Social comparison effects on academic self-concepts-Which peers matter most?

Social Comparisons in the Classroom: The Big-Fish-Little-Pond Effect

Students who learn in high-achieving groups (i.e., schools or classrooms) report lower ASCs than students with similar individual achievement in low-achieving learning groups. This is the central postulation of the BFLPE ( Marsh, 1987 ) and one of the most robust findings in educational psychology ( Seaton et al., 2009 ). Typical studies on the BFLPE use a multilevel design and large-scale observational data of students. The effect is then operationalized as the incremental negative effect of school- or classroom average achievement on ASC beyond the positive effect of individual achievement.

The BFLPE has been interpreted as a result of contrasting social comparison processes ( Marsh, 1987 ). Testing the mechanism underlying this assumption (e.g., by testing boundary conditions of the effect) played a relatively minor role in early BFLPE research. In their critical review, Dai and Rinn (2008) suggested to broaden the research scope. Consequently, an increasing number of studies on the BFLPE began to test specific assumptions from SCT or used innovative research designs to replicate the effect. For example, some studies examined students that experienced a change in their learning environment (e.g., moving between school tracks or courses) and thus, assumingly, their frame of reference. These studies supported that students moving into higher-achieving contexts experienced a drop in their ASC ( Arens & Watermann, 2015 ; von Keyserlingk et al., 2019 ; Wouters et al., 2012 ).

Local dominance effects , initially described in social psychology as the tendency to “rely on the most local comparison information while deemphasizing more general, and typically more diagnostic, forms of comparison feedback” ( Zell & Alicke, 2010 , p. 369), also received recent attention in BFLPE research. Several studies showed that the more local frame of reference of the classroom produced a stronger negative BFLPE than the school frame of reference ( Janssen et al., 2015 ; Liem et al., 2013 ; Marsh et al., 2014 ). This is plausible given that (a) students know more about the achievement of their classmates compared with other peers in their school and (b) some teachers grade “on a curve” within a class and thus teacher-assigned grades, which are an important source of ASC ( Marsh et al., 2018 ), partly represent the class rank.

Probably the most direct tests of social comparison mechanisms in BFLPE research were conducted in the few studies that combined estimations of the BFLPE with items asking students explicitly about their social comparison targets and about their perceived ability rank in their class. The seminal study by Huguet et al. (2009) showed that the substantial negative BFLPE (i.e., the usage of peer average achievement as a frame of reference for contrasting comparisons) coexists with additional single comparisons to specific classmates, which produce small assimilation effects. However, other salient characteristics that might affect the selection of comparison targets, such as gender or ethnic status, were not examined. Furthermore, the study showed that the BFLPE can be explained by students’ perceived rank in class. This study provided the most conclusive evidence for the notion that the BFLPE is based on social comparisons, more specifically “ego-deflating comparisons with the class standard” ( Huguet et al., 2009 , p. 158). The mediating effect of perceived within-class rank between average achievement and ASC has also been found in other studies ( Marsh et al., 2008 , 2014 ; Thijs et al., 2010 ).

This evidence suggests that social comparisons with generalized others (i.e., the average classmate), which lead to stable contrast effects, can be differentiated from social comparisons with specific others and that both processes can have independent effects. However, few previous studies investigated who these specific others are—that is, whether there are particular subgroups within a class that students compare themselves with.

Similarity

The notion that people select similar others as preferred social comparison targets (similarity hypothesis) was already introduced by Festinger (1954) and its importance highlighted by Mussweiler (2003) . Similar others could be peers from students’ social ingroups along salient dimensions, such as gender or ethnicity ( Huguet et al., 2001 ; Tajfel, 1974 ). In addition, social network studies demonstrated that friends are more likely than other peers to show similarities along various dimensions ( homophily ) ( McPherson et al., 2001 ), such as academic achievement ( Gremmen et al., 2017 ; Kretschmer et al., 2018 ; Lomi et al., 2011 ), aspirations ( Lorenz et al., 2020 ), favorite subjects ( Raabe et al., 2019 ), and engagement ( Wang et al., 2018 ). Therefore, friends may serve as “routine standards” for social comparisons ( Lubbers et al., 2009 ; Mussweiler & Rüter, 2003 ). Similarities have also been found among those who study together ( Stadtfeld et al., 2019 ), likely because study partners are also more likely to become friends over time and vice versa ( Palacios et al., 2019 ), making study partners another likely comparison target.

Status and (Perceived) Popularity

Social status has high importance in the life of adolescents. Students of high status can exert social power over other students by serving as opinion leaders or role models ( Coleman, 1961 ), often through a combination of prosocial and aggressive behaviors ( Cillessen et al., 2011 ). Therefore, other students might strive to adapt their behavior to that of peers they perceive as popular ( Helms et al., 2014 ; Laninga-Wijnen et al., 2018 ). In the case of social comparisons, assimilation effects could be a means to increase similarity between oneself and popular students—if popular students had high academic achievement, other students may increase their ASCs. However, this would only apply for students who themselves have a high popularity motive which is not the case for all students ( Dawes & Xie, 2017 ; Jones & Cooke, 2021 ).

Summary and State of Research

Overall, it is unclear whether comparisons with specific peers would result in contrast or assimilation effects. Zell and Alicke (2010 , p. 376) formulated a possible preference for friends within the framework of the above-mentioned local dominance effect and suggested that contrast effects would be present: “Students who, by chance, associate with friends […] who perform terribly on standardized tests may have inflated ability perceptions.” Furthermore, as mentioned above, previous research ( Gerber et al., 2018 ) indicates that contrast effects are much more frequent and stable than assimilation effects. However, a perception of similarity to the target should increase the likelihood of assimilation—thus if assimilation effects do occur, it should be when comparing oneself with friends ( Mussweiler, 2003 ; Suls et al., 2002 ).

Very few studies have tested the role of friends for within-class social comparisons and students’ self-concepts empirically. In longitudinal studies, having friends with high grades seemed to have a small detrimental effect on self-evaluations of low-achieving students ( Altermatt & Pomerantz, 2005 ; Guay et al., 1999 ). However, the studies did not control for class-average achievement.

The probably most comprehensive study of Wouters et al. (2013) tested two competing hypotheses: a local frame of reference should result in “friendship dominance,” but the salient nature of comparisons with the class mean should lead to “classroom dominance.” The authors also tested whether both standards could be used simultaneously. Using a large sample of primary school students from The Netherlands, they included both predictors in the same model and found only a negative effect of classroom average achievement. A recent study based on Finnish primary found similar classroom-dominance when comparing classroom average and the average of peers students spend time with during and after classes ( Koivuhovi et al., 2020 ). It should be noted that, even though Koivuhovi et al. (2020) used techniques from social network analysis, they did not differentiate social selection and influence mechanics. Furthermore, the study did not examine friendships.

The Present Study

The central research question of this study was how social comparisons with different (overlapping) groups of peers in the classroom affect students’ ASCs. We juxtaposed different frames of reference and peer groups that students might use for their social comparisons using longitudinal social network analysis (SAOMs; Snijders et al., 2010 ) based on two measurement waves of secondary school students in Germany. Our working model is shown in Figure 1 . In addition to the classroom average achievement, we examine friends, study partners, peers the student perceives as popular as well as same-gender and same-ethnic classmates as possible frames of reference.

Without considering other frames of reference, we expect the classic BFLPE to be replicated—that is, we expect positive effects of individual achievement on ASC and additional negative effects of classroom average achievement (Hypothesis 1).

Regarding the effect of friend average achievement on ASC, two conflicting expectations can be formulated. The first argument is based on BFLPE research and the idea of a “local dominance” effect ( Koivuhovi et al., 2020 ): There should be a negative effect of friend-average achievement on ASC indicating contrasting social comparison. In this case, friends would function as a more local frame of reference than the classroom ( Zell & Alicke, 2010 ). In contrast, following arguments made in social psychology, a high similarity between the source and the target of social comparisons should make assimilation effects more likely ( Mussweiler, 2003 ). Thus, a positive effect of friend-average achievement on ASC could be interpreted as an assimilative social comparison effect with students’ ASCs benefiting from high-achieving peers through identification and “basking in reflected glory” ( Koivuhovi et al., 2020 ). On the grounds of previous empirical work on the prevalence of contrasting comparisons, we expect contrast effects be more likely—the effect of friend-average achievement on ASC should thus be negative (Hypothesis 2).

Study partners might serve as viable social comparison targets as students know more about their ability and achievement (owing to studying together) than about the achievement of other students. Furthermore, a study group might be perceived as a local frame of reference by the students ( Zell & Alicke, 2010 ). Therefore, as for friends (who substantially overlap with study partners; Stadtfeld et al., 2019 ), we expect the average achievement of study partner peers to show a negative effect on ASC (Hypothesis 3).

There are no previous studies on social comparison effects with student-perceived popular peers on ASC and it has been shown that popularity cannot easily be inferred from friendship status, but should be assessed separately ( Vörös et al., 2019 ). We expect the achievement of student-perceived popular peers to be meaningful for ASC owing to popular peers’ visibility and status (and thus being likely social comparison targets). One argument for assimilation effects could be that students perceived as popular serve as role models for the academic behavior within a class. However, this would only be the case if (a) students think of students they perceive as popular as similar to them in the first place ( Mussweiler, 2003 ), which is less likely than for friends, and/or (b) students have a popularity motive ( Jones & Cooke, 2021 ). Furthermore, contrast effects are generally dominant in most situations (see above). Therefore, we refrain from making a prediction on the direction of social comparison effects with student-perceived popular students and examine whether the effects of the average achievement of student-perceived popular peers on ASC are contrastive/negative (Hypothesis 4.a) or assimilative/positive (Hypothesis 4.b).

Finally, salient ingroups of same-gender or same-ethnic peers might constitute a more “local” frame of reference than the class. In that case, students would compare themselves with peers from their ingroups rather than all classmates, but still in a contrasting way. Therefore, the achievement of same-gender and same-ethnic peers should show a negative effect on ASC (Hypothesis 5).

We tested the effects for specific peers both with and without controlling for classroom average achievement, and our Hypotheses 2 to 5 assume that the effects of specific peers are incremental to possible effects of the classroom average. Through this juxtaposition of different frames of reference (see Figure 2 for an illustration how these different frames of reference can look like in the social network of one example classroom), we could examine whether there is “classroom dominance” or “friendship dominance” (or perhaps dominance of other specific peers) in social comparisons with one frame of reference possibly prevailing ( Wouters et al., 2013 ). We extended the studies by Wouters et al. (2013) and Koivuhovi et al. (2020) , who found evidence for “classroom dominance” (i.e., no incremental effect of friend-average achievement beyond a persisting BFLPE) in several ways. First, we separated social selection and influence processes (see below). Second, we considered several types of peers that could be valid comparison targets in addition to friends. Third, both studies were based on primary school samples. However, friendships become much more important as students reach adolescence where peers become their primary social contacts ( Larson & Richards, 1991 ). Therefore, we formulated no hypothesis on the dominance of different social comparison effects. Still, the relative strength and possible coexistence of these effects are essential research objectives of this study.

Finally, we note that, using terminology from social network research, the research questions we examine are examples of social influence —that is, an individual’s behavior or characteristic (here: students’ ASC) being affected by the behaviors or characteristics (here: achievement) of specific

Analysis Sample

Our study relied on SOAMs to study social networks longitudinally. These models have particular requirements with regard to the data structure. Thus, following the recommendations in the sociometric fieldwork reports of the CILS4EU ( Kruse et al., 2016 ; Kruse & Jacob, 2016 ) and previous applications of SAOMs ( Boda, 2018 , 2019 ; Lorenz et al., 2020 ; Raabe et al., 2019 ), we only analyzed classes with a sufficient participation rate across all waves and without major compositional changes. More specifically, we included classes only if (a) students in those classes participated in the sociometric questionnaire at both measurement points and students did not change the class between the two survey waves (resulting in N = 3.858), (b) not more than 25% of students dropped out between the measurement points (resulting in N = 2.506) and (c) the class included at least 10 students (resulting in N = 2.467). Finally, one class was excluded for which all grade variables were missing.

These exclusions resulted in an analysis sample of 2,438 students from 117 school classes. They were on average 15.18 years old ( SD = .66) and 47% were male. More than half of the students (56%) had an ethnic minority background with the biggest origin groups being students whose families emigrated from Turkey (19%) and Eastern Europe (10%). Further details on the sample can be found in Supplement 1, Table S1 , which includes the sample statistics.

Academic Self-Concept

Because there are no domain-specific, multiitem ASC scales in the CILS4EU dataset, we constructed a proxy measure for general ASC. We combined three indicators referring to self-perceived domain-specific achievement in the three main subjects (“How well are you doing in the following subjects?” asked for German, English and mathematics with replies given on a 5-point scale ranging from not well at all to very well ) with two Likert-style indicators of general academic self-efficacy (“I am sure that I can do well at school” and “I am sure that I can get good grades at school” with agreement indicated on a 5-point-scale from strongly disagree to strongly agree ) to create a latent factor. All indicators were assessed in 9th (T1) and 10th (T2) grade. The resulting general ASC factors showed a reasonable, though not ideal fit (which is plausible given its nature as a combined proxy) to the data in a first-order one-factor confirmatory factor analysis (CFA) model (T1: CFI = .921, RMSEA = .091, SRMR = .042, df = 5; T2: CFI = .891, RMSEA = .131, SRMR = .057, df = 5) as well as high reliability (T1: McDonald’s ω = .82, T2: ω = .86). The model fit could be improved by adding two correlated residuals between the two self-concept measures in the language domains and between the two self-efficacy measures. This led to better model fit (T1: CFI = 1.00, RMSEA = .005, SRMR = .005, df = 3; T2: CFI = .999, RMSEA = .015, SRMR = .007, df = 3). However, the reliability was lower (T1: ω = .71, T2: ω = .81).

Therefore, factor scores from the CFA models without correlated residuals were used in the further multivariate analysis. Descriptive statistics of the indicators are provided in Supplement 1, Table S1 . However, we replicated model 3 for the friendship network using the latent self-concept variable with the two correlated residuals. The results were very consistent across all parameters (see Supplement 5 for details).

Academic Achievement

Similar to the procedure for ASC, we computed latent achievement scores based on five indicators that were then used in further analyses. This is in line with the recommendations by Dicke et al. (2018) to use several indicators when computing peer-averages to control for measurement error. We used reverse-scored teacher-assigned grades in German, Mathematics and English (which, in Germany, range from 1 [ excellent ] to 6 [ insufficient ]) as well as the sum scores of a cognitive ability test and a language ability test. Cognitive abilities were measured using a language-free test based on solving figural problems. The language ability tests were country-specific (given that different languages were being tested), but all had a focus on lexical knowledge using synonym- or antonym tests. In Germany, the verbal subscale of the KFT 4–12+ R ( Heller & Perleth, 2000 ) was used which is a well-validated and frequently used ability test for student samples. The achievement tests were only conducted at T1. The latent factor showed a reasonable fit (CFA = .964, RMSEA = .127, SRMR = .028; correlated residuals between the grade indicators were included) and reliability (ω = .67) given that it can be considered a composite achievement measure. We additionally created two separate measures (a) using only the two achievement tests and (b) using only the three grade variables. In a two-factorial CFA model, the fit was adequate (CFI = .964; RMSEA = .090; SRMR = .027) and the latent correlation between the achievement and the grade factor was .49. The results of the main models (model 3 for each of the social networks; see data analysis section) could be replicated with those alternate achievement measures (i.e., separate models for grades and test-scores were estimated), and they were consistent for all hypotheses (see Supplement 4 ).