Tracking and political engagement: an investigation of the mechanisms driving the effect of educational tracking on voting intentions among upper secondary students in France

ABSTRACT

Many scholars argue that the practice of educational tracking exerts a distinct effect on young people’s political engagement. They point out that students in academic tracks are becoming more politically engaged than those than those in vocational ones, and suggest that this may be due to differences across tracks in the curriculum, pedagogy, peer environment or student self-confidence. The current paper aims to investigate whether tracking is related to political engagement through any of these four mechanisms. It uses survey data collected among students in the final year of upper secondary education in France and employs a stepwise multilevel analysis to explore this question. It finds little differences between tracks in the curriculum and in pedagogy relevant for political engagement. Students in academic tracks nonetheless express a stronger commitment to vote than those in vocational ones. This difference between tracks disappears when the social composition of the school population is taken into account, suggesting that the peer environment is the primary mechanism driving the effect of tracking in France. However, in contexts with greater variation between the tracks in curriculum and pedagogy, the latter may well be equally or more important mechanisms.

KEYWORDS:

• Educational tracking
• political engagement
• upper secondary
• France
• citizenship education

1. Introduction

Equal participation in politics for all social groups is an important precondition for effective democratic rule. If some groups are markedly less inclined to vote and are less active in other ways, democratically elected governments will be less responsive to their needs, which will affect the public legitimacy of liberal democracy (Bartels 2008; Levinson 2010). Tracking is often mentioned as a feature of education systems that is counterproductive in mitigating these inequalities (Gamoran and Mare 1989; Janmaat, Mostafa, and Hoskins 2014; Hoskins and Janmaat 2016; van de Werfhorst 2017).

Tracking refers to the practice of educating young people in different vocational or academic tracks, usually in upper secondary but sometimes also at the beginning of or midway through lower secondary. Allocation to the different tracks occurs at the end of the preceding phase and is almost always done on the basis of educational achievement. As a rule, entry requirements for academic tracks are more demanding than those for vocational ones. Although tracking is usually justified by the argument that young people need specialised training to prepare them for specific positions in the labour market, it can, at the same time, generate inequalities in active citizenship by offering different learning opportunities for civic and political engagement. In this regard, scholars have pointed out that academic routes offer much better preparation for active citizenship than vocational tracks (Dewey [1916] (1966); Whitty 1985; Ichilov 2002). These different learning environments are likely to enhance pre-existing inequalities in engagement because the children allocated to the different tracks are not the same in terms of family background and prior engagement. Because of selection on the basis of achievement, children from disadvantaged backgrounds, who tend to be less politically engaged from the onset (Verba, Lehman Schlozman, and Brady 1995; Achen 2002), are disproportionately allocated to the vocational tracks (Hallinan 1994; Loveless 1999). Once in the vocational tracks, they will experience a less stimulating environment than their middle-class peers in the academic track, who are already more engaged. In this way, tracking is likely to exacerbate pre-existing social disparities in political engagement (Janmaat, Mostafa, and Hoskins 2014).

Scholars have suggested a number of reasons why tracking could have an independent effect on political engagement. These concern cross-track differences in the curriculum, pedagogy, pupil composition and political efficacy, which the next section will discuss in greater detail. Interestingly, the literature has so far, to our knowledge, not indicated which of these reasons could be more important and why. Neither has it tested them. Consequently, we currently do not know whether tracking has an effect on political engagement through the mechanisms suggested by these explanations and how important these mechanisms are. The present paper aims to address this omission. It identifies a number of mechanisms through which tracking is said to influence political engagement and tests these with survey data collected among young people in their final year in upper secondary in France. In view of the mounting evidence about the inequality enhancing impact of tracking (Eckstein, Noack, and Gniewosz 2012; Janmaat, Mostafa, and Hoskins 2014; Hoskins and Janmaat 2016; van de Werfhorst 2017; Witschge, Rözer, and Van de Werfhorst 2019; Witschge and Van de Werfhorst 2019), it is important that we come to know precisely how tracking shapes political engagement. Clarity on this could help to reform the system in ways that reduce this effect and thereby mitigate the inequality-augmenting side effect of tracking.

Considering the paper’s aim, France is an interesting country to focus on for three reasons. First, it offers a series of general courses, including citizenship education, in all tracks in upper secondary (see below for further explanation), which suggests that curriculum differences between tracks are small compared to countries or regions with more specialisation, such as England. Consequently, if tracking mainly fuels political inequalities through the curriculum, one would not expect France to show large differences between tracks in political engagement. Second, and on the other hand, while the education system in France is formally founded on the ideals of equality and meritocracy, it has also been identified as a system that merely reproduces social inequality by making middle class children succeed and failing those of working class backgrounds (Bourdieu and Passeron 1990). It is the subtle ways in which the cultural capital of middle class children is rewarded and that of working class children is punished that makes the county’s education system have this effect (ibid.). The latter suggests that tracking might produce the said inequalities through mechanisms other than the formal curriculum. Third, France rarely conducts surveys among youth regarding their civic values and it did not participate in 1999, 2009 and 2016 editions of the international civic education study organised by the International Association for the Evaluation of Educational Achievement (IEA). Thus, little is known about the political engagement of French school-going youth overall.

The paper starts with a discussion of the mechanisms through which tracking influences political engagement. This section also addresses the issue of selection effects. Subsequently, we will discuss France’s system of upper secondary education, and notably the aspects relevant for political engagement, in order to briefly assess how distinct France’s system is relative to that of other countries. We then proceed with a discussion of the data source, the variables of interest and the analytic approach. The last two sections offer the findings and a discussion of them in the light of previous research.

2. The centrifugal effect of tracking on political engagement

As already noted, the literature offers various reasons as to why a differentiated system with several academic and vocational tracks increases cross-track inequalities of political engagement. The first of these concerns the curriculum. While academic routes typically include a number of general subjects aimed at fostering active citizenship, such as citizenship education, social studies and history, vocational trajectories as a rule offer practical courses aimed at developing job-specific skills. These general subjects do not only seek to enhance knowledge and understanding of contemporary political issues and institutions, but also the skills to navigate the political world and participate effectively in it, such as critical analysis, developing and defending an argument, taking the initiative and persuading others (Niemi and Junn 1998; Hillygus 2005). In contrast, the practical courses in vocational education, if they at all touch on issues of citizenship, predominantly train young people how to be loyal workers and good followers by emphasising discipline, conformism and good manners (Whitty 1985; Apple 1990; Ichilov 2003; Ten Dam and Volman 2003; Leenders, Veugelers, and De Kat 2008; Eckstein, Noack, and Gniewosz 2012). By thus depriving young people of their own voice, they foster disengagement and alienation, it is concluded.

Another explanation of track differences in political engagement focuses on learning through practice. Based on the notion that young people only start to identify with the political process when they are active agents in their own learning (Sfard 1998; Torney-Purta 2002a; Biesta, Lawy, and Kelly 2009), scholars have noted that academic tracks generally offer more opportunities for such participatory forms of learning than vocational ones. In the former teachers have high expectations of students (Kelly and Carbonaro 2012). They encourage them to take part in debates on sensitive social and political issues and to participate in school decision-making as these activities are seen as important for their formation as engaged and responsible citizens willing to take on leadership roles in the future (Ichilov 2002). In the vocational tracks, by contrast, teachers are less likely to encourage a free discussion of political issues as they fear a disruption of order within the class (Hurn 1978). As teachers have lower expectations of students, they are less likely to offer them a voice in their own learning and in school matters more generally (Ichilov 1991).

A third explanation considers the social composition of schools. It argues that the concentration of relatively disengaged and disadvantaged young people in the vocational tracks (as a result of selection on the basis of ability, as noted previously) gives rise to a peer group culture marked by a rejection of the world of politics, alternative status symbols and a contempt for the educational process (Willis 1997; Ichilov 2002; van de Werfhorst 2007; Jacobsen, Frankenberg, and Lenhoff 2012). In this counter culture, politics is seen as a world for ‘the others’ and politicians are portrayed as arrogant, unreliable and self-serving people out of touch with and indifferent to the needs of the ordinary person (Hoskins and Janmaat 2019). Norms arise that sanction a lack of understanding of and participation in politics, as Forsberg (2011) has found for male-dominated vocational tracks in Sweden.

Finally, tracking has been said to exert its effect on political engagement by undermining the self-confidence of students allocated to the vocational track, particularly in contexts where a large status difference exists between academic and vocational education and where the latter is associated with failure (Hoskins et al. 2016). This lack of confidence or self-efficacy not only has consequences for these students’ educational aspirations (Van Houtte and Stevens 2009), but also for their engagement with the world of politics. The educated and sophisticated jargon of this social domain, which they struggle to understand, could well turn their lack of self-confidence into low levels of political efficacy and hence a lack of motivation to participate (Hoskins et al. 2016; van de Werfhorst 2017).

Studies looking at the consequences of tracking, such as the present one, always need to take selection effects into account. In case of the present study, such effects refer to the possibility that any difference in political engagement between tracks is not the result of tracking but simply the reflection of pre-existing differences in political engagement, i.e. differences prior to track enrolment. In that case any relation between track attendance and political engagement is a spurious one where both track attendance and political engagement are caused by a third factor. Some scholars argue that family socialisation represents this third factor. They consider early childhood to be a key formative stage for young people to develop a propensity to participate and believe that education in late adolescence has little to add (e.g. Kam and Palmer 2008; Persson 2012). Education in their view merely proxies for factors capturing family background (e.g. Persson 2014).

However, this view has been contested by others who believe that young people only begin to become interested in political matters during late adolescence and early adulthood (Flanagan and Sherrod 1998; Costa and McCrae 1994). As a disposition to participate in political affairs is thus still in the making during these stages, education and other experiences will have a considerable influence on it (Flanagan and Levine 2010; Jennings and Stoker 2004; Sherrod, Torney-Purta, and Flanagan 2010; Sondheimer and Green 2010). In our case, selection effects could have been optimally assessed if we had had panel data at our disposal that included measures of political engagement prior to enrolment in tracks. However, we could only use cross-sectional data collected in the last year of (tracked) upper secondary (see further below). We nonetheless believe we can at least partly address selection effects because these data include several items allowing us to control not only for social and ethnic background but also other relevant family characteristics, such as parental interest in politics, that shape young people’s political engagement (see further below). Having said this, we are well aware that the dataset we use does not allow us to make any causal inferences and we will consequently refrain from doing so. We will only use terms implying causal relations, such as ‘effect’ and ‘impact’, when remarking that the findings are ‘suggesting’ or ‘indicative of’ an influence of tracking.

3. Upper secondary education in France

In this section we provide some contextual information on France’s system of upper secondary education not only to offer insight into the characteristics and status of academic and vocational education but also to develop some conjectures regarding the different mechanisms through which tracking might influence political engagement.

The upper secondary phase in France lasts three years and starts at age 15. As education is compulsory between the ages of 6 and 16, students have the possibility to leave school one year into upper secondary. As in many other countries, upper secondary education in France is divided in several tracks. Academic tracks concern the pathways leading up to a baccalauréat général or technologique, a qualification comparable to A-Levels in England. These tracks are offered in a lycée d’enseignement général et technologique (academic college; equivalent to a Sixth Form college in the United Kingdom¹). Vocational pathways, which are provided in a lycée professionnel (vocational college), can be followed at two levels: those preparing for the baccalauréat professionnel and those culminating in a certificat d’aptitude professionnelle, or CAP (professional aptitude certificate), a two-year vocational qualification. In addition, there are lycées polyvalents, which are colleges that offer pathways for all three baccalaureates – general, technological and vocational (henceforth called a mixed college; roughly equivalent to further education colleges in the United Kingdom). Such colleges cannot be compared to comprehensive schools with mixed ability classes as students remain in their distinct pathway during their time at the college.

All baccalaureate qualifications offer direct access to higher education, but the degree to which baccalaureate graduates make use of this opportunity varies substantially across the different types of baccalaureate. While almost 95% of students passing the general baccalaureate and 85% of those passing the technological baccalaureate enrolled into higher education in 2013, only 25% of those having obtained the vocational baccalaureate did so (Ucas 2016). Moreover, when the latter do pursue higher education, their non-completion rates are higher and they take a longer time to obtain the bachelor’s degree (Cnesco 2018). Entry into the different tracks at age 15 depends on the orientation decision taken at the end of lower secondary. This decision is the responsibility of the school head and is informed by the conseil de classe (class council), a body composed of teachers and class representatives which reviews students’ results and grades (Eurydice 2018; OECD 2004, 18).

As in many other countries, there is a large status gap between academic and vocational tracks in France to the detriment of the latter. Indeed, the image of vocational education is more negative in France than in any other EU member state except Hungary (Cedefop 2017, 33). This may be related to the limited value of vocational qualifications for occupational attainment in France: seven months after graduation from school 41% of vocational baccalaureate holders are still unemployed (Cnesco 2018). It may also emanate from the demographic composition of vocational education, with vocational tracks mainly attracting students from disadvantaged and migrant backgrounds, and from problems of absenteeism and disorder. In 2014, for instance, the student absenteeism rate in vocational colleges stood at 11.5%, which was more than twice as high as in academic colleges. In addition, during the year 2014–2015, the former experienced, on average, 5 times more serious incidents than the latter (Cnesco 2018; Di Paola et al. 2016). This negative image is also likely to account for the relative unpopularity of vocational education in France: among the 15–19 year olds 23% are enrolled in vocational tracks and 37% are enrolled in academic ones; the equivalent figures for the OECD as a whole are 29% (vocational) and 35% (academic) (OECD 2017, 258). The latest figures, moreover, show that there are more than twice as many students enrolled in an academic than in a vocational college (MENJ 2016, 13).

The Republican tradition of curricular encyclopeidism means that students have to study a wide variety of subjects whatever the track they are enrolled in (Green, Preston, and Janmaat 2006; Mons 2007). Thus, the vocational tracks also offer a number of general subjects relevant for civic and political engagement, such as enseignement moral et civique (EMC) (citizenship education), history and geography (MENJ 2019). This broad curriculum with its emphasis on civic education lies at the heart of state and nation-building, which historically have been major objectives of public schooling in France (Déloye 1994). No other European state assigns so much weight to civic education, in terms of teaching hours in both primary and secondary education, as France.

In 2015, in a context marked by the Paris bombings and challenges to national cohesion, the effort to support civic education was redoubled. Its content was made more similar across tracks in secondary education and it received the same status as history and geography in the curriculum. Its inclusion in the final exams of the vocational baccalaureate meant that it acquired an even greater prominence in vocational than in academic tracks. Yet this reform did not fundamentally change the low status of the subject, especially among teachers (Bozec 2016). In upper secondary education, teachers of history, geography, French or philosophy are in charge of civic education and some of them use the teaching hours allocated to civic education for their own subjects (ibid).

In addition to these courses directly related to civic education, both vocational and academic tracks provide French, a foreign language, prevention-health-environment and several other general courses. In thus ensuring a certain degree of curricular uniformity across tracks, France contrasts strongly with England where education at the upper secondary level is highly specialised and where vocational tracks as a rule provide only practical, job-related training (Dehmel 2005).

As noted previously, we have not identified an argument in the literature about the effect of tracking on political engagement running through one of the aforementioned mechanisms in particular. However, the relatively high degree of curricular uniformity of the French system of upper secondary education does lead us to propose that any independent effect of track (taking family background characteristics into account) is likely to reflect one or several of the other three proposed mechanisms mentioned above (pedagogy, peer group culture, or self-confidence). Among these mechanisms, peer group culture could well be an important one in view of possible social sorting effects emanating from the selection procedure for admission to academic tracks (as explained above). Considering the relatively large status difference between the tracks, allocation to the vocational track could well affect students’ self-confidence, in which case the effect of tracking could also run through this mechanism.

4. Data source

We use data of the Enquête école et citoyenneté (EEC) (School and Citizenship Survey) (Cnesco 2018) to investigate the main research question. This nationally representative survey was organised by the Conseil national d’évaluation du système scolaire (Cnesco), a research agency of the French Ministry of Education, and was held among 8146 students in the last grade of upper secondary (terminale) in 2018.² These students were drawn from 175 colleges. In each sampled college all the students from two randomly selected classes were surveyed. Of the 8146 students selected, 6682 participated, thus resulting in a response rate of 82%. The study also surveyed principals and teachers in charge of teaching citizenship in the selected schools. As the sample was clustered and oversampled students with certain individual characteristics and those in vocational colleges to increase robustness, the database includes weights to make the data representative of the national population of pupils. We used the ‘studentmerge’ weight (ponderation_merge_el) as our analytic sample consists of both student and school data.

The EEC is a rich data source that dramatically expands the possibilities to link school characteristics to civic outcomes. Unlike the IEA Civic Education Study among Upper Secondary Students (Amadeo et al. 2002), it includes important information on the participating schools, such as the type of college (academic, vocational or mixed). It combines this with a wealth of data on students’ civic attitudes and family background characteristics. Its limitation is that it does not include data on students in vocational tracks leading to a CAP qualification (first vocational degree, below the vocational baccalaureate) nor on early school leavers. Thus the data only represent those studying for the baccalaureate (general, technological or vocational).

5. Measures

5.1 Dependent variable

As most of the respondents are not eligible to vote yet, we relied on questions tapping voting intentions to represent political engagement. Of course intentions need not be lived up to and previous research has found that voting intentions indeed overestimate actual electoral participation (Achen and Blais 2010). However, this research also found that intentions and actual participation are closely correlated, suggesting that intentions are a good predictor of the latter (ibid.; Quintelier and Blais 2015). We specifically focussed on voting because it is the most common and accessible way of participating in politics. According to Dahl (1998), voting is key to the legitimacy of a liberal democracy because it enables equal and effective participation. Although this traditional form of participation is less popular among young people (Lichterman 1996), it still is the most frequently reported political activity among this age group (Keating, Green, and Janmaat 2015). We constructed voting intentions as a scale composed of three items asking respondents how likely they are to vote in local, national and European elections (alpha = 0.85). It represents the average of the item Likert scale responses, which consist of the options (1) ‘certainly not’, (2) ‘probably not’, (3) ‘probably’ and (4) ‘certainly’. It thus ranges between 1, denoting minimal intention to vote, and 4, denoting maximum intention to do so (see Appendix for the full wording of the items). Scales constructed in this straightforward fashion allow for a richer and more transparent interpretation than those based on the output of a factor analysis as they enable an assessment of both relative and absolute scores. The scales discussed below were constructed in a similar fashion.

5.2 Key independent variables

We measure track with three dummy variables at the school level denoting the type of college attended: (1) academic, (2) mixed or (3) vocational (as explained in the previous section). The average age of the sampled students going to these colleges is 18.0, 18.3 and 18.6, respectively, which indicates that students in vocational colleges have a higher rate of grade repetition (this practice is common in France to regulate the attainment of educational standards (Green, Preston, and Janmaat 2006).

We have already mentioned that the different tracks share a number of key courses central to political engagement, including EMC as a separate school subject (not a programme integrated in several subjects as it can be in various European countries). We therefore do not expect any effect of tracking to operate through the curriculum, not because the curriculum is ineffective in enhancing political engagement but because there is too little variation across tracks in this mechanism. Nonetheless schools, and therefore tracks, can show some differentiation in other aspects of the curriculum, such as the content covered in a particular course. The relatively low status of EMC contributes to this differentiation. The database has information on the content covered in EMC, as arguably the most relevant of these courses. From the student questionnaire we take 7 items from a battery of 17 items on the topics addressed in EMC to construct a measure of course content. The topics queried include voting in elections, political institutions, the judiciary, symbols of the Republic, the functioning of democracy, political and social organisations, and the media. The measure represents the sum of these items (which have the response categories 0 = no and 1 = yes) and thus has a minimum of 0 (EMC covers none of these topics) and a maximum of 7 (EMC covers all of them).

To capture the mechanism of pedagogy/learning through participation, we created three variables, open climate, student influence on pedagogy, and civic project. The first of these represents a scale that measures how open the climate of discussing political and social matters is in class, as perceived by students. It synthesises six items asking among other things whether the teacher encourages students to express their opinions and engage in a debate (alpha = 0.89) (see Appendix for all items). Constructed in the same way as the dependent variables, it has a minimum of 1 (never) and a maximum of 4 (always). An open climate of classroom discussion has been identified as a powerful agent of political engagement by many studies (e.g. Torney-Purta 2002b; Campbell 2008; Hoskins, Janmaat, and Villalba 2012). Student influence on pedagogy (henceforth ‘student influence’) is a scale based on seven items asking whether student views are taken into account in decisions on all kinds of school matters, such as the content of the lessons, teaching approaches, modes of assessment and school trips (alpha = 0.86) (see Appendix for all items). It ranges between 1 (not at all) and 4 (very much so). Civic project is based on a single question asking whether the respondent has participated in a civic project organised by the school (1 = no; 2 = yes, in previous years; 3 = yes, this year; 4 = yes, this year and in the previous years). These projects include a great variety of activities and often involve cultural exchange and aid missions to developing countries (MCM 2019).

School social composition, as the third suggested mechanism through which tracking might exert an effect on political engagement, is measured by the school average of social background (see below for the construction of this variable). It ranges between 0 (maximally disadvantaged) and 3 (maximally privileged). This way of measuring school social composition (or status) is common to studies interested in exploring peer effects (e.g. Gamoran 1996; Wilkenfeld 2009). Wilkenfeld (2009) found school social status to have a strong positive effect on political engagement, supplementing that of individual social background.

Finally, as an indicator of self-confidence relating to political issues (reflecting the last of the aforementioned mechanisms) we created the variable of political efficacy. This variable represents a scale that is based on six items tapping self-assessment on political knowledge, participation in discussions and ability to understand politics (alpha = 0.93) (see Appendix for all items). As these items all refer to assessments of one’s own ability in dealing with political matters, they represent internal political efficacy. Gallego and Oberski (2012) found this type of efficacy to have a strong influence on both traditional and alternative forms of political participation. Political efficacy was constructed in the same way as the other scales and has a minimum of 0 (not confident at all) and a maximum of 4 (very confident).

5.3 Control variables

We add a number of individual-level controls to the analyses concerning individual and family background characteristics to assess, to the greatest degree possible, whether the effect of track merely proxies for these characteristics. We measured social background with parental education, which represents the average of the mother’s and the father’s highest level of education (with the categories 0 = no qualification, 1 = lower secondary, 2 = upper secondary, 3 = higher education).³ We chose to use parental education rather than occupation or income as indicator of social background because of the strong parent-child correspondence on education (Beck and Jennings 1982) and because it is a strong determinant of children’s political engagement (Gidengil, Wass, and Valaste 2016; Smets and van Ham 2013; Lahtinen, Erola, and Wass 2019). Thus, including parental education in the model helps us to at least partially address selection effects. Household composition is measured with three variables: living arrangement [0 = other arrangement; 1 = living with both parents], brothers [0 = no brothers; 1 = one or more (half-)brothers]; sisters [0 = no sisters; 1 = one or more (half-)sisters]. We included these controls because single parenthood has been associated with lower political engagement (Shore 2019) and because more siblings have been argued to reduce the amount of time parents can devote to their children with consequences for their political learning (Brady, Verba, and Schlozman 1995). We further added two variables capturing parental political engagement as a major determinant of young people’s political participation (Beck and Jennings 1982; Jennings, Stoker, and Bowers 2009): Political interest, which is based on an item asking whether one’s parents are interested in current political, economic and social affairs [1 = not interested at all ≪≫ 4 = very interested] and political discussion, which is represented by an item asking how often the respondent engages in discussion with parents about political and social affairs [1 = never ≪≫ 4 = every day]. As immigrants have been found to show much lower rates of political participation than the native population (de Rooij 2012), we also control for immigrant status, which is captured with three variables: birth place respondent [0 = abroad; 1 = France], birth place mother [0 = abroad; 1 = France], birth place father [0 = abroad; 1 = France]. We further add controls for gender and age. Table 1 offers the basic descriptive statistics of all the variables.

Table 1. Descriptive statistics of the variables included in the analysis.

Variables	Valid N	% missing	Minimum	Maximum	Mean	SD
Voting intentions	6182	7.5	1.00	4.00	3.11	.81
LEGT	6682	0	0	1	.53	.50
LPO	6682	0	0	1	.18	.39
LP	6682	0	0	1	.28	.45
Course content	6160	7.8	.00	7.00	4.14	2.50
Open climate	6161	7.8	1.00	4.00	2.53	.79
Student influence	6110	8.6	1.00	4.00	2.1	.70
Civic project	6524	2.4	1.00	4.00	1.59	.91
School social composition	6619	1.0	0.33	2.85	1.82	.42
Political efficacy	6603	1.2	.00	4.00	2.15	.92
Gender (1 boy; 2 girl)	6603	1.2	1	2	1.49	.50
Age	6495	2.8	15.2	28.3	18.24	.98
Social background	5870	12.2	0	3	1.83	.91
Living arrangement	6576	1.6	.00	1.00	.63	.48
brothers	5745	14.0	0	1	.60	.49
sisters	5630	15.8	0	1	.57	.49
Birth place respondent	6445	3.6	.00	1.00	.93	.25
Birth place mother	6460	3.3	.00	1.00	.82	.39
Birth place father	6372	4.6	.00	1.00	.80	.40
Parental interest	6521	2.4	1	4	3.08	.70
Political discussions	6117	8.5	1	4	2.76	.84

6. Analytical approach

As the sample is clustered and the analysis will include variables at the individual and school level, we conducted a two-level multilevel analysis (MLA) using SPSS. MLA adjusts for the bias produced by the non-independence of observations in clustered samples and ensures that the higher level variables are estimated accurately (Snijders and Bosker 1999). We use a linear model in view of the continuous nature of voting intentions as the dependent variable. All independent variables were grand-mean centred before entering them in the analysis.

Aside from sisters (16%), brothers (14%) and social background (12%), no other variables included in the analysis had a level of missing values of more than 9% (see Table 1). Nonetheless, to prevent data loss due to item non-response, we imputed data using the default option in SPSS and using all the variables in the analysis as input variables. The default produces five datasets with imputed values and a pooled one representing the average of these datasets. We used the pooled dataset for the analyses (N = 6682).

The analysis will proceed in a stepwise fashion. We first offer the null model to assess the partition of the variance of voting intentions across the two levels of the analysis. If less than 4% of the variance is at the class or school level, it is unlikely that explanatory variables at these levels have much of an effect (cf. Duncan and Raudenbusch 1999). We then offer a series of models to assess what drives the link between track and political participation. The basic idea is to explore how the coefficient of track changes with the consecutive inclusion of variables representing the four mechanisms in the model. If the strength of the coefficient is reduced substantially after the inclusion of one of such ‘mechanism’ variables, we infer that the relation of tracking runs mainly or for an important part through this mechanism (cf. Verba, Lehman Schlozman, and Burns 2005; Semyonov, Raijman, and Gorodzeisky 2006; Stubager 2008,; Semyonov and Glikman 2009, who use a similar logic). We can assess the degree of significance of this reduction by comparing the confidence intervals of the coefficients across two models. If these show a less than 50% overlap, we can infer that the reduction has been significant (cf. Cumming 2009). Model 1 only includes track as predictor variable; Model 2 adds the control variables to assess whether the ‘effect’ of track is not solely due to selection.⁴ Models 3 to 6 then introduce the variables representing the four mechanisms one after the other. Model 7 includes all the explanatory variables.

7. Results

We first present descriptive statistics to see how marked the differences between the tracks are in voting and in the proposed mechanisms driving the impact of track (Table 2). As could be expected, the mean level of intended voting is significantly higher among students in academic colleges than among those in mixed and vocational ones (3.22 compared to 3.08 and 2.85 on a scale from 1 to 4). It is worth pointing out, however, that all groups score above the midpoint of the scale (2.5), indicating that on balance young people are rather more than less prepared to vote, whatever the track they are enrolled in. Thus, while students in the vocational colleges are relatively disengaged, they are not disengaged in an absolute sense. Braconnier (2013) similarly found high levels of engagement overall, with intended participation in presidential elections being notably higher than in local or European ones.

Table 2. Differences across tracks in levels of voting intentions and the four mechanism associated with the effect of tracking (means and mean differences).

	Academic	Mixed	Vocational	Difference academic-mixed	Difference academic-vocational	Difference mixed-vocational
voting intentions	3.23	3.08	2.85	0.15***	0.38***	0.23***
Curriculum
course content of EMC	4.05	4.33	4.25	−0.28***	−0.20***	0.08
Pedagogy
open climate	2.59	2.51	2.52	0.08***	0.07***	0.01
student influence	2.04	2.07	2.14	−0.03**	−0.10***	0.07***
civic project	1.62	1.48	1.57	0.14***	0.05*	0.09***
Social composition	2.06	1.80	1.41	0.26***	0.65***	0.39***
Political efficacy	2.26	2.14	1.89	0.12***	0.37***	0.25***

* P < 0.05; ** P < 0.01; *** P < 0.001.

A mixed picture emerges when we compare the three tracks on the curriculum and pedagogy variables. Students in the mixed and in the vocational colleges report higher levels of relevant curriculum content than those in the academic ones. This could indicate that these schools make an extra effort to improve the political knowledge and engagement of their intake. It could also indicate the generally lower status of EMC in academic colleges. An inconclusive pattern shows up on the learning through participation variables. While students in the academic colleges are more positive about the openness of the climate for political discussions and report a greater frequency of civic projects, they are least positive about the possibilities for students to have a say in school matters. Moreover, the differences between the tracks, although significant, are not large: neither of the three groups is very pronounced in their views of an open climate (hovering just above the midpoint of the scale), all of them are fairly negative about the possibilities for students’ voice, and engaging in a civic project is a rare experience for all them (as indicated by the mean values below the midpoint of the scale for both variables). The higher levels of open discussions in the academic colleges may be related to the 2000 civic education reform (of L’Éducation civique, juridique et sociale (ECJS)), which introduced classroom debate as a source of learning and was mainly applied to academic tracks. The limited possibilities for student voice across the board echo the studies of Bozec (2016) and Condette-Castelain (2009) who similarly found students’ participation in school governance to be weak, whichever track they follow, despite official policy calling on schools to give students a greater say in matters of teaching and learning.

Thus, on neither of the two types of variables are there pronounced and consistent differences between the academic and vocational pathways. This provisionally suggests that mechanisms other than the curriculum and pedagogy are driving the difference between the tracks in intended voting.

The tracks do show a marked difference in their social composition. While the average social background of students in academic colleges is 2.06 (on a scale from 0 to 3), that of students in mixed colleges is 1.80 and that of students in vocational colleges is 1.41, and these differences are highly significant. These average scores mean that the parents of students in academic colleges score slightly above upper secondary in terms of highest level of education, those of students in mixed colleges score slightly below upper secondary and those of students in vocational colleges score halfway between lower and upper secondary. A similar pattern is observable for political efficacy. On a scale from 1 to 4, students in academic colleges on average score 2.26, those in mixed ones 2.14 and those in vocational ones 1.89. Although these values are all below the midpoint of the scale, thus denoting less rather than more self-confidence in dealing with political matters, the differences between them are highly significant as shown by the t ratios. The patterns in these variables thus suggest that track could well exert its effect through school social composition or students’ sense of efficacy in dealing with political matters.

We proceed now with the results of the multilevel analysis, as displayed in (Table 3). The empty model shows that 11.9% of the variance in voting is between schools (see Note 2 below the table). This is a sufficiently high proportion to expect school-level variables, such as track, to show a link with voting (cf. Duncan and Raudenbusch 1999). Track is indeed significantly related to voting, as shown in the model without any controls: levels of intended voting are significantly lower among students in the vocational (particularly so) and mixed colleges (academic colleges are the reference category), as noted before.

Table 3. The determinants of voting intentions.

	Model 1		Model 2		Model 3		Model 4		Model 5		Model 6		Model 7
	b	t	b	t	b	t	b	t	b	t	b	t	b	t
Track
Academic colleges (ref cat)
Mixed colleges ^a	−.20**	−3.5	−.09*	−2.0	−.10	−1.9	−.09	−1.9	−.06	−1.3	−.08	−1.8	−.04	−0.8
Vocational colleges ^a	−.37***	−7.7	−.18***	−4.4	−.19***	−4.5	−.19***	−4.6	−.09	−1.86	−.12**	−3.1	−.05	−1.0
Gender (0 = m; 1 = f)			−.18***	−18.9	−.19***	−19.7	−.18***	−19.0	−.18***	−19.0	−.23***	−24.0	−.24***	−25.1
Age			−.02*	−2.4	−.01	−1.6	−.02*	−2.6	−.02*	−2.4	−.02*	−2.5	−.01	−1.8
Social background			.06***	8.2	.06***	8.5	.06***	9.5	.06***	8.1	.06***	7.4	.06***	8.2
Brothers			−.07***	−6.2	−.06**	−5.4	−.07***	−5.5	−.07***	−6.2	−.06***	−5.1	−.05**	−4.0
Sisters			−.09**	−4.3	−.09**	−4.7	−.10**	−4.4	−.09**	−4.3	−.08**	−4.8	−.09***	−5.4
Birth place respondent			.11**	4.9	.15**	6.3	.12**	5.0	.11**	4.9	.12***	5.5	.15***	6.7
Birth place mother			.08***	4.7	.07***	3.9	.08***	5.1	.08***	4.7	.09***	5.4	.08***	5.1
Birth place father			.14***	8.3	.14***	8.6	.13***	8.0	.14***	8.3	.13***	8.2	.12***	8.2
Parental interest			.17***	19.4	.17***	18.9	.16***	18.0	.17***	19.4	.13***	13.6	.12***	13.0
Political discussions			.18***	18.6	.17***	16.0	.17***	17.3	.18***	18.5	.18***	18.1	.15***	14.8
Living arrangement			.10***	10.3	.10***	10.3	.10***	9.9	.10***	10.2	.11***	11.4	.11***	11.6
Course content					.02***	8.1							.00	1.3
Open climate							.06***	5.7					.05***	5.9
Student influence							.06***	8.3					.07***	8.4
Civic project							.02**	2.9					.00	0.1
Social composition									.13*	2.4			.14*	2.5
Political efficacy											.20***	31.3	.20***	32.1
Explained variance Level 1 (%)	0		12.7		12.8		13.6		12.7		16.3		17.5
Explained variance Level 2 (%)	36.0		57.3		49.3		58.7		58.7		61.3		62.1

* P < 0.05; ** P < 0.01; *** P < 0.001

Note 1: N = 6682 individuals and 175 schools for all models.

Note 2: Model 0 (the empty model) shows that 0.553 (88.1%) of the variance in voting intentions is at the individual level and 0.075 (11.9%) is at the school level.

^aThe confidence intervals are as follows for the estimates of mixed and vocational colleges, respectively: Model 1 [−.315; −.087] [−.459; −.273]; Model 2 [−.189; −.001] [−.259; −.098]; Model 3 [−.196; −.006] [−.280; −.110]; Model 4 [−.183; .004] [−.267; −.107]; Model 5 [−.158; .033] [−.199; .006]; Model 6 [−.173; .006] [−.200; −.046]; Model 7 [−.140; .059] [−.157; .054].

Model 2 shows that a significant part of this link can be explained by individual and family background characteristics. When controls for these characteristics are introduced, the coefficient for mixed colleges changes from −0.20 in Model 1 to −0.09 and is still but barely significant at the 0.05 level. This reduction in the size of the coefficient is also significant judging by the less than 50% overlap of the confidence intervals corresponding to the two coefficients (see below the table). The coefficient for vocational colleges decreases significantly as well (see the completely non-overlapping confidence intervals) but remains significant at the 0.001 level. This residual significance suggests that tracking does not merely proxy for pre-existing characteristics relating to family background but also exerts a genuine influence on young people’s political engagement. All the control variables show strong and significant links with voting. Parental interest in politics, discussing political issues with parents and gender show the strongest links. As expected, the more interested one’s parents are and the more one engages in political discussions with parents, the more one plans to vote in the future. Girls are less enthusiastic about voting than boys and so are older students by comparison to younger ones and students from disadvantaged backgrounds by comparison to those from privileged ones. Voting intentions are further stronger among those who live with both parents, are lone children, have been born in France and have parents who have been born in France.

To explore this further, we looked in greater detail at students from disadvantaged backgrounds (both parents have lower secondary as highest level attained) and privileged backgrounds (both parents have a degree in higher education) in the different tracks. We found that disadvantaged students in academic colleges had a significantly higher mean level of voting intentions (3.15) than disadvantaged students in vocational colleges (2.92) (at a 0.001 level of significance) (not shown in Table 3). Likewise, privileged students in academic colleges had a significantly higher level (3.44) than privileged students in vocational colleges (3.08) (again at a 0.001 significance level). Interestingly, students from privileged backgrounds in vocational colleges are thus slightly less engaged than students from disadvantaged backgrounds in academic colleges, although this difference is not significant. These additional analyses thus offer more indications of an influence of tracking that is independent of and complements that of social background, although the caveat also applies here that we have not been able to assess the voting intentions before track allocation.

Adding variables to the model representing aspects of the curriculum or pedagogy does not alter the link of vocational colleges and only just makes the link of mixed colleges become insignificant (see how the t statistic changes from −2.0 to −1.9) (see Models 3 and 4 in Table 3). Thus, the relation of track can hardly be explained by differences between tracks in the curriculum or pedagogy. This is not to say that these educational conditions are not strongly related to intended voting. We see, for instance, that EMC course content shows a strong positive relationship. In other words, the more relevant topics EMC covers (as reported by students), the more students plan to vote in the future. We also see that open climate and student influence show strong positive relations with intended voting, confirming the proposition that these participatory forms of learning are effective in fostering political engagement. Engaging in a civic project only shows a weak positive link. The reason why course content, open climate and student influence cannot explain the link of tracking, despite being strong predictors, is most likely that the tracks vary too little in these conditions, as suggested previously.

When we add social composition to the analysis, the coefficient of vocational colleges changes from being highly significant and negative to insignificant (compare Model 2 to Model 5), which is a significant reduction as shown by the less than 50% overlap in the confidence intervals. This result suggests that the influence of tracking runs mainly through social composition. In other words, it is the differential peer environment that can explain the difference between academic and vocational colleges in intended voting. Social composition also shows a significant positive link itself (b = 0.13; t-ratio = 2.4; p < 0.05). In other words, the more privileged the background of your peers in school, the more you plan to vote in the future, whatever your own social background.

Introducing political efficacy to the analysis boosts the explained variance at both the individual and school level as it turns out to be the most powerful predictor of intended voting (with a t-ratio of 31.3). It also reduces the effect of vocational colleges, but only slightly with the coefficient changing from −0.18 to −0.12 and losing one level of significance. Thus, even though the tracks show appreciable differences in levels of political efficacy (as noted before), it is largely not because of these differences that tracking is related to intended voting. In any case, it is difficult to assess whether the effect of tracking runs through this mechanism because it cannot be ascertained with the present data source how political efficacy has developed over the life course. If political efficacy stabilised before the allocation to different tracks at age 15, it is not a consequence of this allocation (as suggested in the literature) but possibly one of its causes. If so, it would contribute to the idea that the influence of tracking merely represents a selection effect.

When all the variables tapping the proposed mechanisms are included in the analysis (Model 7), the link of vocational colleges becomes even weaker than it was in Model 5. As one might expect, this model also performs best in terms of explained variance. School social composition retains its significant and positive link with intended voting. In other words, the link of tracking can wholly be explained by the variables representing the different mechanisms. This link can mainly be accounted for by social composition and to some extent by political efficacy as well.

8. Discussion

The current paper has shown that tracking is related to young people’s political engagement in France. Using the type of college attended as an indicator of track, we found that students in academic colleges expressed significantly stronger intentions to vote than those in mixed or vocational ones. This difference became smaller but did not disappear when we controlled for a range of conditions tapping family background and individual characteristics, suggesting that the effect of tracking is not only due to students from more endowed backgrounds or with other specific properties entering the academic track (i.e. a selection effect). Testing four mechanisms through which tracking has been postulated to influence political engagement, we found that school social composition, as one of these mechanisms, was key to explaining the relation of tracking to voting intentions: once this variable was included in the analysis, the difference between academic and vocational colleges in intended voting levels disappeared.

At the same time we showed that curriculum, pedagogy and political efficacy as the other suggested conduits of the effect of tracking, could not explain the difference between the tracks. These conditions were strongly related to intention to vote, however. The reason why they nonetheless could not explain the differences between the tracks in intended voting, we proposed, is that tracks vary too little in these conditions. Descriptive analyses indeed showed that differences between the tracks, although significant, were small, particularly on curriculum and pedagogy, and sometimes not in the expected direction, with students in the vocational colleges, for instance, reporting higher levels of student influence in school matters. The latter suggests that schools and teachers in France make a consistent effort to foster citizenship qualities among vocational students by offering a variety of relevant learning opportunities. This contradicts studies that find teachers in the vocational track to be merely fostering discipline and compliance and to deny students the opportunity to have a say in things (e.g. Whitty 1985; Ichilov 2003; Ten Dam and Volman 2003; Nieuwelink 2016).

Our findings thus show that even in a national context with a strong tradition of curricular uniformity compared to other western countries and where, as our analyses provisionally suggest, there are no major differences between tracks in pedagogical approaches, the academic and vocational track still have a marked gap in political engagement because of their differential social make-up. As the latter results from the (unintended) social sorting effects of track allocation on the basis of prior achievement, the implication is that equality of citizenship outcomes will likely be enhanced if this form of selection is abolished. Aiming for a complete abandonment of such selection is a tall order, however, given the prevalence of the idea that selection by ability is necessary to optimise the development of knowledge and skills and to prepare young people for jobs requiring different skills levels (van de Werfhorst 2014). Indeed, all OECD states practice some form of ability selection to regulate entry to different tracks at the upper secondary stage, even those with comprehensive institutions combining different tracks within the same college, such as the United States, Norway and Sweden (Green and Pensiero 2017).

Instead, it might be more productive to focus on curriculum and pedagogy and ensure that students in the vocational track experience at least as many, if not more, relevant learning opportunities as those in academic tracks. We have three reasons to suggest this. First, our own findings are in agreement with those of other studies (see Geboers et al. 2013 for a good review of these studies) in suggesting that both the content of citizenship education and the learning through participation pedagogies are effective in fostering political engagement. It is therefore likely that differences between tracks in intended voting would have been much larger in France if there had been less uniformity in these aspects of the formal and taught curriculum. Second, a number of studies drawing on various national contexts have shown that students from disadvantaged backgrounds benefit more from citizenship education and participatory pedagogies than students from privileged families in terms of becoming more politically confident and engaged (Gainous and Martens 2012; Campbell 2008; Sohl and Arensmeier 2015; Neundorf, Niemi, and Smets 2016; Hoskins, Janmaat, and Melis 2017). As such students are concentrated in vocational tracks, offering such learning opportunities in vocational education is likely to be particularly effective, and more effective than in academic pathways. Third, although these opportunities thus enable low SES students to (partially) catch up with their middle-class peers in political engagement, equal provision of such learning sources across tracks (as in France) is unlikely to completely neutralise the inequality enhancing effect of social composition. After all, in that case we should have seen no difference in intended voting across tracks after controlling for the civic learning opportunities and family background variables. A case could therefore possibly be made to apply positive discrimination for the vocational tracks and provide more relevant learning opportunities in such tracks as a means to achieve greater equality in political participation.

We do have to highlight three important limitations of the present study. First, and most importantly, the cross-sectional nature of the dataset made it impossible for us to control for political engagement prior to track enrolment and thereby to assess whether the effect of tracking on political engagement does not simply reflect selection into academic tracks of those with higher engagement levels from the start (i.e. a selection effect). Because of this limitation we have consciously avoided using terminology suggesting causation (such as effect, impact or influence) when reporting the findings of our analyses. However, it is not the case that selection effects cannot be addressed at all with cross-sectional data. We have been able to at least partially address them by including as many variables as we could relate to family background characteristics, as these are important determinants of both selection into different tracks and political engagement. We cannot be sure whether we have taken on board all conditions capturing selection effects. Yet, we are likely to have included most of them. After all, we were able to fully account for the statistical effect of tracking with all the variables representing the proposed mechanisms of the influence of tracking included in the model. If there had been a residual effect of tracking after the inclusion of all variables, then we would have had a reason to believe we had omitted some condition producing a selection effect. Moreover, our results tally with those of Hoskins and Janmaat (2016), who could control for prior political engagement and nonetheless found an independent effect of tracking. Interestingly, the coefficients of the different tracks hardly changed when they entered prior engagement in their models on intended voting and protest activities (ibid. pp. 81–82).

Secondly, the data only included respondents studying for the baccalaureate, academic, technological or vocational. Thus, it has no records of those in vocational colleges studying for the CAP, the less prestigious two-year vocational qualification that does not give access to university. Some 17% of students in vocational colleges (and about 6% of all students in upper secondary education) are in this pathway (MENJ 2018), which is a non-negligible number. The percentage of students from working-class backgrounds in this pathway is higher than in all other tracks (COS 2019), increasing the likelihood that political engagement is lowest among the students in the CAP. Our study is therefore likely to have underestimated the actual track differences in the student population.

Thirdly, we acknowledge that voting intentions, as the outcome of interest in this study, are only one facet of political engagement. Using the same survey data as the present study and looking beyond voting, a recent study found that students in vocational tracks think it more likely they will become a member of a political party, do voluntary work for candidates or political parties, and stand as a candidate in elections than those in academic tracks (Cnesco 2018). Similarly, Hoskins and Janmaat (2019) found illegal forms of protest, such as blocking traffic, occupying buildings and spraying graffiti, to be more popular among the former. This shows that cross-track patterns in voting cannot be generalised to other forms of engagement. On the other hand, the aforementioned report also found that the level of total disengagement (in terms of not seeing oneself taking part in any political activity) is higher among vocationally trained students (10%) than among academically educated ones (4%). Future research could delve deeper into these differences and explore qualitatively why vocationally trained students prefer different forms of participation or why many of them abstain from participation altogether.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Jan Germen Janmaat

Jan Germen Janmaat is Professor of Political Socialization at UCL Institute of Education. His research focuses on the role of education, broadly conceived, in fostering civic values. His latest book is Education, Democracy and Inequality: Political Engagement and Citizenship Education in Europe.

Nathalie Mons

Nathalie Mons is a professor of sociology specialized in education public policies. She devotes her research to education policy analysis from an international comparative perspective. She holds the national academic chair dedicated to the assessment of educational policies at Conservatoire national des arts et métiers (CNAM). She is the founder and Managing Director of CNESCO (the French national council for school system assessment).

Notes

1. We use the term ‘college’ to refer to schools specifically providing upper secondary education. It should thus not be confused with the French collège, which offers lower secondary education.

2. Ethical approval for this data collection was given by the ethics committee of CNESCO.

3. The education level of the mother or the father only was used in case of single parent families.

4. We give the word effect in quotation marks to highlight that we are not speaking of a causal relation but of a statistical one.

Appendix.

Composition of five scales used in the analyses

Voting intentions:

‘There are different ways to engage in public life. To what degree do you think you will take part in the following activities when you are an adult?’

• vote in local elections (municipal/departmental/regional)

• vote in national elections (presidential and parliamentary)

• vote in European elections

Categories: 1 = certainly not; 2 = probably not; 3 = probably; 4 = certainly.

Open climate:

‘According to you, during the course of EMC [civic education] … ’

• students ask questions regarding current political affairs

• students express their opinion in class even when the majority of students do not agree

• the teacher encourages students to make up their own minds about issues

• the teacher encourages students to voice their opinions

• the teacher encourages students with different opinions to engage in debate

• the teacher presents different points of view on political and social issues

Categories: 1 = never or almost never; 2 = sometimes; 3 = often; 4 = always or almost always

Student influence:

‘In your class, the views of student are taken into account regarding … ’

• materials of teaching and learning (textbooks, handbooks.);

• teaching content (when there is a choice of subject matter);

• the content of excursions;

• the organisation of excursions;

• pedagogy (work in groups/individual, research, debate, time devoted to a particular task);

• rules and regulations

• assessment

Categories: 1 = not at all; 2 = mostly not; 3 = mostly; 4 = completely

Political efficacy:

‘Regarding your attitude towards politics, do you agree with the following statements?’

• I know more about politics than most people my age

• When political issues or problems are being discussed, I usually have something to say

• I am able to understand most political issues easily

• I have political opinions that are worth listening to

• I feel capable of participating in politics

• I have a good understanding of political issues relating to France

Categories: 1 = strongly disagree; 2 = mostly disagree; 3 = mostly agree; 4 = strongly agree