Socioeconomic status biases among children and adolescents: The role of school diversity and teacher beliefs in Nepal

Investigating socioeconomic status (SES) biases, Nepalese children and adolescents ( N = 605, 52% girls, M age = 13.21, SD age = 1.74) attending schools that varied by SES composition were asked to anticipate whether a peer would include a high or low SES character as a math partner. Novel findings were that students attending mixed SES schools were more likely to expect inclusion of a low SES character than were students attending high SES schools. With age, high SES participants attending mixed SES schools increasingly expected the inclusion of the low SES character. Moreover, teachers' democratic beliefs in high SES schools predicted inclusive expectations. Teacher beliefs and school diversity play a significant role for fostering students' inclusivity in educational contexts.

Socioeconomic status (SES) inequalities create status hierarchies which often translate into exclusionary practices in both political and social life. Families and children at the low end of the socioeconomic ladder experience numerous societal obstacles for moving up the SES hierarchies (Duncan & Murnane, 2011). Economic segregation in schools contributes to these obstacles with less access to opportunities and resources for schools at the low end of the economic spectrum (Reardon & Firebaugh, 2002). Moreover, school segregation restricts social interactions between students from different social backgrounds. Such interactions would provide important social capital for students at the lower social end as well as opportunities to change negative group biases (Ruck et al., 2019). Biases about SES are often the source of social exclusion and ostracism, particularly of children and adolescents from lower status groups (Elenbaas, 2019). Social exclusion, in turn, renders children vulnerable for negative long-term health outcomes, which affects academic achievement, lowering students' chances to improve their social status . A central aim of this study was therefore to provide novel insights on children's and adolescents' beliefs and biases about low and high status peers by studying their expectations and reasoning about whether a low or high status peer would be included in an academic group task.
In particular, the current study investigated children's and adolescents' judgments, expectations, and reasoning about SES and social inclusion in Nepal. Investigating youth's conceptions of social inequalities requires conducting research in varied sociopolitical contexts, which often reflect different social status hierarchies (Wainryb & Turiel, 1994). Recent research examining adolescents' and young adults' beliefs about social inequalities revealed their desire for greater economic equality, whereby similar patterns evolved in various social and cultural contexts, such as Argentina (Barreiro et al., 2019), Europe (Niehues, 2014), and the United States (Arsenio & Willems, 2017;Flanagan & Kornbluh, 2019). The current study examined whether these findings also applied to a novel, non-Western context. Nepal was chosen as a focus for this study given its long history of societal inequalities and social exclusion based initially on the caste system and more recently on SES. In the past, the caste system organized most aspects of life and divided society into different social groups within a social hierarchy, with restrictions on marriage and occupation. Due to the extensive social, political, and economic discrimination, the caste system was deemed illegal in 1963. Social status differences persist, however, and are currently reflected in individuals' SES, which perpetuate social inequalities (Gurung et al., 2012).
Previous work from another sociopolitical context (i.e., Northern Ireland) suggests that adolescents' awareness of hostile intergroup relations in societies with persistent intergroup divisions may negatively affect their prosocial development . Moreover, in divided societies, children often acquire knowledge about intergroup tensions early in development, including negative stereotypes, which may have long-lasting effects on their attitudes pertaining to intergroup relationships (Bar-Tal et al., 2017). With regards to Nepal, we have proposed that the long-term discrimination of low status groups may contribute to children's and adolescents' awareness of social hierarchies. Thus, it remains an important question whether their beliefs about peer inclusion reflect group biases in line with the restricted social mobility that emerged from the caste system or whether they would express a strong desire for equal opportunities, resulting in higher expected peer inclusion. In fact, a recent study revealed Nepalese adolescents' desire for greater economic equality (Grütter, Dhakal, et al., 2021). Extending these first insights, the current study investigated children's and adolescents' expectations and reasoning about including a high or low SES peer in an academic context.
Further, an important aim of this study was to identify potential factors of the social climate within schools and classrooms that may contradict biases rooted in status hierarchies and promote inclusion Turner & Cameron, 2016). Hence, we sampled adolescents who attended high SES schools (i.e., schools with predominantly high SES students) and mixed SES schools (i.e., with high diversity by SES, including both high and low SES students), and examined the role of teachers' beliefs about teaching styles that target equality, responsibility, and fairness along with potential biases about students from low SES backgrounds (i.e., democratic teacher beliefs). The current study provides new data on the variation of experiences that lead to SES-based social exclusion from an understudied population in psychological science.

Children's and adolescents' reasoning about SES
Youth's SES-related reasoning has been identified as a significant area for developmental science research (Ruck et al., 2019). Socioeconomic status reflects a social intergroup category, defined by an individual's access to material resources and the perception of their standing in society relative to others (Kraus & Keltner, 2013). To date, little is known about whether and how SES impacts children's and adolescents' reasoning about intergroup relations, particularly in their peer world (Ruck et al., 2019). Only a handful of studies have examined how children conceptualize SES and its relationship to peer group social exclusion (Burkholder et al., 2019;Elenbaas, 2019;Elenbaas & Killen, 2018). These studies, which were conducted in the United States, suggest that children and early adolescents (ages 8-14 years) are sensitive to wealthrelated biases and deem exclusion of low SES peers from social activities as unfair. For example, when examining children's expectations about including wealthy or poor children to a summer camp, children preferred to include poor children more than wealthy children based on lack of access and opportunity (Elenbaas, 2019).
What has not yet been investigated are contextual factors such as how the role of school diversity and teachers' beliefs are related to children's and adolescents' predictions and reasoning about inclusion of low and high SES peers, specifically in an academic context. While research has shown that children who experience exclusion from academic contexts based on biases about race and ethnicity are at risk for low motivation to succeed in school (Rivas-Drake et al., 2014), much less is known about children's and adolescents' beliefs and biases about SES in academic situations.
Theoretical framework: Social reasoning development approach The current study was based on the social reasoning development (SRD) approach (Killen & Rutland, 2011;, which provides a developmental intergroup framework for understanding how children and adolescents reason about intergroup encounters and relationships. This framework focuses on the interplay between morality and group processes and integrates theories on both the development of moral reasoning (Smetana et al., 2014) as well as social identity and group dynamics (Rutland et al., 2010). When facing complex social situations, such as including or excluding a peer, children's and adolescents' reasoning involves weighing multiple considerations, such as moral concerns (e.g., fairness, others' welfare), societal and group concerns (e.g., group norms), and personal concerns (Killen et al., 2016;. Empirical research findings based on the SRD perspective suggest a developmental shift, starting in late childhood and increasing into early adolescence, when peer groups gain in salience and individuals increasingly focus on group identity and norms; hence, a growing understanding of intergroup relations and need for group affiliation contributes to being more susceptible for group biases during late childhood and early adolescence . However, at the same time, developmental increases in perspective taking, understanding of abstract concepts, and the ability to coordinate multiple aspects of complex social situations (i.e., moral and nonmoral considerations) lead to more multifaceted, sophisticated, and comprehensive reasoning in adolescence (Smetana et al., 2014). Based on these previous findings, we theorized that children's and adolescents' expectations and reasoning about social inclusion based on SES would include how they consider aspects of fairness, SES-related group concerns and biases, as well as attributions related to social status. SES-related group concerns, biases, and attributions related to social status Concerning SES-related group concerns and biases and attributions related to social status, a growing body of research demonstrates that biases based on SES emerge early in childhood and continue through adolescence. Specifically, children and adolescents ascribe more positive attributes to high rather than to low SES individuals (e.g., Mistry et al., 2015;Sigelman, 2012;Woods et al., 2005). Such attributions reflect stereotypes if they are generalized over a group of people, irrespective of intragroup variations. Based on the stereotype content model (Fiske et al., 2002), low SES people are seen as low in warmth and competence while high SES people are seen as high in competence and low in warmth. However, recent work has shown that these attributions are context-dependent. In academic contexts, high wealth individuals are rated higher in academic and work competence (i.e., smart, responsible) than low wealth individuals (i.e., dumb, lazy). Yet for social contexts, high wealth individuals are rated more negatively (i.e., cold, exclusive) than are low wealth individuals (i.e., generous, charitable; Burkholder et al., 2019;Elenbaas & Killen, 2018;Sigelman, 2012). In addition, recent work with adult samples suggests that under the condition of high social inequality, low SES people are rated as even less competent, while high SES people are perceived as even lower in warmth (Durante et al., 2017).
Moreover, according to social identity theory (Tajfel & Turner, 1979) an individual's own group status predicts how groups are evaluated. Negative views about low status characters can serve as a strategy of high status individuals to maintain their status in the social hierarchy and to prevent social mobility of low status groups (Kraus et al., 2012). With regards to the current study, it seems likely that adolescents from high SES backgrounds would be more inclusive regarding high SES individuals than low SES individuals. In contrast, in-group bias may serve students from lower SES to protect their own group from negative evaluations. And yet, students from lower SES backgrounds may also want to belong to a positively valued group which may lead to positive out-group evaluations (Mistry et al., 2015;Verkuyten & Martinovic, 2006). To date, there are limited findings regarding the role of participants' own SES in their reasoning about SES, which indicate that older children and early adolescents (ages 8-14 years) from higher SES backgrounds perceive exclusion of low SES characters as less wrong than their peers from lower SES backgrounds (Burkholder et al., 2019). Another recent study pointed to higher perceptions of in-group biases based on wealth in individuals (ages 8-14 years) of higher SES backgrounds (Elenbaas & Killen, 2018).

Considerations of fairness with regards to social inequalities
Based on the SRD perspective, we theorized that children's and adolescents' reasoning about SES would be informed by considerations of fairness. Children and adolescents ages 8-14 have been shown to be sensitive to the needs of economically disadvantaged peers and to negatively evaluate the preferential treatment of high wealth peers, particularly in the context of societal inequalities (Elenbaas, 2019). Further, a study with 12 to 19-year-olds suggests that older as compared with younger adolescents are more likely to understand contextual factors that determine life outcomes and to realize that structural obstacles disproportionally affect low SES groups (Flanagan et al., 2014). A recent study conducted with Nepalese adolescents (ages 12-18) revealed a higher likelihood of understanding the role of social hierarchies in restricting friendships between low and high SES students among older as compared with younger adolescents (Grütter, Dhakal, et al., 2021). Thus, with age, adolescents may hold more negative evaluations of exclusion based on SES as compared with younger participants.
However, to date little is known about potential agerelated differences in adolescents' reasoning about social exclusion. Research on exclusion based on sexual orientation (Horn, 2019;Horn & Sinno, 2014) points to more negative evaluations among late adolescents as compared with early and mid adolescents, whereby they more likely refer to concerns of fairness and equal rights. Similarly, a scarce study on how adolescents at various ages reason about social exclusion based on race  showed that tenth graders were more likely to recognize that stereotypes may lead to race-based exclusion than seventh and fourth graders, whereby both seventh and tenth graders perceived exclusion based on race as more wrong than fourth graders.
The current study aimed to identify factors that motivate children and adolescents to focus on fairness and inclusion rather than stereotypes and group biases when SES-related inclusion decisions are being considered. By studying participants in late childhood up to late adolescence, the current work adds novel insights about potential developmental differences in reasoning about social exclusion. With regards to contextual factors, we theorized that school diversity and teacher beliefs would be related to students' evaluations and reasoning.

School diversity and teachers' democratic beliefs
Schools can provide opportunities for positive interactions between students from different social backgrounds, particularly when positive intergroup contact is feasible. Positive intergroup contact is associated with positive social, emotional, and academic development and less group bias (Turner & Cameron, 2016). Previous research focused on school diversity in the United States and Europe with regards to race, ethnicity, nationality, or special educational needs ; however, less is known about schools that differ in terms of socioeconomic diversity (for an exception see . Moreover, experiences with diversity may differentially predict how youth reason about social inequalities, depending on their group status. The conceptual model of the psychosocial benefits of ethnic and racial school diversity (Graham, 2018) assumes different explanations for changes in intergroup attitudes and school adjustment among children and adolescents from social minority and majority groups. These are opportunities for cross-group friendships, exposure to multiple ethnic groups, and numerical balance of power. Related evidence from different studies conducted with sixth graders shows that, among social majority group students, cross-race friendships predicted more positive racial attitudes over time, but not among social minority group students (Graham, 2018). Regarding social minority group students, a more even representation of different ethnic groups and thus, a higher numerical balance of power among social groups, predicted higher feelings of safety at school and less victimization of ethnic minorities. However, if students from social minority groups also represented numerical minorities in their classroom, their risk for negative peer experiences increased as group interactions were less likely to be based on equal status (Graham, 2018;Juvonen et al., 2019). Experiences as a numeric minority may also apply to SES, whereby individuals from low SES backgrounds may face social exclusion and become targets of negative biases .
The current project examined the role of school diversity by comparing students enrolled in predominantly high SES schools to those enrolled in mixed SES schools regarding their SES-related biases. Based on prior evidence related to school diversity (Graham, 2018), we assumed that in mixed SES schools, students from low SES backgrounds would benefit from more power balance, while students from high SES groups would benefit from cross-group interactions with low SES students. In addition, we theorized that socioeconomic diversity would differentially relate to students' experiences depending on their age.
Evidence based on a meta-analysis suggests agedifferences in the effectiveness of intergroup contact, whereby its effectiveness becomes increasingly contextdependent in adolescence (Raabe & Beelmann, 2011). Hence, we assumed that older adolescents from social majority (i.e., high SES) groups would benefit more from contact experiences in mixed SES schools as compared with younger participants, expecting higher inclusion in mixed SES as compared with high SES schools. In contrast, we assumed that younger participants from high SES backgrounds in late childhood and early adolescence would be more susceptible for group biases (McGuire et al., 2021;, focusing more on group concerns and expecting less peer inclusion. More specifically, we anticipated that their biases would be harder to change even when attending mixed SES schools. When considering students' intergroup relations, recent research also pointed to the important role of teachers as moderators of their experiences Thijs & Verkuyten, 2014). One important factor is teachers' democratic beliefs, capturing beliefs about teaching styles that target equality, responsibility, and fairness. Whether authority figures, such as teachers, are on board with the goals of mutual respect, fairness, and equality represents one of the several conditions of Allport's (1954) theory about prejudice reduction. Teachers' attitudes can translate into modeling the equal treatment of students and thereby shape their students' attitudes (Vezzali et al., 2012). In line with the idea of teachers as socialization agents, previous work suggests that students in a classroom with teachers who express a higher value for diversity and equality are more inclusive as compared with students in classrooms with more biased teachers (Grütter & Meyer, 2014). Providing additional evidence, more inclusive peer relations have been observed in classrooms in which teachers do not display biases against students from social minority groups (Farmer et al., 2019;Juvonen et al., 2019). On the flip side, teachers who hold biases create negative classroom climates, for example, by applying harsher discipline to minority group students, which in turn may translate into more exclusive behavior (Okonofua et al., 2016).
Despite their critical role for social inclusion, research has only recently focused on teachers (Farmer et al., 2019;Juvonen et al., 2019). Surprisingly, little attention has been given to teachers' beliefs and behaviors in intergroup development. The current study examined whether students in classrooms with teachers who endorse beliefs about equality, responsibility, and fairness would express less SES-related biases and expect more peer inclusion. In addition, we investigated whether the role of teachers would be particularly important in schools with a greater risk for peer exclusion, such as schools with low socioeconomic diversity, in which students from lower SES backgrounds represent a numerical minority. Thus, we theorized that teachers who perceive it as important to treat everybody equally would moderate whether these minority group students would be included.

The current study
In order to investigate children's and adolescents' judgment, expectations and reasoning about SES and social inclusion, participants were presented with a hypothetical scenario about an academic group activity in which two unfamiliar students (their SES was not revealed in the vignette) had to invite a third student to join them to work on a difficult math problem. They could choose one of two students, one was identified as from a low SES background and the other from a high SES background. The scenario was adapted from previous studies using inclusion decisions in both academic (e.g., Gasser et al., 2017) and social (Elenbaas, 2019) contexts. Since Nepal represents a new context, focus groups were conducted with children and adolescents from the area (N = 30). The academic group task was chosen based on children's and adolescents' feedback for subject matters where stereotypic expectations exist concerning competency beliefs for low SES students.

Expected inclusion, reasoning, and moral judgment about inclusion based on SES
Based on prior evidence from the SRD perspective (Elenbaas & Killen, 2018;Horn, 2019;Killen et al., 2010;, the stereotype content model (Fiske et al., 2002) and children's and adolescents' understanding of social inequalities (Elenbaas, 2019;Flanagan et al., 2014), we expected that students' reasoning for decisions to include the low SES character would be based on considerations of fairness and attributions of positive traits regarding low SES students. In contrast, reasoning for decisions that include the high SES character would be based on positive attributions of traits about high SES students and social hierarchies (H1a). Moreover, we expected that with age, participants would be less likely to consider generalized trait attributions when reasoning about inclusion (H1b). In addition, we assumed that when participants would use reasons based on fairness, they would evaluate inclusion based on SES as wrong; when they would use reasons based on social hierarchies and generalized trait attributions, they would in turn evaluate it as more legitimate (H1c).

SES background
Based on findings related to social identity theory (Verkuyten & Martinovic, 2006) and two prior studies on students' expectations about exclusionary behavior of high wealth peers (Burkholder et al., 2019;Elenbaas & Killen, 2018), we assumed that students from low SES backgrounds would expect more inclusion of the low SES character than students from high SES backgrounds (H2a). Importantly, prior evidence showed a high awareness of status differences between low and high SES adolescents in Nepal (Grütter, Dhakal, et al., 2021). Moreover, previous research documented that older children and adolescents are aware of their own and others' relative economic status (Goodman et al., 2001;Mistry et al., 2015).

School diversity and teacher beliefs
Based on the conceptual model of the psychosocial benefits of ethnic and racial school diversity and related evidence (Graham, 2018;Juvonen et al., 2019;, we predicted that participants in mixed SES schools would expect more inclusion than participants in high SES schools (H2b). Based on different explanations for how students of minority and majority groups experience diversity (Graham, 2018) and based on age-differences in the effectiveness of intergroup contact during adolescence (Raabe & Beelmann, 2011), and age-differences in reasoning about social exclusion (Horn & Sinno, 2014;Killen et al., 2010) and considerations of group norms , we expected that this relation would be moderated by students' SES and age (H2c). For low SES minority group students, we expected that experiencing diversity would have a protective effect, independent of their age. For high SES students, we expected that school diversity would be more beneficial with increasing age, rendering older participants more inclusive in mixed SES schools as compared with high SES schools.
Lastly, we hypothesized that students in school classes with teachers expressing higher democratic beliefs would expect more inclusion than those in school classes where teachers express lower democratic beliefs (H2d). We also expected that the role of democratic teacher beliefs would be more predictive of inclusion in high SES schools as compared with mixed SES schools (H2e). These assumptions were based on the idea that teachers' beliefs may moderate intergroup interactions between students (Grütter & Meyer, 2014;Okonofua et al., 2016) and be more relevant in high SES schools, where low SES students represent a numerical minority.

Participants, region, and procedure
The sample consisted of 605 children and adolescents in Nepalese grades 5-10 (52% girls, M age = 13.21; SD age = 1.74; range = 9-18, 75% of the sample was between 12 and 15 years) attending schools in the Kathmandu Valley, Nepal. For each of the 27 classrooms that participated in the study, Nepalese teachers who held the primary responsibility for each school class filled out a teacher questionnaire (56% female, age range: 20-50, with 41% in the age group 36-40 years, followed by 26% between 25 and 30 years).
Participants were sampled from two types of schools: (1) Mixed SES schools were characterized by high socioeconomic diversity (N = 356) with a high number of scholarships distributed to socially disadvantaged students in order to provide access to education (percentage of students with scholarships = 52%); and (2) high SES schools were characterized by students from predominantly high socioeconomic backgrounds (N = 249). As requested by Nepalese law, low diversity schools provided scholarships to a minimal number of students from disadvantaged social backgrounds (percentage of students with scholarships = 10%). Thus, high SES schools had low levels of diversity by SES while mixed SES schools had high levels of diversity by SES with almost equal representations of students from low and high SES backgrounds. SES calculations relied on students' reports regarding their housing situation and family property, based on which we estimated the real per capita consumption (RPC) of each student's family (for details, see Appendix S0). The inflation corrected mean RPC was 37,716 NPR (ranging from 4,323 NPR to 88,539 NPR).
The study procedure was approved by the ethics committee of the Teacher University Lucerne and was in line with the ethical recommendations of the Helsinki declaration, the American Psychology Association, and the World Health Organization (WHO). The study was first discussed at the local village council who gave permission to approach schools within the community. Next, principals and teachers were informed during visits of the research assistants and, if approved, students were informed about the study during class. As the data collection heavily relied on their approval, the sample size was based on opportunity sampling. Primary caregivers were asked to provide active consent for their child's participation, whereby home visits were made by the research assistants if parents were illiterate (14% of fathers, 33% of mothers, and 25% of other primary caregivers than parents). Moreover, to reach illiterate parents, research assistants attended parent meetings of parents whose child received a scholarship. For illiterate parents, the additional signature of a witness was requested as recommended by the guidelines of the WHO. Only 2% of parents did not give their consent. In addition to parental consent, child and adolescent assent was requested and they were given the opportunity to terminate the interview at any time (all students participated and completed the interview). Data collection took place between June and October 2017. The local research assistants conducted individual face-toface interviews (average duration of the task was 15 min) in Nepalese. The interview coding system was created in English, translated into Nepalese, and back-translated for verification purposes.

Social inclusion task
Participants were individually interviewed in a quiet room at school. The interviewer used a flip book that depicted pictures accompanying the story (Figure 1). The social inclusion task was a modification of prior work on children's inclusion decisions in academic contexts (e.g., Gasser et al., 2017). In the inclusion scenario, participants were told about a school assignment in which three students from a different (unfamiliar) school have to solve a difficult math problem in groups of three. The protagonist (e.g., Durga) and their friend, depicted as neither low or high SES, are looking to complete their group with an additional student, and two students ask to participate: One student is from a low SES background (e.g., Radha) and one student is from a high SES background (e.g., Devoki). The descriptions of the low and high SES students were gained from focus groups (for more information on the focus groups, see the Appendix S1) and evaluated in another group of children and adolescents from the area where the research was conducted. In the evaluation group, the characters were rated on a social ladder, ranging from 0 (very poor for the area) to 10 (very rich for the area). Two versions were administered, one with female names and one with male names (gender matched for participants). The female version was as follows: a. "This is Radha, she lives in a small house and sleeps on a simple mattress on the floor in the same room with the rest of her family. She does not have many clothes and her clothes look old and some have holes in it. She usually eats dhal bhat [local food] with vegetables. Sometimes Radha plays a marble game with her siblings when she finishes her schoolwork." b. "This is Devoki, she lives in a house with two floors and sleeps in her bed in her own room. She sometimes goes to Kathmandu with her mother to buy new clothes for herself. She usually eats dhal bhat with vegetables, and sometimes also with chicken. Sometimes Devoki plays a game with her siblings on the computer when she finishes her schoolwork." The two story characters were introduced in randomized order before presenting the academic group task to the participants. Participants answered a control question after listening to both story descriptions (i.e., "Who lives in a small house?") and the descriptions were revealed a second time if the answer was not correct (seven participants needed a second explanation and then answered correctly). In order to explain the story, Figure 1 was presented. Children and adolescents were asked the following questions about the scenario: (1) Expectation of inclusion: "Who do you think will Durga pick to join the group?" (0 = inclusion of the high SES child, 1 = inclusion of the low SES child), (2) Reasoning about inclusion: "Why would Durga choose X (i.e., the character that was chosen)?", and (3) Moral judgment about inclusion: "What do you think, is it ok or not ok that Durga chose X (i.e., the character that was chosen)?" (1 = extremely not ok, 6 = extremely ok). The last question asked the students, how much they liked math (1 = not at all, 6 = extremely) and was used as a control variable in the analyses.

Reliability coding
The reasoning data were transcribed by the research assistants and translated by professional translators from Nepalese to English. A team of five research assistants from Nepal, Switzerland, and the United States coded the answers based on the coding scheme, which was developed based on the SRD approach (Killen & Rutland, 2011), and previous research about social exclusion based on social group membership (Gasser et al., 2017;Grütter, Dhakal, et al., 2021;Killen et al., 2010). The coding scheme differentiated fairness concerns from social hierarchies and generalized trait attributions. Fairness concerns included reasons about fairness, equality, and others' welfare (e.g., "Devoki [high SES character] is already popular. She has everything, is talented, but no one wants to take Radha [low SES character]. One should not discriminate poor people."). Social hierarchies included references to the social structure, social status, and power (e.g., "Most people would include people from the same level as they are. If they are high class, they would show high class behavior."). Generalized trait attributions reflected assumptions of how low or high SES students are based on the limited information provided about the SES of the character (e.g., low SES: "She is poor, so she is honest and understandable. The poor try to be friendly."; high SES: "Because he is rich. That is why he is very talented."). Only four participants (0.7% of the sample) referenced negative traits; thus, we considered only positive trait attributions. The initial coding scheme also included personal reasons (e.g., "It is her own choice whether she wants to include X"); however, since only 4% of the students mentioned this category, it was not further considered.
The research assistants received extensive training on how to code the data and participated in multiple sessions to discuss about the codes and the coding scheme. Since the sample was large, two teams of research assistants coded the same 25% of the sample independently after the training was completed. Light's Kappa was calculated with the "irr"-package in R for each pair of coders, whereby the arithmetic mean of all these estimates represents the overall agreement. An agreement above .80 is considered as almost perfect agreement (Hallgren, 2012). For the two research teams, Light's Kappa was κ = .91 (three researchers coding a subset of 54 cases) and κ = .88 (four researchers coding 97 cases).

Democratic teacher beliefs
Teachers rated a subset of 10 items from the Democratic Teacher Beliefs Scale (Shechtman, 2002). The subset of items included beliefs about students from low income F I G U R E 1 Social inclusion task. Hypothetical scenario adapted for the Nepalese context, based on focus group feedback, for boys and girls. The presentation was gender matched. © Nicole Raquinio, Illustrator.
backgrounds and teaching strategies that target equality, responsibility, and fairness, and thus, was relevant for the current study. To adapt the original measure that consisted of 34 items to the Nepalese context, a focus group was conducted with teachers, and the 10 items most relevant to their context was selected (e.g., "The most important attributes of effective teachers are authority and power.", "Students should be discouraged from criticizing textbook content because it is written by the ministry of education.", "Students from lower socio-economic backgrounds have more difficulties in science than students from higher socio-economic backgrounds.", "Students from lower socioeconomic backgrounds are less determined to work hard.", 7-point scale: 1 = strongly disagree, 7 = strongly agree). All 10 items were recoded in order to reflect higher democratic beliefs. The reliability of the scale was α = .67 (M = 2.68, SD = 0.89; high SES schools: M = 2.99, SD = 0.77; mixed SES schools: M = 2.48, SD = 1.00). A full list of all items and additional information regarding the democratic teacher beliefs scale can be found in the Appendix S2.

Data analytic strategy
There were three dependent variables with three different metrics. Thus, depending on the metrics of the dependent variable, we chose different models. The variable reasoning about inclusion was a nominal variable that included the four categories: fairness concerns, generalized trait attributions about low SES students, generalized trait attributions about high SES students, and social hierarchies. Therefore, hypotheses were analyzed with a multinomial logit model (MLM). The variable expectation of inclusion had a binary answer format (expected inclusion of the low SES character: no = 0, yes = 1) and was thus analyzed with generalized linear models (GLM). Both, MLM and GLM estimate the probability that one category is chosen over the other. Since MLM compare multiple categories, one reference category is chosen and compared with all other categories (in this study fairness concerns were the reference category and the likelihood of referencing generalized trait attributions and social hierarchies was compared with referencing fairness concerns). The third dependent variable, moral judgment about inclusion, was a metric response variable and analyzed with a linear model.
In all models, hypotheses regarding age-related differences were investigated with age as continuous mean-centered predictor variable. As the range of the SES variable was very high (ranging from 4,323 NPR to 88,539 NPR), this predictor variable was z-transformed in order to facilitate convergence. Control variables included gender and students' liking of math, as previous work on peer inclusion and reasoning about social inequalities has revealed gender differences (e.g., Flanagan et al., 2014;Grütter & Meyer, 2014). Students' liking of math was considered as an indicator for how much personal relevance students would ascribe to the academic group task which might be related to their inclusion decisions. Similarly, previous work on children's reasoning about social inequalities controlled for the role of personal interest in the scenario that was used to measure their reactions to inequality (e.g., Elenbaas, 2019).
In this sample, students were enrolled in 27 different school classes. To account for the hierarchical data structure, we first investigated whether there was significant between-group variance between classrooms. Therefore, for each dependent variable, we compared the model fit of a model without a random effect to a model with a random effect, allowing the dependent variable to vary between classrooms (Bliese, 2000). For all dependent variables there was significant between-group variance; thus, we conducted hierarchical models. For more details on the specific analyses, see the Appendix S3.
In addition to the confirmatory analyses described above that were derived from previous work on intergroup relations and moral development in other social contexts, we also performed an exploratory analysis. Specifically, as we were interested in the content of the reasoning data, we conducted an exploratory post-hoc analysis, in which the content of participants' generalized trait attributions was analyzed based on Fiske et al.'s (2002) stereotype content model.

Expectation of inclusion: Descriptive information
Across all participants, there were no significant differences for expecting the character to choose the high or low SES character for the math project. Children and adolescents were equally likely to pick either character when asked "Who do you think will Durga pick to join the group?" (51% for the low SES and 49% for the high SES character). A central question was whether the likelihood of inclusion would differ depending on students' SES, school diversity, and teacher beliefs. Next, we report on the underlying reasoning about inclusion and the moral judgment of certain reasoning types (Hypotheses 1a-1c).

Reasoning about inclusion
We first analyzed whether participants referenced different reasons ("Why would Durga choose X?") depending on whether they expected the low SES character to be included or not (H1a). Thus, "expected inclusion" served as a predictor of participants' reasoning, along with age, SES, and gender (Table 1). The results showed that participants' reasoning significantly depended on whether the low SES character was expected to be included. To better understand this effect, we plotted the estimated likelihood for referencing each category in Figure 2. As hypothesized, when participants expected inclusion of the low SES character, they most likely referenced positive trait attributions about low SES children (expected likelihood: 74%), followed by fairness concerns (expected likelihood: 23%). In contrast, when participants expected inclusion of the high SES character, participants most likely referenced positive trait attributions about high SES children (expected likelihood: 85%) and social hierarchies (expected likelihood: 14%).
Furthermore, supporting H1b, the analyses showed age-related differences in students' reasoning about inclusion (Table 1). As predicted in H1b, with increasing age, participants were significantly less likely to refer to generalized trait attributions and were more likely to reference fairness concerns when reasoning about the inclusion of the low SES character (for a detailed description of the follow-up analyses concerning the control for the classroom level, see the Appendix S4).

Exploratory post-hoc analyses on the content of generalized trait attributions
Since the largest proportion of reasoning data about inclusion focused on positive trait attributions about low and high SES students, we conducted exploratory posthoc content analyses. For the 436 students who mentioned generalized trait attributions, we analyzed the content of the attributions based on Fiske et al.'s (2002) stereotype content model. Thus, we distinguished between competence (knowledge, skills, intelligence) and warmth (prosociality, agreeableness, and social responsibility) and added a third category, namely assumptions about effort as these were cited by a majority of the participants. Subsequently, we analyzed whether the participants were more likely to reference these specific trait attributions when expecting either the low or the high SES character to be included. Providing further evidence for H1a, these more fine-grained post-hoc analyses showed that expected inclusion of the low SES character was most likely based on assumptions of the student working hard and being prosocial and responsible, while expected inclusion of the high SES student was most likely based on assumptions of knowledge and skills that could be contributed to the group.
In detail, the results from Fishers' exact test with central confidence intervals (Fay, 2018) revealed that when students expected inclusion of the high SES character, they were nearly four times more likely to reference competence as compared with when the low SES story protagonist was included, p < .001, odds ratio (OR) = 3.60 [CI 95 = 2.34, 5.55], Φ = 0.30. In contrast, when expecting the inclusion of the low SES character, participants were more than six times more likely to reference hard work as compared with when the high SES story protagonist was included, p < .001, OR = 6.36 [CI 95 = 3.96, 10.22], Φ = 0.39. Similarly, traits associated with warmth were more likely attributed to the expected inclusion of the low SES as compared with the high SES story character, p < .001, OR = 3.41 [CI 95 = 2.13, 5.56], Φ = 0.26 (for a detailed description of the categories, see the Appendix S5).

Moral judgment about inclusion
To analyze the moral judgment about inclusion (i.e., "What do you think, is it ok or not ok that Durga choose X?"), we investigated whether students evaluated inclusion of the low and high SES character differently. Moreover, we tested our prediction that students would evaluate reasons concerning intergroup relations, namely positive trait attributions and social hierarchies, more negatively compared with fairness concerns. We computed a T A B L E 1 Multinomial logit model on participants' social reasoning about expected inclusion of the low SES character with the model predicting the likelihood to choose trait attributions (regarding low and high SES students) and social hierarchies over fairness concerns  Guisan & Zimmermann, 2000). Of the 605 participants, 38 reasons were missing or not codable, and the 4 respective 21 participants who referenced negative trait attributions respective personal reasons were excluded from these analyses.
hierarchical linear model, predicting the moral judgment about inclusion by the inclusion decision and the four reasoning categories (i.e., fairness, positive trait attributions low SES, positive trait attributions high SES, and social hierarchies), controlling for gender, age, and SES (for the complete model, see the Appendix S6, Table S6). The results showed that participants evaluated the inclusion of the low SES character as significantly more ok than the inclusion of the high SES character ( = 0.79, [CI 95 = 0.24, 1.32]). Moreover, in line with H1c, when participants referred to reasons of social hierarchies, they evaluated the inclusion decision as significantly less ok as compared with when they referred to fairness reasons ( = −0.96, [CI 95 = −1.57, −0.35]); however, H1c was only partially supported as there was no significant difference regarding positive trait attributions. Students did not evaluate expected inclusion as significantly less ok when they assumed that the protagonist decided based on generalized trait attributions rather than considerations of fairness reasons.

Expectations of inclusion: Participants' SES, school diversity, and teacher beliefs
We tested our second set of hypotheses regarding the role of participant SES, school diversity, and democratic teacher beliefs for expected inclusion in a stepwise procedure using a hierarchical GLM with a binomial distribution. In a first step, we entered the predictor and control variables at level 1 (i.e., students: gender, age, SES, and math liking). Supporting H2a, high SES students were less likely to expect the inclusion of the low SES student ( Table 2, Step 1). However, when the level 2 classroom variables were entered in the next step, SES was not significantly associated with expected inclusion, but instead the SES diversity of the school significantly predicted expected inclusion. Supporting H2b, students in mixed SES schools were significantly more likely to expect the inclusion of the low SES student than students in high SES schools ( Table 2, Step 2). School diversity accounted for 50% F I G U R E 2 Reasoning about inclusion of the low SES student. Predicted likelihood (based on the MLM displayed in Table 1) for referencing each of the categories depending on whether participants expected inclusion of either the high or low SES character. The bars represent transformed log-odds of the likelihood of referencing a specific reasoning type. The 95% CIs are depicted. MLM, multinomial logit model; SES, socioeconomic status. of the differences in expected inclusion between school classes (i.e., between-group variance). In a next step, we investigated whether there were differential findings for low and high SES students in schools that varied in SES diversity. Moreover, we investigated whether these effects depended on age.
The results from Table 2, Step 4 show that students' expectations of inclusion depended on their age, SES, and school diversity, partially supporting H2c (i.e., the hypothesized significant three-way interaction; see Figure 3). As age was entered as a mean-centered continuous predictor variable, we created an interaction plot (Figure 3) with the predicted values based on 1 SD above and below the mean age (i.e., comparing younger participants, 11.47 years old, with older participants, 14.95 years old).
For younger participants, the results showed that participants from low SES backgrounds expected more inclusion of the low SES character when they attended mixed SES schools as compared with high SES schools (OR = 3.42 [CI 95 = 1.61,7.27]). This difference in inclusion between mixed and high SES schools was not found for younger students from high SES backgrounds (OR = 1.08 [CI 95 = 0.56, 2.13]). Thus, with regards to younger participants, the results supported our assumption that experiencing diversity would have a protective effect for students from low SES backgrounds.
In contrast, when more closely examining the results for older participants, the interaction graph and posthoc contrasts showed that low SES students were equally expecting inclusion in mixed SES and high SES schools (with both groups being more likely to expect that the protagonist would include the low SES character). Next, we examined whether there was a significant difference in expected inclusion between older participants from high SES backgrounds when they attended mixed versus high SES schools. Even if the expected means were in this direction (Figure 3), this specific comparison was only significant for participants who were two standard deviations above the mean in their age and SES (OR = 5.27 [CI 90 = 1.01, 27.78]). Still, and importantly, Figure 3 and the post-hoc contrasts show that older participants in mixed SES schools expected high rates of inclusion of the low SES student, independent of their own SES (OR = 1.07 [CI 95 = 0.78, 1.47]). In contrast, when looking at high SES schools, older participants with high SES expected less inclusion than older participants from low SES backgrounds (OR = 0.66 [CI 90 = 0.44, 0.99]). T A B L E 2 Results of the multilevel generalized linear models predicting children's expected inclusion of the low SES character Step 1 Step 2 Step 3 Step 4 Note: N = 593 students in 27 classrooms (12 cases were deleted as there was no information on income available). SES diversity of school, 0 = high SES school, 1 = mixed SES school). Control variables are gender (0 = male, 1 = female) and math liking (0 = not at all, 6 = extremely). The variable democratic teacher beliefs was centered at the grand sample mean. We report odds ratios with their 95% confidence intervals (CI) for effect sizes. We report marginal and conditional R 2 GLMM as an estimator of the explained variance, whereby R 2 GLMM(c) can be interpreted as the variance explained by the entire model while R 2 GLMM(m) represents the variance explained by the fixed and random factors (Nakagawa & Schielzeth, 2013).
Inspecting the results regarding democratic teacher beliefs, there was no direct effect (contrary to H2d); however, the hypothesized interaction between school diversity and democratic teacher beliefs (Table 2, Step 3) revealed that teacher beliefs played a different role in mixed and high SES schools (H2e). Students in mixed SES schools displayed higher rates of expected inclusion than students in high SES schools and they were equally likely to expect inclusion, independent of democratic teacher beliefs (OR = 0.95 [CI 95 = 0.75, 1.21], see Figure 4). Importantly, students in high SES schools expected more inclusion when their teacher expressed higher support for democratic teaching strategies than when their teacher reported lower democratic beliefs (OR = 1.57 [CI 95 = 1.10, 2.23]). Thus, teacher beliefs played a supportive role for students' expectations about inclusion in high SES schools.

DI SC US SION
The current study revealed novel findings about Nepalese children's and adolescents' judgments and reasoning about the inclusion of peers from low and high SES backgrounds in an academic context. Recent research in the United States has shown that children's and adolescents' views about SES are associated with expectations about academic performance, personality traits, and decisions about whom to include in peer groups (e.g., Elenbaas, 2019;Mistry et al., 2015). However, very little research has examined the development of attitudes about SES in other sociopolitical contexts with respect to variables such as school diversity and teachers' beliefs. The current study was conducted with Nepalese children and adolescents and provided new evidence regarding F I G U R E 3 Students' expected inclusion depending on school diversity and student SES. Students' expected inclusion of the low SES character (predicted values) as a function of students' SES (±1SD), age (±1SD), and SES diversity of the school (i.e., high SES vs. mixed SES). Transformed log-odds of choosing the low SES story-character over the high SES character with 95% confidence intervals are shown. SES, socioeconomic status.

Social reasoning about inclusion based on SES
Overall, expectations of inclusion were guided by generalized traits associated with low and high SES peers. When considering the content of these attributes, participants associated traits regarding effort and warmth with the low SES student while competence beliefs were more significantly related to the inclusion of the high SES student. These findings extend recent research on the stereotype content model regarding social class stereotypes (Durante et al., 2017) with adult samples to the study of children and adolescents. Our findings integrate in this prior work, which suggests that with increasing levels of social inequalities, low SES peers are perceived as less competent while high SES peers are perceived as lower in warmth. Similarly, previous developmental research conducted in the United States revealed that competence beliefs were more strongly associated with high SES than low SES characters (Sigelman, 2012). At the same time, however, high wealth individuals were perceived as low in warmth in a social context where children and adolescents negatively evaluated group exclusivity of high SES peers (Burkholder et al., 2019;Elenbaas & Killen, 2018). Hence, children's and adolescents' beliefs about low and high SES individuals may not only entail group biases, but also considerations about existing social inequalities.
In the current study, expected inclusion of the low SES character was evaluated as significantly more legitimate than the inclusion of the high SES character, even though the high SES character was more likely perceived as competent. Thus, biases in competence beliefs did not predict participants' evaluations. Moreover, when expecting the inclusion of the low SES character, some participants also mentioned fairness considerations (e.g., "Helping such children is our responsibility. It is not fair to discriminate them."). Yet, when the high wealth character was expected to be included, a quarter F I G U R E 4 Students' expected inclusion as a function of democratic teacher beliefs. Students' expected inclusion of the low SES character (predicted values) as a function of democratic teacher beliefs (±1SD) and SES diversity of the school (i.e., high vs. mixed SES). Transformed logodds of choosing the low SES story-character over the high SES character with 95% confidence intervals. SES, socioeconomic status. Democratic teacher beliefs Predicted inclusion of the low SES student School diversity high SES schools mixed SES schools of the participants provided reasons of social hierarchies (e.g., "Because now to the poor … see, the poor would be discriminated a bit, whether they are good in school or not."). With regards to the context of the study, Nepalese children and adolescents grow up within a system of unequal treatment. Taken together, these findings indicate that growing up with social inequalities does not necessarily translate into acceptance of the observed disparities.
The results of the current study contribute to the literature on adolescents' negative perceptions of societallevel inequalities concerning wealth disparities (for research conducted in the United States, see Arsenio & Willems, 2017, for Europe, see Niehues, 2014 for Argentina, see Barreiro et al., 2019). Moreover, older children and early adolescents are willing to address economic disparities, particularly in the context of peerbased exclusion (Elenbaas, 2019). With age, however, adolescents develop an even more detailed understanding of how contextual factors in the social system limit life outcomes for disadvantaged groups (e.g., Flanagan et al., 2014;Grütter, Dhakal, et al., 2021), which may translate into a more pervasive concern for fairness. The current study aligns with this prediction given that, with increasing age, participants were more likely to provide reasons of fairness when justifying their expectation about inclusion in an academic context. This finding supports the well documented increase in adolescents' capacity to coordinate concerns about various aspects of the situation (Smetana et al., 2014) and extends the literature on adolescents' reasoning about social exclusion based on race or sexual orientation to the context of socioeconomic disparities. Hereby, our findings integrate with previous work showing that with age, adolescents expressed more negative evaluations about exclusion based on sexual orientation (Horn & Sinno, 2014) and a higher awareness of negative consequences of exclusion based on stereotypes about race .

School diversity, SES, and age-related differences regarding expected inclusion
Schools are important agents for reducing group biases Turner & Cameron, 2016). Participants in mixed SES schools were more likely to expect inclusion of the low SES peer for the math activity than were students in high SES schools. This finding bears on intergroup contact research, which has shown that intergroup contact based on race and ethnicity has reduced prejudice and bias (e.g., Graham, 2018;McGlothlin & Killen, 2006). The findings of the current study expanded on these previous insights by demonstrating that students attending schools that were integrated by SES benefitted from this experience in terms of inclusive orientations, even in a societal context defined by a long history of social hierarchies.
The current work also points to potential developmental differences regarding whether school diversity is related to students' expectation about inclusion. Older participants were more likely to expect inclusion when attending mixed SES schools, independent of their SES. In contrast, in schools that were not diverse in terms of SES (i.e., high SES schools), older participants from high SES backgrounds expected less inclusion than older participants from low SES backgrounds. Moreover, when specifically comparing older participants from high SES backgrounds in high SES and mixed SES schools, they were more likely to expect inclusion when attending mixed SES schools. However, this specific effect only existed for late adolescents who belonged to the highest SES groups in the sample. Taken together, these results revealed that diversity of the school was a significant factor for whether older participants from high SES backgrounds expected inclusive peer behavior.
This pattern of results regarding students from high SES majority groups was not found for younger students, even if they attended mixed SES schools. In the context of strong group boundaries such as Nepal, increasing diversity with regards to SES may not be enough to target intergroup relations in older children and early adolescents. Participants from high SES backgrounds mostly expected the inclusion of the high SES character. Thus, younger participants from high SES backgrounds may have preferred to affiliate with other students from high SES backgrounds. As a consequence, positive contact, which is needed to foster positive attitudes (e.g., Juvonen et al., 2019), may have been limited. Another potential explanation for the non-significant difference between younger participants from high SES backgrounds in mixed and high SES schools may be that, during late childhood and early adolescence, group boundaries increasingly gain in salience, rendering in-group affiliations to be an important and rigid part of group identity which can increase the display of in-group biases (Horn, 2003;McGuire et al., 2021;. Thus, younger participants may have benefitted less from intergroup contact experiences than older participants. For students from low SES backgrounds, the findings partially supported our hypothesis that students from low SES backgrounds would benefit from a more equal balance of power in mixed SES schools. Previous evidence documented that students from ethnic and racial minority groups report less peer victimization in schools with higher ethnic and racial diversity (Graham, 2018;Juvonen et al., 2019). Thus, we assumed that, similar to racially and ethnically diverse schools, mixed SES schools would promote a safe environment with regards to potential peer discrimination. However, in contrast to our assumption that this effect would be independent of participants' age, the results revealed age-related differences. While older participants from low SES backgrounds more likely expected the inclusion of the low SES character, independent of the school environment, younger participants' expected inclusion strongly differed depending on the school environment. When attending mixed SES schools, younger participants from low SES backgrounds were more likely to expect the inclusion of the low SES character. However, in high SES schools, they expected lower peer inclusion. Thus, diverse school environments may have a protective effect for students from low SES backgrounds, but may be more relevant for younger students.
A potential explanation for this finding can be derived from social identity theory (Tajfel & Turner, 1979) which posits that individuals from lower status groups may protect their social identity from negative evaluations by evaluating their own social group more positively. Alternatively, social minority group students may also want to belong to a positively valued social group and identify with higher status groups (e.g., Verkuyten & Martinovic, 2006). Social identities of minority group students may be more accepted in schools with higher diversity (Graham, 2018). Hence, students from low SES backgrounds in mixed SES schools may have been motivated to identify with the low SES character, while low SES students in high SES schools may have more strongly internalized negative perceptions about low SES individuals. As group membership gains strongly in salience in older children and early adolescents (Abrams et al., 2003;, this may be a sensitive period for protecting their social identity. Future research is needed to clarify how social identities develop depending on school diversity and how schools and teachers may help to protect identity threats (e.g., Umaña-Taylor et al., 2018).

Democratic teacher beliefs and expected inclusion in schools
Students in the high SES schools were more likely to pick the low SES peer when their teachers espoused beliefs about teaching styles that target equality, responsibility, and fairness. In contrast, democratic teacher beliefs were not a significant factor in mixed SES schools. Hence, democratic teacher beliefs may be an important buffer for peer inclusion in high SES schools with little diversity. This finding extends previous work on teachers as socialization agents of their students' positive intergroup attitudes (Grütter & Meyer, 2014;Vezzali et al., 2012). We propose that teachers' beliefs translate into their classroom behavior, whereby teachers with high democratic beliefs may express more egalitarian behavior toward students. Hence, teachers with high democratic beliefs may have more positive relations to students, independent of their social background, which helps to promote social acceptance (Farmer et al., 2019). In contrast, if teachers hold biases themselves, they are more likely to negatively interact with students from minority groups, potentially lowering their peer acceptance (Okonofua et al., 2016;Warikoo et al., 2016). Future research should explore the consequences of teachers' democratic beliefs regarding intergroup relationships, including investigating direct connections between teacher beliefs and peer interactions in the classroom.
Recent research has demonstrated that intergroup relations depend on the quality of student-teacher interactions (e.g., Farmer et al., 2019;Grütter, Meyer, et al., 2021). Teachers who hold democratic beliefs about teaching may foster higher student participation and responsibility (Shechtman, 2002), providing more opportunities for intergroup contact . Importantly, teacher attitudes may contribute to the perception of inclusive peer norms, which in turn predict inclusive peer behavior (Gasser et al., 2018). The results from the current study suggest that these processes related to teachers' beliefs may be more important in schools that have a numeric minority of students from low status backgrounds.
In contrast to high SES schools, students in mixed SES schools were more likely to expect inclusion, independent of teachers' democratic beliefs. With regards to Nepal, mixed SES schools are more the exception than the norm, as most students from high SES backgrounds are sent to the most prestigious private schools. Due to Nepalese law, all schools have to include a minimum of students from low SES backgrounds, whereby most schools stick to this lower limit (Devkota & Bagale, 2015). With the idea to foster inclusion, some schools in Nepal have recently changed their policies to become more diverse with regards to SES. Therefore, these schools may communicate to their students that social inclusion is warranted and create perceptions of inclusive norms. However, the current study did not include students' perceptions of teacher beliefs, however, nor were analyses conducted on the specific processes that may explain how teacher beliefs translate to specific student behavior. Thus, future research is necessary to fully understand the role of teachers' attitudes and behaviors on adolescents' inclusion and exclusion behavior.

Practical and policy implications
As has been argued in the case of race and ethnicity (Frankenberg & Orfield, 2007), inclusive classrooms help to foster positive school climates and reduce prejudice and bias under the right conditions. These conditions include opportunities for cross-group friendships, teacher support of mutual respect and tolerance, and the principles of fair and equal treatment. The current findings provided support for this position with regards to SES, even in a country with very strong social hierarchies. Furthermore, the findings point to heightened risks for climates of social exclusion in schools with little diversity, particularly for students from social minority groups. Thus, in order to promote equality and inclusion, school policies should include considering the balance of high and low SES in schools, along with the support from teachers for the goals of inclusive school environments. Future research is warranted to better understand the connections between SES-based diversity in schools and the development of positive intergroup attitudes, fairness reasoning, and social biases. Furthermore, teacher education could help to foster inclusion by targeting teachers' democratic beliefs; relatedly, future research could shed more light on how democratic teacher beliefs are acquired.

Limitations and future directions
The cross-sectional data of this study did not allow for testing specific assumptions regarding why students in mixed SES schools expressed more positive expectations about inclusion. For example, diversity can be beneficial for intergroup relations as it provides opportunities for cross-group interactions that can become cross-group friendships over time (Graham, 2018). However, we do not know whether positive effects were due to crossgroup friendships between students from different SES, due to higher perceived norms of inclusion or due to reduced rates of discrimination of minority group students. Thus, it would be important to investigate different mechanisms of change with longitudinal data.
Moreover, parental beliefs were not collected or analyzed, and future research could understand how parental beliefs about inclusion play a role in adolescents' decisions and reasoning, or whether parents with more positive beliefs might be more likely to send their child to mixed SES schools. In order to understand parental motivations for choosing a specific school, a subsample of parents of children in a mixed SES school was interviewed as part of the research project. The results revealed that parents were mostly concerned about the quality of education and proximity of the school, while very few parents explicitly mentioned inclusive education (for details, see the Appendix S7). While inclusive education may not be an explicit reason, it is still possible that parents of students in mixed SES schools express more positive beliefs about inclusion and equality. Thus, future research could focus on the role of parental beliefs, not only as a source of influence, but also from a bidirectional viewpoint with the possibility that inclusive educational settings may change parental beliefs. Relatedly, while students may have positive experiences in mixed SES schools, they may still experience neighborhood segregation which could also be studied in future research.
We sampled responses about inclusion of a high or low SES peer in an academic context given the negative consequences that this type of exclusion has for the classroom context for learning. Future research could more directly analyze whether gender stereotypes about math abilities are related to the exclusion of girls and ethnic minority peers in math-related tasks, contributing to evidence about science, technology, engineering and mathematics (STEM)-related biases (Bian et al., 2018). In addition, to test for the role of group boundaries, it would be interesting to vary the SES of the protagonist and the peer group in the hypothetical scenario in order to determine whether portraying a group that reflects high or low SES would differentially affect participants' expectations for inclusion.

Conclusion
This study contributed new empirical evidence regarding biases based on SES in academic peer interactions with an understudied sample. Given the long-term negative consequences of being excluded on the basis of one's SES, further research is warranted on this topic. Furthermore, studying perceptions of wealth inequalities in a country with a long history of status hierarchies, such as Nepal, provided novel information for understanding students' concerns about the (un)fairness of social inequalities, intergroup biases based on SES, and contextual factors, such as school diversity and teachers' democratic beliefs. The study highlights the potential of inclusive schools with high socioeconomic diversity in reducing biases related to SES. More research is needed in order to better understand how to design schools that enable children and adolescents to reject unfair biases based on social status hierarchies and to promote inclusive social attitudes, interactions, and relationships.