Introduction

1The propagation of education in the modern era permitted to open status and power positions outside hereditary privilege. It led to longer years in formal learning, a growing average education level and an increasing participation rate of lower social classes at all education levels (Moore, 2004) and, eventually, induced the massive increase of social mobility. However, the access to increasingly demanding educational qualifications has always been unequal. Thus, it stratified the impact of this propagation and converted education to the foundational process of the modern socially unequal meritocracy (Dubet, 2004; Duru-Bellat, 2009).

4This paper explores the relation of educational performance with social and urban inequalities in the Athens Metropolitan Area during the 2000s. It draws on discussions about education as a mechanism of social reproduction and on work about education inequalities in Greece, and in particular in Athens, where a socially stratified secondary education –compared to much less stratification in the rest of the country– leads to a rather open, but at the same time socially unequal higher education. The focus is on the transition from secondary to higher education. We relate the performance of candidates in the Greek national admissions examination (Panelladikes Exetaseis) and its outcome in terms of the Department of studies they eventually accessed to the socially diversified options within higher education and therefore we consider performance not only in terms of score but also as a predictor of different occupational and social futures. We analyze how this performance relates to the social position of candidates’ families, to the type of secondary schools they attended, to the social profile of candidates’ residential neighborhoods (tentatively) and to their demographic features (age and sex). The object is to illustrate and roughly quantify the function of social reproduction in this socially selective process of transition to higher education.

The social selectivity of school systems and the impact of residential location

State of the art

5National education systems manage social selectivity in various ways. Unique curricula throughout secondary school, as in most Scandinavian countries, are in principle less selective compared to socially differentiated curricula from an early age, like in Germany or the Netherlands; or to options representing privileged paths to the occupational elite, like the filières to the Grandes Écoles in France (Felouzis, 2009).

7The wave of change in educational policies since the 1990s in many countries was driven by the idea of promoting parental choice. These policies are in fact related to the (actual or presumed) support of the middle classes for more ‘consumer choice’ in education. The reforms of New Labour (‘Excellence in Cities’, ‘Educational Priority Areas’, ‘Five-year Strategy’) increased parental choice (Oria et al., 2007), while pro-choice policies justified as boosters of educational attainment eventually increased educational inequality (Ball, 1993 and 2006; Ball et al., 1995 and 1996; van Zanten & Kosunen, 2013; Power et al., 2003; Seppänen, 2003; Bosetti, 2004; Denessen et al., 2005; Riddell 2005; Butler and Van Zanten, 2007; Dubet et al. 2010; Dronkers et al. 2010; Merle, 2012). More parental choice characterizes also education policies in the US, like G.W.Bush’s NCLB (no child left behind) or Obama’s RTTT (race to the top) and the proliferation of Charter schools that ‘have contributed to the privatization and non-profitization of urban schools across the country’ (Patterson and Silverman 2013).

8The proliferation of Charter schools in the US can be related to processes of urban revitalization and gentrification at least in some places, like Chicago and Philadelphia (Davis and Oakley, 2013). The link between gentrification and school segregation is also discussed for London by Butler et al. (2013) and for Amsterdam by Boterman (2012 and 2013) who identifies significant differences in the spatial location preferences among middle-class groups with different amounts of cultural capital. These preferences are also significantly related to gender for middle-class parents in both Amsterdam and London according to Boterman and Bridge (2015). In the UK, where parental choice was promoted since 1988, the competition for allocation to good schools reinforces the relation between school and residential segregation by reducing the distance for eligibility to good schools (Hamnett & Butler, 2013) and reintroduces the criterion of distance, which is relaxed as school quality decreases. Increased parental choice has also increased educational inequalities in Sweden with pupils from lower social profile and immigrant backgrounds travelling smaller distances, and therefore, exercising less their choice for better schools (Andersson et al., 2012, Östh et al., 2013). The relaxation of catchment areas in France, following pro-choice policies, had ambiguous social consequences. It was advertised as a tool for working class and other underprivileged families to access better schools than those in their areas; it eventually served families from classes that are more informed and more driven by educational objectives (Oberti et al., 2012; Merle, 2012). Increased parental choice in Berlin introduced a tension between parents who prioritize this choice as a right –and whose behaviour eventually increased school segregation especially in terms of evading ethnic diversity– and those who filter their strategies through other concerns (Noreisch, 2007a and 2007b). Similarly, in another German region (North Rhine-Westphalia) less educated parents take much less initiative and well-educated ones consider school choice not only as an opportunity, but also as a duty (Ramos Lobato and Groos, 2019).

9Residential segregation and school segregation are therefore closely related: social groups are unevenly distributed in residential space and in schools in ways that usually reproduce their advantage or disadvantage. School and residential segregation usually work in tandem: areas with better schools attract more middle-class residents that eventually improve local schools’ performance increasing further their attraction. This relation is, however, contextually diversified. In US metropolitan areas, schools with greater social, financial and instructional resources are serving high rather than low-income neighborhoods (Owens and Candipan, 2019). In Dutch cities it seems that school segregation is largely the effect of residential location patterns (Boterman, 2019), while in Paris school segregation appears to be stronger than residential segregation (Oberti and Savina, 2019).

10On the other hand, residential segregation is also important per se due to its assumed impact on living conditions and on chances of social mobility. There is a substantial literature on the neighborhood or area effect, mainly developed in the U.S. (Ellen and Turner, 1997). This literature addresses issues related to education, like the lack of role-models related to the absence of successful middle class groups; the forms of social capital that constrain rather than enable social mobility, and the poor quality of services (e.g. schools) (Atkinson and Kintrea, 2001, p.2278).

12Neighborhood effects vary in different contexts. Enforced spatial isolation, as in the black ghetto, obviously reduces opportunities for social mobility to a much higher degree (Massey and Denton, 1993; Wilson, 1987) than spatial separation in comparatively low segregation environments and relatively evenly serviced residential areas, as in Dutch cities. In the latter, neighborhood effects may be found to be of considerably less importance for social mobility than the personal/household characteristics of the relatively isolated and deprived groups (Ostendorf et al., 2001; Musterd et al., 2003). Neighborhood effects in South European cities can be expected to be somewhere in-between due to the contradictory influence of, on the one hand, the absence of highly segregated areas and groups and, on the other, the relatively poor and unevenly distributed social services.

13High levels of residential segregation usually induce high levels of school segregation. The opposite, however, is not necessarily true. Low levels of residential segregation are not necessarily combined with low levels of school segregation. As we will see in the following - where we only partly address the intricate relations between school segregation, residential segregation and social origin in Athens - low levels of residential segregation may not preclude relatively high levels of school segregation and the formation of clear socially unequal educational trajectories.

Educational inequality and residential segregation in Athens

14The Greek educational system has a unique curriculum up to lower secondary school (Gymnasio/Gymnasium) that concludes the nine years of compulsory education and a dominant general option in the upper secondary (General Lykeio/Lyceum) which comprises about 70% of enrolment compared to a smaller (30%) vocational option (UNESCO, 2012).

15In Athens, more than anywhere else in Greece, there is an important private segment operating in secondary education (8,5%) without financial assistance from the State. According to Dronkers et al. (2010) Greece and the U.K. are the only EU countries where private schools do not receive public funds and, therefore, can determine their recruitment policy. Most private schools perform better than average and especially the few elite schools that also offer options (like International Baccalaureate) related to prospective studies abroad (Valassi, 2008). Public schools are much more socially mixed and of variable performance. The transition from secondary to higher education is organized through a national admissions examination since the 1960s (Maloutas, 2019). Performance in these examinations gives access to higher education following the demand for each Department. Mere access to higher education is long out-dated as an indicator of social distinction and, therefore, socially separate paths within higher education have to be taken into account.

16The foundational work of Tsoukalas (1977) stressed the democratic character of Greek secondary and tertiary education, in terms of the massive access provided to students of lower social origin and of the important wave of social mobility it has supported for quite a long period; Frangoudaki (1985) discussed the ‘hypertrophy’ of higher education as an important factor that led to its internal social diversification, which Lambiri-Dimaki (1974) had initially depicted.

17Kontogiannopoulou-Polydorides (1999) showed that, since the 1960s, the chances of candidates from families of professionals and office employees were much higher than those from farmers and the working class; and that, at least since the 1980s, these inequalities are not limited to the acquisition of degrees, but are closely related to the unequal ways that graduates with similar degrees fare subsequently in the labor market. Thanos (2011) drew similar conclusions regarding the important differences amongst socio-professional categories in terms of access to different types of higher education Departments.

18Panayotopoulos (2000) showed that the Faculties of Medicine, Law and most Schools of the National Technical University of Athens (especially Architecture and Mechanical Engineering) are reserved, to some extent, for upper and upper-middle social strata, while those of Theology or Education are mainly relegated to lower and lower-middle ones. Highly demanded Faculties, like Medicine and Law, remain very unequally accessible by students from different social backgrounds. Moreover, they harbor a far higher rate of endogenous reproduction (i.e. within the family’s occupational line) compared to other occupations that also require university degrees (Maloutas, 2007b).

19Sianou-Kyrgiou (2008) stressed the importance of extra-curricular preparation for the admissions examination to higher education. Candidates massively participate in this preparation, sometimes for several years before the event. The process is privately organized on collective or individual basis and the cost is high, especially for its most individualized forms. The cost, as well as the socially uneven awareness of the importance and the workings of this preparation, lead to systematic social differences that mitigate the socially equalizing impact of the predominantly public character of secondary education.

20Hadjiyanni and Valassi (2009) claim that the rapid development of postgraduate studies and the unequal prospects offered by different types of institutions, academic disciplines and specialties have created new social inequalities within higher education and/or reinforced existing ones. They conclude that the attenuation of inequalities in accessing higher education has been counterbalanced by new divisions at the postgraduate level.

21The 'democratization' of higher education did not necessarily lead to more social justice. Thus, researchers shifted their interest to inequalities within higher education and revealed new divisions and hierarchies (Sianou-Kyrgiou, 2010, Spyrellis, 2013), confirming that the education system produces and reproduces inequality in changing forms that may be increasingly difficult to identify (Bourdieu and Passeron, 2000).

23Residential segregation in Athens is relatively reduced both in terms of class and ethnicity for a host of reasons related to its urbanization model, to the local welfare system and the role of family networks (Allen et al., 2004); to the workings of the housing market and the structure of housing supply; to the reduced presence of a foreign corporate elite in its labor and housing markets etc. (Maloutas 2007a, Maloutas et al. 2012). Affluent families are hindered from relocation strategies to good school areas and induced to seek school segregation in other ways in order to gain educational advantage.

Data analysis

The effect of social origin

Table 1: Departments of higher education institutions clustered following the education level and the occupation of students’ parents (percentages). Final Cluster Centers.

Parents Education level	Cluster of Faculties and Departments
	1	2	3	4	5	6	7	all
Higher education (F+M)	55.9	42.5	34.1	23.4	23.2	14.6	13.0	26.8
Compulsory to post-secondary (F+M)	39.1	49.9	56.4	64.7	60.3	66.8	65.1	59.6
Less than compulsory (F+M)	5.0	7.6	9.5	11.9	16.5	18.6	21.9	13.7
Occupation
Managers-Professionals (F)	52.0	39.4	27.3	21.1	25.4	12.8	13.7	24.6
Intermediate professions (F)	30.2	35.9	44.2	46.8	33.4	48.9	34.1	41.1
Working-Class (F)	12.0	16.7	20.4	22.8	28.9	27.3	39.5	24.5
N of Depts and Facs	36	55	66	86	51	97	54	445
N of admitted candidates (2005)	1475	2115	4047	5018	1887	4976	3092	22610
Average score of admitted candidates (max=2000)	1837	1623	1536	1269	1098	1059	1017	1292

F =Fathers; M = Mothers

27We introduced this clustering of Departments as an additional variable in our main dataset and examined the correlation between this social hierarchy of departments and the score of admitted candidates. The correlation appears to be strong (Spearman’s rho = -.786) witnessing that the socially more exclusive Departments are accessed through higher scores and, therefore, that children of higher social categories obtain systematically higher grades at the admissions contest. This correlation is also clear when we compare the average scores of admitted candidates in each cluster of departments (table 1).

28We calculated the socially uneven chances to get admitted in each of these clusters of Departments by comparing the percentage of enrolled students from different socio-educational backgrounds in each of the seven clusters to the distribution of education levels in the whole population aged between 40 and 75 (roughly corresponding to the expected age of students’ parents). Figure 1 shows that candidates from a highly educated family have over four times more chances than the average candidate to get admitted to one of the most demanded Departments (cluster 1) and over 34 times more than the average candidate from a poorly educated family. The range of inequality decreases as we move down the hierarchy of Department clusters. The higher social strata seem to lose interest in the less prominent part of the hierarchy (their chances compared to those of candidates from poor educational background decrease from 34 times [cluster 1] to 2 [clusters 6 and 7]) while those of candidates from intermediate educational backgrounds decrease less steeply, but remain much higher than those of the less privileged candidates even at the end of the clustered hierarchy (respectively from 7 times [cluster 1] to 3 [clusters 6 and 7]).

29Higher socio-educational groups maintain a systematically uneven access to the best segments of higher education through their systematically higher performance in the admissions examination –the magically unequal distribution of educational merit in favor of higher social strata according to Duru-Bellat (2009). According to Sianou-Kyrgiou and Tsiplakides (2011) the effective outcome in terms of chosen options within higher education is socially unequal even when performance is similar.

The school environment

30The quality of secondary schools is the second parameter we assumed important in explaining uneven access to higher education. One way to assess this quality is to measure the average performance of candidates from each secondary school in the admissions examination. However, this measure does not necessarily –or entirely– reflect the quality of schools per se (i.e. the quality of educational work, the educational credentials of the teaching staff, the state of the infrastructure or the organization efficiency) as differences in performance amongst schools may be due to the uneven social profile of their clientele. Since our dataset does not comprise information on the social origin of candidates, our measures of school quality involve both quality per se and the effect of the different social composition of their pupils.

31The correlation between candidates’ performance in the admissions examination and the average performance of the school they attended is important (R = .362). This means that 13% of the variance in candidates’ performance (R2 = .131) is ‘explained’ by school’s performance.

32Renowned private schools are singled out for their high performance as well as for their social selectivity (Valassi, 2012). On the contrary, the small lower tier of private schools –comprising 1.6% of students, compared to 7.0% for the rest of private schools– is of low educational performance and serves a clientele of a different social profile. Evening schools –mostly public– accommodate working students, either from lower socioeconomic backgrounds or mature students. These schools account for 3.2% of the student population and have usually a low rate of admission to higher education. The bulk of secondary schools are daytime public schools, which account for 88.3% of the student population. Among these, a limited number of ‘experimental’ schools used to select students on performance and to implement innovative education methods. Although the selectivity of experimental schools has been tampered, they continue to have a systematically higher performance amongst public schools. They account for 3.3% of the student population.

33We produced a 9-category hierarchical variable (table 2) taking into account schools’ public or private status, their daytime or evening operation and their experimental or regular character. Where necessary, we subdivided these categories according to average school performance in the admissions examination to higher education.

34This hierarchical variable is significantly correlated with the hierarchy of higher education Departments where candidates were eventually admitted (R = -.333).

Table 2: Student population (seniors) and average performance in the admissions examination to higher education by type of school in Athens (2005)

Type of school	Number of students (seniors)	%	Average performance
evening school	1014	3.2	876
private – low performance	495	1.6	944
public – low performance	7572	23.5	973
public – mid-low performance	11029	34.5	1109
public – mid-high performance	3631	11.3	1192
public – high performance	5006	15.6	1276
public experimental	1048	3.3	1323
private – mid performance	1212	3.8	1404
private – high performance	1011	3.2	1543
total	32018	100.0	1142

35Figure 2 shows that there are substantial differences in the access to higher education depending on the type of secondary school. More than 80% of those who graduated from high performance private schools had access to highly or averagely demanded Departments (clusters 1 to 4) and only 4.2% were not admitted. On the contrary, those graduating from evening schools were not admitted at a rate of 75% and only 15% had access to a highly or averagely demanded Department.

36Figure 3 shows the percentage composition of the student population in the different clusters of Faculties and Departments in terms of the type of school they come from. The overwhelming importance of students from public schools affects student composition within most clusters. Even in the most exclusive cluster of Departments (cluster 1), more than 40% originate from medium and low performance public schools, while only 32% originate from high and medium performance private and experimental public schools; this is respectively reduced to 23% for cluster 2.

37This double sided picture (figures 2 and 3) shows that elites and upper middle classes have privileged, but not exclusive, access to higher education in Greece. Thus, these social groups often use more exclusive alternatives, like studies abroad to academically and often financially demanding institutions in Western Europe and the US.

38The broad area of studies which candidates select during the last two years of secondary school in view of the admissions examination is an important parameter that differentiates schools: ‘science’, ‘technology’ and ‘humanities’ account respectively for 13%, 51% and 36% of enrolment (2004-5). Average performance differs amongst these options leading to different success rates (87%, 69% and 67% respectively). The ‘science’ option is much more present in high-performance private schools (27%) against approximately 10% in the bulk of schools at the lower end of the school type hierarchy.

The role of the neighborhood

40At first sight, the social profile of the neighborhood seems to matter, even though the correlation of candidates’ performance with its attributes is not particularly high (R ≈.150 to .200) (table 3).

41In table 3 all area attributes –mostly those assumed related to difference in social rank– are significantly correlated with candidates’ performance in the admissions examination. Higher indices appear for occupational or other groups at the extremes of the social hierarchy contrasting with lower indices for groups around the middle as well as for variables related to location within the metropolitan region (central / suburban / peripheral), to housing tenure and to immigrant presence.

** significance level .95

42Average school performance is correlated to neighborhood attributes much more than individual performance since all variance within schools disappears; Table 3 indicates that schools situated in neighborhoods of higher social profile usually perform better. This is also depicted by the comparative reading of maps 1 and 2. Although there is no absolute correspondence between type and performance of schools and neighborhood status, these two maps show that there is a concentration of low/mid-low performance public schools in low/mid-low status areas and a concentration of high/mid-high performance private and public schools in high/mid-high status areas. Low and mid-low performance schools are much more concentrated in the working class municipalities of Keratsini, Nikaia and Perama at the west of Piraeus) while high and mid-high performance schools are much more concentrated in the wealthy suburbs of Psychico-Filothei or Ekali. This is, in fact, an expected outcome since neighborhoods with a higher social ranking contain a larger percentage of middle and upper-middle class students who systematically perform better in education (table 4).

Table 4 Percentage distribution of students by type of school and by social type of neighborhood in Athens (2005)

	Social type of neighborhood
Type of school	upper	upper-middle	mixed	lower-middle	lower
private low	0.0	1.1	1,8	1.9	0.5
evening	0.0	1.1	4.1	3.5	3.8
public low	9.7	7.7	11.6	33.4	58.3
public mid-low	0.0	22.6	36.1	41.9	31.3
public mid-high	0.0	19.1	10.3	11.1	6.1
public high	15.0	34.2	23.9	4.9	0.0
public experimental	11.6	4.4	4.7	1.6	0.0
private medium	18.2	8.0	4.1	1.2	0.0
private high	45.5	1.8	3.4	0.4	0.0
all schools	100.0	100.0	100.0	100.0	100.0

43Figure 4 shows the average performance of candidates in the admissions examination to higher education by broad social type of residential area. This performance is shown for all candidates (purple line), and separately for those from public or private schools (bars). It also shows the predicted value of this performance based on its relation with the composite index of deprivation (see note 14) in the areas around the 447 schools, where more than 32 000 candidates completed their secondary education in 2005. Comparing the actual with the expected performance for all candidates in different social settings we observe no substantial difference, except in the higher social type of neighborhoods. This could mean that there is a positive neighborhood effect in those neighborhoods, which would be consistent with similar observations elsewhere (Gordon and Monastiriotis, 2006 and 2007). However, this higher than expected performance in upper class neighborhoods is most probably related to the concentration of privileged private schools, which attract highly performing students from a broad range of upper-middle and socially mixed areas. The systematically higher performance of candidates from upper class areas seems, therefore, much more related to the concentration of private schools (table 4) and to school segregation rather than to some form of neighborhood effect.

44For the much larger number of candidates from public schools, performance decreases –in respect to expected values– as we move from lower to higher social types of neighborhood. This is probably the effect of draining performant (i.e. middle class) students by private schools, which culminates in the higher status areas and relegates public schools to a residual role.

45Another way of controlling the assumption that school segregation is more significant than residential segregation for educational performance is by comparing performance in similar segments of the school and the neighborhood hierarchies.

Table 5 Candidates’ average performance by school and neighborhood social type (2005)

School type	Mean*	N	Coefficient of variance	Neighborhood type	Mean*	N	Coefficient of variance
private low evening school public low	944.4 875.5 972.9	443 304 7207	52.0 47.7 41.1	lower	1014.2	2585	40.5
public mid-low	1109.1	10565	36.7	lower-middle	1081.2	11937	38.3
public mid-high	1192.2	3494	34.0	mixed	1176.0	9296	36.5
public high public experim. private medium	1275.8 1323.4 1403.9	4841 1036 1205	32.3 30.7 27.3	upper-middle	1225.6	5141	34.9
private high	1543.2	1008	21.1	upper	1409.5	1144	28.4
Total	1141.9	30103	37.6	Total	1141.9	30103	37.6

^* Maximum possible score = 2000

47Table 5 also shows that there is a declining diversity of performance as we move from lower performance and public schools to higher performance and private schools. The same applies to neighborhoods, with candidates from lower status neighborhoods having a more diverse performance than those from higher status ones. This is due to the much higher percentage of candidates with very low (failing) performance at the bottom of both hierarchies. Thus, in both hierarchies of schools and neighborhoods, educational performance is lower and more diverse at the bottom of the hierarchies and higher and less diverse at the top. Moreover, the range of this diversity is higher for the school hierarchy, providing an additional indication that school segregation is higher than neighborhood segregation.

48In sum, table 5 shows a wider performance range within the hierarchy of schools than within the hierarchy of neighborhoods, supporting the assumption of a higher degree of segregation within the former than the latter. Moreover, the higher difference of scores at the upper –compared to the lower– end of the social hierarchy may be an indication that school segregation, as the outcome of middle and upper middle-class strategies, practically functions more as an advantage for higher social groups than as a disadvantage for lower ones.

The importance of demographic features

49Candidates in the admissions examination are normally 17 to 18 years old. They are older if they failed to be admitted on their first attempt; if they were not satisfied with the Department they were admitted to and chose to retry the admissions examination for a second time; if they opted to finish first with their military service (boys), postponed their attempt to study for any other reason or were delayed in completing secondary school. The correlation between candidates’ performance and their delay in taking the examination (R = -.142) means that there is a non-negligible negative effect of this delay on performance. The average general score of those taking the examination when they are 18 or younger is 1 195/2 000, while for those with at least two years of delay it is 890/2000. Delay is also related to social origin: Students from schools in lower social status neighborhoods take the examination at the expected age at a lower rate (70%) than those from schools in upper social status neighborhoods (90%).

50Gender seems also important for educational performance. In the late 1920s the participation of women in the student population was 4.9% (Katsikas and Kavadias, 1994) against 59.9% in 2010 (ELSTAT, 2010). For several years now, girls have been performing better. According to our dataset there is a positive correlation between girls and performance in the admissions examination (R = .058); also, a larger number of girls have taken this examination (17 246 versus 14 785 boys in 2004-05 for the Athens Metropolitan Area).

51The average score for girls in the examination (1 165/2 000) is 4.5% higher than the score for boys (1 115/2 000). This applies to every cluster of the Faculties and Departments’ typology presented in figure 1. Following their higher scores, more girls were admitted to all clusters of Faculties and Departments, with higher gender differences observed at the middle of the hierarchy (clusters 3 to 5).

The combined impact of school, neighborhood and demographic features on accessing higher education (a regression model)

52We have examined individually the different factors we assumed to be important for educational performance because our dataset did not comprise evidence on the most important one –the social origin of candidates in the admissions contest to higher education– and, therefore, could not support a comprehensive multivariate model. However, our dataset could support the examination of the effect of school segregation on educational performance jointly with a number of individual features of candidates, as well as with the social type of their residential neighborhood.

54The variables considered belong to two different levels: the school or the candidate. Therefore, we built a two-level mixed model mainly addressing the quality/type of schools and the personal features of candidates. This model was intended to refine our approach on the impact of school segregation discussed previously in the section ‘the school environment’.

Table 6 Coefficients for linear mixed models. Dependent variable: Candidates performance in admissions examination to higher education (2005)

	M0: Empty model		M1: M0+School type+Deprivation Index		M2: M1+Sex+Studies field+Age of candidate
	Estim.	Sig.	Estim.	Sig.	Estim.	Sig.
Fixed effects
Intercept	1 116.97	.000	1 529.41	.000	1 402.04	.000
Sch Type Ref categ: private high performance
private low performance			-587.68	.000	-523.81	.000
evening school			-647.50	.000	-528.35	.000
public low performance			-549.65	.000	-466.45	.000
public mid-low performance			-417.33	.000	-344.87	.000
public mid-high performance			-337.37	.000	-284.34	.000
public high performance			-262.95	.000	-206.90	.000
public experimental			-207.00	.000	-178.41	.000
private medium performance			-134.03	.000	-128.41	.000
Deprivation Index			-11.92	.000	-14.15	.000
Sex Ref categ: girls
boys					-55.53	.000
Studies field Ref categ: Social sciences and hunanities
science					381.56	.000
technology					46.63	.000
Age of candidate					-91.44	.000
Random effect variances
Residual (within schools)	162.03	.000	161.93	.000	140.56	.000
Intercept (between schools)	25.63	.000	0. 65	.001
Sex					0.89	.002
Age of candidate					4.64	.000
Studies field					1.94	.000
-2 Log-likelihood	447.47		444.32		440.66

55The models presented in table 6 refine our grasp on the effect of school segregation by identifying the main parameters that explain the variance of candidates’ performance between schools. The empty model (M0) shows that the variance between schools is 13.7% of the total variance [25.63*100/(25.63+162.03)]. This part of the variance is ‘explained’ almost entirely by the type of school attended by the candidates. The school type variable comprises the distinction between public and private schools, daytime and evening schools, experimental and ordinary public schools as well as high / intermediate / low performance schools. The use of this variable minimized the contribution of all other variables related to school quality (average performance, percentage of stuents taking the admissions examination, percentage of students with a migrant background, percentage of teaching staff with a post-graduate degree) and excluded them from the model (M1). The type of school explains almost 95% of the variance between schools and the differences between the reference category of high performance private schools and the other eight categories show a progressive decline in performance as we move from private to public, from daytime to evening, from experimental to ordinary and –obviously– from high to low performance schools.

56The alternative option (not shown in table 6) of using a variable indicating the social type of neighbourhood the school is located in –the neighbourhood’s socioprofessional profile or its composite deprivation index– explains much less of the variance (24% and 26% respectively). In model 1, we have used as variables for fixed effects both the school type and the area deprivation index, increasing the explanation of the between schools variance to 97%. The increase in the value of the deprivation index understandably leads to a decrease in the performance score of candidates.

57In model 2 we added three more variables (sex, field of studies and age of candidate) for random effects since they refer to the candidate rather than the school level. Their inclusion reduced the variance within schools –since the added variables account for part of it– and all three are significant for candidates’ performance. Girls perform systematically better than boys and older candidates are increasingly doing worse than those who pass the examination at the expected age. Performance varies also in terms of the field of studies. This does not reflect necessarily that those in the science option perform better than those in technology and, even more, in social sciences and humanities. The assortment of tests is different for the three fields and this makes final grades not strictly comparable.

58This exercise proved that the variable ‘type of school’ attended by candidates is overwhelmingly explaining the variance of performance in the admission examinations to higher education between schools. Several other variables, related to the social profile of candidates’ neighbourhoods and to individual characteristics, also significantly contribute to explaining the variance of this performance. Other variables –like the share of immigrant pupils in each school or the percentage of post-graduate degree holders among members of the teaching staff– have not provided significant contributions when included in the model. We can assume that immigrant status or origin would have made a significant contribution, if relevant data at the individual candidate level were available. This would have reduced the variance within schools along with the class differences among candidates that must account for the largest part of this variance following our analysis in the section ‘the effect of social origin’.

Concluding remarks

59The dataset of candidates’ performance in the 2005 national admission examinations to higher education gave us the opportunity to investigate and roughly measure the social reproduction function of the transition to higher education in Greece. We focused on the Athens Metropolitan Area, where school segregation and residential segregation are much more developed than anywhere else in Greece and, therefore, where we could most clearly relate them to social reproduction through educational performance.

60We have taken into account all crucial parameters for the transition from secondary to higher education contained in this dataset in order to assess the statistical explanation of educational performance. The main shortcoming was that our dataset did not contain any variables on candidates’ social profile. This induced us to use data from another source –data provided by the Hellenic Statistical Authority (ELSTAT) on the socioeconomic and educational profile of students in every Faculty and Department of all Greek higher education institutions– to address the relation between candidates’ performance and social origin. The outcome was clear and compatible with the role educational performance plays in social reproduction, briefly discussed in the section ‘The social selectivity of school systems and the impact of residential location’: the social hierarchy of Faculties and Departments we were able to produce using the ELSTAT data is very highly correlated with the performance of candidates in the admissions examination to higher education. Educational performance fabricates, therefore, an enrollment with a clear social hierarchy amongst the different Faculties and Departments that clearly reproduces existing social inequalities.

61Then, using the dataset from the 2005 admission examinations to higher education we provided further insight on the role of the school, the residential area and of some of the candidates’ individual features, like age and gender. The role of school segregation appears to be very important, although quite different from cases where residential segregation is its foundation (e.g. Boterman, 2019). In Athens, school segregation is higher than residential segregation and, to some extent, this is the outcome of middle-class strategies of school choice to overcome the paucity of advantages provided by the ‘automatic’ conversion of residential to school segregation through the allocation of students to local schools on the basis of distance. The typology of upper secondary schools in Athens shows that performance is highly related to their public / private status, their experimental character and their daytime or evening operation. Moreover, the distribution of schools by type and performance level in Athens appeared clearly uneven, with high performance private schools concentrated in affluent suburbs and low performance public schools overrepresented in working class neighborhoods.

63Finally, the mixed linear model we used confirmed that the variance of performance between schools is mainly explained by the type of school (public/private, experimental/ordinary, daytime/evening, high/intermediate/low performance) attended by the candidate and to a much lesser extent by the social profile of the school area –assumed to be the same with the candidates’ residential area. Age and gender proved significantly related to performance, while other variables like the percentage of immigrants among pupils or the percentage of highly qualified teaching staff in the schools attended by the candidates did not.

65At the same time, however, Greek society remains highly unequal and educational mechanisms contribute substantially in reproducing inequalities. Privileged groups follow educational strategies to create advantage for their children: investment in private schooling; selection of better schools within the public sector; investment in continuously longer education; investment in studies abroad and in highly rated institutions and degrees. These strategies seldom involve residential relocation, which would be rather ineffective within a context of limited residential segregation and consequently of limited school segregation based on spatial distance.

66The structure of the educational system may be, to some extent, the reflection of a weakly polarized social structure, a situation that has probably changed significantly in the last 20 years and, especially, during the recent crisis. The waning social mobility for lower social classes undermines the broad social compromise based on massive aspirations to access middle-class positions. Educational policies during the crisis further destabilized this compromise as they clearly drove towards more social inequality and so did the impact of the crisis (Maloutas, 2016): with youth unemployment having reached over 50% in 2013 (ELSTAT, 2014) –affecting the highly educated as well– educational achievement looked increasingly futile, especially to those from lower social groups with degrees that used to support social mobility through public employment. At the same time, however, the demand for degrees has not declined. Higher education degrees acted as protective shields during the crisis: in case of redundancies in the public sector, degree holders were not immune, but they were better protected than the rest being usually preferred to those without degrees for the same position. Educational achievement continues to provide social advantage, its function in reproducing and legitimating social hierarchy becomes stronger, and the illusion about the decisive role of personal merit remains largely unchallenged.