Tackling inequalities in political socialisation: A systematic analysis of access to and mitigation effects of learning citizenship at school

This article tackles the issue of social inequalities in voting and identifies how and when differences in learning political engagement are influenced by social background in the school environment between the ages of 11e16 in England. Using Latent Growth Curve Modelling and Regression Analysis on the Citizenship Education Longitudinal (CELS) data this research identifies two elements that influence the political socialisation process: access to political learning and effectiveness in the form of learning in reducing inequalities in political engagement. The results show that there is unequal access by social background to learning political engagement through political activities in school and through an open classroom climate for discussion. However, there is equal access by social background to Citizenship Education in schools and this method of learning political engagement is effective at the age of 15e16 in reducing inequalities in political engagement. © 2017 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Almost by default, democratic institutions lose their responsiveness and subsequently lose their legitimacy for groups in society who have low levels of political engagement. The subgroup of disadvantaged youth in the UK are the most politically disaffected (Hoskins and Janmaat, 2015) and it is this group who have born the recent effects of austerity policies the most. Political apathy not only affects disadvantaged youth, but this apathy then appears to be passed on to their children through the transfer of economic, human, cultural and social capital (Brady et al., 2015: Schlozman andBrady, 2012). Political mobility, the opportunities for young people from non-engaged families to become political, is as problematic as other forms of socioeconomic mobility (Verba et al., 2003: Schlozman andBrady, 2012). Yet at the same time political mobility has received only a tiny fraction of the amount of research attention in comparison to other forms of socioeconomic mobility (Brady et al., 2015) and relatively little is known about socioeconomic effects on political socialisation processes, particularly in relation to what is happening in schools. Some research has addressed the effect of different pathways in the education system (Hoskins and Janmaat, 2015;Janmaat et al., 2014), but this research has not explored the learning methods within the school environment. In this context, this article will explore the political socialisation process in the school environment, how social inequalities in intended voting are reinforced, as well as those approaches in schools that can actually mitigate these differences.
There is a considerable body of knowledge that exists on how students, regardless of their social background, learn citizenship at school. Methods include open classroom climate (Hoskins et al., 2012;Torney-Purta, 2002;Hahn, 1998;and Campbell, 2008), citizenship education (Whiteley, 2012) and political activities themselves (Hoskins et al., 2012). Although a small but rapidly growing body of research (Campbell, 2008;Hooghe and Dassonneville, 2013;Castillo et al., 2015;Persson, 2015) has started to explore the ways in which these approaches affect inequalities in political engagement, to the best of our knowledge no study has yet looked at this issue in the UK. In addition, the current literature in the field has also lacked a comprehensive and systematic approach to identifying the effects of different forms of learning on aspects of inequalities in political engagement. In this paper we develop such an approach by suggesting two ways in which learning sources influence disparities of political engagement, namely through access and mitigation effects. We are also adding to the previous literature by reviewing a wider range of sources of learning within the school environment.
To address the issue of access we assess whether young people from more advantaged social backgrounds gain greater access to the learning of political engagement. We hypothesize that, for the learning methods of political activities at school and open classroom climate, young people from more disadvantaged backgrounds are likely to undertake less quantity of the learning needed to gain the knowledge, skills, attitudes and behaviour for political engagement. We hypothesize that, for the learning method of citizenship education, the quantity of learning is not affected by social background. We test access by measuring the association between social backgrounds and students undertaking the different forms of learning of political engagement. The second set of hypotheses that we test is whether the different learning experiences mitigate the effect of social background on political engagement. We hypothesise that young people from disadvantaged backgrounds are likely to gain more from these experiences. We assess mitigation effects through identifying the extent to which a learning source, once undertaken, diminishes social disparities in intended political engagement. A learning source may be very effective in reducing social gaps in political engagement but still not be able to bring overall inequalities down because of unequal access to this learning source across different social groups. Thus, both issues are important. Existing research has insufficiently taken both these issues into account as we will explain further below.
Using the Citizenship Education Longitudinal dataset, in this article we will investigate, for a range of sources for learning political engagement, how access and mitigating effects influence inequalities in voting intentions. We will begin by reviewing the literature on inequalities, political engagement, and political socialisation. Then, for the first step of the analysis, we use Ordinary Least Squared regression (OLS) analysis to investigate access to sources of learning political engagement. In a second step we then perform Latent Growth Curve modelling (LGCM) using interaction effects to establish whether participation in a number of learning activities has a mitigating effect on social inequalities in political engagement.

Inequalities in political engagement
Inequalities in political engagement according to age and socioeconomic background are an increasingly important issue within the UK (Hoskins and Janmaat, 2015). In the 2015 UK general election there was an approximate 20% difference in voter turnout between the highest and lowest socio-economic groups (Ipsos Mori, 2015). Voter turnout for the lowest socioeconomic groups has been in decline between the 1995 and 2015 general elections, decreasing by about 20%, whilst voter turnout for the highest socio-economic group has only declined just under 10%. These trends suggest that the UK is heading in the direction towards the high levels of inequality in voting patterns found in the US (Schlozman and Brady, 2012), and that research is urgently needed to find out how this process of political alienation is occurring, as well as to shed light on possible strategies for addressing these issues. These immediate concerns add to more long-term anxieties regarding youth apathy and voice in political decision making (Sloam, 2013). In the 2015 UK general election there was a 35% difference in voter turnout between the 18e24 and the 55e64 age group, representing a 5% increase in the gap by comparison to the 2010 general election. The decline in youth voter turnout in general elections in the UK has been about 30% since the 1970s, bringing the differences to a level similar to that of the US and making the UK stand out vis-a-vis its European neighbours, such as Germany, in terms of the size of this gap (Sloam, 2014). Both the trends on socioeconomic background and age highlight the importance of identifying strategies to enhance the political involvement of disadvantaged youth.

Inequalities in access to learning opportunities for political engagement
The inequalities that we focus on in this article are social inequalities. Our understanding of social inequalities is informed by Bourdieu's concept of cultural capital (Bourdieu and Passeron, 1990), and we therefore use indicators of parental education and cultural artefacts for the purpose of measurement. Consequently, we understand disadvantaged youth as those whose families have below average levels of cultural capital. Social inequalities affect both the level of political engagement of parents and that of their children (Brady et al., 2015: Schlozman andBrady, 2012). The process in which these social inequalities in political engagement are transferred to the next generation is the focus of this article.
The first step towards tackling inequalities in political socialisation is to understand how they occur. There are a number of possibilities. First, we will address the issue of access to learning opportunities (Kahne and Middaugh, 2008). The issue of the importance of access is quite a straightforward hypothesis, yet notably under-researched. The people who have researched this area are Verba et al. (2003) and Niemi et al. (forthcoming) who have focused on access to learning opportunities and resources within the home environment, and demonstrated how this is crucial for learning political engagement. However, they have failed to investigate the issue of access to learning beyond the home and how social stratification within the school system may influence inequalities in political engagement. To the best of our knowledge, Kahne and Middaugh (2008) are the only researchers who have performed analysis exploring access to learning opportunities in the schools. Their research examined the IEA CIVED data showing that in the US disadvantaged youth had significantly less access to service learning opportunities and debates than students from more privileged groups. In the first step of our analysis for this article we will explore the extent to which there is differential access to political learning opportunities in schools according to the basis of students' social background.
Access to learning opportunities for political engagement takes different forms, depending on the type of learning. The forms of learning that have been shown to have an effect on political engagement and that we will examine in this article are citizenship education (Whiteley, 2012 Gainous andMartens, 2012), open classroom climate of discussion (Alivernini and Manganelli, 2011;Campbell, 2008;Ichilov, 2003Ichilov, , 2007Quintelier and Hooghe, 2012;Solhaug 2006;Torney-Purta, 2002) and political activities in school (Hoskins et al., 2012;Keating and Janmaat, 2015). As explained in detail below, the noncompulsory nature of some of these learning opportunities means that the question of access is of greater pertinence to them than to learning activities that are mandatory. Table 1 provides an overview of the learning methods and how access and mitigating effects may influence the learning of political engagement (see Table 1). We will now discuss the explanations given in the table, first for access and then for mitigating effects.
An example in England of a compulsory political learning activity is citizenship education. Definitions of citizenship education are contested but at its very basic level the subject can be understood as the study of political institutions, ideologies and allegiances, which can show marked cross-national variation as a result of different needs and priorities of national governments and international organisations (Keating et al 2009). In England the main aims for this discipline were, 'social and moral responsibility, community involvement and political literacy' (DFEE/QCA, 1998, p. 11) between 2003and 2007 (the period of study that this research covers). In these years there was a strong emphasis on both political engagement and voting, alongside volunteering and participation in the community (Kisby and Sloam 2011). As citizenship education became an obligatory part of the National Curriculum for compulsory education in 2002, all students aged 11 to 16 in England, the target population of this research, should theoretically have had access to this learning opportunity.
Nevertheless, due to the ambiguity and ad hoc nature regarding the implementation of citizenship education in schools this may not be the reality (Andrews and Mycock, 2007). The considerable school autonomy regarding the implementation of the National Curriculum, moreover, enables individual head teachers to select from a number of strategies to teach Citizenship in their schools, including teaching it as a separate subject, implementing it as a cross-curricular programme, embedding it in the modus operandi and decision-making structure of the school or implementing it in individual lessons. The design and the level of importance placed on the subject by the head teacher may well influence how much citizenship education a student receives. Nevertheless, the reported low value placed on citizenship education by teachers, parents and  (Burton et al., 2015) would suggest there is little pressure on students to engage with this subject. Taking this into account we would not expect there to be pronounced socioeconomic differences in the quantity of citizenship education received. Another form of widely reported learning of political engagement at school is an open classroom climate of discussion (Alivernini and Manganelli, 2011;Campbell, 2008;Ichilov, 2003Ichilov, , 2007Martens and Gainous 2012;Quintelier and Hooghe, 2012;Solhaug 2006;Torney-Purta, 2002). An open classroom climate is a self-reported perception of the student of their experience in a classroom and their relationship with the teacher. Students score highly on this scale when they perceive that they and their fellow students are able to introduce issues for discussion in class, feel encouraged to make up their own minds on issues, feel free to express their own opinions, feel able to disagree with the teacher, and that teachers respect their opinions. Access to this form of learning may well be differentiated as students from more educated backgrounds may have more political discussions with their parents at home and their peers in other environments, which means that they are better able to take the opportunities available in the classroom (Eckstein and Noack, 2016).
Participatory activities in schools such as debates, student councils, electing school councils, and holding mock elections have equally been found to be associated with higher levels of political engagement Janmaat, 2015: Hoskins et al., 2012). These activities are not mandatory for schools to undertake, which means that access could be an issue. In addition, within each school these activities may well not be mandatory for students to participate in, suggesting that there could be a self-selection effect for partaking in these activities. The choice to participate could also be influenced by the students' peers or by their teachers.
The second step of our research is to test the hypothesis that some forms of political learning in schools are effective in offsetting inequalities in political engagement. The mitigating effect or 'compensatory effect' (Campbell, 2008) is based on the idea that socioeconomic inequality in political engagement is transmitted from one generation to the next through political socialisation in the home (Schlozman and Brady, 2012). The premise is that disadvantaged youngsters have less access to political learning in the home and as a consequence they are then more likely to benefit from these learning experiences when offered in the school (Campbell, 2008). Therefore, access for less advantaged students to citizenship education, an open classroom climate and political activities inside school will actually reduce inequalities in political engagement as these students will be able to catch up in terms of their political learning (see Table 1).
There are a number of scholars who have identified compensatory effects (Castillo et al., 2015: Gainous and Martens, 2012: Campbell, 2008. Gainous and Martens (2012) found that citizenship education in the US mitigated the effect of social background on civic knowledge whilst Neundorf et al. (2016) found that it compensated for interest in politics but not actual levels of political participation. Campbell (2008) research in the US suggests that classroom climate had a compensatory effect on civic engagement. Nevertheless, Castillo et al (2014), using panel data based on the Chilean ICCS sample, and Persson (2015), using the Swedish sample of the IEA CIVED study, find no differential effects of classroom climate on political outcomes. Persson (2015) suggests that the difference between his results and those of Campbell was due to the fact that Sweden is a much more homogeneous country than the US. However, this would not explain the Chilean results, as Chile has one of the highest rates of inequality in the world and has large inequalities built into the education system (Castillo et al., 2015). Our research will examine if any of the three forms of learning have compensatory effects for disadvantaged young people in England.
The accelerating hypothesis introduces the alternative explanation that young people from more advantaged homes (where parents are more likely to discuss politics and be politically engaged themselves) may gain more from the political learning experiences on offer as they have the foundation of knowledge, skills, attitudes and values from which to build on (Campbell, 2008) and the familiarity with these experiences. The only research that we have been able to identify that finds an accelerated effect is Hooghe and Dassonneville (2013) who found that students who had a higher level of political knowledge gained more from civic education classes in terms of enhanced political engagement. As this field of political mobility is quite under-researched we did not want to discount the possibility that other forms of political learning in schools could, potentially, have such exacerbating effects on inequalities in political engagement so we include this scenario as the alternative hypothesis.
For the purpose of clarity in this section we will draw together the hypotheses that will be tested in this article.

Access to learning political engagement at school
Hypothesis 1 is that young people from more disadvantaged backgrounds have less access to the learning for political engagement in the form of political activities at school and open classroom climate.
Hypothesis 2 is that social background will not matter regarding access to citizenship education as a learning activity for political engagement.

Mitigation effects
Hypothesis 3 is that young people from disadvantaged backgrounds are likely to gain more from political learning experiences at school in terms of the effects of these experiences on political engagement than young people from more privileged backgrounds.
Hypothesis 4 (the alternative hypothesis) is that young people from more advantaged backgrounds gain more from political learning experiences at school. These hypotheses will be tested by analysing the Citizenship Education Longitudinal Study (CELS) data with the methods described below.

Data
We used CELS as the data source for this article. The CELS dataset is unique in that it combines a panel study of adolescents with a questionnaire tapping a wide range of civic attitudes. The study includes data from a cohort of young people aged 11 and 12 (Year 7; first year of secondary school) when they were surveyed for the first time in 2003. This cohort was then surveyed every two years until 2014 (Round 6). The data was collected from a nationally representative sample of 112 state maintained schools in England e representative in terms of region, GCSE attainment and percentage of students on free school meals (Keating and Benton, 2013). Within each school all the students of target grade were selected. We used the data from Rounds 1, 2 and 3, when pupils were aged 11/12, 13/14 and 15/16, respectively. At the oldest age group, the young people are students in Year 11, which is the final year of lower secondary education and the last year of citizenship education as a statutory component of the curriculum.
The CELS dataset like other longitudinal datasets has considerable attrition. While the study started out with 18,583 respondents in Round 1, only 6155 of this original sample (33%) took part in Round 3. These 6155 respondents constitute our analytical sample. Following the method adopted by Eckstein et al. (2012), we compared respondents who only participated in the first round with those who took part in both rounds in order to test whether this attrition was selective. Significant attrition effects were found for the Round 1 variables Books at home and Parental educational attainment, with respondents participating in both rounds reporting significantly higher numbers of books and higher levels of parental education. To prevent this selective attrition from skewing our results we included a combined measure of Books at home and Parental education in the model (see Table 2). These two variables are used in this work in a combined index as the measure for family's socio-economic status (SES), thus representing key explanatory factors in the prediction of the outcome, i.e., the students' intention to vote (see next section for a thorough description of the measures used in the present study). It is unlikely there will be any remaining attrition bias if the variables associated with the attrition are included in the analysis as the key independent or control variables (cf. Little, 2013;Paterson, 2013). Selective attrition, however, also occurred on voting intentions as the dependent variable of this study. Students participating in Rounds 1 and 3 showed significantly higher levels of voting intentions than those who only participated in Round 1. This problem cannot be remedied. Caution is therefore required as to the national representativeness of the sample. Reassuringly, however, this selective attrition has not resulted in a loss of variation in intentions to vote. In Round 3 there were as many students saying they would probably or definitely not vote in the future (23.7%) as there were saying they would definitely do so (23%). Thus, it is certainly not the case that only the engaged students have continued to participate.
In addition to unit non-response (i.e. attrition) there is item non-response on the variables included in the model. As Table  2 shows, there is a range of missing responses. For 5 variables there are 5% or less missing responses, for 4 variables there are less than 25% missing responses, for another 4 variables there are between 25 and 50% missing responses. The highest nonresponse was for political activities in school in Round 2 with half the responses missing. In order to prevent further data loss due to item non-response, we imputed data using the Bayesian estimation technique of multiple imputation (MI) in Mplus (see Muth en andMuth en, 1998-2012). This procedure has the advantage of using all the variables in the analytical model (both independent and dependent ones) to impute the missing values and generate a defined number, 10 in our case, of imputed datasets. The final parameter estimates are then obtained from the averages of the parameter estimates over the 10 imputed datasets. MI is able to address non-random missingness as it is based on the observed values for a given individual and on the relations observed in the data for other respondents (cf. Schafer and Graham, 2002). MI occurred at the modelling stage as an integral part of the analyses we carried out with Mplus (see further below).

Dependent variable, intention to vote
Voting intention is the outcome on which this study focuses. Behavioural intentions of young people have been widely used to offer an insight into future political behaviour (Hooghe and Stolle, 2004;Hooghe and Wilkenfeld, 2008), and theories of political psychology would suggest that young people's early development of political attitudes and behaviours have persistent lifelong effects (Sears and Levy 2003). In addition, empirical evidence shows that there is an association between voting intention and actual voting (Achen and Blais, 2010), and recent longitudinal research has found youth intention using the same data source as our study prove to be a strong determinant of actual levels of adult voting (Whiteley, 2012). Furthermore, voting is an activity which has fewer demands in terms of resources with respect to time, money or education than, for example, contacting a politician (Verba et al., 2003) and, at least in theory, should be more open to wider participation.
Voting intentions were tapped with item asking respondents how likely they would be to vote in general elections in the future. This item was asked in Rounds 1e3. Respondents could state their answers on a Likert scale with the following categories: Definitely not do this; Probably not do this; Probably do this; Definitely do this. In all 3 rounds a solid majority of the students opted for the final two more positive responses. Students appeared to be slightly more engaged in the last round (see Table 3).

Key independent variables
The literature review identified three sources of learning within the school environment that supported young people towards becoming politically engaged citizens. They are the volume of citizenship education, participation in political activities in school and an open classroom climate for discussion. The first of these was measured with an item asking respondents about the amount of citizenship education they had received (not at all; a little; a lot). As noted before, citizenship education is mandatory in lower secondary schools in England, but this does not mean that it is taught in a uniform manner. Schools are free to decide on the content, mode of delivery and volume of citizenship education and vary substantially in these aspects, accordingly (Kerr et al., 2007). A scale was created to gauge participation in school-based political activities. This scale represents the sum of the responses to four items asking students whether they have participated in debates, the student council, elections for the school, or in mock elections in the last year. It has a minimum of 0 (not participated in any clubs or events) and a maximum of 4 (participated in all clubs and events). The third school based source of learning, open classroom climate, was measured with a scale consisting of six items about freedom of expression, open discussions and teachers' facilitation of this (see Table 1 for the wording of the items). The scale showed a sound level of internal coherence (Cronbach's alpha ¼ 0.811) and represents the saved output of a factor analysis (i.e. the factor scores). These variables were constructed in the same way for each of the three rounds of data.
The temporal ordering of the learning activities in relationship to the dependent variable of voting intentions uses the following logic; we take the students' stated prior experience of political learning over the last year (R0-1) and then this is related to their future attitudes towards voting (dependent variable R1). This is repeated for each year of the study. Voting intentions from round 1 are then related to voting intentions in round 2 and voting intentions in round 2 are subsequently related to voting intentions in round 3. The inclusion of prior measures of the outcomes (voting intentions) into each step of the model allows one to assess the degree of autocorrelation in the outcomes of interest, an important step in ensuring that the effect of school-based learning sources, such as citizenship education and open climate, on the engagement outcomes is genuine and does not reflect a prior propensity of individuals to be engaged. Controlling for these initial levels of engagement thus enables us to assess the extent that school-based learning sources have contributed to changes in engagement from the moment when pupils entered secondary schools at age 11 (cf. Finkel, 1995;Kahne and Middaugh, 2008). It thus allows us to more accurately assess the influence of such sources during lower secondary.
It could be expected that some of the sources of learning would be interrelated. If, for instance, open discussions mainly take place during citizenship education lessons because the teacher deems debate to be important for the formation of civic competences, then the first and third source of learning will be strongly linked. Similarly, it could be anticipated that schools offering many opportunities to take part in political activities in school are also more inclined to create space for open discussions than schools that do not offer such opportunities. We checked these interlinkages by examining the degree of multicollinearity in regression models. None of these relationships exceeded critical multicollinearity levels.
As highlighted above, we consider social background to be a key influence on political engagement both in a direct and indirect sense. Indirectly, social background shapes political engagement by determining access to learning sources. We measure social background with a scale based on the number of books at home and parental educational attainment. This measure focuses on Bourdieu's concept of cultural capital that may be understood as an individual's gain from parental level of education and cultural artefacts such as books (Bourdieu and Passeron, 1990). The number of books at home has also been found to be strongly correlated to parental income (Schuetz et al., 2008;Baird, 2012), making it a reasonable replacement for parental income and occupation considering that more obvious measures of social background, are not available within the CELS dataset (Keating and Benton, 2013).
Finally for both the regression analysis and the latent growth curve modelling we have added controls for gender and ethnic identity (White British; other) as these background characteristics have been found to be related to political engagement by previous research (e.g. Uhlaner et al., 1989;Hooghe and Stolle, 2004).

Access
To measure access to learning sources we analyse the links between students' social background and the three learning sources using linear regression in Mplus. Due to the clustered structure of the sample (respondents in grades; grades in schools), the observations are not independent. We accounted for this in our analysis in Mplus by selecting the robust maximum likelihood estimator and using the school grade variable to indicate the clustering of the data (cf. Muthen and Muthen 2009).

Mitigating effects
We then proceeded to use latent growth curve modelling (LGCM) to evaluate the mitigating effects of each learning method for reducing inequalities in political engagement.
LGCMs enable the analysis of how predictor variables are related to both initial levels and rate of change in the outcome variable over time. In our case, it allows us to investigate whether the three learning sources are associated not only to higher initial levels of political engagement at age 11/12 but also to a steeper growth in political engagement between ages 11 and 16, over and above the effect of SES. Indeed, it enables us to explore possible compensatory effects of these learning sources in a dynamic way: is it the case that students of low SES families show a steeper growth in political engagement than their peers from more privileged backgrounds when they experience high levels of these learning sources? In other words, are they able to catch up with their more privileged peers when they receive this socialisation input? OLS regression, even when autoregressive parameters are introduced to account for serial correlations and change over time, only allows estimation of population-average parameters, whilst the advantage offered by a LGCM over these models is on individual-level estimates of change, that is, on individual trajectories of change and on interindividual difference on these trajectories. Then, the addition of time-invariant (Gender, Ethnic group, and SES in our case), and of time-varying covariates aims to explain the inter-individual differences/variability in both the initial levels (Intercept) and the rate of change (Slope) of the outcome Intention to vote across the three measurement points (Bollen and Curran, 2006;Little, 2013;McArdle et al., 2009).
Hence we first estimated the intercept and slope for the dependent variable of voting intentions, based on the three available measurement occasions in Rounds 1, 2 and 3. This model represents individual trajectories of voting intentions over three rounds as a function of random intercepts and random slopes. Once this unconditional growth model had been estimated, we proceeded with the conditional LGCM by introducing both time-invariant and time-varying covariates (respectively, TICs and TVCs) in our model in a stepwise fashion. We consider gender, ethnic background and social background to be TICs, and citizenship education, open climate and political activities in school to be TVCs, as we hypothesise that the three sources of learning change over time. 1 While TICs can be directly modelled on the slope, it is more appropriate to relate the TVCs to the outcome measured at the same point in time (e.g. link Round 3 citizenship education to Round 3 voting intentions) (Bollen and Curran, 2006). Our stepwise approach is as follows: first, we added voting intentions of Rounds 1 and 2 to the Round 2 and 3 models of voting intentions, respectively, to take autocorrelation in the dependent variable into account; second, we added the time invariant control variables (gender, ethnicity and social background) to the analysis on the intercept and the slope. The next step was to add the variables of interest, the learning sources (citizenship education, political activities and open classroom climate) and the interaction between the learning sources and social background to the model to evaluate the hypothesis regarding the mitigation of inequalities in political engagement. Fig. 1 below represents the model of political socialisation of voting intention in adolescence. It is a graphic representation of the analysis of Model 1 in Table 5.
We performed the analysis using a step by step process of model construction to ensure the development of a parsimonious and reliable model. To evaluate the fit of the different models to the data we have used the threshold values suggested by Hu and Bentler (1999) for the comparative fit index (CFI), the Tucker-Lewis index (TLI), the desirable values of which are above 0.950, as well as the root mean square error of approximation (RMSEA), which is expected to be below 0.06. In this article, in order to remain concise, we focus on reporting the fit and results of model 1 as this model has the better fit and its enables us to test our hypothesis on the mitigating effects of the learning methods. For the full results and model fit for each of the prior steps please contact the authors. Lastly, to test for reciprocal links between the outcome and the predictors, we included additional paths between voting intentions of the previous round and the three learning methods in the subsequent rounds (see Model 2 in Table 5). 2 6. Results

Access
The results of the regression analysis (Table 4) show that for each of the 3 rounds there are no significant differences in access to citizenship education by social background. The R 2 confirms these results with a very small explained variance in the model at less than 1%. This finding is in accordance with Hypothesis 2. In contrast, political activities inside school are significantly related to social economic background for each round and the strength of the coefficient increases in size across the three rounds. Although the percentages of explained variance in all the models and for all the learning activities are small (perhaps because of small number of predictors included in each of the models), those relating to the models for political activities in school are highest going up to 3% in Round 3, indicating a strong and increasing effect of social background on participating in political engagement in schools. It should be noted here that social background was the only variable that was significantly associated with political engagement in this model. Classroom climate was not measured in Round 1 of the study. In Rounds 2 and 3 social background has a significant effect on access, and this time the coefficient remains fairly constant over these two time points and the R 2 remains stable at about 2% of the explained variance. The findings on political activities in school and classroom climate are thus in line with Hypothesis 1 (see Table 4).
From the controls, ethnicity does not appear to be a factor in accessing any of the three learning sources of political engagement in schools at any point of time. In contrast, gender, appears to have some influence regarding access. For citizenship education and classroom climate, female students appear to have a small but significantly greater access to both these forms of learning. For political activities in school in Round 1 of the study, boys appear to be able to access this form of learning more than girls, nevertheless, gender stops being significant in the later rounds. These gender differences are complex to explain. One explanation for higher quantities of citizenship education for female students could be that social science subjects are less frequently taken up by male students (Whitehead, 1996) and access to citizenship education is not perceived as competitive advantage in the labour market (Burton et al., 2015). Greater access to open classroom climate could be due to the fact that females in general are better able to access this experience as their behaviour could be interpreted by teachers as being less aggressive, and therefore more acceptable in the class. Finally regarding political activities, male *P < 0.05; **P < 0.01; ***P < 0.001. **Gender 1 ¼ male and 2 ¼ female.
2 No diagram of Model 2 is given because it is not possible to display the many relations of this model in a graphically transparent manner. students could prefer the competitive forms of learning political activities in school such as debates and class councils. These results suggest the need for further research on gender and access to political learning.

Mitigating effects
First of all we note the model fit indices which show an acceptable degree of fit for Model 1 (RMSEA ¼ 0.04; CFI ¼ 0.914) (see Table 6). This means that the model represents the observed data reasonably well. The fit indices are less favourable for Model 2, i.e. the model that allows us to explore reciprocal effects between the outcome and the predictors (see further below). As the fit indices of Model 1 are better, we focus on this model in reporting the results of the LGCM analyses.
Looking at the time invariant factors, we observe that social background matters for the initial level of voting intentions but not for the subsequent change in these intentions (see Model 1 in Table 5). This suggests that family SES primarily shapes political engagement during early childhood. Once children have entered puberty, engagement gaps between social groups appear to stay the same. Interestingly, neither gender nor ethnic identities have an impact on either initial levels or the ensuing change in voting intentions.
Regarding the time variant factors, we can first observe that Round 1 voting intentions has a positive and significant effect on Round 2 voting intentions and that Round 2 voting intentions in turn is a powerful positive predictor of Round 3 voting intentions. This does not come as a surprise, as one might expect some continuity in the outcome over time. What happens to the effects of the learning sources if we take this autocorrelation in the outcome into account? We see, in fact, that all three of them have positive effects on young people's intention to vote in each of the three rounds of data collection (with the exception of the nonsignificant link between Round 1 political activities in school and Round 1 intentions to vote). In other words, citizenship education, an open climate of classroom discussion and participation in political activities in school are all likely to help in making young people more politically engaged, irrespective of previous levels of political engagement. The latter two learning sources appear to have the strongest effects, judging by the standardized coefficients. For instance, an increase of one standard deviation in Round 3 open climate leads on average to an increase of 0.11 standard deviation in Round 3 voting intentions, while the corresponding figure for Round 3 citizenship education is only 0.05 (Model 1 in Table 5).
Our interest in this article, however, is in the question whether these learning sources help in making young people from disadvantaged backgrounds catch up with their more privileged peers in political engagement or whether they only enhance the engagement gap across social groups. This question can be addressed by looking at the interaction effects of SES with each of the three learning sources. The only interaction effect that is significant is in Round 3 where the interaction between SES and citizenship education has a significant effect on Round 3 voting intentions. This effect is negative, meaning that the greater the amount of citizenship education is (as reported by students), the smaller are the differences between social groups in their intentions to vote. Citizenship education would thus indeed appear to be able to compensate for the disengaging effect of social deprivation. Thus, for this learning activity Hypothesis 3 is confirmed. Interestingly, the interaction effect also has a clear impact on the slope as the amount of explained variance in the slope falls considerably once it is removed from the model (from 5.9% to 3.3%). 3 This suggests that among those who reported having received a lot of citizenship education the social gap in political engagement has become smaller over time. This gap has narrowed significantly more than for those who reported less exposure to citizenship education. As said, the other interaction effects are not significant. Hence, while the other learning sources may be good in enhancing political engagement in general, they do not seem to be able to reduce social gaps in engagement.
Finally, in order to assess the possibility of reverse causation between predictors and the outcome (cf Keating and Benton, 2013;Keating and Janmaat, 2015), we added relationships between voting intentions of the previous round to learning sources of the subsequent round (see Model 2). This is important as we need to know how dependent the three learning sources are on prior political engagement. If those who are already engaged disproportionately participate in the learning sources, these sources will, in all likelihood, only exacerbate inequalities in political engagement. In addition to prior voting intentions, gender, ethnicity, social background and prior learning sources are added as predictors of the three learning sources (see Model 2).
We can firstly see that none of the effects shown in Model 1 change when these relationships are added to the analysis. Citizenship education is still showing the compensatory effect for socially disadvantaged students. In terms of the added relationships, we can observe a positive link between voting intentions in Rounds 1 and 2 and subsequent levels of political activities and perceptions of an open classroom climate. Thus these learning sources are affected by reverse causation: i.e. students who begin with higher levels of intended political engagement are reporting higher levels of participation in political activities and open classroom climate. However, the amount of citizenship education reported by students was not associated with the prior levels of voting intentions in either round suggesting that this predictor is a genuinely exogenous force. These patterns make good sense if we consider that participating in the first two learning sources is mostly voluntary, while experiencing citizenship education is mandatory. The voluntary nature of the former provides the already engaged with a good opportunity to self-select into these learning sources while the compulsory nature of citizenship education prevents the disengaged students from avoiding further participation.

Discussion and conclusion
In England in recent years schools have been placed at the centre of research and policies for enhancing socioeconomic mobility in the labour market but at the same time little attention has been placed on the role of school in reducing socioeconomic inequalities in political engagement. In this study we have attempted to fill this gap by investigating differences by social background in access towards political learning at school and the effectiveness of specific methods for political learning in mitigating differences by social background in political engagement. It is necessary to take into account the limitations of our study firstly in terms of the attrition in the longitudinal data that has to some degree affected the representativeness of the sample, and therefore the degree to which the results can be generalised to the whole population of English students. A second issue in the data is the self-reported nature of the learning experiences, which implies a certain level of subjectivity in students' attempts to recall their political learning in school. Nevertheless, the CELS data reflects the best possible resource in the UK for examining this topic.
The results of our analyses show that political activities in school and open classroom climate enhance political engagement in general, but in contrast to Campbell (2007) in the US and Eckstein and Noack (2016) in Germany, we do not find that these methods are effective in reducing social inequalities in political engagement. In fact, the situation for young people from disadvantaged backgrounds appears to be unfavourable as the regression analysis revealed disadvantaged youth to have significantly less access to these highly important ways of learning political engagement compared with their more advantaged peers. This access, moreover, was influenced by social background in every round that these methods of learning were measured. 4 There is no justifiable reason why young people from different social backgrounds should have different levels of access to these forms of learning during their compulsory years of education. Perhaps it is this uneven access that can explain why open classroom climate and political activities are not able to reduce inequalities in political engagement. In other words, there may be too few children of low SES backgrounds experiencing these learning sources for these sources to have a mitigating effect.
One reason for different levels of access to political activities in school is that these opportunities are likely to be voluntary. For example, participating in debates or students councils may be a choice and be influenced by an individual's existing levels of political efficacy, political skills and interest. These qualities could already be higher amongst students from more privileged social backgrounds as these aspects of competences could be cultivated within the home environment. Indeed, when we explored reciprocal effects between learning sources and voting intentions we found political activities and perceptions of an open climate of discussion to be dependent on prior levels of voting intentions. In addition, adolescents' participation levels could equally be influenced by the interest in politics of the school that they attend, their peers and their teachers, which can all be influenced by social background. There is clearly a case for making political activities at school compulsory  for all students, across all classes and all schools. In addition, getting teachers to encourage students from less advantaged backgrounds to stand for positions in school councils could also play an important role in reducing political inequalities. As open classroom climate is a learning process rather than a specific activity, it is perhaps more complicated to ensure access for all students. Again the home environment could be important. Bernstein (2003) seminal research has shown that middle class students tend to be brought up in a family environment where open debate is the norm and young people's autonomy in decision making and language skills are enhanced. As a result these students may well be better placed to participate in open discussions of political issues in the classroom regardless of the skills of the teacher or the methods applied. We would suggest that these more advantaged students may well have learnt the rules of the game, i.e. how a young person makes their disagreement on an issue known with the teacher and how they express their opinions will inevitably shape the response of the teacher. It could well be that there is a miscommunication between disadvantaged students and middle class teachers in the school context. Loud and passionate disagreements between disadvantaged students or students' abrupt challenges aimed at the teacher have been found to be interpreted by both the teacher and their fellow more advantaged students as confrontational and disruptive to the class rather than an opportunity for student voice and student learning (Hooks 1994). Adding to this, Ratcliff et al. (2010) found that teachers who had difficulty with students that they identified as 'troublesome' in class asked these students less questions and tended to involve them less in discussions of subject matter. Finally, students from working class backgrounds have also been identified as feeling anxious and not fitting in with the context of classroom discussions and debates (Brookfield and Preskill, 2012). Thus young people from less privileged social backgrounds may feel less able to have their voice heard. Better training of teachers to support open discussion across the whole curricular and school environment for all children regardless of their social background could be one method for enhancing access to this form of learning. Another approach would be to raise awareness with parents and carers regarding the importance of the role of family in developing political engagement from a very early age onwards. This could include informing them how they can encourage these activities by engaging in political activities with them and how to involve their children in decision making in the home. This would need to be undertaken for all parents in order to avoid stigmatising parents from disadvantaged homes.
Our findings on citizenship education, on the other hand, diverge markedly from those on open classroom climate and political activities. Despite the recorded weaknesses in the implementation of citizenship education in schools in England (Andrews and Mycock, 2007), there was no evidence to suggest unequal access by social background or prior levels of political engagement. We also found that citizenship education is beneficial for learning political engagement in every year studied. Most importantly, however, we found citizenship education to be particularly effective in enhancing the political engagement levels of those from lower social background in England at aged 15e16, indicating that it has the ability to reduce social disparities in political engagement.
That we see such an important effect of citizenship education at this age is not surprising as mid to late adolescence has been identified as a key formative period for political engagement (Flanagan and Sherrod, 1998). At this age young people begin to take up an interest in societal affairs after having explored family relations and parental authority in their early teens (Keniston, 1968). Further research that explores whether these compensatory learning processes could also be successful if citizenship education was made compulsory post-16 would also be a useful next step. In England education is now compulsory until the age of 18. If citizenship education could also be made statutory until that age, it would offer the prospect of further reducing social gaps in political participation.
At this time citizenship education is compulsory in England between the ages 11e16 (the period that this study observes). We provide evidence that a mandatory approach is needed in order to maintain equal access for different social groups, as non-compulsory activities like political activities in school are taken up mainly by more privileged students who are more politically aware and engaged in the first place. In finding that compulsory approaches not only offer more equal access but also result in more egalitarian outcomes than those relying on voluntary participation, our study is fully in line with Boudon's (1974) observation that the more choice and 'branching points' a system has the more social class differences are reproduced. In this sense our findings clearly illustrate the tension between freedom of choice and equality, which is a key theme within sociology of education more generally.
The promising finding regarding citizenship education builds on the research from Gainous and Martens (2012) who found similar results in the US and who then proposed on the basis of their research that only disadvantaged students should take compulsory citizenship classes. In England citizenship education could be targeted at disadvantaged students, for example, within vocational education and training courses that in general include a higher intake of socially disadvantaged students. Research on the mitigating effects of citizenship education as part of vocational education and training would be another angle for future research. Nevertheless, we would not argue that citizenship education should only be for less advantaged students as not all more privileged young people are socialised into political engagement in the home. We suggest that compulsory citizenship education until 18 and ensuring the availability of quality citizenship education in vocational education and training are promising policy directions to support young people from more disadvantaged backgrounds to gain a greater political voice in the democratic system.