The relationship between school policy for citizenship education and the citizenship knowledge of adolescents in the Netherlands

(1)

The Relationship between School Policy for Citizenship Education and the Citizenship Knowledge of Adolescents in the Netherlands

Master thesis

H.L. van Gulick (10769137) Master Onderwijskunde University of Amsterdam

Thesis supervisor: dr. M. Coopmans Amsterdam, july 2018

(2)

Abstract

Dutch schools are stimulated by the government to pay attention to citizenship education, without a notion of the extent to which school policies on citizenship education are related to the citizenship competencies of students. The current study will explore the relationship between citizenship knowledge of students and the school’s perceived importance of

citizenship education, insight in the citizenship competencies of students and the vision, goals and agreements on citizenship education by making use of data from the Understanding the Effect of Schools on students’ Citizenship project (ESC), a large-scale data collection that took place in 2016, collecting information on citizenship education and citizenship outcomes amongst 5301 students in over 80 secondary schools in the Netherlands. Extensive multilevel models show that a relationship between school policy for citizenship education and

citizenship knowledge of adolescent students in the Netherlands was virtually non-existent. As it is still possible that school policy has an indirect effect on citizenship outcomes, it is recommended that the school policy influence on the processes and climate in classrooms is taken into account in future research. A recommendation to the educational field is to make sure school policies are actually put into practice, as merely formulating a vision and learning goals and prioritizing citizenship education is not enough to effectively make a difference in citizenship competencies of students.

Keywords: citizenship education, citizenship knowledge, school policy

The Relationship between School Policy for Citizenship Education and the Citizenship Knowledge of Adolescents in the Netherlands

In recent years citizenship education has received increased attention all over the globe by policy makers and researchers alike (Ten Dam, Geijsel, Reumerman, & Ledoux, 2011). A reason for this is a growing dissatisfaction about the decrease of social cohesion in an increasingly individualistic and diverse society (Ministerie van Onderwijs, Cultuur en Wetenschap, 2005). On top of this, concerns for the civic participation and political involvement of young people fuel the particular attention paid to citizenship education (Biesta, 2008). The increased individualism of our society is reflected in the modern way of raising children, which is focused on individual goals, talents, career and identity (De Winter, 2011). The emancipation of the individual instead of the general interest of society has been the main focus of the upbringing of our children. In Western societies, the general interest would be the nurturing of the democracy (De Winter, 2011). This general interest of democracy already

(3)

provides a framework for citizenship education by the democratic values that are embedded in the institutional practices of our Western society (White, 1999). When citizens are not knowledgeable of and engaged in the democratic society, the democratic structures will collapse (De Winter, 2011). A democratic society cannot be preserved or advanced if the citizens of that society are passive and politically inactive (Levinson, 1999). Therefore, according to Levinson (1999), students should be equipped with the competencies needed to sustain the democratic society they live in through socialization, which is defined as ‘the process in which somebody learns the values, norms and cultural characteristics of a group’ (Eidhof, 2016, p. 257). A part of this socialization of young people is educating students to be active and contributing democratic citizens, which has recently been revitalized as citizenship education (Biesta, 2008). The aim of citizenship education is accordingly formulated by Schuitema, Veugelers, Rijlaarsdam and Ten Dam (2009) as ‘to prepare students for participation in a democratic and multicultural society’. What this participation entails can be found in the definition of active citizenship according to the Dutch Ministry of Education, Culture and Science (2005): ‘the willingness and the ability to be a part of a community and to actively contribute to it’.

Nurturing democracy does not just entail having knowledge on the constitution, politics, human rights et cetera, but also acting in a democratic manner (De Winter, 2011 ; Ten Dam & Volman, 2007). Gutmann and Thompson (2004) propose that a modern pluralistic, democratic society often entails moral disagreements. Acting in a democratic way involves being able to solve these conflicts in a peaceful manner using dialogue instead of violence. It also requires the equal treatment of others in a pluriform society and the disapproval of discrimination. In a democratic society, citizens of different religions, cultural background and political views are able to live together in peacefulness, freedom and with equal rights. To be able to act in a democratic and socially competent way, knowledge of the world is necessary (Ten Dam & Volman, 2007). This includes knowledge of the self, knowledge of social rules and knowledge of society as a whole. This knowledge is needed in order to make an active contribution to society, and in that way, nurture the democracy (Ten Dam et al., 2011). Social knowledge is relatively independent of social skills and behavior (Ten Dam & Volman, 2007), and one could argue that knowledge on how to act in a socially accepted, democratic way is needed to develop the skills and behavior to put the knowledge to action.

Education is seen as the solution to the decreasing social cohesion in an individualistic society (Dijkstra, Ten Dam, Hooghoff, & Peschar, 2010). As schools are primarily knowledge institutions, they could mainly contribute to the social knowledge needed to nurture democracy.

(4)

Indeed, controlled for individual differences, schools have a small influence on citizenship knowledge compared to almost no difference to citizenship skills and behavior in the study by Isac, Maslowski, Creemers, and van der Werf (2014). In order to improve on the citizenship knowledge of students, it is important to identify effective components of citizenship education. Previous research has shown that schools can make a difference to students’ citizenship outcomes after controlling for individual characteristics such as socioeconomic status (Isac et al., 2014; Isac, Maslowski, & van der Werf, 2011; Dijkstra, Geijsel, Ledoux, van der Veen, & ten Dam, 2015). However, previously studied characteristics of citizenship practices in schools, such as the formulation of a vision and goals, did not sufficiently account for the differences among schools (Dijkstra et al., 2015). Thus, research has shown that there are differences in citizenship competencies of students among different schools, but we do not know how schools can make a difference. Dutch schools are stimulated by the government to pay attention to citizenship education (Ministerie van Onderwijs, Cultuur en Wetenschap, 2005). School policies around citizenship education are formulated accordingly, without a notion of the extent to which school policies are related to citizenship knowledge of students. It is very well possible that schools only make a difference through the general atmosphere and school or class climate (Geboers, Geijsel, Admiraal and ten Dam, 2013). In the daily interactions between the student and school professionals and among students themselves socialization can already take place (Dijkstra et al., 2010). The socialization function of education has therefore always been present in the hidden curriculum (Veugelers, 2010), and prioritizing it in school policies may not even be necessary. Indeed, many studies on citizenship education have shown that school climate is an important factor (Geboers et al., 2013). Whether giving specific attention to citizenship education in school policies has an effect on the socialization of students is still unclear.

The current study will fill this gap in research by exploring to what extent the prioritization of citizenship education by schools is positively related to the citizenship knowledge of students. This study identifies three subcomponents of school policy for citizenship education, namely the perceived importance of citizenship education, insight in the citizenship competencies of students and the vision, goals and agreements on citizenship education. formulation of vision and goals. This study will explore to what extent these components are effective in improving citizenship outcomes in Dutch secondary school students. Accordingly, the main research question of the current study is: ‘What is the relationship between school policy on citizenship education and the citizenship knowledge of adolescents in the Netherlands?’.

(5)

Theoretical model of hypotheses

Figure 1 shows the hypotheses in a theoretical model, listing variables for the citizenship education policy as well as the control variables on individual and on school level. It is expected that students have more citizenship knowledge if their schools deem citizenship to be more important, have formulated a clear vision and learning goals, have agreements on how the goals can be achieved and have insight in the citizenship competencies of their students and adjust their citizenship education policy accordingly. In other words, it is expected that the effort a school puts into citizenship education will pay off and result in more citizenship knowledge of the school’s students.

In addition, an interaction effect between a school’s citizenship education policy and the frequency with which students participate in cultural activities outside school is expected to exist. It is expected that the relationship between school policy on citizenship education and citizenship knowledge is stronger for students who participate less frequently in cultural activities. Excursions to a museum, for example, are often part of the citizenship educational program of secondary schools in the Netherlands (Inspectie van Onderwijs (2016). These activities will be more impactful for students who do not already participate in these activities at home. Studies have shown that participating in cultural activities such as visiting museums and theatre shows is beneficial for the academic success of students (Dumais, 2002; De Graaf, De Graaf & Kraaykamp, 2000). It is likely that a school’s effort toward citizenship education is mainly beneficial for students who participate in less cultural activities from home. Students are expected to gain more from a school excursion to a museum if they have never gone to a museum with their parents. Van de Werfhorst and Mijs (2010) similarly found that students with a more disadvantaged background gain more from schooling than students from an advantaged home environment. In this study, a disadvantaged background implies, for example, a low socioeconomic status and thus having parents with a lower educational level who are unable to help with schoolwork. The goal of citizenship education, however, is not learning academic subjects, but gaining knowledge about the world to become a contributing, democratic citizen (Ten Dam & Volman, 2007; Ten Dam et al., 2011). This knowledge can partially be gained through cultural activities, such as visiting a museum. In the context of citizenship education, students with a ‘disadvantaged’ background are students who do not participate in cultural activities outside of school and have, as a result, less knowledge about the world to start with.

School policy on citizenship education is expected to have more impact on students with certain characteristics (Isac et al., 2014). These characteristics are expected to influence

(6)

the relationship between school policy and citizenship knowledge and will therefore be controlled for. Students with a high socioeconomic status and no migration background are expected to have more citizenship knowledge (Isac et al, 2014). As citizenship education is

Figure 1. Theoretical model on the relationship between citizenship education policy and citizenship knowledge, controlled for student characteristics.

Interaction

- School policy x Cultural activities

Control variables Student characteristics

- Socioeconomic status (SES) - Religion

- Cultural capital (number of books at home, cultural activities and news reading) - Age

- Gender

- Migration background - Educational level School characteristics

- Average cultural capital of student population

- Average educational level of the student population

- Average migration background of student population

Citizenship education policy

- Perceived importance - Vision, goals and

agreements

- Insight in citizenship competencies

(7)

value-laden, religion could also play a role in how citizenship education is perceived, although it is unknown in what way.

School characteristics that will be controlled for are the school averages of the individual control variables that measure socioeconomic status, migration background and cultural capital. Previous research has shown that the average socioeconomic status of peers at school has an impact on academic achievement (Van Ewijk & Sleegers, 2010). Due to the ‘little fish, big pond’ effect, students who have a lot of peers with a higher socioeconomic status are likely to underachieve. Whenever the average abilities of the peers is higher than your own, you are more likely to develop low self-concept and underachieve (Mash & Parker, 1984). Especially for citizenship competencies that are acquired partially through friends and family (Dijkstra, De la Motte and Eilard, 2014), it can be expected that the characteristics of the peer group influence citizenship competencies. For these underachieving students, citizenship education might be of more influence than for other students. Additionally, it is possible that denomination, urbanity and school size influence the amount of attention a school pays to citizenship education. It is expected that religious schools pay attention to religious education instead of citizenship education, that urban schools deem citizenship education to be of more importance than rural schools and that large schools also pay more attention to citizenship education.

Citizenship competencies: citizenship knowledge

There has been previous research that showed that schools make a positive difference in the citizenship competencies of adolescents (Isac et al., 2014; Isac et al., 2011; Dijkstra et al., 2015). Even though differences in citizenship competencies among adolescents are mainly explained by individual differences (Isac et al., 2014; Dijkstra et al., 2015), such as socioeconomic status, there are also differences in citizenship competencies among schools. Isac and colleagues (2014) made use of the International Citizenship Competencies Study dataset of 2009, which includes 4,078 schools in 31 countries. Overall, they found that schools, including Dutch schools, mainly account for differences in civic knowledge compared to other, non-cognitive citizenship competencies. Even after controlling for

individual characteristics, fifteen percent of the variance in civic knowledge was explained at the school level. In contrast, there were hardly any systematic differences among schools for the non-cognitive citizenship competences of attitude and intended participation.

Unexplained variance on the individual level applied mainly to non-cognitive citizenship outcomes. Dijkstra and colleagues (2015) found that schools account for an even higher

(8)

percentage in differences in citizenship knowledge of Dutch primary school students. The effects of school variables on citizenship knowledge were 21 percent, which is a relatively large effect size. It is possible that primary schools have a higher influence on the citizenship knowledge of their students than schools for secondary education. Nonetheless, as there has been an increase of attention for citizenship education in the Netherlands, it is plausible that secondary schools currently contribute more to citizenship competencies of their students than in Isac et al. (2014), for which a 2009 dataset was used.

School policy on citizenship education

Research on the effect of school policy on citizenship education is limited. Dijkstra et al. (2015) conducted a study on the effect of, inter alia, citizenship policy on citizenship competencies of primary school students. In this study, citizenship policy consisted of four components: whether or not the school had developed a vision on citizenship education, the frequency with which attention was paid to various aspects of citizenship, the extent to which the school had formulated learning objectives for these aspects, and whether or not the school had experienced difficulties in implementing citizenship education. Overall, the effect of citizenship policy was small or insignificant, and differences in citizenship competencies were mainly explained by student characteristics. The significant variables were the extent to which the school had a vision for citizenship education and specific objectives for learning about other cultures, and emphasis on learning social skills. In contrast, Isac et al. (2014) did not find any significant results for the variable of the amount of attention secondary schools pay to citizenship education. However, it is still possible that school policy for citizenship education accounts for the variance in civic knowledge on school level by positively influencing classroom practices. It is important to conduct more research to determine whether or not school policies have any impact on citizenship education beyond primary school. In conclusion, due to limited research (Dijkstra et al., 2015) there is still unclarity about the influence school policy on the citizenship knowledge of adolescents.

In the present study, the vision and goals for citizenship education of the school as reported by the school leader will be taken into account, as this was significant in Dijkstra et al. (2015). In addition, two measures for the amount of attention schools pay to citizenship education will be used: the perceived importance of citizenship education and whether or not the school has insight in the citizenship competencies of their students and adapt the school policy accordingly. It is likely that the more schools pay attention to citizenship education, the more it will be interwoven in their curriculum, and the more students’ citizenship

(9)

competencies will increase. Evaluating student civic outcomes and adjusting the school policy on citizenship education accordingly is also expected to have a positive impact on citizenship knowledge of students (Dijkstra et al., 2014). There has not been previous

research to test these hypotheses, which is why the current study will provide new insights in the effect of school policy on the citizenship of adolescents.

Socialization processes in schools

The current study investigates the relationship between school policy on citizenship education and the citizenship knowledge of students. School policy is expected to be related to citizenship knowledge based on the assumption that school policy leads to processes in the school in which learning takes place. In order for the citizenship knowledge of students to increase with the formulation of a school policy, learning needs to take place in between. School policy could, for example, be geared toward news discussion in classrooms and stimulate teachers to engage with their students on current problems in society. Theoretical ground for the different types of school processes that can take place while practicing citizenship education can be found in the works of Émile Durkheim (1858-1917) and John Dewey (1859-1952). Although they both believed that moral education would bind the pluralistic and fragmented modern, industrialized society together, they differed in their view on the role of education in the moral development of young people. Roughly put, Durkheim believed society could transfer its current morals to the next generation through the means of education, while Dewey saw schools as a place where students could experimentally develop morals in interaction with others (Dill, 2007). Figure 2 shows a simplified model for the interaction between school and society according to both Dewey and Durkheim. The school is situated in society as a microcosm for society, which socializes young people to participate outside of school (Dill, 2007). The school has, Durkheim and Dewey both agree, a social function through which cultural and societal transmission can take place.

Durkheim was mostly focused on the idea of bridging between schools and the outside world (Putnam, 2000). He considered the transmission of the values of society to young people to be an important function of education (Dill, 2007). It was not possible for schools to transmit the values of society if they merely function as communities on their own islands without relating much to other groups or society as a whole (Putnam, 2000). This would result in a lot of bonding, but in little bridging. Durkheim did not believe that society could merely be the sum of individuals (Durkheim, 1925/2012). Individuals’ self-interest lack morality, thus the collective interest, if it only consists of sum of self-interests, would also be amoral. In the

(10)

Figure 2. The interaction between school and society according to Dewey and Durkheim.

context of citizenship knowledge, students would develop knowledge on how to make their way through society that leads to as many benefits for themselves as possible. Functioning in society would consist of knowing how to survive and gain benefits as an individual, instead of how to progress society as a whole. Therefore, society needs own being, the conscience collective, which is a social character that is distinct from the individuals society consists of. To act morally is to act in terms of a collective interest that is beyond individual goals. This conscience collective would bring upon social cohesion needed to sustain society (Dill, 2007). The relationship between school and society according to Durkheim would be that the conscience collective would be educated in schools as a moral authority. The individual is shaped to conform to the moral ideal of society in the social context of school. In terms of teaching students citizenship knowledge in our modern society, Durkheim would suggest to teach all students the same values of a democracy in order to establish a common ground. This would result in the continuation of democratic society and maintaining the status quo. For Dewey, merely maintaining the status quo would not be enough. As Dewey believed there was no real distinction between the individual and society, it is made up of individuals, the moral development of the individual would bring upon the moral progression

(11)

of society as a whole (Dill, 2007). The relationship between the school and society is that the moral progression of the individual, which takes place in the school community, becomes a part of society. This way, the individual can add something to society rather than mere living according to its rules. In the context of teaching students citizenship knowledge this implies that students should not only be taught existing knowledge, but also how to develop their own opinions in a social context. Citizenship knowledge is more than knowledge on the constitution and human rights, it is also knowing how to conduct oneself in a democratic manner (Ten Dam & Volman, 2007). In addition to behaving according to set values and norms learned in school, students should be able to flexibly respond to situations and change perspective in order to put themselves in the position of the other. Knowledge on acting democratically entails, for example, how to solve conflicts among students with different political, religious or moral views.

School effectiveness in citizenship education

A tangible way of displaying the relationship between school policy, school processes and citizenship knowledge is to frame it in a school effectiveness model consisting of an input, throughput and output. Dijkstra and colleagues (2014) have done exactly this and propose that the quality of education becomes visible in the outcomes, even in the social domain when outcomes are partially dependent on factors outside of school. They deem it important to determine the net effect of schools on social outcomes of students by finding out the effect that can be attributed to the school. This calls for a model in which is controlled for student characteristics and the context in which a student grows up, which can be seen as the input. The effect of the student characteristics is in this case even expected to be larger than the effect of the school. A downsized school effectiveness model for the social domain copied from Dijkstra et al. (2014) is shown in Figure 2. This framework is a simplified version of reality because it only includes the components that are relevant for the current research. The processes that take place at school, or the throughput, are fueled by the input and partially produce the output. Meanwhile the input, the characteristics of the students, now only influences the processes at school but it can also directly influence the outcomes.

Examples of school processes in the throughput component that Dijkstra et al. (2014) mention are the ‘school ethos’, in other words the school’s vision, values and practices, and the school’s focused approach. The focuses approach consists of the formulation of clear goals, the coordination and coherency of the approach and the evaluation of the student outcomes.

(12)

Figure 3. Simplified school effectiveness model for citizenship knowledge. Adjusted from Dijkstra, De la Motte & Eilard (2014). Social outcomes of education: Concept and

measurement. In: A.B. Dijkstra & P.I. de la Motte (Eds.), Social outcomes of education: The assessment of social outcomes and school improvement through school inspections. Amsterdam University Press.

The current study mainly includes the formulation of a vision on citizenship

education, goals and the evaluation of student outcomes as important school processes that are important to bring about the desired output. In Dijkstra et al. (2015) a school’s vision and goals was related to citizenship outcomes, thus it will be taken into account in this study. The evaluation of student outcomes will be studied in the current research, as it has not previously been researched in the context of citizenship education, even though it is considered an important part of school effectiveness in line with the learning goals (Reynolds et al., 2014).

Input Student (background) characteristics Throughput School processes

• ‘School ethos’: vision, values, practices • Focused approach o Clear goals o Coordination & coherent approach o Evaluation of student outcomes Output Citizenship knowledge

(13)

Method

Participants

This study made use of data from the Understanding the Effect of Schools on students’ Citizenship project (ESC), a large-scale data collection that took place in 2016, collecting information on citizenship education and citizenship outcomes amongst over 80 secondary schools in the Netherlands. Data was collected using stratified sampling on the distribution of educational level of the Dutch population of grade 9 students. Students as well as school leaders, team leaders and teachers filled in a questionnaire for this study. For the current study, the questionnaires of students and school leaders will be used. At the time the ESC project took place, the International Civic and Citizenship Education Study (ICCS) was also in progress. This lead to the exclusion of 403 schools that were included in ICCS. Additionally, 44 schools were also included in ICCS and were excluded from the current study because they do not have a ninth grade. Table 1 shows the distribution of educational level of schools with a ninth grade and the samples in ICCS and the current study. For the current study, a total of 300 schools were selected, with 200 schools as a reserve.

Table 1 Stratified sample based on the educational level of schools.

Educational level n population n in ICCS n current study n reserve current study

Vocational 524 122 45 90

General 512 220 43 86

Mixed 139 61 12 24

Total 1175 403 100 200

The response rate was 18%, indicating 54 of the 300 selected schools were willing to participate. In addition to these 54 schools, 28 schools were recruited through acquaintance with the researchers. This brings the total of schools who participated in the current study to 82. Table 2 shows the characteristic of participating schools. From this table, it is important to note that vocational schools are slightly underrepresented.

(14)

Table 2 Characteristics of participating schools (n=82). Characteristic Urbanisation Not urban 10(12%) Limitedly urban 10(12%) Moderately urban 18(22%) Strongly urban 21(26%)

Very strongly urban 23(28%)

Educational levels Vocational 30(37%) General 40(49%) Mixed 12(15%) Denomination Public 2205(41%) Christian 1935(36%) Mixed denomination 655(12%) Neutral denomination 557(10%)

In each school all students of three grade 9 classes were selected, which led to a total number of 6317 students who were selected to participate. A total of 5301 Dutch students (51.5% women, 87.2% 14 or 15 years old) eventually filled in the questionnaire. On top of this, 51 school leaders, 64 team leaders, and 672 teachers from the same schools filled in a questionnaire. The characteristics of the participants are presented in Table 3.

Measures

Citizenship knowledge.

Citizenship knowledge will be measured with a questionnaire based on the four social tasks citizenship consists of according to Ten Dam et al. (2011). The total scale consists of 27 multiple-choice questions with three possible answers, only one of which is correct. The percentage of correctly answered questions is the student’s score on citizenship knowledge. The first social task is dealing with differences, of which there are four items in the scale. An example question for dealing with differences is: ‘Statements such as “Fat people are

(15)

Table 3 Characteristics of participating students (n=5301)

Characteristic

Male 2572 (48.5%)

Female 2729 (51.5%)

Age

13 years old or younger 14 years old

15 years old 16 years old

17 years old or older

86 (1.6%) 1898 (35.8%) 2727 (51.4%) 546 (10.3%) 44 (0.8%) Socioeconomic status Low 473 (8.9%) Average 2758 (52.0%) High 2069 (39.0%) Religiousness Not religious 3352 (63.5%) A little religious 832 (15.8%) Religious 1092 (20.7%) Religiousness parents Not religious 2855 (54.1%) A little religious 984 (18.7%) Religious 1436 (27.2%) Migration status No migration background 4059 (76.6%)

Western migration background 312 (5.9%)

Non-Western migration background 928 (17.5%)

Seven items in the total scale. An example item for dealing with conflicts is: ‘Your friends are bullying the friendly neighbor. What is the best thing to do?’. The third social task concerns democratic behavior, and has eight items in the citizenship knowledge scale. An example question for knowledge on democratic behavior is: ‘That there are different political parties is important because…’. The last social task is societal responsibility, and this scale

(16)

consists of another eight items. An example of a question on societal responsibility is: ‘Why is it illegal to smoke on the bus or in the train?’.

School policy on citizenship education.

Vision, agreements and goals. Aspects of school policy on citizenship education, such as whether or not a school works on a vision and goals, evaluate their citizenship education and discuss citizenship in staff meetings, is measured with ten statements on a 5-point Likert scale (1 = does not fit at all and 5 = fits well). With these statements school leaders are not only asked whether or not specific goals for citizenship education are formulated, but also to what extent the school systematically works toward those goals. An example of an item is: ‘The development of citizenship competencies of students is being systematically followed.’. The minimum score of this scale is 10 and the maximum score 50. The reliability measured with Cronbach’s alpha is .899.

Insight in citizenship competencies. Whether or not a school has insight in the citizenship competencies of its students and uses this information to adjust its policy is measured with three statements on a 5-point Likert scale (1 = does not fit at all and 5 = fits well). An example of a statement is: ‘The school has good insight in the views, attitudes and behavior of students around citizenship.’. The minimum score on this scale is 3 and the maximum score 15. The reliability measured with Cronbach’s alpha is .767.

Perceived importance. To measure the extent to which schools perceive citizenship education to be important, two statements on a 5-point Likert scale will be used (1 = does not fit at all and 5 = fits well). An example of a statement is: ‘Citizenship education is an

important topic at our school.’. The minimum score of this scale is 2 and the maximum score 10. The reliability measured with Cronbach’s alpha is .837.

Cultural activities. The scale for cultural activities outside of school consists of two items on a 5-point Likert scale. In the two questions, students are asked how frequently they visit a museum in their spare time and how frequently they go to a classical concert or theatre show in their spare time (1 = never and 5 = at least once a month). The minimum score of this scale is 2 and the maximum score is 10. The reliability of the two items is moderate at a Cronbach´s alpha of .607.

(17)

Control variables.

News reading. To measure news reading four items on a 4-point Likert scale (1 = (almost) never and 4 = very often) and one item on a 5-point Likert scale were used (1 = never and 5 = every day). This last item (‘Approximately how often do you read the

newspaper in you spare time?’) was originally from a different scale (Cultural activities) but more appropriate here, was added in order to improve the reliability. Because this item was measured on a different Likert scale the items were standardized. An example of the other four items is: ‘How often do you follow the news on the internet or social media?’. The minimum score of this scale is 5 and the maximum score is 25. Together, the five items have a reliability of .832. The variable of news reading was also aggregated to the school level, to create a variable for the average amount of news reading of peers.

Educational level. Educational level of the students is measured with seven categories of the Dutch educational system. For the sake of conciseness the three variations of

vocational education without a theoretical focus were combined into one category, and the two variations of pre-university education were combined. This leaves the four categories of vocational education, theoretical vocational education, general higher education and pre-university education. The variable of educational level was aggregated to the school level to form the average educational level of a school as a control variable.

Socioeconomic status. To measure socioeconomic status, students were asked the educational level of both of their parents. The parent with the highest educational level is used as the socioeconomic status of the student. Seven categories were used, ranging from ‘only primary education’ to ‘university’, with one category ‘I don’t know’.

Number of books at home. To measure the number of books a student has at home, they were asked to estimate the number in five categories ranging from ‘0 to 25’ to ‘more than 500’.

Migration status. To assess the migration status of the students, they were asked the birth country of both their parents. Students were coded to have a migration background when one of their parents was born outside of the Netherlands. Students with a migration background were sorted into two categories: Western and non-Western migration background using the classification of the Dutch Central Bureau of Statistics (Centraal Bureau voor Statistiek, 2018). The variable of migration status was aggregated to the school level to measure the average migration status of peers.

(18)

They could choose from three answer categories: yes, a little bit and no.

Gender. Gender was measured with two categories, students could either answer they were a boy or a girl.

Scale construction

The quality of the scales in the questionnaire was assessed using reliability and principal component analyses. To improve the reliability of the scales, items with an inter-item correlation smaller than .30 and inter-items that decreased Cronbach’s alpha were deleted one by one. After that, principal component analysis was conducted to examine whether the scales measured one construct. Principal components with an eigenvalue higher than one were examined. Additionally, items with a loading higher than .50 on a principal component were retained.

Principal component analysis showed that the school policy scale consists of two principal components with an eigenvalue higher than 1. Items with a loading higher than .50 on the first components reflect the vision, goals and agreements on citizenship education of the school. This component explains 54.64 percent of the total variance of the scale. The second component reflects the knowledge the school has on citizenship competencies of their students and whether or not they use this knowledge to alter their citizenship program or the behavior and attitudes of their students. This component explains another 15.99 percent of the variance. To increase the interpretability the two components were made into two separate scales. The vision, goals and agreements scale consists of six items with a Cronbach’s alpha of .899, and the insight in competencies scale consists of three items with a reliability of .767. More details on the scales can be found in Table 4.

After deleting items with an inter-item correlation lower than .30, the principal

component analysis on perceived importance of citizenship education shows two components with an eigenvalue larger than one. The first is easily interpretable as the perceived

importance of citizenship education, whereas the second component is not as easily interpretable as such. To increase the interpretability of the results, only the items with a loading higher than .50 on the first component were kept in the scale. This lead to a scale of only two items, nevertheless with a reasonable reliability of .837.

Scales were used for two of the control variables, namely whether or not students follow the news and the extent to which students participate in cultural activities outside of school. One item of the news scale was deleted due to a low inter-item correlation. In its

(19)

place, one item, originally from the cultural activities scale, but in hindsight more appropriate here, was added in order to improve the reliability. This item was measured on a different Likert scale, meaning the items had to be standardized. Together, the standardized items have a reliability of .832. Principal component analysis showed they all measure the same

construct. The three-item scale of cultural activities lost one item due to a low inter-item correlation. The remaining two items are skewed to the right, indicating that most students never or rarely go to classical music concerts, the theatre or museums. The two items have a reliability of .607.

Table 4 Descriptive statistics and reliability of the scales

Scale Number

of items* M SD Skewness Kurtosis Cronbach’s α

Vision, goals and

agreements 6 16.66 5.36 .02 -.44 .899

Insight in competencies 3 8.68 2.19 -.02 -.69 .767

Perceived importance 2 6.66 1.97 -.26 -.23 .837

Student news reading* 5 .00 3.83 .46 -.28 .823

Student cultural activities 2 3.33 1.54 1.25 1.29 .607

*Standardized

Procedure

Students filled in the questionnaire online during school hours. After finishing the

questionnaire, students could enter their email address to have a chance to win a gift card.

Analytical strategy

As there is no reason to assume the dropout of participants is not random, maximum likelihood estimation will be used in the multilevel models. On all knowledge scales, 104 students had missing values, which leaves 5248 valid cases. Missing values were likely caused by students being absent from school when the study took place. Of the 51 school leaders who filled in the questionnaire, only one had missing values on the measures for citizenship policy. An advantage of the maximum likelihood estimation method is that it is generally robust against violations of the assumptions.

(20)

will be estimated. Each model will build on the last by adding explanatory variables and an interaction, as specified in the hypotheses. Categorical variables will be added into the models as dummies, and all variables will be centered around their grand mean to improve the interpretation and make sure 0 is a possible value of all variables. The empty intercept-only model will not include explanatory variables. Each model estimates the parameters for individual predictors, as well as the total explained variance compared to the previous model. Each model will be compared to the previous model by comparing deviances with a

likelihood ratio test. If the deviance is significantly lower than the deviance of the previous model, the fit of the model improved due to the added parameters.

Results

In this study, the factors that determine citizenship knowledge are examined. In this section, the correlation analyses and multilevel analyses will be discussed.

Correlations among explanatory variables

The correlations among explanatory variables on the school level are shown in Table 5. From this table it becomes clear that the measures for school policy on citizenship

Table 5 Correlation matrix for school policy variables

1. 2. 3.

1. School policy: Vison, goals and agreements

2. School policy: Insight in competencies

.54**

3. School policy: Perceived importance

.74** .67**

* Correlation is significant at the 0.05 level (2-tailed) ** Correlation is significant at the 0.01 level (2-tailed)

education all highly correlate among one another. This is not surprising, as it could be expected that schools that perceive citizenship education to be more important are more likely to have a vision, goals and agreements on citizenship education. It is also likely that

(21)

those schools have more insight in the citizenship competencies of their students, because they pay more attention to citizenship. The correlation between insight in competencies and vision, goals and agreements on citizenship education can be explained in a similar way; schools with a vision, goals and agreements generally pay more attention to citizenship education and are therefore more aware of the citizenship competencies of their students. Correlations among control variables and between control variables and citizenship knowledge are shown in Appendix A.

Assumptions

The assumptions of linearity and homoscedasticity were tested with residual plots for fixed effects and a combination of random and fixed effects for all final models for the four types of citizenship knowledge. The residuals are scattered around the horizontal regression line in a fairly symmetric pattern, indicating the relationship between the explanatory variables and dependent variable is linear and the variance of the residuals is reasonably constant. A few data points in the final model for knowledge on democratic behavior and societal

responsibility have low standardized residuals and are potential outliers, but there is no substantial reason to remove them from the dataset. To check the normality of the distribution of the residuals a Q-Q plot and histograms were used. From this it can be concluded that the distribution is close to normal. The intraclass correlation of the empty model is .09, which indicates an independence of the residuals.

Multilevel modeling

Citizenship knowledge: dealing with differences.

The results of the multilevel modeling for the citizenship knowledge scale of dealing with differences are shown in Table 6. The empty model M0 has a significant random intercept

estimated at 0.74, t(3219)=9.48, p<.001. As for the random part, the within-group variance is larger than the between-group variance, indicating that the variance among individuals is larger than the variance among groups. The intraclass correlation as calculated from the null model is .173, indicating that the proportion of the total variance explained by the group-level variance is estimated to be relatively large in an educational context (Hox, Moerbeek, & van de Schoot, 2017). It was expected that schools could make a difference in citizenship

(22)

Table 6 Models for citizenship knowledge: dealing with differences. Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction

Fixed part Estimate, p Estimate, p Estimate, p Estimate, p Estimate, p

Intercept 0.74, p<.001 0.70, p<.001 0.70, p<.001 0.70, p<.001 0.70, p=<.001

School level

Citizenship policy: vision, agreements, goals

-0.00, p=.744 -0.00, p=.743 -0.00, p=.790 -0.00, p=.746

Citizenship policy: insight in competencies

-0.01, p=.575 -0.01, p=.571 -0.01, p=.590 -0.01, p=.575

Perceived importance of citizenship education

0.01, p=.721 0.01, p=.722 0.01, p=.742 0.01, p=.722

Control variables

Average school news reading 0.16, p<.001 0.16, p<.001 0.16, p<.001 0.16, p<.001 Average school level: vocational

theoretical2

-0.77, p<.001 -0.77, p<.001 -0.77, p<.001 -0.77, p<.001

Average school level: general2 _{-0.83, p<.001} _{-0.83, p<.001} _{-0.83, p<.001} _{-0.83, p<.001}

Average school level: pre-university2 _{-1.09, p<.001} _{-1.10, p<.001} _{-1.09, p<.001} _{-1.09, p<.001}

Average school Western migration3 _{1.83, p=.029} _{1.83, p=.030} _{1.78, p=.036} _{1.83, p=.030}

(23)

Table 6 (Continued). Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction Student level Control variables Gender -0.44 p<.001 -0.44, p<.001 -0.44, p<.001 -0.44, p<.001 Vocational theoretical education2 0.68, p<.001 0.68, p<.001 0.68, p<.001 0.68, p<.001 General education2 _{1.29, p<.001} _{1.29, p<.001} _{1.29, p<.001} _{1.29, p<.001} Pre-university education2 _{1.72, p<.001} _{1.72, p<.001} _1.71,p=<.001 _{1.72, p<.001} Low SES4 _{-.16, p=.020} _{-0.16, p=.020} _{-0.16, p=.020} _{-0.16, p=.020}

Number of books at home 0.03, p=.099 0.03, p=.099 0.03, p=.096 0.03, p=.099

Cultural activities 0.16, p=.002 0.03, p=.039 0.28, p=.049 0.03, p=.040

News reading 0.03, p<.001 0.03, p<.001 0.03, p<.001 0.30, p<.001

Non-western migration background3

0.03, p=.652 -0.03, p=.652 -0.03, p=.632 -0.03, p=.652

(24)

Table 6 (Continued). Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction Interaction

Vision, agreements, goals x Cultural activities -0.00, p=.897 Insight in competencies x Cultural activities 0.01, p=.263 Perceived importance x Cultural activities 0.00, p=.938

Random part Estimate Estimate Estimate Estimate Estimate

σ²e 1.36 1.19 1.19 1.19 1.19 σ²u0 0.28 0.02 0.02 0.02 0.02 Explained variance 0.264 0.000 0.000 0.000 Deviance 10,419.71 9,880.20 9,880.19 9,878.94 9,880.20 Δχ² 539.51** 0.01 1.26 0.00 df 3 22 23 23 23

(25)

attended, which does seem to be the case. Comparing the random intercept model to a linear model using generalized least squared, the random intercept model is a significantly better fit for the data, χ²(3)= 486.90, p<.001.

The first model M1 contains the explanatory variables on school level as well as the

level of the individual1. This model is able to explain 26.4 percent of the variance in

citizenship knowledge of dealing with differences. This is mainly due to the control variables, as none of the school policy variables were significantly related to citizenship knowledge on dealing with differences. Whether the school has a vision, agreements and goals on

citizenship education did not make a significant difference (-0.00, t(40)=-0.33, p=.744), and neither did the school’s insight in its students’ citizenship competencies (-0.01, t(40)=-0.57, p=.575) nor the school’s perceived importance of citizenship education (0.01, t(40)=0.36, p=.721). School policy for citizenship education does not explain the variance in knowledge on dealing with differences among groups. Overall, comparing the deviance of M1 to M0, the

likelihood ratio test is significant, χ²(22)= 539.51, p<.001. This shows that the model with explanatory variables is a better fit for the data than the empty model. Notably, only 0.02 of the variance between schools is left unexplained by the explanatory variables in the model. In the second model M2 the interaction between a school’s vision, agreements and

goals and the frequency with which its students on average participate in cultural activities outside school was added. This interaction was not significant (-0.00, t(3209)=-0.16, p=.871) indicating that the relationship between school policy and citizenship knowledge on dealing with differences is not different for students who frequently participate in cultural activities from students who do not. In the third model the interaction between a school’s insight in the citizenship competencies of its students and cultural activities is added, and was again not significant (0.01, t(3209)=1.11, p=.268). In the fourth model the interaction between a school’s perceived importance of citizenship education and cultural activities was also not significant (0.00, t(3209)=0.04, p=.965). None of the models that contain an interaction explain any more of the variance than the first model, and none improved the model fit.

Citizenship knowledge: dealing with conflicts.

The results of the multilevel modeling for the citizenship knowledge scale of dealing with conflicts are shown in Table 7. The empty model M0 has a significant random intercept

estimated at 0.79, t(3219)=10.79, p<.001. As with the last model of dealing with differences and the models still to come, the within-group variance is larger than the between-group variance. The intraclass correlation as calculated from the null model is .139, indicating that

(26)

the proportion of the total variance explained by the group-level variance is estimated to be relatively moderate in an educational context (Hox et al., 2017). The intraclass correlation for dealing with differences was higher, which indicates that students’ knowledge on how to deal with conflicts is less dependent on the school they attend than their knowledge on how to deal with differences.

In the model selection, the same procedure was followed as for citizenship knowledge on dealing with differences. The first model was able to explain 23.4 percent of the variance due to the control variables. Overall, comparing the deviance of M1 to M0, the likelihood ratio

test is significant, χ²(22)= 506.01, p<.001. The school policy variables were not related to citizenship knowledge on dealing with conflicts. Vision, agreements and goals on citizenship education did not make a significant difference (0.01, t(40)=0.72, p=.478), and neither did the school’s insight in its students’ citizenship competencies (-0.03, t(40)=-1.68, p=.101), nor the school’s perceived importance of citizenship education (0.02, t(40)=1.01, p=.320). Only a mere 0.01 of the variance between schools remains unexplained by the model.

In the upcoming models, none of the interactions were related to citizenship

knowledge on dealing with conflicts (M2: -0.00, t(40)=-0.24, p=.808; M3: 0.00, t(40)=0.40,

p=.688; M4: -0.00, t(40)=-0.71, p=.475). None of the models explained any more variance

than the first model or improved the model fit.

Citizenship knowledge: democratic behavior.

The results of the multilevel modeling for the citizenship knowledge scale of democratic behavior are shown in Table 8. The empty model M0 has a significant random intercept

estimated at .91, t(3219)=10.21, p<.001. The intraclass correlation as calculated from the null model is .074, indicating that the proportion of the total variance explained by the variance at school level is estimated to be low in an educational context (Hox et al., 2017). The students’ knowledge on democratic behavior is not highly dependent on the school they attend.

The same procedure as for the previous models was followed again in the model selection. The first model was able to explain 36.2 percent of the variance due to the control variables. The likelihood ratio test comparing the deviances of the first model to the empty model is significant, χ²(22)= 700.82, p<.001. The school policy variables were mostly not related to citizenship knowledge on democratic behavior. Vision, agreements and goals on citizenship education did not make a significant difference (-0.00, t(40)=-0.54, p=.592), and neither did the school’s perceived importance of citizenship education (0.03, t(40)=1.28,

(27)

Table 7 Models for citizenship knowledge: dealing with conflicts. Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction

Intercept 0.79, p<.001 0.74, p<.001 0.74, p<.001 0.74, p<.001 0.74, p<.001

School level

0.01, p=.478 0.01, p=.477 0.01, p=.464 0.00, p=.491

-0.03, p=.101 -0.03, p=.099 -0.03, p=.103 -0.03, p=.098

0.02, p=.320 0.02, p=.319 0.02, p=.326 0.02, p=.314

Control variables

theoretical2

-0.66, p<.001 -0.66, p<.001 -0.66, p<.001 -0.67, p<.001

(28)

Table 7 (Continued). Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction Student level Control variables Gender -0.53, p<.001 -0.53, p<.001 -0.53, p<.001 -0.53, p<.001 Vocational theoretical education2 0.64, p<.001 0.64, p<.001 0.64, p<.001 0.64, p<.001 General education2 _{1.18, p<.001} _{1.18, p<.001} _{1.18, p<.001} _{1.18, p<.001} Pre-university education2 _{1.54, p<.001} _{1.55, p<.001} _{1.54, p<.001} _{1.55, p<.001} Low SES4 _{-0.10, p=.187} _{-0.10, p=.189} _{-0.10, p=.186} _{-0.09, p=.191}

-0.10, p=.144 -0.10, p=.144 -0.10, p<.142 -0.10, p=.146

(29)

Table 7 (Continued). Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction Interaction

Vision, agreements, goals x Cultural activities -0.00, p=.808 Insight in competencies x Cultural activities 0.00, p=.688 Perceived importance x Cultural activities -0.00, p=.475

(30)

p=.209). The school’s insight in its students’ citizenship competencies was only significant in the second model for knowledge on democratic behavior (-0.03, t(40)=-2.04, p=.048). The relatively small negative estimate of -0.03 indicates that the more schools have insight in the citizenship competencies of their students and adjust their policy accordingly, the less citizenship knowledge on democratic behavior the students possess.

In the upcoming models, none of the interactions were related to citizenship knowledge on democratic behavior (M2: -0.00, t(40)=-1.01, p=.314; M3: 0.00, t(40)=0.84,

Citizenship knowledge: societal responsibility.

The results of the multilevel modeling for the citizenship knowledge scale of societal

responsibility are shown in Table 9. The empty model M0 has a significant random intercept

estimated at 0.89, t(3219)=12.40, p<.001. The intraclass correlation as calculated from the null model is .142, indicating that the proportion of the total variance explained by the variance at school level is estimated to be relatively moderate in an educational context (Hox et al., 2017). The students’ knowledge on democratic behavior is dependent on the school they attend.

The same procedure as for the previous models was followed again in the model selection. The first model was able to explain 24.4 percent of the variance due to the control variables. The likelihood ratio test comparing the deviances of the first model to the empty model is significant, χ²(22)= 554.72, p<.001. The school policy variables were, again, not related to citizenship knowledge on dealing with conflicts. Vision, agreements and goals on citizenship education did not make a significant difference (-0.00, t(40)=-0.24, p=.814), and neither did the school’s insight in its students’ citizenship competencies (-0.01, t(40)=-0.85, p=.401), nor the school’s perceived importance of citizenship education (0.02, t(40)=0.85, p=.403). In the upcoming models, none of the interactions were related to citizenship knowledge on dealing with conflicts (M2: 0.00, t(40)=0.49, p=.622; M3: 0.00, t(40)=-0.36,

(31)

Table 8 Models for citizenship knowledge: democratic behavior. Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction

Intercept 0.62, p<.001 0.86, p<.001 0.86, p<.001 0.86, p<.001 0.86, p<.001

School level

-0.00, p=.592 -0.00, p=.588 -0.00, p=.620 -0.00, p=.568

Citizenship policy: insight in competencies -0.04, p=.053 -0.03, p=.048 -0,04, p=.056 -0.04, p=.051 Perceived importance of citizenship

education

0.03, p=.209 0.03, p=.206 0.03, p=.216 0.03, p=.203

Control variables

theoretical2

-0.69, p<.001 -0.69, p<.001 -0.69, p<.001 -0.69, p<.001

(32)

Table 8 (Continued) Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction Student level Control variables Gender -0.32, p<.001 -0.32, p<.001 -0.32, p<.001 -0.32, p<.001 Vocational theoretical education2 0.71, p<.001 0.71, p<.001 0.71, p<.001 0.71, p<.001 General education2 _{1,25, p<.001} _{1.25, p<.001} _{1.24, p<.001} _{1.25, p<.001} Pre-university education2 _{1.86, p<.001} _{1.86, p<.001} _{1.85, p<.001} _{1.86, p<.001} Low SES4 _{-0.11, p=.066} _{-0.11, p=.068} _{-0.11, p=.065} _{-0.11, p=.069}

Number of books at home .03, p=.082 0.03, p=.080 0.03, p=.080 0.03, p=.081

Cultural activities .04, p<.001 0.04, p<.001 0.04, p=.001 0.04, p=.001

News reading .86, p<.001 0.86, p<.001 0.03, p<.001 0.03, p<.001

-0.25, p<.001 -0.25, p<.001 -0.25, p<.001 -0.25, p<.001

(33)

Table 8 (Continued). Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction Interaction

Vision, agreements, goals x Cultural activities -0.00, p=.314 Insight in competencies x Cultural activities 0.00, p=.400 Perceived importance x Cultural activities -0.01, p=.225

(34)

Additional analyses.

In additional analyses all control variables were removed from the model to check whether or not the school policy variables are at all related to citizenship knowledge on dealing with differences. In each of the four additional models for each measure of citizenship knowledge, the three policy variables together were added as explanatory variables. The school policy variables remained to seem unrelated to all four types of

citizenship knowledge. Adding the interactions to these models with no control variables lead to one significant interaction between a school’s insight in citizenship competencies and cultural activities a student participates in outside school for citizenship knowledge on dealing with differences (0.01, t(3217)=1.99, p=.047). This relatively small effect of 0.01 loses its significance once explanatory variables of educational level or migration background on the school level are controlled for.

Conclusion and discussion

This paper explored the impact of school policy for citizenship education on the citizenship knowledge of adolescents in the Netherlands. It was expected that students have more citizenship knowledge if their schools deem citizenship to be more important, have formulated a clear vision and learning goals, have agreements on how the goals can be achieved and have insight in the citizenship competencies of their students and adjust their citizenship education policy accordingly. This was expected to be the case even if student characteristics such as socioeconomic status, cultural capital, religion and migration

background were controlled for on individual and school level. Additionally, an interaction effect between a school’s citizenship education policy and the frequency with which students participate in cultural activities outside school was expected to exist. It was expected that school policy would benefit students with low cultural capital more than students with a high cultural capital from home. It can be concluded that a relationship between school policy for citizenship education and citizenship knowledge of adolescent students in the Netherlands was virtually non-existent. Having insight in student’s citizenship competencies and adjusting the policy for citizenship education accordingly was negatively related to citizenship

knowledge in one of the models for democratic behavior. The interactions between school policy on citizenship education and the frequency with which students participate in cultural activities outside school were also never found to be related to citizenship knowledge.

(35)

Table 9 Models for citizenship knowledge: societal responsibility. Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction

Intercept 0.89, p<.001 0.84, p<.001 0.84, p<.001 0.84, p<.001 0.84, p<.001

School level

-0.00, p=.814 -0.00, p=.814 -0.00, p=.830 -0.00, p=.806

-0.01, p=.401 -0.01, p=.413 -0.01, p=.405 -0.01, p=.399

0.02, p=.403 0.02, p=.403 0.02, p=.408 0.02, p=.401

Control variables

theoretical2

-0.59, <.001 -0.59, p<.001 -0.59, p<.001 -0.59, p<.001

Average school level: general2 _{-0.52, <.001} _{-0.52, p<.001} _{-0.52, p<.001} _{-0.51, p<.001}

Average school level: pre-university2 _{-1.04, <.001} _{-1.03, p<.001} _{-1.03, p<.001} _{-1.04, p<.001}

(36)

Table 9 (Continued). Model M0: Intercept-only M1: With predictors M2: With cross-level interaction M3: With cross-level interaction M4: With cross-level interaction Student level Control variables Gender -0.55, p<.001 -0.55, p<.001 -0.55, p<.001 -0.55, p<.001 Vocational theoretical education2 0.45, p<.001 0.45, p<.001 0.45, p<.001 0.45, p<.001 General education2 _{0.84, p<.001} _{0.84, p<.001} _{0.84, p<.001} _{0.85, p<.001} Pre-university education2 _{1.44, p<.001} _{1.44, p<.001} _{1.44, p<.001} _{1.44, p<.001} Low SES4 _{-0.09, p=.223} _{-0.09, p=.220} _{-0.09, p=.222} _{-0.09, p=.225}

-0.10, p=.134 -0.10, p=.135 -0.10, p=.132 -0.10, p=.135