Elite Schools, Elite Ambitions? The
Consequences of Secondary-Level School Choice
Sorting for Tertiary-Level Educational Choices
Magnus Bygren
1,2,* and Erik Rosenqvist
31
Department of Sociology, Stockholm University, Stockholm 106 91, Sweden,
2Institute for Future Studies,
Stockholm 101 31, Sweden and
3Department of Management and Engineering, Institute for Analytical
Sociology, Linko¨ping University, Norrko¨ping 601 74, Sweden
*Corresponding author. Email: magnus.bygren@sociology.su.se Submitted March 2019; revised December 2019; accepted January 2020
Abstract
We ask if school choice, through its effect on sorting across schools, affects high school graduates’ appli-cation decisions to higher eduappli-cation. We exploit a school choice reform that dramatically increased achievement sorting across secondary schools in the municipality of Stockholm, employing a before– after design with a control group of students in similar schools located outside this municipality. The re-form had a close to zero mean effect on the propensity to apply for tertiary educational programs, but strongly affected the self-selection by achievement into the kinds of higher educational programs applied for. Low achievers increased their propensity to apply for the ‘low-status’ educational programs, on aver-age destining them to less prestigious, less well-paid occupations, and high achievers increased their propensity to apply for ‘high-status’ educational programs, on average destining them to more presti-gious, well-paid occupations. The results suggest that increased sorting across schools reinforces differ-ences across schools and groups in ‘cultures of ambition’. Although these effects translate into relatively small increases in the gender gap, the immigration gap, and the parental education gap in educational choice, our results indicate that school choice, and the increased sorting it leads to, through conformity mechanisms in schools polarizes educational choices of students across achievement groups.
Introduction
Since the early 1990s, more than two-thirds of the OECD countries have increased school choice opportu-nities. Research is more or less unanimous in finding that children from more affluent homes are more likely to exercise school choice and opt out of their local school, and equity concerns have been raised with re-gard to this development. Critics argue that school choice can exacerbate inequalities, as it increases the
sorting of students across schools based on their socioe-conomic background, their ethnicity, and their ability, and that the quality of schooling can become increasingly unequal across schools as a consequence (Musset, 2012). There is also a worry that increased sorting deprives less able students of opportunities to benefit from positive peer effects; that attitudes related to educational choices will diverge more across schools, with students in popular schools acquiring a culture of
VCThe Author(s) 2020. Published by Oxford University Press.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
doi: 10.1093/esr/jcaa008 Advance Access Publication Date: 16 March 2020 Original Article
high achievement/high ambition, while students in less popular schools instead get socialized into a culture of
low achievement/low ambition (Oakes, Gamoran and
Page, 1992;Kelly, 2009).
However, from a theoretical standpoint, the effects of increased sorting on such frames of reference are not clear, and because of that, the effects of increased sorting on educational choices are not clear. In this art-icle, we elaborate on the assumptions underlying the hy-pothesis that an increase in achievement sorting brings about a polarization of attitudes and choices. In the course of that endeavour, we tease out the arguments for a contrasting hypothesis—namely, that an increase in achievement sorting should rather bring about hom-ogenization of such attitudes and choices.
To identify causal effects of achievement sorting, we exploit the effects of an admission reform implemented in the municipality of Stockholm in 2000. Before the re-form, admission was based on residence; students living close to a school had priority in the admission process. After the reform, admission was based on grades; students with high grades were given priority in admis-sions to schools. A consequence of the reform was that achievement sorting across schools increased dra-matically, but socioeconomic and ethnic sorting was un-altered. Further, the formal curriculum remained the same, as did the population from which students were recruited. Usually, it is extremely hard to disentangle the effect of the school achievement composition from many confounding influences. In the present case, the change in sorting across schools seems to have occurred more or less in isolation from sorting dimensions that usually go hand in hand with achievement sorting, implying that it can be treated as an exogenous shock, making it possible to identify the causal effects of peer compos-ition on educational choices with a relatively high degree of confidence.
Conforming and Contrasting Effects
Humans make social comparisons when judging them-selves or their situation, and this seems to be a funda-mental aspect of their nature. By comparing themselves with their social surrounding, individuals learn about their standing in relation to something or someone else. Depending on the characteristics of what is compared, different outcomes are to be expected. If it is a positively valued achievement, relative overachievement will pre-sumably give rise to a positive emotion, and relative underachievement will give rise to a negative emotion. In this way, comparisons lead to satisfaction, happiness, frustration, deprivation, etc. This human trait has been
observed widely and used to help explain a diverse range
of phenomena (Jasso, 1990;Garcia, Tor Schiff, 2013).
Ambitions and aspirations are formed in interplay be-tween individual achievements and the reference group’s achievements, and reference group-oriented explana-tions help us understand choices and outcomes that at first sight may appear as paradoxical to the outside
observer (Merton and Rossi, 1968).
An implicit assumption in the perspective that increased sorting across schools is problematic is that
there are potential conformist peer effects (Sacerdote,
2001;Owens, 2010;Goldsmith, 2011) that are ampli-fied through a social multiplier process in high-sorting contexts. Students are influenced by their peers’ attitudes, achievements, and choices, and if low-achievement students get concentrated in certain schools, a culture of
low achievement/low ambitions develops there (Farkas,
Lleras and Maczuga, 2002; Kelly, 2009). With adoles-cents’ well-known tendency to conform to peer pressures in attitudes, norms, and choices, these will become more similar within schools with increased sorting across
schools (Coleman et al., 1966). Feelings of belonging are
perhaps particularly important among adolescents. Adolescents tend to maintain positive relationships with their peers by conforming to prevailing group norms, for example, norms on ‘appropriate’ educational ambitions
(e.g.,Frank et al., 2008;Ream and Rumberger, 2008).
Previous findings suggest that the conformist mech-anism is at play when students decide on college majors,
and the like (Lyle, 2007;Frank et al., 2008;De Giorgi,
Pellizzari and Redaelli, 2010), especially in comprehen-sive educational systems with less curricular tracking (Buchmann and Dalton, 2002). If conformist pressures dominate other kinds of influences of sorting by achieve-ment, the expectation is that sorting by achievement will polarize of educational choices across achievement groups: Low achievers conform to the educational norms formed among other low-achieving students while high achievers conform to prevailing norms and ambitions formed among other high-achieving students. In a less-sorted environment, there is a more random mix of influences, and consequently less polarization.
Self-fulfilling teacher expectations is a related mech-anism that potentially could generate similar outcomes as the conformist mechanism, or reinforce these, when
achievement sorting increases (Ready and Wright, 2011;
Becker, 2013). Teachers may expect the students sorted into high-achieving schools to be more ambitious, and in response communicate ambitious post-secondary educational pathways as the obvious future choices for these students. Conversely, teachers may also expect students sorted into low achieving schools to be less
ambitious, and in response communicate low expecta-tions by abstaining from encouragement, or even out-right discouragement, of students to pursue ambitious post-secondary educational pathways.
Lacking in this perspective, however, is potential dis-couraging peer effects. Students compare themselves to others in their immediate social surrounding and, depend-ing on how these relative comparisons turn out, effects in the opposite direction of conformity can be expected on students’ self-concepts and the educational choices related
to these (Marsh and Hau, 2003; Jonsson and Mood,
2008;Goldsmith, 2011). All else equal, the more success-ful the school environment, the lower the students’ aca-demic self-concepts, and the less ambitious their
educational choices tend to be (Marsh, 1991). The
so-called ‘big-fish-little-pond’ research demonstrates that school-average achievement has a negative effect on aca-demic self-concept after controlling for individual
achieve-ment (Davis, 1966; Seaton, Marsh and Craven, 2010;
Nagengast and Marsh, 2012;Rosenqvist, 2018). This so-cial contrast mechanism suggests that changing environ-ments, from a setting where a student is a relative high achiever to a setting where the student is a relative low achiever, will depress the student’s educational ambitions.
If social contrast effects are important, the expect-ation is that increased achievement sorting attenuates differences between groups: if a student’s relative stand-ing with regard to achievement matters for their educa-tional choices (e.g., the choice to go on to higher education or not, or the choice to make an ambitious higher educational choice), these should become more similar across groups in settings with higher achieve-ment sorting.
Consequences for Stratification
The question of how achievement sorting affects these choices is important because it gets at fundamental stratifying mechanisms in processes generating gaps in educational outcomes between groups. As pointed out byBoudon (1974), between-group differences in educa-tional attainment can be conceptualized as the combined
outcome of between-group differences in grades
(primary effects) and between-group differences in edu-cational choices given grades (secondary effects). An additional aim of ours is to investigate if and, if so, how achievement sorting modifies such secondary effects of social background. More specifically, we want to see if achievement sorting, through its effects on choices of higher education, increases or decreases between-group gaps (groups defined by gender, social background, and country of birth) in choices.
For this reason, our study is partially related to the literature on educational tracking. Educational institu-tions differ in the degree to which they sort students into different tracks with different curricula based on their
achievements (Woessmann, 2009; Van de Werfhorst
and Mijs, 2010; Bol et al., 2014). Some countries track students into differing-achievement and curricu-lum tracks as early as age 10 (e.g., the Austria and the Germany), while other countries keep their lower sec-ondary school system comprehensive (e.g., the Sweden and the USA). In addition to stratifying learning activ-ities in a very direct way, curriculum tracking affects the social comparison points, or reference groups, to which students belong. In contexts with early tracking, these reference groups should be more homogenous compared to contexts without early tracking. Research suggests that early curriculum tracking generally tends to in-crease educational inequality as well as inequality
between social class and race/ethnicity groups (Pfeffer,
2008; Buchmann and Park, 2009;Woessmann, 2009; Van de Werfhorst and Mijs, 2010;Bol et al., 2014). A few scholars have studied consequences of educational reforms within countries, from differentiated to compre-hensive systems. This research suggests that a develop-ment toward more comprehensive schooling reduces
educational and labour market inequalities (Meghir and
Palme, 2005;Pekkarinen, Uusitalo and Kerr, 2009). To the extent that these studies capture secondary effects of sorting, they give indicative evidence that achievement sorting gives rise to increases in between-group gaps in choices, which is also the prediction that would be made assuming conformity to peers to be the more important mechanism in peers’ influence on educa-tional choice. Assuming the contrast mechanism to be more important, though, gives the opposite prediction. Because social background correlates with achievement, increasing the achievement sorting would, on average, discourage students with a privileged background and encourage students with a non-privileged background if contrast effects are important, leading to decreases in between-group gaps in choices.
The Context: A School Choice Reform for High Achievers
In Sweden, the educational system is goal oriented with the government deciding on the framework of laws and regulations, but operations are decentralized to its 290 local, autonomous municipalities. As mentioned, there is a global trend of increasing school choice
opportuni-ties for parents (Musset, 2012), endowing them with
certain rights to make independent choices regarding
where their children will be educated, and Sweden is no
exception in this regard (seeBjo¨rklund et al., 2006). The
‘liberation model’ of school choice, where school choice is seen as a means of counteracting undesirable effects of
residential segregation and empowering parents
(Archbald, 2004), together with ideological develop-ments that favour the New Public Management para-digm of promoting market-like competition in the
public sector (seeGunter et al., 2016), has guided
educa-tional policymakers in many Western countries for the past couple of decades. In the municipality of Stockholm, an upper secondary school choice reform was implemented in 2000. Until 1999, residential close-ness to a school determined a student’s priority in admis-sion. A student only applied for an upper secondary school program (there are 18 national programs; 12 vo-cational and six academic), with grades deciding admis-sion. Once a student had been admitted to a specific program, he or she was referred to the closest school that offered it. Applicants could express wishes about which school to go to, but the ones living in the school’s catchment area had priority. The cohort that enrolled in the fall of 2000 was the first that was required to apply to both program and school. Applicants were ranked according to their grades, and those with the highest grades among the applicants to a program at a school were admitted first, regardless of their residential
close-ness to the school (USK, 2002).
That the new rules had consequences for student composition across schools is clear, and some schools even had to close down because they failed to attract students when school catchment areas ceased to be used. As with most school choice reforms, its influence on the choices made was, in practice, selective. In this case, only the high achievers got a truly free choice. One clear (and intended) consequence of the reform was a dramat-ic increase in achievement sorting across schools (So¨derstro¨m and Uusitalo, 2010;Bygren, 2016); how-ever, sorting by socioeconomic class and birth country
did not change (Bygren, 2016).So¨derstro¨m and Uusitalo
(2010) found that segregation on immigration back-ground and parental education initially increased but this effect was explained to a large extent by differences in achievements. Analytically, it is important to distin-guish between changes in the observed and unobserved sorting of students across schools. It is clear that sorting based on students’ grades increased with the reform. It is plausible also that the sorting on unobserved phenom-ena (e.g., educational aspirations) increased as well—an issue we will return to in the interpretations of our findings.
Analytical Strategy
A well-known challenge for social research that aims to estimate causal effects of social environments is that individuals select into these based on unobservables. This is why it is hard to disentangle the effects of pre-selection factors from those of post-pre-selection environ-ment factors. Schools are no exception to this rule as students are far from randomly selected into these. For the present project, the observed naı¨ve effect of a par-ticular school context might be a spurious effect of pre-selection unobserved differences between individuals, not the school context per se.
We will circumvent this problem by raising the level of analysis and compare schools in different institutional settings (municipalities) varying in their degrees of achievement sorting. The reform affected a set of schools in just one municipality. Therefore, we can com-pare student outcomes before and after the reform in this municipality, and use students in similar schools outside this municipality as controls. The effects of the reform will be evaluated with a so-called ‘difference-in-difference’ (DD) design to assess the average effect and a ‘difference-in-difference-in-differences’ (DDD) design to
assess heterogeneous reform effects (Angrist and
Pischke, 2008). The logic of DDD is to examine the ef-fect of a treatment by comparing the development of a difference in the treatment group—exposed to the treat-ment in the latter part of the observation period—to the development of a difference in a control group (never exposed to the treatment). For the present case, the ‘treatment group after treatment’ consists of students enrolled in upper secondary schools in the municipality of Stockholm in the year 2000 and onward, whereas the ‘treatment group before treatment’ consists of students enrolled in upper secondary schools in the municipality of Stockholm prior to 2000.
We select the control group school cases strategically in a manner similar to the so-called ‘synthetic control
method’ (Abadie, Diamond and Hainmueller, 2010)
using a data-driven (nearest neighbour matching) method based on pre-treatment characteristics, reducing the arbitrariness in the choice of control group cases. The treatment effect on the choice gap between students
belonging to a high-achievement category (Xi¼ 1)
ver-sus students not belonging to this category (Xi¼ 0) is
captured with the triple interaction coefficient d4below.
P yð ijtÞ ¼ b0þ b1Xiþ b2TREATjþ b3POSTt
þ d1XiTREATjþ d2TREATjPOSTt
þ d3XiPOSTtþ d4XiTREATjPOSTtþ cjtþ oijt
þ uijt:
(1)
In this equation, yijt measures an educational choice indicating a certain level of ‘ambition’ made by student i in school j at time t (more details on the operationaliza-tion of this and other variables are provided in the
sub-sequent section). TREATj is a dummy for Stockholm
schools, with value 1 if an individual went through the upper secondary level in a school in the municipality of
Stockholm, and 0 otherwise. Xi is an individual-level
achievement dummy. POSTt is a dummy for the
post-reform period. As the admission post-reform was imple-mented in 2000, we defined individuals with upper secondary level enrolment in the years 1997–1999 as the
pre-reform group (with POSTt ¼ 0), and individuals
with enrolment years 2000–2002 as the post-reform
group (with POSTt¼ 1). cjtrepresents time-varying peer
characteristics and oijt represents time-varying
individ-ual characteristics. This equation describes a simplified generic empirical model, and we will estimate variants of this depending on the specific question we seek to answer.
Because the reform made the context within schools more homogenous with regard to prior achievements, the conformity hypothesis predicts that the treatment should have made choices more different across achieve-ment groups: high achievers should have increased their propensity to apply for high-status educational programs, and low achievers should have increased their propensity to apply for low-status educational programs.
The contrast hypothesis predicts that the treatment should have had the opposite effect, making choices more similar across achievement groups. Once low achievers are placed in a surrounding of, as it were, ‘equals’, their choices should, on average, become more similar to high-achiever choices. That is, they should in-crease their propensity to apply for higher-status educa-tional programs. The high-achiever choices should, discouraged by a high-achieving environment, become more similar to low-achiever choices. That is, according
to the conformity hypothesis, the expectation is that d4
will be positive; according to the contrast hypothesis,
the expectation is that d4will be negative.
Because we hypothesize that changes in the peer con-text generate the effects, we do a mediation analysis whereby we add measures of the peers’ grades and their
application choices to this equation (cjtÞ. Once the
school peer grade composition and the application behaviours of the peers are accounted for, our
expect-ation is that d4 has attenuated to zero.
First, we investigate mean effects of the reform, then the effect of increased achievement sorting on gaps in educational choices across social and demographic
groups. The way we go about testing this follows the
logic of the empirical equation above, except that Xi
then is an indicator variable representing the social or demographic dimension in focus. Finally, we evaluate the effect of increased achievement sorting on gaps in educational choices across achievement groups, which
corresponds directly to estimatingequation 1.
Data and Variables
We used a compilation of population registers provided
by Statistics Sweden (listed inTable 1, column 3, and
Appendix Table A1, column 2). With regard to the
measurement of the dependent variable, research usually conceptualizes educational choice as realized matches between students and educational programs, conflating the two sides of these matches: the student applications and the admissions decisions based on these tions. As we have access to the students’ actual applica-tions, we were able to isolate the student choice part of this match, and we distinguish between low-, mid-, and high-status choices with three complementary indicator dependent variables. We wanted to combine pecuniary as well as status payoffs of destinations in the larger stratification hierarchy, and we constructed indicator variables based on whether a student applied to a ‘high-status’, ‘mid-‘high-status’, or ‘low-status’ tertiary educational program within 2 years from graduation, i.e., an educa-tional program that, on average, destines its students to certain levels of pay and prestige. To derive these meas-ures, we computed the mean occupational prestige (Treiman, 1977) and mean earnings, by 5-digit
educa-tion codes (so-called SUN-codes)1 among individuals
aged 30–45 in 2001. If a student applied to an educa-tional program with a SUN-code that on average des-tines its students to the two uppermost deciles with regard to prestige and the two uppermost deciles with regard to earnings, we consider the student to have made a high-status educational choice. If a student applied to a tertiary education with a SUN-code that on average destines its students to the eight lowest deciles with regard to prestige and earnings, we consider the student to have made a ‘low-status’ educational choice. ‘Mid-status’ educational choices include all other educa-tional program, as well as applications to single courses or semesters, which are not assigned SUN-codes in the
registers.2
The construct validity of this approach seems to be confirmed when we list the largest educational programs in the different outcome categories. The largest educa-tional programs falling into the high-status category are the following: law, engineering (only certain 4-year
programs), medical school, pharmacy, and architecture. The largest mid-status educational programs are: busi-ness administration, teacher, engineering (only certain 3-year programs), physiotherapy, and certain programs in social sciences (e.g., business administration and political science). The largest programs categorized as low status are: nursing, social worker, media produc-tion, dietician, and journalism. We should perhaps stress that these are relatively low status in the subset of occu-pations requiring post-secondary education, certainly not low status in the larger occupational hierarchy.
Control Group Selection and Control Variables
Researchers often select the control group on the basis of subjective measures of affinity, or simply use ‘untreat-ed’ units hoping that confounding is netted out through the conditioning implied in the modelling and controls. In contrast, we select the control group in a data-driven
manner, considerably reducing arbitrariness. Our
ap-proach is similar toAbadie, Diamond and Hainmueller
(2010). We selected the control group from schools not exposed to the reform with ‘nearest neighbour matching’, where 10 school neighbours were selected for each school exposed to the reform, using the Stata
mod-ule PSMATCH2 (Leuven and Sianesi, 2015). All schools
operated in the pre-treatment period (1997–1999) and the treatment period (2000–2002). While some schools in Stockholm were excluded because they did not oper-ate both in the pre-treatment and treatment periods, 81 per cent of all students that entered an upper secondary school in Stockholm during the period are included in our analytical sample. We drew the control group school neighbours with replacement (i.e., schools out-side of Stockholm may be matched to several schools in Stockholm) and weights between 0.1 and 2 were assigned by the module depending on how similar they Table 1. Yearly means of the school aggregated matching variables in treatment and control groups
Matching variable
Definition Source
Pre-treatment year
Treated Control t-score for test of differences in means
High-status education
School’s share students applying to a high-status tertiary educational program within 2 years from graduation
Swedish higher educa-tion authority register 1997 0.106 0.104 0.1 1998 0.102 0.105 0.1 1999 0.105 0.104 0.04 Mid-status education
School’s share students applying to a mid-status tertiary educational program within 2 years from graduation
Swedish higher educa-tion authority register 1997 0.321 0.329 0.19 1998 0.318 0.319 0.03 1999 0.339 0.330 0.25 Low-status education
School’s share students applying to a low-status tertiary educational program within a year from graduation
Swedish higher educa-tion authority register 1997 0.102 0.092 0.58 1998 0.085 0.091 0.36 1999 0.089 0.101 0.84 Immigration background
School’s share of students who were born abroad, whose parents both were born abroad
Migration register and Multigenerational register 1997 0.369 0.291 1.94 1998 0.359 0.320 0.96 1999 0.393 0.330 1.71 Parents’ education
School’s share of parents with at least tertiary education (domin-ance criterion)
LISA register and Multigenerational register
1997 0.586 0.590 0.09
1998 0.569 0.567 0.06
1999 0.561 0.577 0.35
Parents’ SES School’s share of mid to high service class parents (dominance criterion)
LISA register and Multigenerational register
1997 0.560 0.575 0.35
1998 0.520 0.553 0.75
1999 0.515 0.554 0.89
Final grades School’s average final grades at age 19, based on individual students within-cohort rank
Upper secondary edu-cation register
1997 51.2 55.8 0.87
1998 50.0 54.9 0.94
1999 51.8 55.0 0.64
were to the treated Stockholm schools. The treatment group consists of 33 schools with a total of 31,456 stu-dents attending at some point during the study period, and their weight is set to 1. The control group consists of 75 schools with a total of 99,846 students, drawn from a ‘donor pool’ of 372 schools with 521,921 stu-dents outside of Stockholm.
We matched the treatment group and control group schools on yearly pre-treatment school means of the dependent variables, and the following school-level variables: the share of students with an immigration background (i.e., either born abroad or two foreign-born parents), the share of students with at least one ter-tiary-educated parent, the share of students with at least one service class parent, and the school’s average final
grades. InTable 1, we report a balancing test between
schools in Stockholm and matched schools in the pre-treatment period (1997–1999). There are no statistically significant differences in the means of the matching vari-ables between the treatment and control group, implying that the time trends and levels with regard to the de-pendent variables and the additional school-level varia-bles prior to treatment are such that the control group is indistinguishable from the treatment group in the pre-treatment period. As the control group and pre-treatment group are closely matched in the pre-treatment period, this suggests that the control group schools provide a sensible approximation of the ‘what-if’ educational choices that would have been made in the absence of the
reform in the treatment period. In AppendixTable A1,
we report descriptive statistics for all variables separate-ly for the treatment and control group aggregated over the pre-treatment period and the treatment period.
To make the interpretation of the coefficients as intuitive as possible, we relied on linear probability models to estimate effects. Our significance tests and confidence intervals are based on standard errors robust
to heteroscedasticity.3 We investigate heterogeneous
effects across the following dimensions: school achieve-ment, gender, parents’ education, and immigration background. We operationalize school achievement using the pre-upper secondary level GPA rank age 16, derived from the grade point average (GPA) of the indi-vidual at graduation in the 9th grade. To weed out pos-sible effects of grade inflation and changes in the grading system, we transformed grades into graduation year quintile rank scores for the treatment and control group separately, with 1 indicating the lowest quintile scores, and 5 the highest quintile scores. We based the variable Girl on biological sex. Highly educated parent is based on the highest education level either one of the student’s parents, dichotomized into an indicator for
tertiary education. Immigration background has value 1 if the student and/or both parents are born abroad, 0 otherwise. Additionally, we use the following control variables in the regressions: GPA rank age 19, derived from the GPA of the individual at graduation from the upper secondary level, usually at age 19, in terms of graduation year percentile rank scores. Highest parental education with seven levels of education based on the parent with the highest education level. Parental occupa-tional class based on six class positions, based on the parent with the highest class position.
Mediation Variables
Because we hypothesize that changes in the peer context are responsible for the effects of increased soring across schools, we carried out a mediation analysis where measures of the peers’ grades, and the peers’ educational choices were included as controls in the estimation of effects. We define peers’ grades as (i) the mean GPA rank age 16 (i.e., their achievements prior to enrolling in the school) in the student’s upper secondary level school cohort and (ii) the mean GPA rank age 19 in the
student’s upper secondary level school cohort.
Additionally, we include controls for (i) the share of peers applying to a low-status education, (ii) the share of peers applying to a mid-status education, and (iii) the share of peers applying to a high-status education within 2 years from graduation.
Results
By way of introducing the context for the analysis, we
report in Figure 1 the degree of sorting by entrance
grades for the control group and the treatment group.
.16 .2 .24 .28
T−3 T−2 T−1 T T+1 T+2
Treatment group Control group
Figure 1. Average yearly within upper secondary school stand-ard deviations in lower secondary school grades (prior achievements) for treatment and control groups.
Prior to the reform, the Stockholm schools were more sorted in terms of prior achievements. It is clear that the student composition within Stockholm schools became more homogenous within schools with respect to inflow of students after the implementation of the reform. The average within-school standard deviation in prior achievements fell with about 4 grade rank points in the treatment group but was rather stable in the control group.
To get additional overview, we report unconditional means over time on our three dependent variables, in Figure 2. These give an indication of whether there was any average treatment effect on educational choices. We do not see any clear pattern of changes in choices fol-lowing the reform, and the degree of confidence interval overlap is substantial. However, higher shares of appli-cations to low-status eduappli-cations in the control group in T þ 1 and T þ 2 contributes to significant differences in applications to low-status educations between the control group and the treatment group.
This ‘ocular’ conclusion is also, by and large, con-firmed through a series of regression analyses reported
in Table 2. In these analyses, the time dimension has been reduced to two periods, before and after the reform, with control variables introduced step-wise (models 2 and 3), and the mediation variables have been entered in the final models (4 and 5). For choices of high-status educational programs, we find that the treatment effect is positive, but insignificant, irre-spective of controls and mediators included in the regression model. For low-status educational choices, we find that the treatment effect is negative, but close to zero and insignificant in all models. For choices of mid-status educational programs, all models indi-cate a small positive treatment effect on the probability of applying. However, when controlling for gender, immigration background parental characteristics the treatment effect turns statistically insignificant. In
other words, the pattern seen inFigure 2seems rather
robust to conditioning on a large number of factors, showing no evidence that increased sorting affects mean educational choices.
It is interesting to note that after controlling for peers’ prior achievement to capture selection into
.08 .1 .12 .14 T−3 T−2 T−1 T T+1 T+2 Low−status education .32 .34 .36 .38 T−3 T−2 T−1 T T+1 T+2 Mid−status education .08 .1 .12 .14 T−3 T−2 T−1 T T+1 T+2 High−status education
Treatment group
Control group
Figure 2. Yearly shares of students applying to low-, mid-, and high-status education within 2 years from graduation in treatment and control groups, with 95% confidence intervals around the shares. Notes: Each data point represents a coefficient (combination) from a regression including yearly dummy variables interacted with a dummy variable indicating the control group.
Table 2. Estimation results for linear probability model regressions of applying to high -, mid-, and low prestige education within 2 years from graduation, on independent variables
(1) (2) (3) (4) (5)
Dependent variable: applying to a high-status education within 2 years from graduation
Treatment (Stockholm * T2) 0.010 0.010 0.009 0.009 0.007
Stockholm 0.004 0.002 0.004 0.000 0.000
T2 0.001 0.003 0.005 0.007* 0.007*
Rank grade from graduation 0.296* 0.297* 0.277* 0.276*
Girl 0.027* 0.027* 0.026*
Immigrant background 0.055* 0.055* 0.054*
Avg rank grade from 9th grade in school 0.142* 0.135*
Avg school rank grade in school from graduation 0.027 0.064*
Share applying to prestige education in school 0.257*
Share applying to mid-prestige education in school 0.088*
Share applying to non-prestige education in school 0.003
Control for parents dominant occupation No No Yes Yes Yes
Control for parents dominant education No No Yes Yes Yes
Constant 0.117* 0.038* 0.064* 0.114* 0.089*
R2 0.000 0.070 0.086 0.089 0.092
Dependent variable: applying to a mid-status education within 2 years from graduation
Treatment (Stockholm * T2) 0.016* 0.015* 0.014 0.013 0.014
Stockholm 0.007 0.005 0.005 0.001 0.001
T2 0.012* 0.015* 0.014* 0.018* 0.017*
Rank grade from graduation 0.444* 0.397* 0.362* 0.363*
Girl 0.013* 0.013* 0.014*
Immigrant background 0.031* 0.032* 0.033*
Avg rank grade from 9th grade in school 0.216* 0.225*
Avg school rank grade in school from graduation 0.013 0.004
Share applying to prestige education in school 0.229*
Share applying to mid-prestige education in school 0.115*
Share applying to non-prestige education in school 0.055*
Control for parents dominant occupation No No Yes Yes Yes
Control for parents dominant education No No Yes Yes Yes
Constant 0.361* 0.128* 0.114* 0.027* 0.009
R2 0.000 0.071 0.083 0.086 0.088
Dependent variable: applying to a low-status education within 2 years from graduation
Treatment (Stockholm * T2) 0.004 0.004 0.005 0.005 0.005
Stockholm 0.014* 0.014* 0.014* 0.013* 0.012*
T2 0.011* 0.011* 0.011* 0.010* 0.009*
Rank grade from graduation 0.024* 0.015* 0.021* 0.021*
Girl 0.055* 0.055* 0.055*
Immigrant background 0.018* 0.018* 0.018*
Avg rank grade from 9th grade in school 0.097* 0.096*
Avg school rank grade in school from graduation 0.133* 0.125*
Share applying to prestige education in school 0.012
Share applying to mid-prestige education in school 0.004
Share applying to non-prestige education in school 0.017*
Control for parents dominant occupation No No Yes Yes Yes
Control for parents dominant education No No Yes Yes Yes
Constant 0.105* 0.093* 0.066* 0.081* 0.078*
R2 0.001 0.001 0.012 0.012 0.012
Observations 63,647 63,647 63,647 63,647 63,647
Notes: *P < 0.05 using robust standard errors.
schools, and the student’s own achievements, we can see, firstly, that the higher the peers’ grades (at age 19) are, the less ambitious ego’s educational choice becomes, and secondly, that ego’s educational choice appears to be strongly influenced by the peers’ choices. In particu-lar, there seems to be strong conformity effects for mak-ing mid- to high-status choices. These patterns suggest that the achievements and choices of peers have oppos-ing effects on how students value their own achieve-ments in the process whereby they decide on which educational program to apply for: you are discouraged by peers with high grades, but encouraged by peers mak-ing ambitious choices.
We have thus far established that there was an aver-age effect close to zero with regard to low- and mid- and highly status educational choices. Next, we turn to the question of whether the admission reform dispropor-tionately affected certain sociodemographic groups. To get clean reform estimates, we only condition on grade rank age 19 in these models. That is, we choose model 2 as our baseline model and include socioeconomic and
demographic characteristics interacted with the reform, allowing for heterogeneous reform effects conditional
on grades (i.e., we estimateequation 1without cjt and
with Xiindicating the social or demographic dimension
in focus).
The treatment pattern emerging inFigure 3is such
that the rule is an absence of a statistically significant treatment effect on the probability of applying for a certain type of education. Girls shifted their choices somewhat towards mid- to high-status educations. The probability for girls to apply to a high-status education increased by roughly two percentage points, correspond-ing to a relative increase of 14 per cent, as a result of the reform. Students with highly educated parents decreased their propensity to apply for a low-status education, and increased their propensity to apply for a high-status edu-cation, but the estimated treatment effects are rather small, around one percentage point. The tendency is that students with immigrant background shifted their choices towards mid- and to some extent high-status
educations.4 −.03 −.02 −.01 0 .01 .02 .03 .04 .05
Boys Girls educated pa No tertiary
r
Tertiary
educated pa
r
Native
Background Background Immigrant Low−status education −.03 −.02 −.01 0 .01 .02 .03 .04 .05
Boys Girls educated pa No tertiary
r
Tertiary
educated pa
r
Native
Background Background Immigrant Mid−status education −.03 −.02 −.01 0 .01 .02 .03 .04 .05
Boys Girls educated pa No tertiary
r
Tertiary
educated pa
r
Native
Background Background Immigrant High−status education
Figure 3. Point estimates of admission reform effects on probability of applying to low-, mid-, and high-status education within 2 years from graduation, by demographic and socioeconomic characteristics with 95% confidence intervals. Notes: Each data point represents a group-specific reform effect (d4inequation 1), controlling for grades in upper secondary education.
These findings indicate that the reform tended to reinforce stratification in educational choices, but rather little compared to baseline pre-treatment group
differen-ces in educational choidifferen-ces (see AppendixTable A2). Note
further that as confidence intervals strongly overlap with those for the reference categories, they provide indicative rather than conclusive evidence in this direction.
Given that the reform sorted students into upper secondary schools based on prior achievements we also expect the reform effect to vary over the achievement distribution. To recall: according to the conformity hy-pothesis, the treatment is expected to increase differences in choices across achievement groups, but according to the contrast hypothesis, the treatment is expected to de-crease these differences. We therefore proceed by dividing the student population into five quintiles based on prior achievements from lower secondary school, interact each
quintile with the reform variable (i.e., Xiinequation 1is
a vector of quintiles), and report our estimates inFigure 4
(conditioning on grade rank at age 19).
The treatment pattern strongly points in the direction of support for the conformity hypothesis. With increased sorting, educational choices become more similar within achievement groups: the low-achievement group increases their propensity to choose low-status educa-tional programs, the mid-achievement group increases their propensity to choose mid-status educational pro-grams and the high-achievement group increases their propensity to choose high-status educational programs. These effects are substantial. The probability of applying to a low-status education increased by three percentage points for the lowest-achieving students, which corre-sponds to a rather dramatic relative increase of 67 per cent. The estimated reform effect on the probability of applying to mid-status educational track indicates a bell-shaped treatment pattern where the probability increased by more than four percentage point for stu-dents in the middle of the achievement distribution, which corresponds to a relative increase of 13 per cent, while the probability for the lowest-achieving students −.08 −.06 −.04 −.02 0 .02 .04 .06 .08 Q1 Q2 Q3 Q4 Q5 Low−status education −.08 −.06 −.04 −.02 0 .02 .04 .06 .08 Q1 Q2 Q3 Q4 Q5 Mid−status education −.08 −.06 −.04 −.02 0 .02 .04 .06 .08 Q1 Q2 Q3 Q4 Q5 High−status education
Reform effect
Reform effect conditional on
peer application & grades
Figure 4. Point estimates of admission reform effects on probability of applying to high -, mid-, and low prestige education within 2 years from graduation, by five achievement groups, based on prior achievements, with and without controls for peers’ average grades and peers’ applications, with 95% confidence intervals. Notes: Each data point represents a coefficient from a DDD regres-sion with separate reform effects (d4inequation 1) based on grades from prior achievements (lower secondary education). The
point estimates represented by a circle include controls for grades achieved in upper secondary education. The point estimates represented by squares also include controls for upper secondary school peers’ average grades and the share of school peers applying to a low-status, mid-status, and high-status education, respectively.
were decreased by four percentage points (a relative de-crease by 20 per cent). The highest achievers’ probability of applying to a mid-status education where unaffected by the reform. The treatment pattern for applying to a high-status education instead indicates a clear upwardly sloping treatment pattern over the achievement distribu-tion, where students in the highest-achieving quintile seem positively affected by the reform, while the other students were either unaffected or even (marginally) negatively affected. The relative reform effect corre-sponds to a relative increase of 20 per cent among the highest-achieving students, despite their high baseline probability of applying to a high-status education (23 per cent).
What we see here is evidence that achievement sorting seems to enhance the achievement stratification in educa-tional choices, with high-achieving students adjusting their application choices ‘upwards’ by pursuing more ambitious post-secondary educational programs, and low-achieving students adjusting their application choices in the ‘downward’ direction, pursuing less ambitious post-secondary educational programs.
In the next step of the analysis, we estimate the reform effects conditional on the average peer grades at the point of making the educational choice, and the application be-haviour of the peers at the same time. The estimates at-tenuate somewhat for high-status choices, but not very much, and no change is statistically different from zero. Though attenuated, most reform effects are still positive and statistically significant, indicating that some, but far from all, of the reform effects are mediated by achieve-ments and choices of the students’ peers.
Discussion
In this article, we rigorously study the effects of achieve-ment sorting on educational choices, using an admission reform that exogenously increased grade sorting, but had no or little effect on sorting by socioeconomic back-ground and country of birth, allowing us to convincingly isolate the effect of achievement sorting on educational choices. We find a close to zero mean effect of the re-form on educational choices, but strong diverging effects across achievement groups in the kinds of higher educa-tional programs applied for.
When we divided the students into groups based on their prior achievements (grades in lower secondary education), a heterogeneous effect pattern from the reform appeared, where high-achievers became even more ambitious in their choices with the reform, and low-achievers became less ambitious in their choices with the reform. Mid-achieving students became more
prone to apply to mid-status educations while their probability of applying to other types of education was not affected much. We thus find rather substantial con-formity effects. High-status educational programs are more likely to be chosen the more concentrated the high-achievers are in a school, and low-status educations are more likely to be chosen the more concentrated the low-achievers are in a school. Previous research suggests that one of the main mechanisms that drive the effect of increased achievement sorting in high-achieving con-texts is conformity to peers, making attitudes, and choices, more similar among because of spillover to
school peers (Sacerdote, 2001; Kelly, 2009; Owens,
2010; Rosenqvist, 2018). Cultures of high ambitions and norms appear to evolve in schools with a high con-centration of high achievers and a culture of more mod-est ambitions and achievement norms emerge in schools with a high concentration of low achievers.
The increase in achievement sorting can be linked to an increase in between-group choice gaps; groups that are usually identified as coming from less privileged homes (i.e., students with an immigrant background and students without tertiary-educated parents) became more prone to make mid-status educational choices given their grades. Girls on the other hand tend to out-perform boys in terms of educational attainment on average but are not excessively found in educations
lead-ing to the most prestigious jobs (cf.Charles and Bradley,
2009). We indeed find that girls given grades become
less likely to apply for high-status educations, but that increased achievement sorting actually increased girls’ likelihood of applying to such educational programs in relation to boys.
It is important to note that peer effects go in different directions: you are encouraged by peers making ambi-tious educational choices, but discouraged by peers with high grades. Our results further suggest that the former aspect appears to be more important for high achievers, whereas the latter aspect appears to be more important for low achievers. It should however be stressed that this is what the results suggest, as both rank grades might capture a mix of ability and different types of peer effects, difficult to discern empirically. Some of these effects can be attributed to the reform’s effect on (our measures of) peer composition, but not all, which is problematic given the aims of our study. Our data clear-ly show that sorting based on students’ grades increased with the reform, but as the increased school choice opportunities were left to students to exploit, it is plaus-ible also that sorting on unobserved individual charac-teristics increased as well. Another way of saying this is that we might have measurement error in peers’
achievement sorting (ability selection into schools) proxied by rank grade at age 16. In our case, the school choice reform might have affected certain portions of ‘compliers’, on average positively selected on high edu-cational aspirations, or aspirations to somehow leave their current circumstances and get ahead in society, and some of the effects that we observe are probably gener-ated by such an unobserved selection process.
Self-fulfilling teacher expectations is a related mechanism that potentially could have contributed to generating the outcomes we observe. When achievement sorting increases, teacher expectations and taken-for-granted visions of their students’ future educational at-tainment—which we do not measure—may change accordingly, and reinforce peer effects of observed and unobserved sorting on ambitions.
Our results are in line with findings from the related literature on curriculum tracking, where highly tracked educational systems increase polarization in educational attainment over ability and socioeconomic background (Pfeffer, 2008; Buchmann and Park, 2009;Bol et al.,
2014). We have been able to separate achievement
sorting effects from curriculum tracking effects since the reform increased the sorting by achievements while the curriculum was unaffected, suggesting that some of the polarizing effects of tracking probably are a conse-quence of achievement sorting amplifying differences in educational choices between groups.
With a global trend toward increasing school choice
opportunities (Musset, 2012), the worry that increased
sorting deprives less able students of the opportunity to benefit from positive peer effects has been voiced. The concern is that attitudes related to educational ambitions diverge more across schools, with students in high achieving schools acquiring a high achievement/ high ambition frame of reference, while students in low achieving schools instead get socialized into a low
achievement/low ambition frame of reference (Oakes,
Gamoran and Page, 1992;Farkas, Lleras and Maczuga, 2002;Kelly, 2009). Our results corroborate this conjec-ture, and point to a possibly problematic consequence of school choice policies.
Notes
1 The Swedish education coding system SUN is a
clas-sification system for educations, harmonized with International Standard Classification of Education (ISCED). The first three digits indicate level of edu-cation and the last two specifies field of study.
2 The median destination for this select group of
post-secondary level graduates is around the 70th
percentile for both the status hierarchy and the in-come hierarchy, so drawing the line at the 80th per-centiles is not extreme. The results are relatively robust to different grouping tresholds of the depend-ent variable. One motivation for using the currdepend-ent tresholds is that they make the low-status and high-status categories of similar in size. The mid-high-status group is inherently more heterogenous, why we place applications with less information about future prospects in terms of income and prestige, as well as applications in between low-status and high-status educations, in this category.
3 Using school clustered standard errors is not a
feas-ible approach because clustered standard errors in principle assume clusters to have been randomly drawn from a cluster population. In contrast, we use a population of clusters (schools) suggesting that school clustered standard errors would be too large. In addition, clustered standard errors become
inflated if treament effects are heterogenous (Abadie
et al., 2017), which is clear in our case, see Supplementary Figures SA2–SA7.
4 Increased achievement sorting into upper secondary
educations may also affect the grades achieved in upper secondary education. That is, peer effects on grades should affect applications to tertiary educa-tions. However, excluding grades from upper sec-ondary education does not alter the findings in Figure 3 significantly (see Supplementary Figure SA1).
Supplementary Data
Supplementary dataare available at ESR online.
Acknowledgements
We are indebted to William Carbonaro, Frida Rudolphi, Martin Ha¨llsten, and Jan O. Jonsson for reading and comment-ing on previous drafts of this manuscript.
Funding
Funding from the Swedish Research Council for Health, Working Life and Welfare (grant #2011-0968) is gratefully acknowledged.
References
Abadie, A., Diamond, A. and Hainmueller, J. (2010). Synthetic control methods for comparative case studies: estimating the effect of California’s Tobacco Control Program. Journal of the American Statistical Association, 105, 493–505.
Abadie, A. et al. 2017. When Should You Adjust Standard Errors for Clustering? NBER Working Paper No 24003. National Bureau of Economic Research, Inc.
Angrist, J. D. and Pischke, J.-S. 2008. Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton, NJ: Princeton University Press.
Archbald, D. A. (2004). School choice, magnet schools, and the liberation model: an empirical study. Sociology of Education, 77, 283–310.
Becker, D. (2013). The impact of teachers’ expectations on stu-dents’ educational opportunities in the life course: an empiric-al test of a subjective expected utility explanation. Rationempiric-ality and Society, 25, 422–469.
Bjo¨rklund, A. et al. 2006. The Market Comes to Education in Sweden: An Evaluation of Sweden’s Surprising School Reforms. New York: Russell Sage Foundation.
Bol, T. et al. (2014). Curricular tracking and central examina-tions: counterbalancing the impact of social background on student achievement in 36 countries. Social Forces, 92, 1545–1572.
Boudon, R. 1974. Education, Opportunity, and Social Inequality: Changing Prospects in Western Society. New York: John Wiley & Sons.
Buchmann, C. and Dalton, B. (2002). Interpersonal influences and educational aspirations in 12 countries: the importance of institutional context. Sociology of Education, 75, 99–122. Buchmann, C. and Park, H. (2009). Stratification and the
forma-tion of expectaforma-tions in highly differentiated educaforma-tional sys-tems. Research in Social Stratification and Mobility, 27, 245–267.
Bygren, M. (2016). Ability grouping’s effects on grades and the attainment of higher education. Sociology of Education, 89, 118–136.
Charles, M. and Bradley, K. (2009). Indulging our gendered selves? Sex segregation by field of study in 44 countries. American Journal of Sociology, 114, 924–976.
Coleman, J. S. et al. 1966. Equality of Educational Opportunity. Washington DC: National Center for Educational Statistics. Davis, J. A. (1966). The campus as a frog pond: an application
of the theory of relative deprivation to career decisions of col-lege men. American Journal of Sociology, 72, 17–31. De Giorgi, G., Pellizzari, M. and Redaelli, S. (2010).
Identification of social interactions through partially overlap-ping peer groups. American Economic Journal: Applied Economics, 2, 241–275.
Farkas, G., Lleras, C. and Maczuga, S. (2002). Does opposition-al culture exist in minority and poverty peer groups? American Sociological Review, 67, 148–155.
Frank, K. A. et al. (2008). The social dynamics of mathematics coursetaking in high school. American Journal of Sociology, 113, 1645–1696.
Garcia, S. M., Tor, A. and Schiff, T. M. (2013). The psychology of competition: a social comparison perspective. Perspectives on Psychological Science, 8, 634–650.
Goldsmith, P. R. (2011). Coleman revisited: school segregation, peers and frog ponds. American Educational Research Journal, 48, 508–535.
Gunter, H. M. et al. 2016. New Public Management and the Reform of Education: European Lessons for Policy and Practice. London: Routledge.
Jasso, G. 1990. Methods for the theoretical and empirical ana-lysis of comparison processes. In Clogg, C. C. (Ed.) Sociological Methodology. Oxford and Cambridge, MA: Basil Blackwell Ltd, pp. 369–419.
Jonsson, J. O. and Mood, C. (2008). Choice by contrast in Swedish schools: how peers’ achievement affects educational choice. Social Forces, 87, 741–765.
Kelly, S. (2009). Social identity theories and educational engage-ment. British Journal of Sociology of Education, 30, 449–462. Leuven, E. and Sianesi, B. (2015). PSMATCH2: Stata module to perform full Mahalanobis and propensity score matching, common support graphing, and covariate imbalance testing. Statistical Software Components.
Lyle, D. S. (2007). Estimating and interpreting peer and role model effects from randomly assigned social groups at west point. The Review of Economics and Statistics, 89, 289–299. Marsh, H. W. (1991). Failure of high-ability high schools to deliver
academic benefits commensurate with their students’ ability levels. American Educational Research Journal, 28, 445–480. Marsh, H. W. and Hau, K.-T. (2003). Big-fish–little-pond effect
on academic self-concept: a cross-cultural (26-country) test of the negative effects of academically selective schools. American Psychologist, 58, 364–376.
Meghir, C. and Palme, M. (2005). Educational reform, ability, and family background. American Economic Review, 95, 414–424. Merton, R. K. and Rossi, A. S. 1968. Contributions to the theory of
reference group behavior. In Merton, R. K. (Ed.), Social Theory and Social Structure. New York: The Free Press, pp. 279–334. Musset, P. 2012. School Choice and Equity: Current Policies in
OECD Countries and a Literature Review. Paris: OECD Education Working Paper, Organisation for Economic Co-operation and Development.
Nagengast, B. and Marsh, H. W. (2012). Big fish in little ponds aspire more: mediation and cross-cultural generalizability of school-average ability effects on self-concept and career aspirations in science. Journal of Educational Psychology, 104, 1033–1053. Oakes, J., Gamoran, A. and Page, R. N. (1992). Curriculum
dif-ferentiation: opportunities, outcomes, and meanings. In Handbook of Research on Curriculum. pp. 570–608. Owens, A. (2010). Neighborhoods and schools as competing
and reinforcing contexts for educational attainment. Sociology of Education, 83, 287–311.
Pekkarinen, T., Uusitalo, R. and Kerr, S. (2009). School tracking and intergenerational income mobility: evidence from the Finnish comprehensive school reform. Journal of Public Economics, 93, 965–973.
Pfeffer, F. T. (2008). Persistent inequality in educational attain-ment and its institutional context. European Sociological Review, 24, 543–565.
Ready, D. D. and Wright, D. L. (2011). Accuracy and inaccur-acy in teachers’ perceptions of young children’s cognitive abilities: the role of child background and classroom context. American Educational Research Journal, 48, 335–360.
Ream, R. K. and Rumberger, R. W. (2008). Student engage-ment, peer social capital, and school dropout among Mexican American and non-Latino White students. Sociology of Education, 81, 109–139.
Rosenqvist, E. (2018). Two functions of peer influence on upper-secondary education application behavior. Sociology of Education, 91, 72–89.
Sacerdote, B. (2001). Peer effects with random assignment: results for Dartmouth roommates. The Quarterly Journal of Economics, 116, 681–704.
Seaton, M., Marsh, H. W. and Craven, R. G. (2010). Big-fish-little-pond effect generalizability and modera-tion—two sides of the same coin. American Educational Research Journal, 47, 390–433.
So¨derstro¨m, M. and Uusitalo, R. (2010). School choice and seg-regation: evidence from an admission reform. The Scandinavian Journal of Economics, 112, 55–76.
Treiman, D. J. 1977. Occupational Prestige in Comparative Perspective. New York: Academic Press.
USK. 2002. Fo¨ra¨ndrad Elevsammansa¨ttning Pa˚ Gymnasieskolan I Samband Med A¨ndrad Intagningsprincip [Change in Student Composition following Change in Admission Principle].
Stockholms stad: Stockholm: Utredningsoch statistikkontoret (USK).
Van de Werfhorst, H. G. and Mijs, J. J. B. (2010). Achievement inequality and the institutional structure of educational sys-tems: a comparative perspective. Annual Review of Sociology, 36, 407–428.
Woessmann, L. (2009). International evidence on school track-ing: a review. CESifo DICE Report, 7, 26–34.
Magnus Bygren is Professor of Sociology at the Department of Sociology, Stockholm University. His re-search centers on understanding ethnic, gender, and class inequality processes. Recent work has appeared in European Sociological Review, Social Forces, and Sociology of Education.
Erik Rosenqvist is a post-doctoral fellow at the Institute for Analytical Sociology, Linko¨ping University. Current research interests comprise social stratification, educa-tional inequality, and peer effects. His work has been published in Sociology of Education.
Appendix
Table A1. Descriptive statistics of treatment and control groups divided into pre- and post-reform periods
1997–1999 2000–2004
Variables Register source Treatment Control t-score for test of differences
Treatment Control t-score for test of differences
Apply to a high-status education within 2 years from graduation
Swedish higher education authority register
0.11 0.12 1.05 0.12 0.12 1.72
Apply to a mid-status education within 2 years from graduation
0.35 0.36 1.31 0.36 0.35 1.61
Apply to a low-status education within 2 years from graduation
0.09 0.11 3.95 0.10 0.12 4.97
GPA rank age 16 Lower secondary
education register
0.55 0.58 9.39 0.56 0.60 12.44
GPA rank age 19 Upper secondary
education register
0.52 0.52 1.60 0.53 0.53 1.09
Girl Background register 0.54 0.54 0.39 0.55 0.53 1.84
Immigration background Migration register 0.33 0.29 7.53 0.33 0.30 6.80
Parents’ dominant occupation Population censuses
Unskilled manual worker 0.07 0.06 1.93 0.08 0.08 0.89
Skilled manual worker 0.08 0.08 0.71 0.08 0.09 3.96
Routine non-manual labour lower grade 0.12 0.11 2.49 0.11 0.11 0.61
Routine non-manual labour higher grade 0.23 0.25 2.75 0.23 0.23 0.94
Professional 0.36 0.37 1.91 0.33 0.35 2.28
Self-employed 0.06 0.07 0.97 0.07 0.06 2.11
Unknown 0.08 0.07 4.86 0.10 0.08 5.82
(continued)
Table A1. (Continued)
1997–1999 2000–2004
Variables Register source Treatment Control t-score for test of differences
Treatment Control t-score for test of differences
Parents’ dominant education LISA (longitudinal inte-gration database for health insurance and labour market studies) register Short compulsory 0.01 0.01 0.42 0.01 0.01 1.14 Compulsory 0.04 0.04 0.36 0.03 0.03 3.14 Lower secondary 0.17 0.15 2.85 0.16 0.17 1.14 Upper secondary 0.13 0.12 0.33 0.13 0.13 0.42 Lower tertiary 0.05 0.06 3.02 0.06 0.06 0.65 University degree 0.51 0.51 0.32 0.51 0.51 0.89 Postgraduate 0.04 0.06 7.10 0.05 0.06 4.85 Unknown 0.05 0.04 5.56 0.05 0.04 3.53
Table A2. Probability of applying to educations by sociodemographic and achievement groups before the reform (1997–1999)
Pre-reform probability of applying to education by group
High-status education Mid-status education Low-status education
Treatment group (per cent) Control group (per cent) Treatment group (per cent) Control group (per cent) Treatment group (per cent) Control group (per cent) Boys 11 10 35 32 6 8 Girls 12 11 36 34 12 15 Native background 10 10 35 34 9 11 Immigrant background 13 11 35 32 10 13
Students without tertiary-educated parents
7 6 26 24 9 11
Students with a tertiary-educated parent 14 14 42 41 9 12 Achievement quintile 1 6 2 20 12 5 6 Achievement quintile 2 3 2 19 19 7 9 Achievement quintile 3 7 7 32 32 12 14 Achievement quintile 4 13 12 44 43 11 15 Achievement quintile 5 23 25 51 49 8 12 All students 11 11 35 33 9 12
