School segregation, selective education, and adolescents’ alcohol use – is there a connection?

ABSTRACT This study examines the connection between studying in selective classes and students’ alcohol use using data from the European School Survey Project on Alcohol and Other Drugs (ESPAD) from Finland. Linear probability models were used to investigate whether (1) studying in selective classes is associated with students’ recent heavy episodic drinking, and (2) whether parental education, students’ school success, and their friends’ alcohol use mediate the association. The results show that Finnish students who are selected for classes with a special emphasis engage less in heavy episodic drinking compared to their peers in non-selective classes. Parental education, school success, and friends’ alcohol use mediate part of the association. The results suggest that having the opportunity to apply for classes with a special emphasis might increase segregation between student groups and schools regarding risky behaviour such as heavy episodic drinking. Implications of urban and school policies on segregation are further discussed.


Introduction
In recent years, the educational research community has been paying increased attention to the importance of socio-emotional and health and well-being factors as the "bedrock" of educational success (OECD, 2021, p. 3).As an example, the Icelandic Model for Prevention of Adolescent Substance Use demonstrated empirically the important relationship between students' well-being, health behaviours, and academic outcomes (Kristjansson et al., 2020).However, there is still limited understanding of how health behaviours such as substance use are linked to questions of school context, peer groups at school, and growing social differences between schools in European countries.How are the health behaviours as prerequisites for learning shaped by the school context?In this article, we explore the connection between students' alcohol use, school segregation, and the ways in which students are grouped in selective classes within schools in Finland.Alcohol use provides an interesting perspective on the issue, since heavy episodic drinking (HED) in adolescence is associated not only with negative health outcomes, but also with poor school performance and engagement (Patte et al., 2017).At the same time, it may have an important normative role in structuring peer relations (e.g., Demant & Järvinen, 2006).Like educational performance and attainment (e.g., Pfeffer, 2008), alcohol use is connected to the family background (Gomes de Matos et al., 2017;Pape et al., 2017).However, no previous studies have examined the connection between the social selectivity of classes with special emphasis teaching and youth alcohol use.
Research on the interrelationship between youth alcohol use and school context is particularly important, as there is growing research evidence on increasing school segregation throughout Europe (Boterman et al., 2019), even in Finland (Bernelius & Huilla, 2021), and on the potential effects of school segregation on students' behaviour and outcomes (van Ewijk & Sleegers, 2010).In recent years, there has been growing research interest in the socio-economic and ethnic segregation of neighbourhoods and schools, especially in bigger cities across Europe and the United States, also concerning the Nordic welfare regime (Boterman et al., 2019;OECD, 2018).Increasing segregation has also been documented in Finland.While neighbourhood differentiation used to be very low up until 1990s, both socio-economic and ethnic gaps have widened in more recent decades (Bernelius & Huilla, 2021;Saikkonen et al., 2018).For example, in Helsinki the proportion of adults with master's level or higher tertiary education now ranges from seven percent to 38 percent in primary school catchment areas, meaning a five-fold difference between catchment areas (Bernelius & Huilla, 2021).This difference is reflected in schools when children enter schools based on geographical catchment areas.
As school allocation of students in Finland is based primarily on families' residential location, residential segregation and school segregation are closely connected.In addition to the potential effects of socio-economic school segregation on the educational outcomes of students, segregation is also connected to parental strategies of neighbourhood selection.For example, Venla Bernelius (2013; see also Bernelius & Vilkama, 2019) noted that school reputation and popularity are connected to the socio-spatial characteristics of the neighbourhoods in ways which shape school choices and housing choices made by educationally motivated families, and that the rejection of schools in relatively disadvantaged neighbourhoods is particularly consistent.
School segregation is exacerbated by school choices.When moving to the upper level of comprehensive school at the age of 12-13, there are more opportunities to choose between schools and classes with emphasis on a range of subjects, for example, sport, music, or mathematics.At this stage, many pupils move from their own neighbourhood school to other schools offering specialised subjects.These student flows strengthen the differences between schools, as all families are not equally active in school choices.Studies have demonstrated that well-educated, middle-class parents are more active in school markets, which is reflected in the social selectivity of these choices (e.g., Ball et al., 1996;Kosunen & Seppänen, 2015).As a result, choices increase both socio-economic and ethnic gaps between schools, as well as the differences between specialised classes and classes without an emphasis on a specialised subject (Bernelius & Huilla, 2021;Kosunen, 2016).These options have also expanded during the last few decades, challenging the universalist non-selective principle of the Finnish school system (Kalalahti et al., 2015).
School choice as a middle-class practice has commonly been explained by applying Bourdieu's reproduction theory (Bourdieu & Passeron, 1990), according to which highly educated middle-class parents have the ability and resources to maximise their children's benefits in the formally meritocratic schooling system through the transmission of cultural capital.In the context of publicly funded schooling system with an option to apply for special emphasis classes, special emphasis teaching has been considered to resonate with middle-class practices of "concerted cultivation" (Lareau, 2003), as middle-class parents aim to direct and encourage their children to engage in organised leisure time activities such as music, sport or arts (Kosunen & Seppänen, 2015;Vincent & Ball, 2007).These are usually the same activities, to which teaching in these selective classes is often specialised.As admission into these classes is usually based on aptitude tests, middle-class children may enjoy a head start having the skills, interests, and engagement required for admission compared to lower socio-economic status children (Kosunen & Seppänen, 2015).In addition to children's embodied skills and dispositions (habitus), parental cultural capital may be reflected in the strategic knowledge concerning different educational options (Pfeffer, 2008), through which parents aim to optimise their children's learning environment in order to reproduce their relatively high-class positions in the future (Ball, 2003;Ball et al., 1996).The increased social class-based selectivity of the schooling system has been considered to reflect a shift from meritocratic ideology to "parentocratic" ideology (Rinne, 2000; see Brown, 1990).
Reflecting the selectivity of school choices and middle-class emphasis on special subjects, studies on learning results in Finland show that students in classes with a special emphasis do better compared to students in non-selective "regular classes" (Berisha & Seppänen, 2017;Koivuhovi et al., 2020;Kosunen, 2016).This is connected to the observation that in international comparison, socio-economic segregation between classes within the same school is at an elevated level in Finland (Yang Hansen et al., 2014).While the between-school variation in student background and educational outcomes is still below the OECD average, within-school segregation is noticeably high.This "dual differentiation" makes Finland a particularly interesting case for research on the relationship between school segregation and youth risk behaviour, as we are able to compare groups sharing the common school context, but a highly differentiated peer group within the school.

Alcohol use, youth culture, and school segregation
According to the European School Survey Project on Alcohol and Other Drugs (ESPAD) data, 86 percent of 15-16-year-old Finnish adolescents (9th grade) had consumed alcohol during the last 12 months in 1999.Twenty years later this proportion had fallen to 60 percent.Correspondingly, the proportion of adolescents who had drunk alcohol at least once during the last month fell from 61 percent to 32 percent, and the proportion of those who had drunk at least six portions at least once during the last month decreased from 48 percent to 22 percent.There were no significant differences between boys and girls in levels and trends of alcohol use and heavy drinking.(ESPAD report, 2020;Raitasalo & Härkönen, 2019.)There has been extensive research in Finland and globally on youth alcohol use and the reasons behind its decline (Kraus et al., 2018;Raitasalo et al., 2018).Peer-related factors, such as unsupervised leisure time with peers has been proven to be linked to youth alcohol use (Flannery et al., 1999;Lee & Vandell, 2015;Richardson et al., 1993).Osgood's extension of the Routine Activity Theory (Osgood et al., 1996) posits that spending unsupervised time with peers, places young people at risk of misbehaviour and deviant behaviours because of the lack of structure and the presence of peers who may encourage the risky acts.Further, peer drinking habits have been found to have direct and indirect effects on adolescent alcohol use (Ivaniushina & Titkova, 2021).
Drinking alcohol has been considered to have a contradictory dual role in marking youth identity by being a rite of passage into adulthood and by excessive drinking being a vital part of youth culture itself (Bakken et al., 2017).Youth drinking has been considered to simultaneously reproduce and resist normative ideals of adult society (Bakken et al., 2017), reflecting youth cultures' partially autonomous, but partially subordinate, position in relation to adult society (Corsaro & Eder, 1990).Experiments with alcohol may signal social maturity in adolescent peer culture, being connected to popular and dominant position in a peer group (Demant & Järvinen, 2006), even though some adolescents may also construct maturity by distancing themselves from drunk peers (Bakken et al., 2017).Increasing emphasis on individual responsibility for health and well-being in contemporary culture has been suggested to blur the boundaries between drinking and non-drinking, opening novel ways to challenge the dominant alcohol culture (Pavlidis et al., 2019).
In addition to peer-related factors, parenting, parental norms, and control have been associated with youth drinking (Choquet et al., 2008;Pape et al., 2018;Ryan et al., 2010).Considering family's socio-economic position, higher parental SES has been associated with adolescents' lower probability of heavy episodic drinking (Gomes de Matos et al., 2017).Further, parenting practices and parents' drinking behaviour have been observed to account for the association between adolescent heavy drinking and parental SES (Pape et al., 2017).Also, school success has been proved to be negatively associated with youth alcohol use (Patte et al., 2017;Schulenberg et al., 1994).Some studies indicate that school-related contextual factors also matter, by finding that adolescents in socio-economically advantaged schools and neighbourhoods tend to use substances (including alcohol and tobacco products) more than adolescents in less advantaged contexts (Olsson & Modin, 2020;Pedersen et al., 2015Pedersen et al., , 2017) ) even though advantaged neighbourhood and school environments may potentially have a positive effect on other outcomes such as educational attainment (Chetty et al., 2016;van Ewijk & Sleegers, 2010).Nevertheless, alcohol related problems might still be concentrated in relatively disadvantaged neighbourhoods (Pedersen et al., 2015).

Formulation of the research hypotheses
Based on the theory and findings presented above, we hypothesise that 1. Students in selective classes engage in heavy episodic drinking less often than students in regular classes (H1).
Several mechanisms may explain this connection.Parental socio-economic status is known to be associated with both heavy episodic drinking and educational outcomes of children on one hand, as well as selection into special emphasis classes on the other.As student's family background causally precedes children's sorting into schools and classes, parental education cannot be considered as a mediator for the association between class type and student's alcohol use in a theoretical sense.Thus, focusing on the class selection process, class type should be considered as a mediator for the association between parental education and heavy episodic drinking.Our second hypothesis, therefore, is that.

Class type mediates the association between parental education and HED (H2).
In other words, children from high-educated families tend to get selected into special emphasis classes, which is further connected to their lower probability of recent heavy episodic drinking (Figure 1).Considering class type as a mediator, we assume that the association between parental education and HED does not vary according to the class type, i.e., that class type does not moderate the association.
The causal role of grade point average (GPA) is more complicated compared to parental education, as it can both precede (selection effect) and supervene (contextual effect) the class selection.In this study, there are no empirical means to make this causal separation.Therefore, without strong causal claims, we assume that.

School success accounts for part of the association between class type and HED (H3).
In other words, high-performing students tend to get selected into special emphasis classes, which is further connected to their lower probability of recent heavy episodic drinking (Figure 1).School success can also mediate the association between parental education and heavy episodic drinking through class selection (Figure 1).In addition to selection effects, class "ethos", teaching practices and peer group norms in selective classes may differ from those of regular classes, further encouraging students to engage in schoolwork, which is reflected in their school success and through school success connected to their lower probability of recent heavy episodic drinking (Figure 2).
Finally, peer relations and peer group norms related to alcohol use may vary between schools and classes, encouraging students to either abstain from excessive alcohol use or engage in heavy drinking.Highlighting the predominantly social nature of adolescent drinking, our last assumption is, that.4. Peers' drinking behavior mediates part of the association between class type and a student's recent heavy episodic drinking (H4).
Concentration of drinking peers in a class may expose students to peer pressure, considered as a contextual effect (Figure 2).However, it should be recognised that peers' drinking behaviour may also relate to class selection processes.Acknowledging the dual role, peer pressures are illustrated in Figure 2 as contextual effects.

Sample
The analysis was based on data from the European School Survey Project on Alcohol and Other Drugs (ESPAD) from Finland on a representative cross-section of young people born in 2003.The data were collected in spring 2019.The target population was defined as: (1) regular students who (2) turned sixteen in the calendar year of the survey and (3) were present in class on the day of administration of the survey.This definition includes students who were enrolled in regular or general studies but excludes special schools (for students with learning disorders or severe physical disabilities), and students who were not able to reply in one of the two official languages; Finnish or Swedish.
A two-stage systematic probability-proportional-to-size sampling method using the Nomenclature of Territorial Units for Statistics 2 regions (NUTS 2) as strata and schools as clusters (NUTS 2, 2020) was used to collect nationally representative samples by means of a self-administered penand-paper questionnaire.The students answered the questionnaires in a classroom setting.Participation was voluntary and anonymity was ensured (ESPAD Group, 2020;Raitasalo & Härkönen, 2019).The school participation rate and the student response rate were both about 90 percent.Those not belonging to the target group (n = 375) and those who had responded inconsistently or had clearly exaggerated (n = 161) or had not answered over half of the questions (n = 151) were removed from the data.The data were weighted to reflect a representative sample of Finnish ninth graders.
In our analysis, only schools in the Uusimaa region, including the Helsinki capital region, were included in the data as most of the schools with special emphasis in Finland are located in this area and the competition in applying for these classes is harsher than in other parts of the country.The final number (non-weighted) of sampled schools was thus 78 and sampled classes 107.The number of students in the final sample was 1582 (1738 with weights).

Measurements
As a dependent variable, we used heavy episodic drinking during the past 30 days (HED 30).This was measured by the question "How many times have you had six or more drinks on one occasion during the past 30 days?", with five response alternatives ranging from never to 40 times or more.The variable was used as a dichotomy in the analysis (0 = no, 1 = yes, once or more often) because there were very few responses in the upper part of the scale and the greatest part of those who drank (in general or heavily) did it one or two times a month.
Class type was measured by asking the teachers at the point of class sampling whether the sampled class was a "non-selective class" in which students are enrolled based on where they live or a "selective class" having a special emphasis which the students have to apply for.
Plans for post-compulsory education were assessed with the question "Where are you aiming to study after comprehensive school?" with the following response alternatives: 1 = general upper secondary school, 2 = vocational school/training, 3 = vocational school with general upper secondary school courses, 4 = "10th grade", 5 = I'll not continue my studies, 6 = I don't know.Those answering general upper secondary school were coded as 1 and all others as 0.
Heavy alcohol use by peers was measured with the question "How many of your friends would you estimate get drunk?" with the following response alternatives: 1 = none, 2 = a few, 3 = some, 4 = most, 5 = all.The variable was dichotomised so that answers 1 and 2 were coded as 0 and 3-5 as 1.
Parental education was defined by whether either of the parents had completed upper tertiary education, to the best of the respondent's knowledge (yes/no).The original question with the response options was: "What is the highest level of schooling your (a) father/(b) mother completed?",with "comprehensive school or lower" (grades 1-9), "some general upper secondary/vocational education", "completed general upper secondary/vocational school", "some college or university", "completed college or university", "don't know", "does not apply".
Class composition was measured by aggregating the former binary individual-level parental education variable into class-level and dichotomising the resulting variable at class-level median.The resulting variable presents the proportion of children at sampled class whose at least other parent had completed upper tertiary education to the best of the respondent's knowledge (0 = less than 35 percent of students in a class has at least one parent with upper tertiary education, 1 = at least 35 percent of students in a class has at least one parent with upper tertiary education).Thus, the variable represents the educational structure of class and can be considered to be a measure of class advantage.As such, it reflects the already selected student composition, which creates a specific context for learning and social, peer cultural action.

Statistical analysis
Statistical analyses were conducted using linear probability models.The school-and class-clustered nature of the data was taken into account by using Complex samples design in SPSS, conducting analyses twice, with both schools and classes separately as clusters, and NUTS2 regions as strata.Final analyses (Tables 1-3) were conducted with classes as clusters, corresponding to the measurement level of class type, our main variable of interest.Based on linear probability models, a mediation analysis approach was applied to investigate how the association between class type and HED is mediated by school success, plans for post-compulsory education, and heavy alcohol use by peers.In addition, parental education-related selection into these classes, and this potential indirect effect on alcohol use, was examined by including parental education in the models.As the class context may confound or mediate the associations between independent variables and HED, class composition (proportion of students with high-educated parents in a class) was added in the models.Robustness of the results was checked using multilevel linear probability models and multilevel logistic regression with between-school and between-class random intercepts using the "generalised linear mixed models" function.Missing values were excluded using pairwise deletion concerning descriptive statistics (Table 1), and listwise deletion concerning regression results (Tables 2 and  3).All analyses were conducted using IBM SPSS Statistics 25 software.
The study procedures were carried out in accordance with the Declaration of Helsinki.The Institutional Review Board of the Finish Institute for Health and Welfare approved the study (THL/ 1740/6.02.01/2018).

Results
Descriptive statistics presented in Table 1 show that 12 percent of selective class students and 21 percent of students in regular classes had drunk at least six drinks on one occasion during the past 30 days (p = .01).Furthermore, class status was significantly related to parents' education level and student's GPA.Despite apparent between-class type differences in the distributions of friends' heavy episodic drinking, these differences were outside the bounds of statistical significance, when the clustered nature of the data was considered.
Table 2 presents the results of the linear probability models on adolescents' heavy episodic drinking during the past 30 days.The first column on the left presents the gender-controlled associations between the independent variables (class type, educational success, parental education, friends' drinking behaviour) and heavy episodic drinking of students.All the studied variables were statistically significantly associated with past 30 days' heavy episodic drinking.Plans for post-compulsory education were not included in the regression models due to its strong correlation with GPA: it essentially accounted for the association between post-compulsory education plans and HED.Studying in a selective class was associated with a lower probability (B = −0.084,p < 0.01) of recent heavy episodic drinking compared to studying in non-selective classes, being align with our first research hypothesis (H1).High GPA was strongly associated with lower probability of heavy episodic drinking (B = −0.191,p < 0.001).Also, heavy alcohol use by peers was strongly associated with student's own HED (B = 0.263, p < 0.001).High parental education seemed to reduce children's probability of recent HED, as it was associated with lower probability of recent heavy drinking (B = −0.056,p < 0.05).
Indirect and direct effects are more comprehensively presented in Table 3.The association between class type and HED decreased slightly when parental education was added to the model, it accounted for 7 percent of the association between class type and HED.The reduction indicates that students from highly educated families get selected into classes with a special emphasis more often than their peers from lower educated families on one hand, and that these students also use less alcohol than their peers from lower educated family backgrounds, on the other.The reduction in the strength of the estimate was rather small, yet being aligned with our second hypothesis (H2).
Similarly, GPA accounted for 19 percent of the association between class type and HED.Still, the association remained statistically significant at p < 0.05 level after adding GPA to the model indicating that for this part, the connection between class type and HED was mediated by educational success, as assumed by our third hypothesis (H3).When parental education and GPA were added together in the model, the association between parental education and adolescents' HED vanished completely, indicating that the connection between parental education and children's HED was completely mediated by children's educational success, which also related to the family background-based selection into classes with specialised teaching.The association between GPA and HED may thus relate both to selection into classes with specialised teaching and studying in these classes with like-minded peers.
The strongest individual mediator for the association between class type and HED was heavy alcohol use by peers, which accounted for 26 percent of the association, dropping it to being statistically insignificant, being in line with our fourth hypothesis (H4).
When all the mediators were included simultaneously in the model, the proportion of the association between class type and HED that was accounted for by these mediators was as high as 40 percent.The remaining direct effect (B = −0.050)was statistically insignificant.
Additional analyses were conducted to consider the class composition as a contextual factor, and its potential confounding effect on the association between class type and heavy episodic drinking.Including class composition in the models enabled us to consider the differences between high-and low-educated class contexts.If class type is associated with class composition, and class composition is associated with students' alcohol use, class composition may confound the effect of class type on HED.
When the class composition as an aggregated class-level measure was controlled for, the association between class type and HED modestly increased.There was no bivariate association between class composition and students' individual HED.Nevertheless, when class type and student's GPA were controlled for, the association between class composition and HED increased, reaching statistical significance.Interestingly, after controlling for GPA, the effects of class composition and class type on HED pointed in the opposite direction, as studying in a special emphasis class was Note: (1) The cumulative indirect effect percentage for all mediators was calculated by using the formula c-c', in which c refers to total (gender-controlled) effect of class type on HED.
(2) Gender is controlled in all models.
(3) In case of parental education, the causality is reversed between class type and mediator.
associated with a lower probability of HED but studying in a high-educated class context seemed to increase the probability of recent heavy episodic drinking.However, as students in these educationally advantaged contexts tended to succeed at school, high GPA "worked against" or "protected" them against this contextual effect on HED.Concerning class type, the associations were more straightforward as the bivariate association between class type and HED was also statistically significant and pointed in the same direction as the effect of GPA.Thus, in this case, high GPA did not work against or protect from the effect of class type, but rather it contributed to the effect of a special emphasis class on heavy episodic drinking.
In the final model, in which heavy alcohol use by peers was also included, the effects of class type and class composition on HED decreased.
The analyses were repeated using multilevel linear probability models and logistic regression with and without between-school and between-class random intercepts, as a sensitivity analysis.These results were essentially analogous to the complex samples regression results reported in Table 2. Complementary analyses were also conducted by adding the interaction of parental education and class type to the models predicting HED, finding no evidence of class type moderating the effect of parental education.

Discussion
Despite the clear decline in youth alcohol consumption during recent decades, alcohol use in general, as well as heavy episodic drinking, are still relatively common among present-day adolescents (ESPAD report, 2020;Raitasalo & Härkönen, 2019).Family background characteristics, socio-economic position, parenting practices, parental norms, and control have been associated with youth drinking (Choquet et al., 2008;Pape et al., 2018;Ryan et al., 2010).In addition, peer drinking habits (Ivaniushina & Titkova, 2021) and unsupervised time with peers (Flannery et al., 1999;Lee & Vandell, 2015;Richardson et al., 1993) have been found to be linked with youth alcohol use.Some studies indicate that school context also matters (Olsson & Modin, 2020;Pedersen et al., 2015Pedersen et al., , 2017)).Our results mirror these previous findings.This study complements the existing research by examining how tracking between selective and non-selective classes is associated with adolescents' alcohol use.
Our study found significant differences in the prevalence of heavy episodic drinking among Finnish adolescents aged 15-16 years according to whether they study in classes with a special emphasis or in regular classes.As we assumed (H1), students in classes with special emphasis teaching engage less in heavy episodic drinking compared to their peers in non-selective classes.Our results also support our second hypothesis (H2) that students from highly educated families get selected into classes with a special emphasis more often than their peers from lower educated families and that these students also use less alcohol than their peers from lower educated family backgrounds.Further, the association related to family background-based selection is accounted for by school success.School success measured by grade point average explained about one-fifth of the association between class type and HED altogether, being aligned with our third hypothesis (H3).The strongest individual mediator is heavy drinking by peers, which accounted for one-quarter of the association between class type and heavy episodic drinking, being in line with our fourth hypothesis (H4).
All Finnish schools are financed by the state, are free-of-charge and follow a universal, national curriculum.Still, tracking between selective and non-selective classes has become more common during recent decades, as up to one-third of the pupils in the biggest cities are selected into the classes with some special emphasis (Kalalahti et al., 2015).These opportunities appear to attract parents in higher social strata.School and selective group choice have therefore been considered as a middle-class strategy and concern, offering a way to maintain the family's socio-economic position in the future (Ball, 2003;Ball et al., 1996;Kosunen, 2016).The opportunity to apply for classes with a special emphasis seems to resonate with middle-class parenting styles (Lareau, 2003) and cultural capital more generally, while simultaneously making the choice seem natural and justifiable by meritocratic reasoning.
Also, adolescents' alcohol use is connected to parental socio-economic status through parenting practices (Pape et al., 2017).Our results suggest that having "a touch of social class", as Pape et al. (2017) call it, adolescents' drinking behaviour may reflect their parents' middle-class cultural capital in the form of embodied dispositions related to school success and attitudes towards drinking.To continue, parenting practices as well as organised out-of-school activities like music or sports may simultaneously protect children from spending "risky" unsupervised time with peers (Osgood et al., 1996) which may include alcohol use.Activity in organised hobbies and previous school success also increase the probability of being selected into special emphasis classes.In other words, abstaining from heavy alcohol use in adolescence may relate to the set of lifestyle-based social distinctions of middle-class families, which help their children to flourish in the formally meritocratic schooling system.Even though youth cultures are partly independent from adult culture, transmitted cultural capital and internalised dispositions during childhood are reflected in the differentiated student compositions between schools and especially between classes.This selection-based differentiation may further work as a platform for contextual peer cultural variations, in which students feel more or less pressured to engage in heavy episodic drinking (Demant & Järvinen, 2006) or to abstain from it (Bakken et al., 2017;Pavlidis et al., 2019) according to the class type or school context.
Class and school contexts may also differ in the options, opportunities and resources of students to engage in alcohol consumption.As youth alcohol use is predominantly a socially oriented action, the quality and the extent of one's social networks (cf.social capital) may channel drinking habits (Villalonga-Olives & Kawachi, 2017).In socio-economically advantaged school contexts, more opportunities may be open to engage in such acts, for example in terms of schools' geographic location or the amount of pocket money.Even though differences in resources or opportunities may thus potentially explain the GPA-adjusted differences between advantaged and disadvantaged school contexts, they may not be as relevant in explaining the effect of class type.Despite relatively advantageous student compositions of selective classes, heavy episodic drinking in these classes is less frequent compared to non-selective classes.Bernelius (2013) noted that the schools' reputation and popularity are connected to the sociospatial characteristics of the neighbourhoods in ways which shape school choices and housing choices made by educationally motivated families, and that the rejection of schools in relatively disadvantaged neighbourhoods is particularly consistent.Other studies on the topic have demonstrated that school reputation is strongly connected to parental choices and the rejection of schools (Bernelius et al., 2021;Kosunen, 2016).Peer effects and negative health behaviours such as alcohol use may play a role in the construction of school reputations and parental decision-making (see e.g., Bernelius et al., 2021), which can have far-reaching consequences on school choices and gaps that are growing between schools.
Our results also raise questions about the effects of organising specialised educationsuch as music or mathematics emphasisby differentiating between the students in separate classes, feeding into within-school segregation.The differentiation of alcohol use between these different types of classes within schools demonstrates how within-school segregation is connected to multiple levels of relative (dis)advantage, creating gaps not only in students' socio-economic and cultural backgrounds and educational outcomes but also differentiating and potentially exacerbating gaps in health behaviours between the student groups.

Strengths and limitations
A strength of the ESPAD data is that it consists of large samples with high response rates, making it representative of Finnish 15-16-year-old students.However, as only schools in the Helsinki-Uusimaa region were included in our analysis, the generalisability of the results outside this region must be considered with caution.It is possible that the strict competition in the selection between special emphasis and regular classes in the Helsinki capital region results in deeper divides between these class types, providing an "extreme" case for research in the Finnish context.In other parts of the country, student compositions remain more mixed, and lower levels of differentiation between class types may not be reflected in students' alcohol use.
As always with self-reporting data, there is a risk that students consciously or unconsciously fail to give accurate, honest answers about all substances.The direction of these incorrect answers may go both ways, i.e., there may be over-reporting as well as under-reporting depending on what is socially desired/accepted in different contexts.However, a validity report on ESPAD (Hibell et al., 2015) shows that a very small minority (1-2 percent) do not respond honestly to questions about substance use.Another potential source of bias relates to students' self-reports concerning family background characteristics.For example, children may overestimate their parents' educational level (Lehti & Laaninen, 2021), which may lead to underestimation of the connections between parental education, school success, and class type.About 16 percent of the students in the data did not know the educational level of either of their parents.These observations were excluded from the analysis, which may bias the results if the observations are not missing at random and are related to the variables used in the models.
Also, the skewness of the class type and heavy episodic drinking variables must be considered.As the proportions and absolute numbers of those students who studied in selective classes as well as those who had recent heavy drinking episodes was quite small, the number of those who both studied in selective classes and had drunk heavily remained small in the data.The results concerning complex mediational processes between class type, HED, and many other variables simultaneously must especially be considered with caution.A larger sample of individuals, nested within schools and classes, would be needed to investigate simultaneously the effects of school and class contexts and class type further.In addition, longitudinal data with appropriate research design would enable more precise and reliable separation between selection effects and contextual effects on youth alcohol use.

Conclusions
Our results show that Finnish students who are selected into classes with a special emphasis engage less in heavy episodic drinking compared to their peers in non-selective classes.They also often do well at school and come from families with well-educated parents.These same factors are related to lower alcohol use among adolescents, and they partly mediate the association between class type and alcohol use.In addition, friends' drinking behaviour mediates some of the association.The opportunity to apply to study in classes with a special emphasis may support motivated and talented students in their studies, but it might also deepen the inequalities between student groups and schools in terms of risky behaviour such as alcohol use among adolescents.
Socio-economic school segregation creates differentiated school contexts in which health behaviours such as alcohol use may also be differentiated between schools and between classes within schools.These differentiated peer groups and health behaviours may further exacerbate school segregation, feeding into vicious circles of segregation in schools and neighbourhoods.Targeted supportive measures, such as needs-based funding (Bernelius & Huilla, 2021), accompanied by the principles of the Icelandic Prevention Model (Kristjansson et al., 2020), could help not only to secure and maintain optimal learning environments for all students but also promote meaningful social activities that are not structured around substance use.Close connections between schools, families, and local neighbourhood communities could serve to promote social control in the neighbourhood, potentially reducing opportunities for deviant acts and channelling students' leisure time activities with peers.
However, schools as institutions may have limited opportunities to affect students' peer cultural norms in relation to alcohol use.As school and class compositions may affect students' alcohol use as well as their learning outcomes through peer effects, precautionary measures counteracting school and teaching group segregation might have the potential to condition the emergence of peer pressures on both substance use and educational outcomes.It is thus essential to critically consider how student sorting and school choice policies are connected to within-school segregation, between-school segregation, and reflected further in residential segregation of urban areas.It is also important to recognise the multidimensional and complex nature of these interconnected systems and policies, when different practical solutions and alternatives are evaluated by their potential consequences.Our results offer a novel addition to these discussions by focusing on students' alcohol use as a health-related social behaviour.

Figure 1 .
Figure 1.Effects related to selection between special emphasis classes and regular classes (H2, H3) on alcohol use.

Figure 2 .
Figure 2. Effects related to studying in special emphasis classes versus regular classes (contextual effects) on alcohol use (H3, H4).

Table 1 .
The prevalence of heavy episodic drinking and the background factors according to class type, % (weighted) with p (chisq).

Table 2 .
Linear probability models for adolescents' heavy episodic drinking during the past 30 days (parameter estimates with standard errors), uncontrolled and controlled for the other background variables.

Table 3 .
Direct and indirect effects of class type, parental education, GPA and friends' alcohol use on adolescents' heavy episodic drinking in the past 30 days.