School composition and academic decisions

Abstract

Research on the effects of school composition tends to focus on how it shapes school achievement. In this study, we instead examine how school composition shapes children’s educational aspirations, given their achievement, and if children from different socio-economic backgrounds are affected differently. We apply school-fixed effects on Swedish register data, including all 9th-grade students from 2013 to 2017. Being exposed to a high share of low-achieving schoolmates increases the likelihood of applying for academics instead of vocational tracking across socio-economic backgrounds. In contrast, the share of high-achieving schoolmates is negatively associated with academic tracking only for high-SES children. Being exposed to peers with highly educated parents increases the likelihood of applying for academic tracking for low-SES children, whereas the effect is weaker or even negative for some of the high-SES groups. Together, our results suggest that the academic decisions of both high- and low-SES children could benefit from a less segregated school environment.

Introduction

Social scientists and policymakers have had a long-standing interest in examining how various aspects of school composition shape educational outcomes (Coleman, 1966; Jencks and Mayer, 1990; Engberg and Wolniak, 2010). Traditionally, research has mostly focussed on how school composition shapes various learning outcomes, such as test scores and grade point averages, with findings suggesting that aspects of socio-economic composition, such as classmates’ parents’ education and measures related to family income, are positively correlated with test scores and grades (Van Ewijk and Sleegers, 2010). In addition, studies examining the operation of racial and ethnic school composition, net of socio-economic composition, have found small or no effects on learning outcomes (Jencks and Mayer, 1990; Rumberger and Palardy, 2005a; Hermansen and Birkelund, 2015; Brandén, Birkelund and Szulkin, 2019). Over the last decade, there has been an increasing interest in the way school peers and school composition shape educational aspirations and choices rather than learning outcomes (e.g., Jennings et al., 2015; Palardy, 2020; Smith, 2023). Studies have shown that schools that are successful in achieving high learning outcomes (e.g., test scores and grades) are not necessarily successful in promoting high rates of graduation and enrolment in further education(Rumberger and Palardy, 2005b), which suggests the importance of studying the processes of educational achievement and educational aspirations and decisions independently of one another.

This study aims to increase our understanding of the way school composition affects ambition-related educational decisions, with a particular focus on how school composition impacts socio-economic inequalities in such decisions. We study this by examining whether school composition shapes children’s educational decisions differently depending on their socio-economic background. More specifically, we study how the composition of the school in which children are enrolled at age 15 shapes their educational decisions about applying for academic tracking in upper secondary education, given their grades, and we allow the effect to differ by socio-economic background.

The study capitalizes on rich Swedish administrative data that include all students in Sweden who attended 9th grade between 2013 and 2017 and asks two broad questions:

1. How is a student’s decision to apply to an academic upper secondary education program at the age of 15 affected by the composition of the student’s school peers, viewed in terms of socio-economic background and educational achievement?

2. Are students from different socio-economic backgrounds affected differently?

Analyzing the rich longitudinal Swedish registers enables us to construct detailed measures that capture characteristics at the school-, family- and child levels and to include school-fixed effects to adjust for time-invariant school-level confounders that affect both school composition and academic decisions.

In addition to our focus on ambition-related educational decisions rather than achievement, our explicit focus on the heterogeneous effects of school composition by socio-economic background distinguishes this study from much of the other school composition literature, which has disproportionately often focussed on average effects. The findings from the relatively few previous studies that have examined heterogeneous school composition effects by socio-economic background are inconsistent. While some studies suggest that school composition effects can exacerbate educational inequalities (Crosnoe and Muller, 2014; Klugman and Lee, 2019), others suggest that school composition effects may reduce educational inequalities (Murphy and Weinhardt, 2020; Strømme, 2020). If school composition influences students from distinct groups differently, this has implications for whether a less segregated school system has the potential to compensate for socio-economic differences in the likelihood of pursuing an academic upper secondary education or whether it might instead exacerbate the effects of such differences. Another major contribution of this study is that we distinguish between the compositional effects of peer achievement (share of low and high achievers) and the effects of peer socio-economic composition (parental income and educational level). This enables us to compare effects stemming from contagion mechanisms with effects stemming from social comparison mechanisms. By studying these four measures of school composition, we are in a better position to establish which aspects of school composition matter for students’ educational decisions and whether different forms of composition matter for different students. This will enhance our understanding of the link between school segregation, school composition and educational inequalities.

Background

Two main mechanisms for explaining the link between peer composition and educational ambitions

Building on previous literature in the field, we suggest two, in some respects contradictory, social influence mechanisms. These mechanisms are likely to determine how students’ achievements and their educational ambitions are affected by the composition of their school peers.

The contagion mechanism, which could also be labelled a normative influence mechanism, suggests that students are susceptible to cultures, norms, and dispositions towards academic careers within their school. Such cultures and norms tend to emerge when students deploy family norms and dispositions towards schooling within their educational context, suggesting that family norms influence the educational dispositions of other students within that context. These norms are highly determined by socio-economic status. For instance, a high share of students with an academic background may contribute to the creation of learning-oriented peer cultures within a school (Legewie and DiPrete, 2012), while a high proportion of students whose parents have no academic background might potentially contribute to a downward levelling of norms and ambitions (Lauen and Gaddis, 2013).

Previous studies have used various measures of school composition to capture how students bring these kinds of resources to their school contexts and how this, in turn, shapes their peers’ educational behaviour through local cultures, norms, and dispositions towards education (Van Ewijk and Sleegers, 2010). One general finding is that socio-economic composition tends to be more important than racial and ethnic composition for students’ educational outcomes (Rumberger and Palardy, 2005; Hermansen and Birkelund, 2015). A high share of students with relatively poor parents tends to be associated with negative learning outcomes (Caldas and Bankston, 1997), whereas the share of peers whose parents have occupations associated with high socio-economic status and cultural capital tends to be positively associated with applications for and enrolment in academically tracked education programs (Kauppinen, 2008).

In the present study, we examine the effects of school composition with respect to both parental educational level and income. These dimensions of composition tend to be correlated but represent different underlying resources in terms of human, cultural, and economic capital. Previous findings suggest that parental educational composition tends to be more important for educational outcomes than parental income composition (Cherng, Calarco and Kao, 2013; Fruehwirth and Gagete-Miranda, 2019). One suggested reason for this is that parental education, to a greater extent than income, generates academic cultures and norms that are transmitted from parents to their children and between peers (Kim, Tong and Sun, 2021).

The social comparison mechanism, which is sometimes labelled relative deprivation/gratification, focuses more on the achievement of peers than on their socio-economic background. It suggests that student ambition and aspirations are influenced by comparisons with their school peers and that students’ assessments of their own competence are therefore affected by their relative standing in relation to their peers in terms of achievement and ability (Merton, 1968; Richer, 1976). This implies that two students with the same performance levels will view their achievement differently depending on their peers’ achievements. A student who compares unfavourably to the achievements of his or her schoolmates may experience feelings of failure or discouragement, while a student who compares favourably may experience feelings of success or encouragement. Such comparative influences have previously been shown to have bearing on college enrolment (Davis, 1966), choosing an academic track in upper secondary education (Jonsson and Mood, 2008; Rosenqvist, 2018; Skov, 2022) and course selection in secondary education (Murphy and Weinhardt, 2020). In addition, studies rooted in social psychology have shown that peer achievement and ability may have an adverse influence on students’ academic self-concepts or self-esteem, commonly referred to as big-fish-little-pond effects (e.g., Marsh, 1987; Marsh and Hau, 2003).

These two mechanisms are sometimes argued to cancel each other out, attenuating the role of peers in educational contexts (Goldsmith, 2011). However, it is important to note that the two influence functions are driven by different rationales—educational norms and self-evaluation, respectively. This means that different students ascribe different relevance to various aspects of their peers and their context in different situations (Shibutani, 1955; Merton, 1968; Jennings et al., 2015). For instance, students are likely to be differentially affected by their peers’ backgrounds depending on their own background. Therefore, even if these influence mechanisms work in opposing directions, they are likely to affect students with different backgrounds and characteristics in different ways. This is important since it has implications for whether school compositional effects work to reduce or exacerbate pre-existing inequalities in academic decisions. Below, we discuss the mechanisms behind these potentially heterogeneous peer effects.

Heterogeneous peer effects by student characteristics

Heterogeneous effects from the contagion mechanism

The decreasing marginal returns to resources argument builds on the contagion mechanism but further stipulates that students experience different returns from peer resources depending on their own characteristics and that they do so in such a way that being exposed to a large share of high-SES peers has a compensatory effect on less advantaged students (Jennings et al., 2015). For instance, less affluent students may have larger marginal returns from being surrounded by affluent peers than affluent students who already have access to these kinds of resources at home. This suggests that school compositional effects will be compensatory in the sense that a less segregated school composition will increase the similarity of the educational decisions of students across student backgrounds (Crosnoe and Muller, 2014).

An alternative perspective, which could be called a social separation mechanism, suggests that low-SES students will not be able to capitalize on attending schools with a high share of high-SES peers because social interactions within schools tend to be segregated with respect to socio-economic background (Cherng et al., 2013; Malacarne, 2017). Also, the contexts that characterize high-SES schools, such as educational cultures, norms, and practices, may be harder for low-SES students to decipher than their higher-SES peers (Lareau and Weininger, 2003; Calarco, 2011). All this suggests null- or negative effects from exposure to high-SES peers on the likelihood of applying for academic tracking among low-SES children, whereas such exposure is likely to have a positive effect for high-SES children.

The empirical support for heterogeneous contagion effects is inconsistent, partly because many studies lack data that are suited to studying these effects and partly because compositional effects are likely to vary between different educational systems and types of composition (Buchmann and Dalton, 2002). The existing research on educational applications and enrolment lends support to both the decreasing marginal returns to resources argument and the social separation mechanism. In the US context, studies focussed on college enrolment have tended to find support for the latter; being exposed to a high share of high-SES peers is more advantageous for the college enrolment of high-SES students than for that of low-SES students. Examining how school socio-economic composition is related to taking college preparatory courses in high school, Crosnoe and Muller (2014) found that being surrounded by a high share of high-SES students was more beneficial for high-SES students than for low-SES students. Furthermore, Klugman and Lee (2019) examined college enrolment and found that attending a high-SES secondary school was associated with a higher probability of enrolment in both selective and non-selective colleges for mid- and high-SES students but not for low-SES students. The study also found that the relationship between college enrolment and school composition was nonlinear for mid- and high-SES students, with a high concentration of high-SES students having the largest effects on enrolment in selective colleges, whereas a high concentration of low-SES students had larger negative effects on enrolment in non-selective colleges. In addition, several studies have found that students in general benefit from high-SES school composition, but they have been unable to conclude that students of varying socio-economic backgrounds are affected differently by school composition (Owens, 2010; Bifulco, Fletcher and Ross, 2011; Palardy, 2013).

In contrast, studies situated in a European context have generally found support for the decreasing marginal returns to resources argument, with low-SES students tending to benefit the most from a high-SES school composition. A study based on Norwegian population registers found that students in lower secondary education were more likely to apply for an academic track in upper secondary education if they were exposed to a high concentration of students from an upper-class background, whereas such compositional effects were smaller for high-SES students (Strømme, 2020). In addition, Bertoni, Brunello and Cappellari (2020) used register data to compare siblings in Denmark and found that students whose parents had low levels of education benefited from attending schools with high-SES students in terms of their income at age 30–40, while high-SES students were penalized. Moreover, recent studies in Denmark and the Netherlands have also found that low-SES students are more likely to conform to the educational decisions of their high-SES peers (Smith, 2023; Zwier et al., 2023).

Heterogeneous effects from the social comparison mechanism

Early work on how social comparisons affect the way individuals assess their ability suggested that differences in the comparisons that individuals make depend on the individual’s expected outcomes. This means that levels of deprivation or gratification are dependent on the expectations associated with a given situation, which are, in turn, conditional on one’s own characteristics and the characteristics of the reference group (Davis, 1959). This suggests that students with the characteristics of high achievers (e.g., high-SES students) are more likely to be discouraged if they are situated in a high-achievement school context compared to students with the characteristics of low achievers. Some recent studies have provided speculative support for this proposition. For instance, students eligible for free lunches in the United Kingdom are less sensitive to having high-achieving peers and make more ambitious educational decisions if they are among the highest-achieving students in their class, as compared to more affluent students (Murphy and Weinhardt, 2020). Further, students of immigrant background are less negatively affected by high-achieving peers in Sweden than native-born students, while girls are more sensitive to high-achieving peers than boys (Rosenqvist, 2018). Jointly, these findings suggest that social comparison effects might potentially work in a compensatory fashion, with students who are more likely to be high achievers (here: high-SES students) being discouraged by high-achieving peers to a greater extent than students who are more likely to be low achievers.

In summary, theory and previous research lead us to expect that a more equal school composition across schools, with respect to educational achievement and socio-economic composition, will have compensatory effects on socio-economic differences in the likelihood of applying for an academic track. We would expect the advantageous socio-economic composition of a school to particularly increase the likelihood of applying for academic tracking among students who are, on average, less likely to apply for academic tracks via a process of decreasing marginal returns to educational capital from their peers. Students who would not expect to be high achievers, such as low-SES students, are also likely to be less discouraged by exposure to high proportions of high-achieving students via processes of social comparisons.

The Swedish context and school system

The educational system in Sweden is comprehensive and typically considered to be egalitarian. One of the system’s stated goals is that students’ life chances should be equalized by providing all students the opportunity to succeed in school and by compensating for students’ different backgrounds and conditions. During the period examined in this study, 2013–2017, schooling was compulsory from 1st grade, which children start at the age of seven, until the end of 9th grade. The curriculum is nationally standardized. Despite the goal of equality and the educational system’s compensatory obligation, the achievement gap in school results has increased over recent decades between both students of different socio-demographic backgrounds and students at different schools (Holmlund, Sjögren and Öckert, 2019; Yang Hansen and Gustafsson, 2019). In parallel with these increased differences in students’ school results, socio-demographic segregation between schools has also increased (Bunar, 2010; Östh, Andersson and Malmberg, 2013; Yang Hansen and Gustafsson, 2016).

Grading and applications to upper secondary education

During the studied period, grades were assigned by teachers and based on standardized criteria, which involved matching skills to standardized requirements. To be eligible for upper secondary education, students are required to attain at least a passing grade in math, English, and Swedish or Swedish as a second language (Swedish as a second language is offered to students who do not have Swedish as their mother tongue). Students who are not eligible for upper secondary education upon completion of lower secondary school have the option of attending an introduction program focussed on providing eligibility for upper secondary education, and when students become eligible, they can transfer to another upper secondary track.

Students apply to curriculum-tracked upper secondary education during 9th grade, and there is a wide range of educational tracks for which they can apply. There are two main branches in upper secondary education: academic and vocational. The main purpose of the academic branch is to prepare students for post-secondary education. The vocational tracks prepare students for employment in different manual occupations. The vocational track curriculum is not designed to prepare students for post-secondary education, and for students in these tracks to be eligible for post-secondary education, they must take additional elective courses. Consequently, the share of students who proceed to post-secondary education is substantially lower than that of students enrolled in an academic track (Rudolphi, 2013).

Data and analytical design

Our data are based on a collection of administrative registers supplied by Statistics Sweden and provide rich information about the students’ education and their educational and family context. The registers cover the entire population of students who attended 9th grade between 2013 and 2017. The students have been traced across different registers using unique personal identification numbers. Our data contain information about which school each student attended, their school achievement in 9th grade, and their applications to upper secondary education. Furthermore, our data also contain information about the students’ and their parents’ migration background, as well as their parents’ educational level, income, and occupation. Given that the purpose of the study is to examine how social context impacts academic decisions, net of academic performance, our analytical sample only includes those students who were eligible to apply for an upper secondary school with academic tracking, N = 414,410 students.

Variables

The dependent variable is dichotomous and defined as 1 if the student’s firsthand choice in the final application to upper secondary education was an academic track and 0 if the student had applied to another track or had not applied to an upper secondary track.

Students’ grades are based on their leaving certificate from 9th grade, in which the grade score (GS) is based on their 16 highest grades, ranging from 0 to 320. To address grade inflation over time, the GS has been z-standardized on a yearly basis, where the yearly mean is 0 and 1 represents one standard deviation from the mean. We include a variable capturing students’ immigrant background, distinguishing between (i) both the student and at least one of his/her parents born abroad, (ii) both parents born abroad, while the student was born in Sweden, and (iii) the student and at least one parent born in Sweden, or the student born abroad and both parents born in Sweden. In addition, the analysis includes a dummy variable indicating whether the student is a girl.

Socio-economic status is measured by (i) highest parental education, (ii) household income, and (iii) parental occupational status. Parental education is based on SUN-codes (the Swedish equivalent of ISCED), where students’ parents’ highest education is categorized as; (i) no post-secondary education, (ii) post-secondary education 3 years or shorter, and (iii) post-secondary education longer than 3 years. Household disposable income rank is based on the highest observed household income among the students’ parents and is defined in terms of yearly percentile household income ranks among parents with a child in the 9th grade that same year. Parents’ occupational status is based on 3-digit SSYK-codes (the Swedish equivalent of ISCO-08), using a binary variable indicating whether at least one of the parents had a managerial occupation or was in an occupation that generally requires a post-secondary education. We have tested alternative categorizations of these core variables with equivalent results.

We include four main explanatory variables that capture school composition. These are calculated as leave-out annual means, which means that the ego student is excluded from the calculations and composition is based on all other students within the same school in the same grade level and year (regardless of whether they were eligible for academic tracking). The compositional variables capture two different dimensions of school composition, namely socio-economic composition and student achievement composition. Socio-economic composition is captured through (i) the average income rank within the school and (ii) the share of students with at least one parent with a post-secondary education of more than three years. School achievement composition is measured by ranking all of the students within a cohort into quintiles on the basis of their GSs and calculating the share of students within each school who belong to (iii) the highest quintile (with an average standardized GS of 1.15), and (iv) the lowest quintile (with an average standardized GS of −0.77), respectively.

The data do not contain information on school classes, which would have enabled us to study the composition of smaller social circles. The average 9th grader is exposed to a school cohort comprising 88 students (for the distribution of school size, see appendix, Supplementary Table A1). As a sensitivity check, we ran all the models presented below on a subsample that only included schools with fewer than 88 students in the 9th-grade cohort, and the results are almost identical.

Identification strategy

We run linear probability models focussed on applying to an academic track using Stata version 17. Our main models include school fixed effects (using Stata’s reghdfe command (Correia, 2017)), which enables us to adjust for non-random unobserved selection into lower secondary schools that is correlated with aspects of both school composition and the probability of applying to academic tracking, such as overall school quality, school reputation, and the like. The underlying identifying assumption of school fixed effects is that there is no time-variant, unobserved school-level characteristics that are correlated with school composition and student applications to upper secondary education. To relax this assumption, we include yearly dummy variables that interact with 21 regional dummy variables in the models. These interactions capture unobserved regional changes over time, such as variations in educational supply or in local labour markets, which might affect students’ educational decisions. Our general regression framework is specified as

$${\displaystyle \begin{array}{lll}{\displaystyle Appl{y}_{ist}=}& \beta E\left(PC\right){}_{st,-i}+SE{S}_{ist}\mu & +X_{ist}\pi +{\gamma }_s+{\tau }_{rt}+{\epsilon }_{ist}\end{array}}$$

where i, s, t, and r are indices for the individual, the school, the year, and the region, and where the index −i denotes the exclusion of the focal individual from the composition of peer averages. E(PC) is the measure of school peer composition (share of low achievers, high achievers, peers with highly educated parents, and average income percentile). SES is a vector of socio-economic status based on parental educational level, occupation, and income. X is a vector of individual controls (zgpa, gender, and immigration background), ${\gamma }_s$ represents school fixed effects, ${\tau }_{rt}$ represents regional time trends based on interactions between 5 yearly and 21 regional dummy variables. When examining heterogeneous effects, we interact each measure of peer composition with indicators of either parental education, parental income quintile, or a dummy variable indicating whether at least one of the parents had a skilled occupation. For model specifications, see appendix, Supplementary Tables A4, A5, and A7.

Using this strategy, we model how the yearly variation in school composition with regard to socio-economic background and achievement affects students’ applications to upper secondary school tracks, net of the students’ own characteristics and selection into lower secondary schools. The modelling strategy also allows us to assess whether students with different socio-economic backgrounds are differently affected by the yearly variation in school composition, net of which lower secondary school they have attended. One caveat with the school fixed effects approach is that it does not capture the effects of school composition between schools –it only captures effects within schools. This implies that the models cannot assess how systematic sorting of students into different schools influences educational inequalities or the extent to which students from different socio-economic backgrounds in the same schools are selected differently based on unobserved characteristics. In an attempt to address this shortcoming, Table 3 provides descriptive statistics on the degree to which students with different socio-economic backgrounds experience differences in peer exposure.

Results

Between 2013 and 2017, 491,922 students attended 9th grade. Of these, 422,338 were eligible to enrol in upper secondary education, and of these, 414,410 students had no missing variables. These 414,410 students comprise our analytical population. All observed students (491,922) have been included when calculating school compositional variables and when z-standardizing the GS. The descriptive statistics presented in Table 1 show, not surprisingly, that students who were ineligible to attend an academic track and students with missing variables differ from the students in our analytical population. Their average GS is more than 2 standard deviations lower than that of the students in the analytical population. Moreover, the excluded students are more likely to be boys, foreign-born with foreign-born parents, have lower household incomes, and have parents with no post-secondary education.

Table 1

Descriptive statistics

	Population studied			Ineligible for track or missing variables
	Mean	Std. Dev.	Avg within school Std. Dev.	Mean	Std. Dev.
Apply to academic track	0.718			0.178
Native-born with two foreign-born parents	0.198			0.199
Foreign born	0.087			0.359
Girl	0.493			0.428
z-grade score	0.302	0.63	0.599	−1.815	0.9
Highest parental education
No post-secondary education	0.436			0.772
Post-secondary education 3 years or less	0.373			0.174
Post-secondary education of more than 3 years	0.191			0.054
Household income rank	0.534	0.28	0.264	0.317	0.25
Skilled occupation	0.447			0.128
School composition variables
Share lowest achievers in school	0.192	0.12	0.053	0.299	0.18
Share highest achievers in school	0.203	0.12	0.053	0.143	0.09
Avg income rank in school	0.504	0.11	0.035	0.435	0.11
Share with highly educated parents	0.178	0.12	0.043	0.13	0.1
Observations	414 410			77 512

Main effects of socio-economic background and school composition on applying for academic tracking

Table 2 presents estimates from linear probability models on the likelihood of applying for an academic-track upper secondary school program. Models 1–4 introduce our school composition variables one by one, whereas Model 5 includes them all simultaneously.

Table 2

Regression analysis results

	(1)	(2)	(3)	(4)	(5)
Native born with two foreign-born parents	0.106***	0.106***	0.106***	0.106***	0.106***
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
Foreign born	0.148***	0.148***	0.148***	0.148***	0.148***
	(0.003)	(0.003)	(0.003)	(0.003)	(0.003)
Girl	−0.004**	−0.004**	−0.004**	−0.004**	−0.004***
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
z-grade score	0.318***	0.318***	0.318***	0.318***	0.318***
	(0.003)	(0.003)	(0.003)	(0.003)	(0.003)
Highest parental education (ref = No post-secondary education)
Post-secondary education 3 years or less	0.067***	0.067***	0.067***	0.067***	0.067***
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
Post-secondary education of more than 3 years	0.066***	0.066***	0.067***	0.066***	0.067***
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
Household income rank	0.009***	0.009***	0.009***	0.009***	0.009***
	(0.003)	(0.003)	(0.003)	(0.003)	(0.003)
Skilled occupation	0.031***	0.031***	0.031***	0.031***	0.031***
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
School composition variables
Share lowest achievers in school	0.059***				0.043***
	(0.013)				(0.014)
Share highest achievers in school		−0.108***			−0.110***
		(0.013)			(0.014)
Share with highly educated parents			0.035**		0.067***
			(0.017)		(0.017)
Avg income rank in school				−0.010	0.014
				(0.021)	(0.023)
Constant	0.523***	0.556***	0.528***	0.539***	0.529***
	(0.003)	(0.003)	(0.004)	(0.011)	(0.012)
School fixed effects	Yes	Yes	Yes	Yes	Yes
Number of schools	2093	2093	2093	2093	2093
Regional time trends (Year*Region)	Yes	Yes	Yes	Yes	Yes
Observations	414,410	414,410	414,410	414,410	414,410
R-squared	0.319	0.319	0.319	0.319	0.320

School clustered standard errors in parentheses, ***P < 0.01, **P < 0.05, *P < 0.1.

Our variables of primary interest are the school composition variables. The results show that individuals are more likely to apply for academic tracking if they are surrounded by a large share of low-achieving schoolmates (Model 1) and/or a low share of high-achieving schoolmates (Model 2), even after controlling for the focus individual’s own academic achievement and for other unobserved school effects. These findings are in line with the big-fish-small-pond argument, which states that students arrive at assessments of their own ability by reference to their school peers.

Interestingly, even after adjusting for the focus individual’s family background, selection into schools, and any effect that peer composition may have on one’s school results, having a large share of schoolmates with highly educated parents is associated with an elevated likelihood of applying for academic tracking (Model 3). This effect increases even more after adjusting for the effect of the shares of low- and high-achieving students in school (Model 5). Finally, the income levels of the parents of children in school, on average, have no independent effect on the likelihood of applying for academic tracking (Model 4).

In robustness checks (appendix, Supplementary Table A1), we examined how much these results were driven by controlling for the student’s own grades. The effects of socio-economic composition are very similar in models that exclude controls for individual grades, whereas the effects of achievement composition were attenuated when no adjustment was made for individuals’ own grades. This was expected since the effects of achievement composition on educational decisions reflect a comparative influence mechanism, whereby students form educational decisions based on their own achievement in relation to the achievement of their peers.

The school fixed effects model relies on the assumption that variation between cohorts within a school is random. We test this assumption by conducting placebo regressions using peer compositions from adjacent years (Rosenqvist, 2018; Skov, 2022), which are presented in Supplementary Table A3. The results show that all placebo regressions produce smaller and insignificant estimates, except for when the average income rank is based on the cohort above the student, where the effect is positive and significant. It is possible that the income background of students in the cohort above the students examined may have an influence on the students’ applications. However, one should also note that the original specification is insignificant.

Heterogeneous effects by family background

Next, we examine how the effects of these four compositional measures differ from our three measures of the family’s socio-economic status (parents’ education, income, and occupation).

We start by descriptively examining how applications for academic-track programs, GS, and exposure to different forms of school composition differ between students from different socio-economic backgrounds. Table 3 corroborates the findings of Table 2, showing that parental background is a strong predictor of academic decisions and academic achievement, with parental education, income, and occupational skill level positively affecting GS and the likelihood of choosing academic tracking when applying for post-secondary school education. Table 3 also shows that students from different socio-economic backgrounds are exposed to different peer compositions. For instance, students whose parents have no post-secondary education are more likely to be exposed to low-achieving peers and less likely to be exposed to the highest-achieving students than students whose parents have post-secondary education. These students are also less likely to be exposed to peers with highly educated parents or to peers with a high average parental income. Similar differences in peer exposure are also found when examining differences on the basis of students’ household income and their parents’ occupation.

Figures 1–3 display whether the peer composition in schools has a differential effect by parental background (see appendix, Supplementary Tables A4–A9 for full regressions and linear combinations of main effects and interaction effects). All interaction terms between socio-economic background and composition significantly improve the models unless otherwise specified in the text. The figures report the results for two different models, one of which only uses a single peer composition variable (corresponding to Models 1–4 in Table 2), while the other includes the full set of compositional variables (Model 5 in Table 2).

Starting with educational background, we find that in the model that only includes the share of low-achieving students, the size of this share increases the likelihood of applying for academic tracking, mainly for students whose parents have post-secondary education. However, when controlling for other compositional variables, the effect is reduced and is similar across educational backgrounds (a Wald test for the joint significance of the interaction terms gives P = 0.1714). In line with expectations, for both model specifications, the negative impact of the share of high-achieving schoolmates becomes increasingly important the higher one’s parents’ education, and for children whose parents have no post-secondary education, the share of high-achieving schoolmates has no impact on the likelihood of applying for academic tracking. The effect of the average income rank of schoolmates’ parents is sensitive to the model specification. In the model with no other compositional variables, there is a positive effect for children whose parents lack post-secondary education, while there is a strong negative effect for children whose parents have the highest levels of education. When including controls for other compositional variables, the effects go toward zero for these students, while there is a positive and significant effect for children whose parents have a short post-secondary education. The attenuation of the income rank coefficients for those children whose parents have the lowest and highest levels of education in the full model suggests that the initial association is driven by other compositional characteristics, such as peers’ parents’ education.

Perhaps the most striking result is the effect from the share of peers whose parents have a long post-secondary education. This characteristic is of particular importance for children whose parents lack any post-secondary education, and the effect is weaker for children whose parents have a higher level of education. In fact, for children whose parents have a long post-secondary education, the effect is even negative. The baseline probability of enrolling in academic tracking for the group whose parents have a low level of education is 0.59, and a one standard deviation increase in the share of peers with an academic background would increase this probability by 1.8 percentage points or 3 per cent, to 0.61 (0.18*0.1, see Table 3 (SD) and Figure 1 (β)), based on the estimate from the model that includes all the compositional variables. A similar increase for the group whose parents have a high level of education would decrease the probability of applying for an academic track program by 0.7 percentage points or 1 per cent, from 0.9 to 0.89 (−0.05*0.14, see Table 3 (SD) and Figure 1 (β)). This corresponds to a reduction in the initial difference in applications of 8 per cent, which we interpret as a non-negligible effect size.

The results show a similar pattern if we instead allow for heterogeneous effects by parents’ income or occupational skill level. In terms of results by parents’ income (Figure 2), the share of low achievers has a positive effect among students in the tails of the income distribution, while students in the middle of the income distribution are unaffected by the share of low achievers. However, similarly to the effects of parental education (Figure 1), the differential effects of income are attenuated when controls are included for other compositional effects. The share of high achievers has a discouraging effect that increases with income, and this effect is robust to the model specification used, implying that the discouraging effect of exposure to high achievers is stronger for children who are more likely to apply for academic tracking to begin with. For instance, for children with parents in the highest income decile, a 1 standard deviation increase in the share of high achievers is associated with a 2.5 percentage point decrease in the probability of applying for an academic track (−0.18*0.14, see Table 3 (SD) and Figure 2 (β)). The effect from the average income at school does not show any substantial differences over income quintiles, except for a positive effect among children in the highest income quintile in the model specification that controls for other school compositional characteristics. In line with the results by educational background, the share of children whose parents have a long post-secondary education is particularly important for children from the lower income quintiles but unimportant for children in the higher income quintiles.

Figure 1.

School compo sition effects on probability of applying to an academic track by parents’ highest educational level. Notes: Regression models are provided in appendix, Table A4, and point estimates corresponding to the linear combinations of main effects and interactions presented in the figure are provided in Table A5.

Figure 2.

School composition effects on probability of applying to an academic track by parents’ income quintile. Notes: Regression models are provided in appendix, Table A6, and point estimates corresponding to the linear combinations of main effects and interactions presented in the figure are provided in Table A7.

Table 3

Share applying to academic tracks, grades, and exposure to different peer compositions by socio-economic background

					Peer exposure
		Share applying to an academic track	z-GS	Observations	Share lowest achievers in school	Share highest achievers in school	Share with highly educated parents	Avg. income rank in school
Educational background
No post-sec education	Mean	0.59	0.08	180,728	0.22	0.18	0.14	0.48
	Std. Dev.		0.59		0.12	0.11	0.1	0.11
Post-sec education ≤ 3 years	Mean	0.77	0.38	154,380	0.19	0.21	0.18	0.51
	Std. Dev.		0.6		0.11	0.12	0.12	0.11
Post-sec education > 3 years	Mean	0.9	0.65	79,302	0.15	0.25	0.25	0.55
	Std. Dev.		0.58		0.11	0.14	0.14	0.12
Income quintile
Q1	Mean	0.66	0.09	68,362	0.22	0.18	0.15	0.46
	Std. Dev.		0.62		0.12	0.11	0.11	0.11
Q2	Mean	0.67	0.18	79,058	0.21	0.19	0.16	0.48
	Std. Dev.		0.62		0.12	0.11	0.11	0.11
Q3	Mean	0.66	0.24	85,769	0.21	0.18	0.16	0.49
	Std. Dev.		0.61		0.11	0.11	0.11	0.1
Q4	Mean	0.74	0.36	89,536	0.19	0.2	0.18	0.51
	Std. Dev.		0.61		0.11	0.11	0.12	0.1
Q5	Mean	0.85	0.56	91,685	0.15	0.26	0.24	0.57
	Std. Dev.		0.6		0.11	0.14	0.14	0.12
Occupation
Other	Mean	0.67	0.15	209,944	0.21	0.18	0.15	0.48
	Std. Dev.		0.61		0.12	0.11	0.11	0.11
Skilled occupation	Mean	0.77	0.46	204,466	0.17	0.23	0.21	0.53
	Std. Dev.		0.61		0.11	0.13	0.13	0.11

Finally, the results of the parents’ occupational skill level (Figure 3) tell a similar story. The effect of the share of low achievers is dependent on the model specification. In the simple model that only includes the share of low achievers, children who have a parent with a highly skilled occupation are positively affected, while children with no highly skilled parent are unaffected. However, when controls for other school compositional variables are included, this interaction term is no longer significant, which means that children are equally affected regardless of occupational background (P = 0.0785). The share of high achievers has a stronger negative impact on children from resourceful homes. Peer composition in terms of educational capital is particularly important for children whose parents work in occupations where such capital is less needed. Finally, the effect of average income rank on children who have at least one parent with a skilled occupation depends on the model specification. When the model does not include controls for other school compositional variables, the effect is negative, whereas the effects are positive in a model that controls for other peer characteristics.

Figure 3.

School composition effects on probability of applying to an academic track by parents’ occupational skill level. Notes: Regression models are provided in appendix, Table A8, and point estimates for the linear combinations of main effects and interactions presented in the figure are provided in Table A9.

Discussion

This study has examined how students’ academic ambitions, measured in terms of educational decisions, are affected by a combination of their own socio-economic background and the characteristics of their school peers.

Our findings show that, on average, students’ academic ambitions are positively affected by the share of peers with highly educated parents but less affected by the average income of their peers’ parents. This mirrors the idea that peers’ parental education is better at capturing the educational cultures that parents pass to their children, and that are in turn passed on to their peers, than peers’ parental income (Kim et al., 2021). We also find that peers’ grades influence application behaviour and that exposure to a high share of high achievers may discourage students from making ambitious educational decisions, net of these students’ own grades. These findings support both the proposed normative influence/contagion mechanism and the comparative influence mechanism.

Importantly, we find differential effects of peer characteristics by socio-economic background. The effects are compensatory, with exposure to a more advantageous school composition appearing to compensate for the lack of parental resources among children with low socio-economic status. The effects appear to operate through both the normative and the comparative influence mechanisms. Children of lower socio-economic status, in terms of parental educational level, income, and occupation, benefit more from exposure to peers with highly educated parents than students at the same school whose parents have a higher socio-economic status. This suggests that normative influence has a compensatory and equalizing effect on educational inequalities, given that the marginal return of having peers with highly educated parents is larger for students whose parents have no post-secondary education. Our findings also suggest that comparative influence works in an equalizing fashion. Students whose parents have a post-secondary education or a managerial occupation and students with parents in the highest income quintile are more discouraged from applying for an academic-track upper-secondary education if they are surrounded by high-achieving school peers, as compared to less affluent students with similar grades in the same schools. Combined with the fact that students from low-SES backgrounds are disproportionately seldom exposed to high-SES and high-achieving peers, and vice versa, these findings suggest that a more even distribution of students from diverse backgrounds across schools would benefit low- and high-SES children alike. However, the effect of peers’ parental income does not support this pattern of compensatory effects. An advantageous school composition with respect to parental income does not benefit low-SES students but, in some specifications, benefits high-SES students. However, these differences should be interpreted with caution since they are sensitive to how the model is specified. One potential explanation might be that it means different things for students to be surrounded by peers with highly educated parents and peers whose parents have high incomes. For instance, occupation and income might constitute more salient status symbols than educational attainment, and exposure to a high proportion of peers from high-income households, when education is held constant, could work in a discouraging fashion for less affluent students (e.g., Crosnoe, 2009).

In sum, this study contributes to our understanding of school compositional and peer effects in at least two ways. First, our focus on heterogeneous peer effects by socio-economic background is uncommon and represents an important contribution. The finding that school compositional effects often work in a compensatory fashion in Sweden stands in contrast to findings from other less egalitarian educational systems, where high-SES students appear to be more affected by school composition than low-SES students in terms of both educational achievement (e.g., Kim et al., 2021) and educational attainment (e.g., Crosnoe and Muller, 2014; Klugman and Lee, 2019), but the result corroborates findings from Norway (Strømme, 2020).

Second, this study contributes to the social comparison literature in two ways. (i) Previous studies have tended to examine peer achievement by assuming a linear effect of the average achievement level of peers. By examining the share of school peers who are achieving in the top and the bottom of the achievement distribution, we show that the discouraging effect of being surrounded by high-achieving students is substantially larger than the encouraging effect of being surrounded by low-achieving students. (ii) We also show that social comparison effects influence students differently depending on their socio-economic background, with the share of top achievers at a school working in a compensatory fashion such that students from a more affluent socio-economic background are more discouraged than less affluent students at the same school with similar levels of achievement.

It should be mentioned that our identification strategy assumes that there are no unobserved school-specific trends over the period examined. If this assumption is violated, our results cannot be interpreted as causal. Given that the period examined is short, however, and that we adjust for regional time trends, we believe this problem to be minor. In terms of external validity, it is worth noting that our findings mainly pertain to how school composition shapes ambition-related educational decisions among students in lower secondary education, who, to this point, had never experienced educational tracking. At later stages of the educational career, the effects of school composition are likely to affect students’ educational decisions in a different fashion. For instance, Bygren and Rosenqvist (2020) showed that increased ability-based sorting into upper secondary schools mostly influenced what types of post-secondary education the students applied for and not the overall likelihood of applying for a post-secondary education. Given the scope of this study, we have only focussed on composition with respect to achievement and socio-economic status and have only examined how the effect of composition varies by socio-economic status. To develop further insights into how school composition shapes educational inequalities, research should examine how composition with respect to immigration background shapes different educational outcomes and to what extent students with various migration backgrounds are affected differently by different measures of school composition.

Despite these restrictions in terms of scope, we would argue that this study contributes important new knowledge to our understanding of the way school composition shapes educational inequalities. Our results suggest that in a relatively egalitarian educational system, reducing the level of segregation in school composition with respect to socio-economic background and achievement may also reduce educational inequalities in the level of applications for academic-track education programs by allowing less affluent students to capitalize on the resources of their more affluent peers while simultaneously being less discouraged by high-achieving peers compared to their more affluent schoolmates. To better understand why school compositional effects work in a compensatory fashion in some educational systems while increasing educational inequalities in others, future research would benefit from a comparative approach using data and methods that are harmonized across different countries.

Erik Rosenqvist is an Assistant professor at the Institute for Analytical Sociology at Linköping University. He is currently interested in educational inequality, school peer, and compositional effects. He previously published his research in Sociology of Education and European Sociological Review.

Maria Brandén is an Associate Professor of sociology at the Institute for Analytical Sociology, Linköping University, and at the Demography Unit, Department of Sociology, Stockholm University. Her current research interests include social stratification and inequality, school- and residential segregation, and neighbourhood effects. Her work has, for instance, been published in European Journal of Population, International Migration Review, and the Lancet Healthy Longevity.

Author contributions

Erik Rosenqvist (Conceptualization [equal], Data curation [lead], Formal analysis [lead], Funding acquisition [equal], Investigation [lead], Methodology [lead], Project administration [equal], Resources [equal], Software [equal], Supervision [equal], Validation [equal], Visualization [lead], Writing—original draft [equal], Writing—review & editing [equal]) and Maria Branden (Conceptualization [equal], Data curation [supporting], Formal analysis [supporting], Funding acquisition [equal], Investigation [supporting], Methodology [supporting], Project administration [equal], Resources [equal], Software [equal], Supervision [equal], Validation [equal], Visualization [supporting], Writing—original draft [equal], Writing—review & editing [equal])

Funding

This research was funded by the Swedish Research Council (VR) (2013-07681; 2015-01635; 2017-02177; 2019-00245; 2023-00933); the Swedish Research Council for Health, Working Life and Welfare (FORTE) (2019-00915; 2021-01069); and Riksbankens Jubileumsfond (M18-0214:1).

Ethical statement

This project was ethically vetted with no. 2017/2166-31/5.

Data availability

The data used in this study are provided through Statistics Sweden’s remote access system MONA. Restrictions apply to the availability of these data, which were used under license for the current study and are not publicly available. Standard procedures for the release of Swedish administrative register data apply: https://www.scb.se/en/services/ordering-data-and-statistics/ordering-microdata/.