. Learning that another is better off than oneself, provides at least two pieces of information: (a) that one is not as well off as everyone and (b) that it is possible to be better than one is at present. Those who contrast themselves with a better off target, thinking, for instance, that they will never be able to do as well as the target, may feel worse about themselves. In contrast, those who identify with the better off target, thinking, for instance, that they may become as good as the target, may feel better about themselves. In a similar vein, learning that someone else is worse off provides at least two pieces of information: (a) that one is not as badly off as everyone and (b) that it is possible to get worse. Individuals who contrast themselves with someone who is worse off, thinking, for instance, that they are totally different from the target and will never end up like him or her, may feel better about themselves. In contrast, those who identify with the worse off target, thinking that they themselves may become just like him or her, are likely to feel worse about themselves. The third dimension in the models of Buunk et al., 2005 and Smith, 2000 is students’ focus: students may focus either on themselves, the comparison target or take a dual perspective. Depending on their focus, students may experience different types of affect (see Table 1). Table 1. Buunk et al., 2005 and Smith, 2000 models of affective responses to social comparisons. Comparison type Buunk et al. (2005) Smith (2000) Upward Identification Self-focus Hope Optimism Other-focus Sympathetic enjoyment Admiration Dual focus X Inspiration Contrast Self-focus Frustration Depression/shame Other-focus Resentment Resentment Dual focus X Envy Downward Identification Self-focus Worry Fear/worry Other-focus Compassion Pity Dual focus X Sympathy Contrast Self-focus Relief Pride Other-focus Contempt Contempt/scorn Dual focus X Schadenfreude Table options In sum, both Buunk et al., 2005 and Smith, 2000 propose a model in which affective responses to social comparisons are a function of direction, interpretation and focus. Although there are some differences between both models in the labels of the affective responses, and Smith’s model also includes emotions that result from taking a dual focus (i.e., focus on both self and other), whereas Buunk et al.’s model does not, the two models are highly similar. As noted before, social comparisons in the classroom may have far reaching consequences: the way in which students respond to social comparisons with their classmates may affect students’ motivation and satisfaction, and, as a consequence, their performances (e.g., Buunk et al., 2005). The second goal of the present research is therefore to examine the extent to which affective responses to social comparison information are related to actual performances. The present study will examine the affective responses, that according to these models, high school students may experience in response to social comparisons with class mates. In this context it is important to note that Carmona, Buunk, Dijkstra and Peiro (2008) examined the mediating role of students’ social comparison responses in the relationship between goal orientation and self-efficacy. Carmona et al. based their study on the identification-contrast model of Buunk and Ybema (1997) that combines two of the dimensions distinguished by the models described above, i.e., comparison direction and interpretation of social comparison information. In so doing, Carmona et al. did not frame or interpret social comparison responses in affective terms, as the present study did, but merely in terms of identification and contrast and upward and downward comparisons, and the resulting cognitions with regard to self-efficacy. Another difference between the present study and Carmona et al.’s study is that we examined the responses of thousands of high school students, whereas, in Carmona et al.’s study, 120 university students participated. Carmona et al.’s showed that only one of the four social comparison responses they distinguished (upward-identification, upward-contrast, downward-identification, downward-contrast) had a mediating role, i.e., the upward-contrast response: the tendency to contrast oneself with others who were doing better mediated the relationship between a prevention goal orientation and self-efficacy. The present study aims to take a different perspective and takes affect as the central concept, with different affects constituting of a combination of different social comparisons in terms of comparison direction and the interpretation of social comparison information. We chose to do so because, first, according to the models of Smith and Buunk et al., affect is the central outcome of different types of social comparisons. Second, although it is commonly believed that emotions are an important part of learning and teaching, they constitute a severely understudied topic in educational research (Gumora and Arsenio, 2002 and Hazari et al., 2007). Gumora and Arsenio (2002), however, showed the importance of emotions in the classroom: they found negative affect to be negatively related to GPA and achievement scores and to contribute to students’ GPA, over and above the influence of cognitive variables, such as academic self-concept. In sum, taking the emotion perspective seems a worth while undertaking and may uncover new insights with regard to the role of social comparisons in the classroom. 1.3. Gender differences in affective responses We expected boys and girls to differ in their affective responses to social comparisons. In general, women and girls are more concerned with other’s wellbeing than men and boys of the same age: they relate their own feelings more to those of others and have a greater empathic disposition (e.g., Mestre, Samper, Frias, & Tur, 2009). Rueckert and Nyabar (2008) even found a possible neural basis for gender differences in empathy: whereas they found, in women, activation in the right hemisphere to be related to levels of empathy, in men they did not find this association. In contrast, men, in general, are more competitive than women, tending to contrast themselves to others, or at least to take a dual focus (e.g., Räikkönen, Keskivaara, & Keltikangas-Järvinen, 1992). Males’ stronger tendency to compete has been attributed to the reproductive advantages men obtain from having a high status in a group’s dominance hierarchy. In sum, we expected that female students would show more other-focus and more identification with the comparison target than male students, whereas male students would show more contrast with the comparison target and show self or dual focus. In other words, in terms of Buunk et al., 2005 and Smith, 2000 models, we expected girls, more than boys, to experience responses reflecting sympathy and pity, and boys, more than girls, to experience affective responses such as resentment and envy. In line with this expectation, previous research has already shown girls to report more altruistic and empathic responses (e.g., I thought it was nice for the other person) and boys to report more egocentric and hostile responses (e.g., I envied the other person) to comparison targets (Buunk et al., 2005). 1.4. Social comparison orientation Several researchers have theorized that people may differ in their disposition to compare themselves with others. According to Gibbons and Buunk (1999), the extent to which and the frequency with which people compare themselves with others varies from one individual to the next, an individual difference variable they call ‘social comparison orientation’ (SCO). People high in SCO do not seem to have a preference for comparisons with better or worse performing others. Instead, they compare themselves more in either direction than people low in SCO. Relative to individuals with a low SCO, individuals with a high SCO are more interested in comparisons, seek out more comparisons, spend more time engaging in comparisons, and base their personal risk perceptions (more) on comparisons with others (Buunk & Gibbons, 2006). Especially relevant to the present study is that people high in SCO show stronger affective responses following social comparison than people low in SCO (e.g., Buunk et al., 2003 and Buunk et al., 2005). When studying social comparisons, SCO is therefore an important individual difference variable. More specifically, Buunk et al. (2005) explicitly refer to the importance of examining the role of SCO in the importance of social comparisons of grades to students’ performances. Research on the role of SCO in the classroom is, however, extremely scarce. To date only one study has examined students’ SCO, finding that, as students were higher in SCO, they adopted stronger mastery goals, performance-approach goals and performance-avoidance goals (Régner, Escribe, & Dupeyrat, 2007). The present study aimed to fill this void in the literature and further study the role of SCO in the classroom. In line with findings from Buunk et al. (2003), and Buunk et al. (2005), we expected to find stronger affective responses to social comparison among students with a high SCO as opposed to students with a low SCO. 1.5. The present study, and summary of hypotheses The present study examined the affective responses of high school students to social comparisons. We first examined the type of affective responses high school students experienced in response to social comparisons with class mates, regardless of the direction (upward vs. downward) of the comparison. We expected girls to differ from boys in their responses, with girls, more than boys, showing affective responses reflecting empathy and concern for others, and boys, more than girls, showing affective responses reflecting competition with others (Hypothesis 1). Also, we hypothesized that students high in SCO would score higher on all types of responses than students low in SCO (Hypothesis 2). The central hypothesis of the present study was, however, that affective responses to social comparison would be related to performance on objective tests for reading comprehension and mathematics, after controlling for performance on objective pre-tests, with positive affective responses being positively related to future performances and negative affective responses being negatively related (Hypothesis 3). Because of an effect of gender on test results that has been found in previous research (e.g., Van der Werf et al., 1999 and Wehrens et al., in press), we also controlled for gender. SCO was controlled for since we expected it to be related to the experience of affective responses.

3. Results 3.1. Representativeness of the sample First, we tested whether the selected group of 1500 students was representative of the original subsample of 2442 students. Compared to the excluded students, it appeared that the selected group consisted of relatively more girls (55.8% vs. 47.5%) and relatively more students from the three highest tracks (i.e., track A, AB, and B; 52.8% vs. 41.5%). As a result, the average scores on the pre-tests and post-tests were slightly higher for the selected group than for the excluded group. Table 5 shows how the selected group related to the excluded group on all continuous variables of the present study. Cohen (1988, p. 40) classified effect sizes around 0.20 as small and around 0.50 as medium. As can be seen in Table 5, all effect sizes were quite modest. In broad lines, it seems that the selected group consisted of participants who seemed to be relatively conscientious: they had higher test scores and more often had completed the measures of our control and predictor variables. Although our sample may therefore not be entirely representative of the original sample, we feel this sample is large and valid enough to test our hypotheses in. Table 5. Comparison of average scores on the pre-tests, post-tests, SCO, and the response factors between the selected group and the excluded group. Variables Selection Excluded M SD M SD ESa Prior performance Dutch language 13.15 3.64 12.30 3.83 0.23 Prior performance arithmetic 13.03 4.37 12.10 4.54 0.21 Reading comprehension 0.12 0.21 0.05 0.21 0.33 Mathematics 0.11 0.19 0.06 0.17 0.27 Social comparison orientation 3.12 0.53 2.93 0.54 0.36 Factor 1 (empathic response) 3.42 0.97 3.21 0.89 0.21 Factor 2 (constructive response) 3.38 0.80 3.28 0.85 0.13 Factor 3 (destructive response) 1.78 0.65 1.90 0.70 −0.18 a Effect size (Cohen’s d; Cohen, 1988) was calculated to compare the selected group with the excluded group. Table options 3.2. Descriptive results and correlations Table 2 shows the track averages on the two pre-tests and two post-tests. The average performance on the pre-tests and post-tests from highest to lowest track was in decreasing order, as can be expected when achievement tests are used. A few minor exceptions to the decreasing pattern were caused by tracks CD and DE with only 13 and 37 students, respectively. Table 6 shows the zero-order correlations between all the variables measured in the present study, except for track. Because of the large sample size, for the results section a minimum level of significance of .01 was employed. Table 6. Zero-order correlations between control variables, predictor variables, and criterion variables. 1 2 3 4 5 6 7 8 9 1. Prior performance Dutch language – .61 .12 .19 .24 .12 −.09a .53 .55 2. Prior performance arithmetic – −.14 .10 .11 .08a .00b .47 .64 3. Gender – .15 .34 −.04b −.26 .09a −.05b 4. Social comparison orientation – .21 .30 .15 .14 .14 5. Factor 1 (empathic response) – .17 −.21 .20 .16 6. Factor 2 (constructive response) – .30 .06b .07b 7. Factor 3 (destructive response) – −.15 −.07b 8. Reading comprehension – .59 9. Mathematics – Note: Correlations without superscript are significant at p < .001. a p < .01. b p = ns. Table options 3.2.1. Gender effects We tested for gender differences in SCO. It appeared that girls scored significantly higher on SCO than boys (M = 3.2, SD = 0.5 vs. M = 3.0, SD = 0.5; t(1498) = −5.7, p < .001; d = 0.40). With respect to gender differences in the response factors, as hypothesized, we found that girls reported more empathic response than boys (M = 3.7, SD = 0.9 vs. M = 3.0, SD = 0.9; t(1498) = −14.1, p < .001; d = 0.67), and that boys reported more destructive response than girls (M = 2.0, SD = 0.7 vs. M = 1.6, SD = 0.6; t(1278) = 10.3, p < .001; d = 0.63). Boys and girls did not differ in constructive response (M = 3.4, SD = 0.8 vs. M = 3.4, SD = 0.8; t(1498) = 1.6, ns). Thus, our findings provide support for Hypothesis 1, the expectation that boys would show stronger competitive responses than girls whereas girls would show stronger empathic responses than boys. 3.2.2. SCO In order to test whether students high in SCO reported stronger affective responses to social comparisons than students low in SCO, we investigated the correlations between SCO and empathic, constructive, and destructive response. It appeared that SCO was positively correlated with all three affective responses to social comparison (empathic response: r = .21, p < .001; constructive response: r = .30, p < .001; destructive response: r = .15, p < .001), supporting Hypothesis 2. 3.3. Multilevel analyses To test the central hypothesis, i.e., Hypothesis 3 that states that responses to social comparison are related to academic performance, two multiple regression analyses were performed, one for reading comprehension and one for mathematics. In order to take the nested structure of the data into consideration, i.e., the natural grouping of students into classes, we applied multilevel analyses (Snijders & Bosker, 1999). The eight tracks were represented by seven dummy variables which contrasted each track with the track C. It was decided to use C as the reference category since this track represented the median according to the cumulative percentage of students. Gender was also represented by a dummy variable with boys coded as 0 and girls as 1. The other control and predictor variables were standardized, as were the two criterion variables. In the multilevel analyses, we used a stepwise approach. In both analyses the predictors were added in the same order. The first model (Model 0) that was estimated was the unconditional or ‘empty model’. This model included a constant, but no predictors yet, in order to estimate the variance on the class level and student level, and the initial deviance. In the second model (Model 1) the combined effect of the track dummies, students’ score on the pre-test, gender, and SCO was examined. Students’ score on the pre-test was added in order to control for differences in previous achievement within tracks. In the analysis for reading comprehension the Dutch language score served as a control variable; in the analysis for mathematics the arithmetic score was the control variable. In Model 2 empathic, constructive, and destructive response were entered simultaneously.4 In Model 3 the three two-way interactions and the three-way interaction between empathic, constructive, and destructive response were entered. Each model yielded a ‘deviance’ value. This deviance can be regarded as a measure of lack of fit between model and data (Snijders & Bosker, 1999, p.88). To compare subsequent multilevel models and to test whether a group of added variables contributed significantly to a new model, the deviance of a new model was subtracted from the deviance of the previous model. For each difference in deviance between two subsequent models a chi-square test was performed to examine the significance of the difference in deviance. The number of degrees of freedom for a chi-square test was equal to the number of variables added to the preceding model. 3.4. Multilevel results for reading comprehension Table 7 shows the change in the fit of the models, and the parameter values for reading comprehension. Model 1 was significantly better than Model 0 (χ2 = 485.1, df 10, p < .001), Model 2 was significantly better than Model 1 (χ2 = 37.2, df 3, p < .001), and Model 3 was significantly better than Model 2 (χ2 = 13.5, df 4, p < .01). Since standardized variables were entered into the analyses, the coefficients are comparable with standardized regression weights. In the empty model for reading comprehension, 55.4% of the variance was at the individual level, the remaining 44.6% was at the class level. In Model 1 the control variables were entered. This produced a large and significant decrease in deviance, i.e., a better fit between model and data. The control variables accounted for 38.7% of the total variance. Adding the control variables also influenced the relative size of the variance components. Of the remaining variance 84.4% was at the individual level. The decreasing pattern of performance on the post-tests from highest to lowest track mentioned in the descriptive results section, was reflected by the coefficients of the track dummy variables in Table 7. Moreover, the coefficients in Table 7 show that the highest tracks had a positive coefficient, i.e., performed better, and the lowest tracks had a negative coefficient, i.e., performed worse, than the reference category C. Prior performance, i.e., students’ scores on the pre-test Dutch language, had a significant effect on the scores on the reading comprehension test (β = .221, p < .001), which of course was not surprising. However, it did show that it were not only differences between the tracks that mattered for test scores 2 years later, but also differences within the tracks. Gender did not influence scores on the reading comprehension test (β = .106, ns), nor did SCO (β = .020, ns). Table 7. Reading comprehension: change in model fit, and coefficients and standard errors. Variables Empty model Model 1 Model 2 Model 3 Decrease in deviance – 485.1⁎⁎ (df = 10) 37.2⁎⁎ (df = 3) 13.5⁎ (df = 4) Cum.% explained variance 0.0 38.7 40.3 40.9 % Unexplained variance Individual level 55.4 84.4 84.9 85.2 Group level 44.6 15.6 15.1 14.8 β SE β SE β SE A 0.852⁎⁎ (0.073) 0.851⁎⁎ (0.072) 0.848⁎⁎ (0.071) AB 0.626⁎⁎ (0.170) 0.645⁎⁎ (0.167) 0.645⁎⁎ (0.166) B 0.376⁎⁎ (0.070) 0.387⁎⁎ (0.069) 0.386⁎⁎ (0.068) CD −0.365 (0.248) −0.404 (0.244) −0.415 (0.242) D −0.377⁎⁎ (0.076) −0.355⁎⁎ (0.076) −0.360⁎⁎ (0.075) DE −0.385 (0.157) −0.363 (0.155) −0.373 (0.154) E −0.648⁎ (0.243) −0.650⁎ (0.240) −0.647⁎ (0.238) Prior performance 0.221⁎⁎ (0.030) 0.207⁎⁎ (0.030) 0.201⁎⁎ (0.030) Gender 0.106 (0.043) 0.014 (0.046) 0.007 (0.046) Social comp. orientation 0.020 (0.022) 0.040 (0.023) 0.039 (0.023) Empathic response 0.039 (0.023) 0.018 (0.025) Constructive response 0.010 (0.024) 0.032 (0.025) Destructive response −0.129⁎⁎ (0.024) −0.157⁎⁎ (0.026) Emp. × constr. response 0.000 (0.021) Emp. × destr. response −0.045 (0.025) Constr. × destr. response 0.064⁎ (0.023) Emp. × constr. × destr. resp. 0.038 (0.019) ⁎ p < .01. ⁎⁎ p < .001. Full-size table Table options In Model 2 empathic response, constructive response, and destructive response were entered. Together, these predictors produced a significant decrease in deviance and explained 1.6% of the variance. As can be seen in Table 7, only destructive response was significantly related to reading comprehension (β = −.129, p < .001). As students reported more destructive responses to social comparison, the lower their scores on reading comprehension were. Three separate analyses for empathic response, constructive response, and destructive response indicated that empathic response on it’s own also was related to reading comprehension scores (β = .061, p < .01). As students reported more empathic responses to social comparison, the higher their scores on reading comprehension were. Nevertheless, when the three response factors were entered simultaneously, the effect of empathic response was overruled by the effect of destructive response. In Model 3 the three two-way interactions and the three-way interaction between empathic response, constructive response, and destructive response were entered. The interactions produced a significant decrease in deviance and explained 0.6% of the variance. Only the two-way interaction between constructive response and destructive response was significant (β = .064, p < .01). This interaction is shown in Fig. 1. Constructive response appeared to compensate the negative relationship between destructive response and reading comprehension. That is, students who reported destructive reactions to the comparison target and only few constructive responses performed worse than students who reported destructive reactions and constructive responses at the same time. Interaction between constructive and destructive response for reading ... Fig. 1. Interaction between constructive and destructive response for reading comprehension. Figure options 3.5. Multilevel results for mathematics Table 8 shows the change in the fit of the models, and the parameter values for mathematics. Model 1 was significantly better than Model 0 (χ2 = 738.4, df 10, p < .001), Model 2 was significantly better than Model 1 (χ2 = 18.6, df 3, p < .001), but Model 3 did not differ significantly better from Model 2 (χ2 = 7.3, df 4, ns). In the empty model for mathematics, 41.0% of the variance was at the individual level, the remaining 59.0% was at the class level. In Model 1 the control variables were entered. This produced a large and significant decrease in deviance, i.e., a better fit between model and data. The control variables accounted for 54.6% of the total variance. Adding the control variables also influenced the relative size of the variance components. Of the remaining variance 79.8% was at the individual level. The decreasing pattern of performance on the post-tests from highest to lowest track mentioned in the descriptive results section, was reflected by the coefficients of the track dummy variables in Table 8. Moreover, Table 8 shows that the highest tracks performed better, i.e., had a positive coefficient, and the lowest tracks performed worse, i.e., had a negative coefficient, than reference category C. Students’ performance on the pre-test arithmetic had a significant effect on the scores on the mathematics test (β = .305, p < .001). Neither gender, nor SCO had a significant effect on mathematics test scores (β = −.046, ns, and β = .030, ns, respectively). Table 8. Mathematics: change in model fit, and coefficients and standard errors. Variables Empty model Model 1 Model 2 Model 3 Decrease in deviance – 738.4⁎⁎ (df = 10) 18.6⁎⁎ (df = 3) 7.3 (df = 4) Cum.% explained variance 0.0 54.6 55.1 55.3 % Unexplained variance Individual level 41.0 79.8 79.2 79.2 Group level 59.0 20.2 20.8 20.8 β SE β SE β SE A 1.073⁎⁎ (0.067) 1.061⁎⁎ (0.067) 1.060⁎⁎ (0.067) AB 0.890⁎⁎ (0.157) 0.898⁎⁎ (0.157) 0.905⁎⁎ (0.157) B 0.445⁎⁎ (0.065) 0.447⁎⁎ (0.065) 0.451⁎⁎ (0.065) CD −0.461 (0.230) −0.494 (0.230) −0.476 (0.230) D −0.257⁎⁎ (0.069) −0.243⁎⁎ (0.069) −0.234⁎⁎ (0.069) DE −0.385⁎ (0.133) −0.364⁎ (0.133) −0.350⁎ (0.133) E −0.544 (0.220) −0.544 (0.219) −0.542 (0.219) Prior performance 0.305⁎⁎ (0.026) 0.303⁎⁎ (0.026) 0.301⁎⁎ (0.026) Gender −0.046 (0.038) −0.110⁎ (0.041) −0.117⁎ (0.041) Social comp. orientation 0.030 (0.019) 0.043 (0.021) 0.042 (0.021) Empathic response 0.027 (0.020) 0.035 (0.022) Constructive response −0.004 (0.021) −0.004 (0.022) Destructive response −0.076⁎⁎ (0.021) −0.075⁎⁎ (0.022) Emp. × constr. response −0.035 (0.018) Emp. × destr. response −0.007 (0.021) Constr. × destr. response 0.025 (0.020) Emp. × constr. × destr. resp. −0.010 (0.017) ⁎ p < .01. ⁎⁎ p < .001. Full-size table Table options In Model 2 empathic response, constructive response, and destructive response were entered. Together, these predictors produced a significant decrease in deviance and explained 0.5% of the variance. As can be seen in Table 8, only destructive response was significantly related to mathematics (β = −.076, p < .001). As students reported more destructive responses to social comparison, their scores on mathematics were lower. Further, in Model 2 a main effect of gender became manifest after entering the response factors (β = −.110, p < .01), indicating that boys performed better on mathematics than girls. Three separate analyses for empathic response, constructive response, and destructive response indicated that destructive response functioned as a suppressor variable of gender. That is, boys performed better than girls only when the influence of destructive response was eliminated. In Model 3 the three two-way interactions and the three-way interaction between empathic response, constructive response, and destructive response were entered. None of the interactions were significant.5 In sum, our multilevel analyses showed partial support for Hypothesis 3, i.e., the prediction that positive affective responses would be positively related to performances and negative affective responses negatively. Whereas destructive response was negatively related to both reading comprehension and mathematics, empathic response was positively related to performances in reading comprehension only whereas constructive response was related neither to performances in math nor to performances in reading comprehension.