The Impact of Education on the Development of Political Trust: Results from a Five-Year Panel Study among Late Adolescents and Young Adults in Belgium



There is a strong ongoing debate about the impact of higher education experiences on political attitudes and behaviours. While some authors assume a direct socialisation effect of educational experience, others have argued that education should be seen as a mere proxy variable for socio-economic status and pre-adult socialisation experiences. In this article we use a 5-year Belgian panel study that tracked respondents between the ages of 16 and 21. Using a hierarchical linear model of repeated measurements, we are able to demonstrate that differences with regard to political trust between future students and non-students are already present and stable at the age of sixteen. Significant determinants were school track and educational goal. The inclusion of actual educational status in the model (at age 21), however, rendered the relation with educational goal not significant. The results suggest that during secondary education students already anticipate and acquire a value pattern that is congruent with their future status. Ultimately, however, this effect is dependent on whether they actually enrol in higher education or not.

There is little doubt that education is strongly correlated with political attitudes and behaviours in Western societies (Nie et al., 1996; Schlozman et al., 2012; Verba et al., 1995). Highly educated actors participate more intensively in politics, they have higher levels of political interest and knowledge (Hooghe and Dassonneville, 2011) and they are more tolerant and trusting (Bobo and Licari, 1989; Claes et al., 2012). All the available evidence therefore suggests that education is, or has become, the main gateway to political involvement (Bovens and Wille, 2009). There is considerable disagreement, however, on how exactly this gateway operates (Berinsky and Lenz, 2011; Henderson and Chatfield, 2011; Kam and Palmer, 2008). While some authors argue that education has a direct impact on value patterns and skills, others claim that the importance attached to education experiences is overrated. Cindy Kam and Carl Palmer (2008, p. 613), for example, state that education should be considered as a mere proxy variable. They argue that specific sets of background variables such as parental and individual characteristics not only lead to more intense political involvement but also allow for access to higher education. Privileged groups or children of privileged parents are more likely to attain higher education levels, but according to these authors, it would be erroneous to ascribe any causal effect to these educational experiences. In this approach, education is little more than an indication for already existing patterns of stratification and social inequality (Persson, 2012a).

Thus far, this debate about the exact nature of educational effects has not led to any final conclusion, partly because of a lack of data. The question of whether college education experiences have an added effect, beyond the stratification that is already present upon entering higher education, requires panel data that are scarce in political science. Indeed, the entire debate that has followed the publication of the Kam and Palmer (2008) article (Henderson and Chatfield, 2011; Kam and Palmer, 2011; Mayer, 2011) is almost exclusively based on an analysis of panel data that by now are a few decades old. Since that period, patterns of access to higher education have changed dramatically (Reynolds and Johnson, 2011; Torche, 2011). Furthermore, thus far the study of higher education effects is concentrated on political participation as a dependent variable, while we can assume that colleges and university campuses often serve as a recruiting ground for acts of participation, thus blurring lines of causality (McAdam and Paulsen, 1993). To avoid this potential ground of contamination, we focus on a political attitude, namely political trust, that is less dependent on this kind of context effect. We also build on new and recent panel data that were collected in Belgium over the period 2006–11, and these data allow us to determine how exactly higher education experiences are related to political attitudes.

Disentangling the Relation between Higher Education and Trust

Following Arthur Miller and Ola Listhaug (1990, p. 358) we define political trust as ‘a summary judgement that the system is responsive and will do what is right even in the absence of constant scrutiny’. Political trust can be considered as one of the most important resources for a democratic political system (Hooghe, 2011). Citizens with higher levels of political trust are, for instance, more willing to comply with political decisions and to contribute to the public good (Hetherington, 2005; Marien and Hooghe, 2011; Tyler, 2011). Recently, a decline in political trust levels was documented in a number of countries; therefore, the question of what factors determine citizens’ political trust has become all the more salient (Dalton, 2004).

In most of the available literature there is a strong and positive relation between education levels and political trust (Schoon et al., 2010). One can distinguish, however, at least three approaches to explain this strong relation. A first approach builds on the sorting function of education: higher education provides access to more privileged positions in society, and these in turn render it easier to express trust in the system (Campbell, 2009; Newton, 1997; Nie et al., 1996). If one has acquired a high level of education and does gain access to better jobs and housing, it is less likely that one is exposed to the ‘darker side of society’ (Huang et al., 2011; Newton, 1997). The sorting approach does not necessarily make any statements about the inherent effects of education and curriculum, but the main argument is that education will have an instrumental effect, as it enables pupils to gain access to more privileged positions later on in life.

A second approach can be labelled the political sophistication approach. Here it is assumed that educational experiences have cognitive effects that enable pupils to understand social and political life in a more profound manner (Hillygus, 2005). The information that is being transmitted to pupils helps them to play a meaningful role in political life. Education experiences increase pupils’ knowledge about the political, economic and legal system and help them to interpret information about these systems. Because of this better understanding they are more likely to support the system (Huang et al., 2011).

Third, Kam and Palmer (2008) have challenged the claims about the effects of education as they consider higher education as a proxy for socio-economic status. Their main argument is that access to higher education is distributed very unequally across society, and that specific political attitudes and habits are already present at the moment students enter higher education. They list a whole range of pre-adult experiences that might be responsible for this a priori difference in political attitudes and behaviours, ranging from the high education and income levels of parents, through the example set by parents and the level of political discussion within the family, to cognitive skills and academic orientation during high school years (Kam and Palmer, 2008; Persson, 2012b). The Kam and Palmer article is based on an analysis of the Youth-Parent Socialisation Panel Study, 1965–97 of M. Kent Jennings et al. (1997). Using a process of propensity matching, they compared panel respondents who were most likely to pursue higher education (based on the income and education level of their parents and their high school grades) but did not, with those who did attend higher education. Their main conclusion is that the experience of attending higher education was rendered not significant. The results of their analysis are supported by another analysis of the same dataset by M. Kent Jennings and Laura Stoker (2008), showing that the difference between those who would eventually go on to college and those who would not was already clearly present in the 1965 wave of that study, when all the respondents were still enrolled in high school. Relatedly, some authors have argued that education experiences do not contribute to the formation of value patterns, but are dependent on them. Søren Serritzlew and Gert Svendsen (2011), for example, have shown that students already have higher levels of political trust upon entering college, and this attitude can be seen as contributing to successful education careers as trust in educational authorities contributes to the willingness to learn (Coleman, 1988).

Within the debate on the effect of education, the results of the Kam and Palmer study have been challenged on methodological grounds. Propensity-matching techniques are highly dependent on the exact characteristics of the model being used to predict future attendance in higher education. A slightly different operationalisation renders the experience of higher education significant again (Mayer, 2011). The basic problem with this approach is that the propensity to enter higher education is indeed a very good predictor of attending college. A vast majority of all panel respondents who were predicted to attend college actually did, leaving only a limited number of respondents who, for some reason or another, did not. This limited pool of highly exceptional cases is subsequently used to compare with all the other respondents who did attend college (Henderson and Chatfield, 2011). The criticism is that these cases are so highly exceptional that they cannot be used for a valid comparison. Furthermore, it has to be remembered that this analysis was conducted with data covering higher education experiences in the period 1965–73, a historically exceptional period with considerable turmoil in student life across US campuses (Kam and Palmer, 2011). Since that period, patterns of access to higher education have changed dramatically (Reynolds and Johnson, 2011).

Research Design

The main goal of the current study is to investigate whether the attitudinal differences routinely found between adults who have pursued higher education and those who have not can already be found during adolescence and thus before the respondents can pursue higher education (Jennings and Stoker, 2008). In contrast to earlier articles, this analysis is conducted on recent data so we can be certain we do not study a specific historic situation, when access to higher education was still more restricted. The Belgian Political Panel Study (BPPS, 2006–11) which we rely on includes information about respondents up to the age of 21. This is a moment in life when most of them have not entered professional life yet, implying that they have not yet developed their own economic position, and this means that the sorting mechanism can already be excluded. This leaves us with two possible associations that can be tested: either the experience of higher education itself or the existence of pre-adult socialisation experiences has a direct impact on political attitudes and simultaneously encourages future enrolment in higher education.

The first reasoning is straightforward: in the 2011 wave of the panel study (with respondents at the age of 21) almost three-quarters of all respondents pursue higher education, and this subgroup can be compared to the remaining quarter who have already left the school system. If higher education has a direct impact, we should observe that the students in the sample have different political attitudes from the non-students.

In order to test the proxy argument, we include various measurements that capture the pre-adult socialisation experiences that are claimed to be responsible for the effects usually ascribed to education, such as the socio-economic status of the parents and the civic education experiences of the respondents during high school (Kam and Palmer, 2008; Persson, 2012b). Moreover, we also include two variables that thus far have not received sufficient attention in this kind of research.

First, in the Belgian education system adolescents are sorted into educational tracks that are intended to lead either to the labour market or to higher education (Bauer and Riphahn, 2006; Gamoran and Mare, 1989). In a school system with a strong tracking tradition, this practice especially predicts the odds that adolescents will be able to pursue higher education (Van Houtte, 2004). Therefore, it is essential to take into account which school track the pupils are enrolled in.

Second, we include the educational goal expressed by the pupil. There is a vast literature demonstrating that adolescents already have a strong and well-ingrained expectation about the kind of education they will pursue and this education goal guides their school efforts (Andrew and Hauser, 2011). The decision to pursue higher education is not only dependent on achievement but also involves factors such as self-confidence, ambition and external constraints (Ashby and Schoon, 2010; Gambetta, 1987) and, therefore, educational aspiration is a better measurement of the propensity to pursue higher education than the information obtained by test scores or intelligence. Previous empirical research shows that educational goal predicts completed years of education in an adequate manner (Manski, 2004; Sewell et al., 1969). Given these considerations, we can claim that the educational goal of the pupil at secondary school serves as an ideal proxy variable for the propensity of adolescents to pursue higher education.

In this study, we focus on education as a source of trust in political institutions. Within established democracies education is theorised to increase levels of political trust.1 Within corrupt political systems, however, negative effects of education on political trust have been documented (Catterberg and Moreno, 2006; Serritzlew and Svendsen, 2011). It has also been claimed that, in established democracies, higher-educated citizens are more distrustful than lower-educated citizens (Inglehart, 1999; Norris, 2011). However, most of the available studies show a positive relation between education and political trust levels within established democracies (Hooghe et al., 2012; Schoon and Cheng, 2011; Van der Brug and van Praag, 2007). In our analysis, we test two competing claims about the mechanism that could help us to explain this relation:

  • H1. Respondents who are enrolled in higher education will have significantly higher levels of political trust.
  • H2. The difference in political trust between students and non-students is rendered not significant when taking into account pre-adult socialisation experiences.

Data and Measures

These hypotheses will be investigated using the 2006–11 Belgian Political Panel Study (BPPS). In this study, a representative sample of Belgian late adolescents and young adults was surveyed three times about their political and social attitudes and behaviour (Hooghe et al., 2011). Given that previous research shows that Belgium is not an exceptional case in Europe with regard to the level and stability of political trust (Marien, 2011), we have no reason to assume that Belgium would offer a deviant case.

The first wave of the study was conducted in 2006, when respondents were sixteen years old. Respondents were selected through a random sample of schools representative of the type and location of schools. For the first wave, adolescents were surveyed in class; although participation was not obligatory the class setting resulted in an almost universal response rate (99 per cent) within the schools that agreed to participate (66 per cent). As a result, the 2006 survey, in which 6,330 sixteen-year olds participated, was representative for region, school type, sex and educational track (Hooghe et al., 2011). The respondents were surveyed again at school in 2008 and through regular mail in 2011 as they had left high school by then. Respondents who had changed schools or dropped out of school received the 2008 survey by regular mail as well. In the second wave, 4,235 pupils (67 per cent) of the first wave participated and 3,025 respondents or 71.4 per cent of the panel participated again in the third wave (Hooghe et al., 2011).

For the current analysis we only make use of the subsample of Dutch-language respondents within the BPPS 2006–11. We do so first because the impact of education is a central element in this analysis and education is a competence of the language communities in Belgium (Deschouwer, 2009). As a consequence, tracking practices in the two language communities cannot be directly compared. Second, response rates, from the first survey onwards, were somewhat lower in the French-language community (Hooghe et al., 2011). Limiting the focus to Dutch-language respondents who took part in the three waves of the panel study, the sample for the current analysis consists of 1,926 respondents, which is 56 per cent of the original 2006 sample (consisting of 3,455 Dutch-language respondents). Missing values on some of the explanatory variables further reduce the data set to 1,634 respondents for the multi-level analysis.2

Dependent Variable

We analyse (the evolution of) political trust between 2006 and 2011. Following David Easton (1965), we can distinguish between trust in the political community, the regime and political authorities. The focus of this study is on the regime level and in particular on trust in political institutions. Political institutions play an important role in shaping a democratic society, and we can assume that trust in these institutions is strongly related to a more comprehensive evaluation of the political system. Moreover, while a critical attitude towards the current leaders can be seen as healthy for democracy, trust in democratic procedures and institutions is vital for democratic stability (Dalton, 2004). Political trust was measured in exactly the same way in the three waves of the panel study. This was done by means of a measurement scale routinely used in research on political trust (Marien, 2011). Respondents were asked to rate their level of trust in six political institutions on a scale from zero to ten.3

Mean levels of trust were lowest for political parties in all survey years (Appendix 1). The adolescents had most trust in order institutions. The trust items produce a clear one-dimensional scale with high factor loadings for the full panel as well as for each of the three survey waves separately (Table 1). Because the six items load on a single variable, a zero to ten sum-scale is used as the dependent variable in the subsequent analyses. Although in some studies various dimensions of trust in political institutions have been distinguished, it is clear that for the current data set only a one-factor solution can be defended on methodological grounds.4 The mean score on the trust scale was 5.31 in 2006, 5.52 two years later and 5.11 in the 2011 wave (Appendix 1).

Table 1. Factor Analysis for Trust in Political Institutions
 PanelWave I (2006)Wave II (2008)Wave III (2011)
Factor loadCronbach's α without itemFactor loadCronbach's α without itemFactor loadCronbach's α without itemFactor loadCronbach's α without item
  1. Note: Entries are the result of principal component analyses.
Source: BPPS 2006–11.
Federal parliament0.880.840.900.860.880.850.850.82
Regional parliament0.890.840.900.860.890.850.860.81
European Parliament0.820.860.840.870.810.870.800.83
Political parties0.740.870.770.880.770.880.690.85
Explained variance0.630.660.650.59
Cronbach's alpha0.880.890.890.86

Independent Variables

The independent variables included are located at different levels within the multi-level framework of the analysis. At the first level, variables that change over time are included; hence, these variables were measured multiple times. We include the year of the survey as a dummy variable in the analysis. Since all respondents were about the same age this time variable also refers to the age of the respondents. On average, they were 16 in 2006, 18 in 2008 and 21 in 2011. Political interest is based on respondents’ self-rated interest in politics on a 1 (not interested) to 4 (very interested) scale. News consumption refers to respondents’ self-assessment of how often they read, watch or listen to the news (including online) on a 1 (never) to 5 (every day) scale.

At the second level, individual variables are included that were measured once and can be expected to be – or are operationalised as – stable over time (Singer and Willett, 2003). We include a number of demographic variables which are assumed to relate to respondents’ level of political trust, such as sex, religious denomination and religious practice. The latter refers to how often the respondent attended religious services over the past year (1 = never to 5 = more than once a week). Furthermore, an estimate of the number of books at home was included (1 = none to 7 = over 500 books). This proxy variable of the respondents’ socio-economic status was used because it is difficult to question adolescents directly about their parents’ level of income or social class (Claes et al., 2012). The number of books at home was shown to be a good proxy for the socio-economic status of a household and therefore is routinely included in this kind of research (Hahn, 2003). As an additional control for the socio-economic situation of the parental household we include dummy variables for whether or not respondents’ mother and father pursued higher education. Furthermore we include respondents’ rating of how often they discuss politics with their parents, measured on a 1 (never) to 4 (always) scale. All these variables are included as measured in the first wave of the panel study. Although minor shifts in these variables might occur, we assume that the variables and their impact on political trust remain largely stable over the period of observation.

Since we are especially interested in the correlates of educational transitions the respondents experience between the age of sixteen and 21, we include a number of variables measuring educational status. First, we include educational goal at the age of sixteen. We distinguish between the goal to pursue higher education and not aiming to enrol in higher education. Second, we include respondents’ high school education, referring to the school track respondents were in at the age of sixteen. We distinguish a general or art education, a technical and a vocational track. Third, we investigate the educational status at the age of 21. This variable refers to the education respondents are in at the time of the 2011 survey. For this variable, as for educational goal, we include a dichotomous variable: we distinguish between respondents who are still enrolled in the school system at the age of 21 (university or non-university) and respondents who are already in the job market (whether they have a job or not). These variables are included at the individual level as well, since they are only measured once. School track and educational goal were measured in 2006, and educational outcome was measured in 2011.

Since pupils in the same class can be expected to be more alike than respondents in different classes, we should also control for the class the respondents were in at the time of the survey. Therefore, we take into account a third class level in the analysis. At this level, we include classroom instruction, that is, whether six different items had been discussed in courses at school over the past school year. The topics included were the functioning of parliament, the United Nations, the European Union, federalism, elections and recent political events. Because the items are strongly related and load on a single dimension (Cronbach's α: 0.83; Eigenvalue: 3.30; Explained Variance: 55.00) they were included in a 1 to 4 point sum-scale of classroom instruction. Second, respondents’ perception of an open classroom climate was included, measured by means of a traditional three-item scale (Torney-Purta, 2002). The most characteristic item is: ‘students are encouraged to make up their own minds about issues’. These items were also strongly correlated (Cronbach's α: 0.62; Eigenvalue: 1.71; Explained Variance: 56.94) and were therefore used to construct a 1 to 4 point sum-scale of the open classroom climate. Since respondents, depending on their level of interest in politics, can be expected to recall experiences with civic education differently, these two civic education variables were aggregated to the classroom level. Descriptives for all the independent variables are provided in Table 2.

Table 2. Descriptives of Independent Variables in the Analysis
 MeanStd. dev.MinMax
  1. Note: N is 4,784 observations for 1,634 respondents in three waves (2006, 2008, 2011).
Source: BPPS 2006–11.
Level 1 (time varying)    
 Political interest2.240.8314
 News consumption3.921.0315
Level 2 (individual)    
 Religious denomination    
 Religious practice1.700.7115
 Books at home3.781.5217
 Discussions about politics with parents2.070.6214
 Mother higher educated0.480.5001
 Father higher educated0.450.5001
 Educational track (2006)    
 Goal higher education0.800.4001
 Outcome higher education0.720.4501
Level 3 (class)    
 Classroom instruction1.790.2912.83
 Open classroom climate2.690.2123.43


Because we analyse the evolution of political trust over time, it is essential that we have at least three measurements of trust. Not only are estimates more precise with three points of observation (Willett, 2004), three measurements also allow specifications of change other than linear ones (Ployhart and Vandenberg, 2010; Singer and Willett, 2003). The analysis takes the form of a hierarchical linear model of repeated measurements. In such a design, the measures are specified as nested within individuals. This hierarchical model allows us to include observations of which the dependent variable is not measured each time.5 Although we focus on the panel respondents of the BPPS for the current analysis, this implies that we can include respondents who have no full measurement of political trust in one or two of the survey waves (Maas and Snijders, 2003; Snijders, 1996). The analysis takes the form of a three-level multi-level model for change, with observations nested in individuals and individuals nested in classes (Singer and Willett, 2003; Tasca et al., 2009).

The hierarchical design of the analysis allows us to investigate not only differences in levels of trust and the evolution of political trust over time, but also differences between individuals in this change over time (Ployhart and Vandenberg, 2010). We not only investigate whether the level of political trust is different between individuals depending on their type of education, but also whether the evolution of trust differs for respondents in, for example, different types of education.

For the analysis we follow the approach presented by Judith Singer and John Willett (2003) and present a number of models. A first unconditional means model allows us to estimate the amount of variance at the different levels. By means of this base model we can calculate the amount of variance explained by adding more explanatory variables in subsequent models. In a second step, the effect of time is added and time is specified as random. By means of this unconditional growth model, we can assess whether there are significant differences in the trajectories of political trust between individuals. In a third model the other time-varying predictors are added and in the following models we also include level 2 (individual) and level 3 (class) variables.


Before proceeding with the multivariate analyses, we present some descriptive analyses looking at the evolution of political trust by respondents’ educational outcome, that is, whether or not respondents were effectively pursuing higher education in 2011. Figure 1 indicates substantial differences in the level of trust of the two groups. More importantly, these differences appear to be present already in 2006 and therefore long before respondents effectively entered higher education. As a result, the figure provides suggestive support for both of our hypotheses: first, there are substantial differences in trust depending on levels of education; second, these differences are already present before adulthood, hinting at the importance of pre-adult socialisation processes (see also Górecki, 2013).

Figure 1.

The Evolution of Political Trust by Level of Education (2011)

Note: Mean levels of political trust for different groups, n = 1,634.

Source: BPPS 2006–11.

Supported by the descriptive results in Figure 1, we now proceed with the multivariate analyses. The first model, the ‘unconditional means’ model, shows significant variance between individuals in the initial state of political trust (Model I in Table 3). Additionally there is significant variance within persons and therefore there are significant differences in the level of trust of the respondents over time. This base model also reveals that 12 per cent of the variance in political trust is situated at the class level and, therefore, it is necessary to include this third level in the analysis.

Table 3. Multi-level Model for Change with Repeated Measures Nested in Individuals and Individuals Nested in Classes
 Model I Unconditional meanModel II Unconditional growthModel III Time varyingModel IV Socio-structuralModel V Educational goalModel VI Educational trackModel VII Higher educationModel VIII Interaction
  1. Notes: Entries are the result of a hierarchical linear model with repeated measurements. Significance levels: *p < 0.05; **p < 0.01; ***p < 0.001.
Source: BPPS 2006–11.
Female   0.19** (0.07)0.11 (0.07)0.07 (0.07)0.05 (0.07)0.05 (0.07)
Rel. denomination (ref: none)        
Catholic   0.42*** (0.09)0.43*** (0.08)0.43*** (0.08)0.40*** (0.08)0.40*** (0.08)
Other   0.32 (0.16)0.26 (0.16)0.27 (0.16)0.28 (0.16)0.28 (0.16)
Religious practice   0.11* (0.05)0.10* (0.05)0.09* (0.05)0.10* (0.05)0.10* (0.05)
Books at home   0.01 (0.02)0.01 (0.02)0.00 (0.02)0.00 (0.02)0.00 (0.02)
Discussion about politics with parents   0.15** (0.06)0.11* (0.06)0.09 (0.05)0.10 (0.05)0.09 (0.05)
Mother higher educated   −0.03 (0.07)−0.07 (0.07)−0.12 (0.07)−0.13 (0.07)−0.13 (0.07)
Father higher educated   0.22** (0.08)0.18* (0.08)0.15* (0.07)0.14 (0.07)0.14 (0.07)
Goal higher education    0.51*** (0.09)0.24* (0.10)0.10 (0.11)0.10 (0.11)
Educational track (ref: general)        
Technical     −0.40*** (0.09)−0.31** (0.09)−0.31** (0.09)
Vocational     −0.77*** (0.13)−0.57*** (0.13)−0.57*** (0.13)
Outcome higher education      0.40*** (0.09)0.43*** (0.11)
Higher education* 2008       0.13 (0.10)
Higher education * 2011       −0.20 (0.10)
Classroom instruction    0.10 (0.13)−0.10 (0.12)−0.10 (0.12)−0.10 (0.12)
Open classroom climate    0.49** (0.17)0.48** (0.16)0.47** (0.16)0.47** (0.16)
Year (ref: 2006)        
2008 0.22*** (0.05)0.14** (0.05)0.14** (0.05)0.15** (0.05)0.15** (0.05)0.15** (0.05)0.06 (0.09)
2011 −0.20*** (0.05)−0.38*** (0.05)−0.38*** (0.05)−0.36*** (0.05)−0.36*** (0.05)−0.35*** (0.05)−0.21* (0.09)
Political interest  0.35*** (0.03)0.34*** (0.03)0.32*** (0.03)0.30*** (0.03)0.29*** (0.03)0.29*** (0.03)
News consumption  0.06* (0.02)0.06* (0.02)0.05* (0.02)0.05* (0.02)0.05* (0.02)0.05* (0.02)
Intercept5.22*** (0.06)5.20*** (0.06)4.33*** (0.11)3.34*** (0.18)1.66** (0.53)2.69*** (0.51)2.55*** (0.51)2.53***
Within person1.73***1.68***1.67***1.67***1.67***1.66***1.66***1.66***
Within class0.38**0.38***0.25***0.14**
Variance in initial status1.03***1.05***0.99***0.97***0.99***0.99***0.97***0.97***
Variance in rate of change
ICC Class0.1200.1210.0850.0510.0210.0080.0070.007

In Model II, the ‘unconditional growth’ model, we include the effect of time by means of the year dummies. The level of political trust was significantly higher in 2008 compared to the 2006 level, but significantly lower in 2011. Furthermore we add the random effect of time in this model, in order to assess whether there is significant variance in the evolution of political trust between individuals. The variance component of this rate of change, however, is not statistically significant, hence there are no significant differences between respondents in the trajectories of political trust from 2006 to 2011. Despite the fact that, during the period observed, a lot of changes occurred in respondents’ life, the trajectories of political trust of respondents appear to be largely similar. It is especially likely that the decrease in trust between 2008 and 2011 is a period effect, as during this period Belgium was confronted with a long political crisis (Hooghe, 2012). Even so, this might also be an age effect. For the sake of our argument, however, what causes this decline is not relevant given the fact that all groups reacted in exactly the same manner. Hence, this does not invalidate our research design, and in the remainder of the analysis we will focus on differences between groups, not on aggregate changes in the level of trust in society or among age cohorts.

In Model III we add the time-varying variables to the model. As can be seen in Table 3, the more interested a respondent is in politics, the higher his or her level of political trust. Following the news is also positively related to political trust, although the effect is rather small. Adding these time-varying effects has only a marginal impact on explaining the variance at the person level. The intra-class correlation (ICC) at the class level decreases, which indicates that there are strong differences in interest in politics and news consumption between different classes.

Subsequently, we include a number of demographic variables at the second – individual – level (Model IV). As the results show, women, respondents with a Catholic background and respondents with a higher religious practice have significantly more trust in political institutions. The father's level of education and discussing politics with parents also positively affect political trust. However, the number of books at home, our proxy for respondents’ socio-economic status, is not significantly related to political trust. As expected, these socio-structural variables are fairly stable and therefore do not explain changes over time. We do see that adding demographic variables to the model slightly decreases the variance at the initial level of trust. Additionally, the ICC at the class level drops by including the socio-structural variables, indicating that class groups are quite homogeneous with regard to background variables.

Model V additionally includes the educational goal of the respondents in 2006. Respondents who intend to enrol in higher education have a significantly higher level of political trust.6 Furthermore, classroom instruction about politics has no significant relation with levels of political trust. Experience with an open classroom climate, on the other hand, strongly increases respondents’ level of political trust. Including these civic education variables and respondents’ educational goal in the analysis causes a strong drop in the ICC at the class level (from 5 per cent to 2 per cent), as could be expected.

One might question, however, whether the relation with respondents’ educational goal is not a mere reflection of the school track they are in. In order to take this possibility into account, we include both the school track of respondents in 2006 and their educational goal in Model VI. As is clear from the results, there is indeed a strong and significant association with the school track. Respondents in a technical track have significantly less political trust compared to those in a general track. Those that are in a vocational track have even lower levels of political trust. Even when including these school tracks, the educational goal of the pupil remains significant, although the effect size is clearly reduced.

Our first hypothesis, however, states that the experience of being in higher education is positively associated with political trust. To control for this effect, in a subsequent model (Model VII) we also add the relation with educational outcome measured in 2011.7 As the results indicate, educational goal is no longer significantly associated with political trust in Model VII. Respondents in a technical or vocational track still have significantly lower levels of political trust. Furthermore, those who eventually enrol in higher education have significantly more trust. The model suggests that while initially educational goal is positively associated with political trust, including information from the third wave of the panel study shows that the association with being in higher education absorbs this difference. Results therefore suggest that respondents’ educational goal is strongly associated with the experience of higher education. This is also clear from the table in Appendix 2: about 80 per cent of the respondents who did not intend to pursue higher education are indeed not enrolled in higher education in 2011. Of the respondents who indicated their plan to pursue higher education, over 86 per cent effectively do so at the age of 21.8

This does not mean, however, that the students in higher education only start to differ from other respondents at the moment they enter college. In Model VIII we add a cross-level interaction between education in 2011 (whether or not respondents are in higher education) and the time variable (Snijders and Bosker, 1999). If pursuing higher education, independent of previous socialisation and schooling experiences, is positively associated with political trust, we would expect the gap between those in higher education and those not pursuing higher education to increase over time (Persson, 2012a). As the results make clear, the opposite can be observed; while the main effect of being enrolled in higher education is positive, the interaction term for 2011 is negative and not significant. This suggests that while respondents in higher education do have a higher level of trust in politics, the gap between the two groups has not widened in 2011. The experience of higher education thus absorbs the effect of educational goal expressed during secondary education, but it is not associated with a widening of the gap between students and non-students.


The goal of the present study was to arrive at a better understanding of the positive relation between education and political trust. Since we only have data for respondents up to the age of 21, we cannot provide any insights on the sorting mechanism. As the respondents in the BPPS mostly have not yet acquired an independent socio-economic position, access to privileged positions in society cannot be a distinct factor in this analysis. This leaves us with two mechanisms to explore: either the association is a result of the actual experience of higher education, or it can be attributed to pre-adult socialisation experiences which simultaneously have an effect on the odds that one will pursue higher education. At first sight, our findings only seem to add to the confusion in this regard. The analysis shows that at the age of sixteen adolescents can already be distinguished clearly, and these differences remain constant. Even if we distinguish between those enrolling in higher education and those who do not pursue higher education (information that is only available five years later on), at the age of sixteen these groups have distinct and stable patterns of political trust. So here we side with Jennings and Stoker (2008): students in higher education already differ significantly from non-students long before they enter higher education. Two elements were shown to be important in this regard: the school track and the student's educational goal. Already at the age of sixteen adolescents not only have a clear idea about their future role in life, but the school system − at least in Belgium − has already sorted them according to their most likely future outcome.

Do our findings support the sceptical view that higher education should be regarded as a mere proxy variable for higher socio-economic status (Kam and Palmer, 2008)? Not necessarily, as educational goal in Model VI and current enrolment in higher education in Model VII are significant predictors of levels of political trust. Being a student or not does itself make a difference. Hypothesis 1 is thus supported: there is a significant difference between students and non-students. Hypothesis 2, however, needs to be rejected: this difference between students and non-students remains significant even when including a full battery of controls.

The analysis makes clear that students are more trusting in politics compared to non-students. What is not found, however, is an additional association with pursuing higher education. For the most part, the differences between students and non-students already exist at the age of sixteen. This means that our problem is the same as in previous research: educational goal happens to be a very good predictor of one's future academic career (Henderson and Chatfield, 2011). There is, inevitably, a very high degree of self-selection present in higher education and, based on these selection mechanisms, it can be demonstrated that secondary school students already have the value patterns that are congruent with their future role. Results of the analysis, however, suggest that education goal is not much more than what it pertains to be, that is, a future goal orientation. Once the high school students are able to do what they aimed to do, this real life behaviour takes over with regard to explanatory power. The impact of education goal therefore is dependent on actual behaviour and this can be interpreted as a regular selection and adaptation effect. Already during secondary education, students anticipate their future role as students in higher education and the current analysis even suggests that they already acquire the value pattern that is in accordance with their future status. This anticipation process, however, does not make any sense if they are not able to pursue their goals and, once they have reached the appropriate age, enrolment becomes more important.

The debate about the effect of higher education on value patterns should therefore take into account basic sociological processes of selection and adaptation (Brand, 2010). The effect of higher education is not only limited to being in college for four or five years. During a large part of their secondary school career, future students actively prepare themselves, and they are being prepared by the school system, for their future role as college students. Teachers help them to acquire the skills they will need in the future to pursue higher education in a successful manner. The students themselves further develop their attitudes in a way that is supportive of their future goal in life. This entire preparation and self-selection process is part and parcel of the meaning of higher education. While it might sound good to label higher education as a ‘mere proxy variable’, one might wonder how secondary schools would function if there was no higher education, or if access to colleges was highly restricted. Secondary schools would function completely differently: they would stress different goals and students would prepare themselves for a different role in life. Therefore, our suggestion would be that the two proposed associational mechanisms do not exclude one another, but are dependent on one another: the pre-adult socialisation experiences that these adolescents are being affected by are designed to a large extent in order to make their future access to higher education possible.

Appendix: Appendix 1: Descriptives for Political Trust Items

 nMinimum–maximumMeanStd. dev.

Source: BPPS 2006–11.

 Federal parliament1,8960–105.022.28
 Regional parliament1,9020–105.142.22
 European Parliament1,8990–105.672.36
 Political parties1,9080–104.482.28
 Political trust 2006 (sum-scale)1,8770–105.311.87
 Federal parliament1,9150–104.992.24
 Regional parliament1,9150–105.372.19
 European Parliament1,9130–105.912.24
 Political parties1,9190–104.492.15
 Political trust 2008 (sum-scale)1,9020–105.521.79
 Federal parliament1,8860–104.542.14
 Regional parliament1,8950–105.022.19
 European Parliament1,8910–105.572.27
 Political parties1,8960–103.722.10
 Political trust 2011 (sum-scale)1,8740–9.335.111.67

Appendix: Appendix 2: Predictive Power of Educational Goal

 Outcome: No higher educationOutcome: Higher educationTotal

Source: BPPS, 2006–11.

Goal: No higher education80.41% (780)19.59% (190)100% (970)
Goal: Higher education14.34% (547)85.66% (3,267)100% (3,814)
Total27.74% (1,327)72.26% (3,457)4,784


Sofie Marien and Ruth Dassonneville acknowledge support from the Research Foundation Flanders (FWO). An earlier version of this article was presented at the MPSA Annual Meeting (Chicago, 12–15 April 2012). We are grateful for the comments of the participants of this meeting and for the comments of the anonymous reviewers for Political Studies.

  1. 1

    When using participation as a dependent variable, the relatively low frequency of political participation acts among this age group require poisson regression techniques. However, this analysis also leads to the same conclusions with regard to the impact of higher education.

  2. 2

    Missing data analysis indicated that there is some systematic attrition bias. Weighting as a strategy to cope with attrition does not significantly affect the results, however.

  3. 3

    The exact wording was: ‘For each of the following institutions, can you indicate on a 0 to 10 scale how much trust you have in them?’

  4. 4

    To be certain about this effect, we also conducted the entire analysis again for every item in this scale separately. All these analyses, however, did confirm the overall result.

  5. 5

    While this is a valuable property of the method chosen, the current analysis includes only a few respondents for whom information on time-varying variables is missing (for 1,523 respondents complete information is used, so for only 7 per cent of the individuals included some information on level 1 is missing).

  6. 6

    Differentiating between respondents intending to go to university and those planning on enrolling in non-university higher education did not point to significant differences.

  7. 7

    We additionally ran a fixed-effects OLS regression in order to assess whether or not including all these variables in a single model causes multicollinearity problems. The highest VIF was 2.02 and the lowest tolerance statistic 0.50, which does not indicate too much collinearity.

  8. 8

    An additional test on whether respondents were able to predict correctly their participation in higher education did not lead to significant results.


  • Marc Hooghe is Professor of Political Science at the Centre for Citizenship and Democracy of the University of Leuven. He is a Visiting Professor at the Universities of Mannheim and Lille. He currently holds an ERC Advanced Research Grant to investigate the democratic linkage between citizens and the state in Europe. Marc Hooghe, Centre for Citizenship and Democracy, University of Leuven, Parkstraat 45, 3th floor, B-3000 Leuven, Belgium; email:

  • Ruth Dassonneville is a Research Fellow of the Research Foundation Flanders (Belgium) at the Centre for Citizenship and Democracy of the University of Leuven. She is currently preparing a PhD on the topic of electoral volatility. Her main research interests are economic voting and electoral behaviour. Previously, her work has been published in Electoral Studies, Party Politics, Acta Politica and Political Science Research and Methods. Ruth Dassonneville, Centre for Citizenship and Democracy, University of Leuven, Parkstraat 45, B-3000 Leuven, Belgium; email:

  • Sofie Marien was Postdoctoral Researcher at the Centre of Citizenship and Democracy at the University of Leuven, and she is now an Assistant Professor at the Political Science Department of the University of Amsterdam. She received a postdoctoral grant from the Research Foundation Flanders (FWO). Her research on political trust and participation has appeared in, among others, the European Journal of Political Research, Intelligence, Political Studies and European Sociological Review. Sofie Marien, Centre for Citizenship and Democracy, University of Leuven, Parkstraat 45, box 3602, B-3000 Leuven, Belgium; email: