The Changing Impact of Family Background on Political Engagement During Adolescence and Early Adulthood

his paper examines the development of the impact of family background on young people’s political engagement during adolescence and early adulthood in order to test a number of hypotheses derived from the impressionable years and family socialization perspectives. The study analyses data of the British Household Panel Study and Understanding Society to assess these hypotheses. Political interest and voting intentions are used as outcomes of political engagement. The study ﬁnds parental education to have no effect on initial levels of these outcomes at age 11 but to


Introduction
The impact of family background on political engagement has not received the same amount of scholarly attention as the transmission of social status from parents to children (Brady et al. 2015). This relative dearth of attention is surprising as parental socio-economic status (SES) (as indicator of social background) and parental civic and political engagement have been found to be the strongest predictors of young people's political engagement (Beck and Jennings 1982). From the point of view of an inclusive democracy, the intergenerational reproduction of political engagement and the resulting inequality deserve more consideration, given that democratic decision-making becomes more elitist and skewed towards the interest of the engaged and privileged when large sections of the population abstain from participating on a permanent basis (Bartels 2008;Levinson 2010).
Interestingly, this issue should be of particular concern to established western democracies as recent research found the impact of parental SES on the electoral participation of young adults to be the strongest in the United Kingdom, Germany, and Austria among a group of thirty-three European states (Hoskins and Janmaat 2019). Evidently, democracies do not necessarily become more socially inclusive as they age. Indeed, these persistent social inequalities are one of the most intractable problems that mature democracies face because of the catch twenty-two situation that the poor find themselves in: they lack the required resources and skills to fully participate in a democracy but only full participation can ensure that they achieve those resources and skills (Pande 2020).
Existing research on the intergenerational transmission of political engagement is mostly confined to the United States. This research found both parental SES and parental political engagement to show strong links with children's political participation and party preferences (Beck and Jennings 1982;Verba et al. 2005;Jennings et al. 2009), findings that were echoed in the few studies that have examined these forms of intergenerational reproduction in European countries (e.g. Kroh and Selb 2009;Neundorf et al. 2013;Lahtinen et al. 2019).
However, these studies have not explored in detail how the effects of parental SES and political engagement change among teenagers. Jennings et al. (2009) examined the effects of parental political involvement on the political involvement of their children at ages 18 and 26, but this design obviously did not allow them to conclude anything about the development of political engagement during adolescence and the parental imprint on that. Neundorf et al. (2013) assessed the dynamics of parental influences on respondent's political interest between ages 17 and 35 and found that parental characteristics explain levels of political interest at age 17 but are largely unrelated to changes in political interest thereafter. Their data source, the German Socio-Economic Panel study, did not allow them to examine earlier ages.
A fine-grained analysis of the dynamics of parental influences from an early age enables us to identify the moments in the life cycle when these influences wax, wane, or stabilize. One may expect such influences to show considerable variability during adolescence and early adulthood as these life stages have been identified as the "impressionable years" regarding political engagement (Jennings 1979;Kinder 2006). In other words, during these life stages young people develop their political preferences, dispositions, and identities and are particularly receptive to all kinds of influences including those of their parents.
Establishing the moments when parental influences change may, moreover, allow us to provisionally identify the mechanisms through which parental characteristics, such as SES and political involvement, influence young people's political engagement. The next section discusses these possible mechanisms in greater detail. In short, the current study examines how the impact of parental SES and political engagement on young people's political engagement changes during adolescence and early adulthood. Empirically it will contribute to the literature on political socialization by offering a detailed analysis of these dynamics; theoretically it will do so by proposing particular mechanisms through which parents influence their offspring's political engagement.
It explores this question with data from the British Household Panel Survey (BHPS) and its successor Understanding Society (USoc). This data source has two advantages. First, it allows us to assess the development of political engagement over the life course from early adolescence (age 11) into adulthood. Consequently, we will select respondents who took part in the BHPS as teenagers in the 1990s and early 2000s and track them to the latest wave of USoc (Wave 9 of 2017). Second, it gives us the possibility to construct measures of parental SES and political engagement on the basis of parental rather than children's reports. The latter presents a considerable advantage over studies that have to rely on children's accounts of their parents' education and occupation since such accounts are often inaccurate and marred by high levels of non-response.

Perspectives on the Dynamics of Parental Influences
When examining the changing impact of parental characteristics during the early years, two perspectives are relevant in the literature on the development of political engagement across the life course. In the "impressionable years" perspective only experiences early in life matter for political engagement as this is the life stage when political engagement is still fluid (Prior 2010). As political attitudes and dispositions start to stabilize among young adults, experiences and life changes from mid adulthood onwards are unlikely to have much of an effect. Although there is no agreement among scholars broadly endorsing this perspective as to when exactly attitudes and dispositions start to stabilize (Neundorf and Smets 2017), with some highlighting the period between ages 17 and 25 (e.g. Jennings and Niemi 1981) and others identifying ages 7-17 as a crucial formative period (e.g. Bartels and Jackman 2014), these scholars would not expect political engagement to change much after the mid-twenties. The relevance for the impact of parental attributes-and that of other conditions relating to people's early years-is that this impact is likely to be volatile during these crucial formative years (however defined) and stable thereafter (Neundorf et al. 2013).
With respect to parental SES, as one of these attributes, status transmission theory points to the crucial role that parental educational attainment, as a core aspect of SES, plays in sustaining inequalities in political engagement (Gidengil et al. 2016). According to this theory, parental education primarily exerts an indirect effect on political engagement by influencing children's level of education. It is the latter that then shapes engagement profoundly: numerous studies have argued that more educated people are more engaged because they have acquired a greater understanding of the world of politics, more opportunities to develop civic skills through their careers, a greater self-confidence to participate in politics, and a stronger belief in engagement as a norm of good citizenship (Verba et al. 1995;Nie et al. 1996;Lewis-Beck et al. 2008;Persson 2015). In other words, it is through the intergenerational transmission of educational attainment that parental SES shapes political engagement. The relevance of this theory for the dynamics of the impact of parental SES is that this impact is likely to increase in late adolescence and early adulthood as meaningful differences in educational attainment start to emerge only in these life stages. Before the tracked phase of upper secondary (with tracks differing by level and type), adolescents all study at the same level because lower secondary is, at least in theory, comprehensive in the whole of the United Kingdom (Green et al. 2006).
However, parental SES does not only shape offspring's educational attainment. Through their social status parents also influence the choice of school for their children, their children's choice of school friends, and their children's choice of school subjects (Jennings and Niemi 1974;Henderson et al. 2018), leading to very different educational experiences of young people already in lower secondary. In turn, aspects of the school, such as its social and ethnic composition, its curriculum, its pedagogical approach, and the degree in which it offers students a voice in school affairs, have been shown to be closely related political interest, intentions to vote, and other dimensions of political engagement (Torney-Purta 2002;Campbell 2007;Campbell 2008;Hoskins et al. 2012;Kahne et al. 2013;Geboers et al. 2013;Hoskins and Janmaat 2019). Particularly in England, where school choice is guaranteed, where schools have been given more autonomy since the academization reform and where a small but influential private school sector contributes to educational segregation (Green et al. 2006;Janmaat 2018), one would expect that schools vary greatly in these aspects of education and therefore that these aspects are important in shaping political engagement. Parental SES could also have a more subtle indirect impact by giving children the skills to exploit the learning opportunities offered in schools. Middle class parents are often associated with negotiation households in which children are encouraged to express their opinions, make independent decisions, and provide reasons for their preferences (Calarco 2018). In turn, the dispositions and skills learned in these interactions are not only beneficial for school performance but also for making the most of the student voice opportunities in the school environment, such as classroom discussions and school councils (Hoskins and Janmaat 2019). To sum up, if parental SES influences political engagement through these other educational mechanisms, one might expect its effect to grow already during lower secondary and therefore in early adolescence.
Yet, there are also reasons to surmise that the stamp of parents on their offspring diminishes during the impressionable years. According to Jennings et al. (2009, p 787), young people are characterized by "labileness," i.e. indecisiveness and capriciousness linked to major choices they have to make concerning education, partner, and career. These preoccupations consume so much energy that there is little space left to be involved in politics (Glenn and Grimes 1968). They may be so salient that they overwhelm the impact of family background features that are associated with early childhood socialization and learning, such as parental political engagement (see further below). Indeed, Jennings et al. (2009) found the parent-child correspondence on a range of political engagement indicators to actually decline between ages 18 and 25.
Based on these reflections, we propose the following hypotheses with respect to the "impressionable years" thesis: 1.a. The impact of parental SES, and particularly that of parental education, on political engagement will grow during adolescence and early adulthood and stabilize afterwards; 1.b. The impact of parental political engagement on that of their children will start high, then diminish in adolescence and early adulthood and stabilize afterwards.
We note here that these hypotheses are not necessarily contradictory as parental SES and parental political engagement are likely to influence the development of children's future political engagement at different points of time, as explained above.
We now turn to the school of thought that attributes a vital role to parents in early childhood, which we dub "the family socialization thesis." Although this perspective may be seen as an extreme version of the impressionable years perspective (and therefore part of it) we think it is useful to identify it as a separate perspective since it concentrates on the socialization and learning within the family at very early ages. Of relevance here is the tenet of social learning theory that parents are the source of direct modelling, instructions, cue giving, and reinforcement processes that are particularly effective towards the learning of political engagement (Jennings 2007;Gidengil et al. 2016). By cultivating norms and values and by acting as role models parents can pass their preferences and behaviours onto their children (Kam and Palmer 2008). Social learning theory further suggests that parental SES does not only influence political engagement through the education pathway (as explained above) but also in a more direct manner (Gidengil et al. 2016). The specific child-rearing practices and expectations of middle class parents, for instance, have been argued to be conducive for political engagement (Beck and Jennings 1982). Such parents introduce their children to social circles that encourage involvement and promote it as a norm of good citizenship (Dalton 1982). However, the theory accords a greater role to parental political engagement as a conduit of parental influence because it offers more possibilities for both socialization (i.e. cultivating engagement as a commendable disposition and norm of good citizenship) and emulation (i.e. children imitating their politically active parents) (Beck and Jennings 1982;Gidengil et al. 2016). As socialization and emulation are transmission processes that are happening from early childhood onwards (Cohen 1987), we should expect to see strong parent-child linkages already in early adolescence.
Another reason to identify family socialization as a separate perspective is that it has given rise to the influential education-as-proxy argument. Scholars supporting this argument contend that the "effect" of education can wholly be accounted for by family background characteristics that determine both educational attainment and political engagement (e.g. Kam and Palmer 2008;Burden 2009;Persson 2014). Thus, these scholars attribute a decisive impact to parental characteristics and one that should be manifested early in life. We derive the following hypothesis from the family socialization perspective: 2. Both parental SES and parental political engagement already have a distinct impact in early adolescence, with the latter showing a stronger effect.
We consider the hypotheses from the family socialization and impressionable years perspectives to be compatible. In other words, that parental characteristics already have an impact in early adolescence (Hypothesis 2) does not rule out that their influence changes during adolescence and early adulthood (Hypotheses 1a and 1b).
We further note that we will not be testing the mechanisms through which the parental characteristics are said to influence political engagement since the data source (BHPS/USoc) has insufficient information on these mechanisms. For instance, the database does not include data on the schools the respondents have attended, on their peer groups, on the subjects they have chosen, or on the learning processes they have experienced at school. Neither does it include information on practices that are known to support the learning of political engagement such as an open climate of classroom discussions (Campbell 2008;Hoskins et al. 2012;Knowles et al. 2018), school councils, or alternative involvement in school decision-making (Hoskins et al. 2012;Keating and Janmaat 2016;Hoskins et al. 2021). Equally the learning experiences and socialization within the family are not sufficiently addressed with useful variables missing such as political discussions in the home and parental political knowledge (see McIntosh et al. 2007 for how these variables influence the learning of political engagement).
Finally, we note that the life stages we focus on (adolescence and early adulthood) do not allow us to engage with the "lifelong openness" perspective, which is often presented as antithetical to the impressionable years one. Unlike the latter it sees political engagement as malleable throughout the adult years and subject to key life cycle events and experiences such as getting married, starting a family, buying a home, and career progression ( van Deth 1989;Tyler and Schuller 1991; see also the discussion in Prior 2010 andNeundorf et al. 2013). We cannot test this perspective properly as these events usually occur in mid-adulthood, which is beyond the age range of this study.
So far we have used political engagement as a container term capturing not only actual participation, but also the motivation, willingness, and perceived ability to participate. In other words, in terms of familiar political outcomes, it can be said to include electoral and other forms of participation, interest in politics, the intention to participate, and political efficacy (Solt 2008). In this paper we will focus on political interest and the intention to vote for a particular party as two dimensions of political engagement. Interest in politics is worthy of more scholarly attention as it is strongly linked to all forms of political participation, conventional, alternative, and influential ones (Finkel 2002;Prior 2010). Indeed, Keating and Melis (2017) found it be more strongly related to online political activity than socio-demographic characteristics. They explicitly called for more research on the evolution of (social inequalities in) political interest during adolescence and early adulthood, which is precisely the aim of this paper. Voting intentions are also worth examining. Although such intentions nearly always overestimate actual levels of voting, they are strongly correlated with validated voting (Achen and Blais 2010), suggesting that the voting intentions of adolescents can be an important predictor of actual electoral participation in adulthood. Quintelier and Blais (2015) found the same close link between participatory intentions and reported behaviour for other forms of political participation. Verba et al. (2005) argue that motivational outcomes, such as interest in politics and voting intentions, are greatly shaped by parental socialization in early childhood. This is another reason for us to expect parental attributes to already show a strong link with our two indicators of political engagement in the early teenage years (i.e. Hypothesis 2).

Data Source
As mentioned before, we will explore the changing impact of parental SES and political engagement with the BHPS and USoc. BHPS started with 5500 households (or 9912 respondents) in 1991 and has tracked this group and their offspring with an annual survey to the present. This group has been incorporated in USoc from 2009. The latest available data are that of USoc Wave 9 (2017). BHPS/USoc interview all household members aged 16 and older, including new members becoming 16, as well as members of newly formed households once members of the original households start to live by themselves. Importantly, from 1994 it also includes a youth survey of 11-15 year olds with an N of approximately 750 in each wave. Youth taking part in this study are transferred to the Main Survey when they turn 16 (University of Essex et al. 2017). To create a sufficiently large sample, we pooled the data of 11 year olds of the first ten waves of the BHPS Youth Survey (1994-2003, with respondents born between 1983 and 1992), resulting in an analytical sample of 1664 respondents. All the data of later waves including the last available wave have been merged with this sample. The youngest group in this sample, the 11 year olds in 2003, will have turned 25 in 2017, whereas the oldest group, the 11 year olds in 1994, will have become 34 in that year. This means that we have complete data from all the ten batches of 11 year olds between ages 11 and 25 in the analytic sample. After age 25 the sample includes progressively less batches until at age 34 only the ones born in 1983 are still included in the sample. We reorganized the data and the individual variables by age to facilitate the trend analyses shown below. For instance, the variable "interest in politics at age 11" includes data on the political interest of all 11 year olds, i.e. of those who were 11 in 1994, 1995, 1996, etc., up until 2003. This variable thus combines data collected in ten different years. We further included data on the SES and political engagement of the respondent's parents from the 1998 and 2003 waves of the Main Survey in the analytical sample, using the household identification number as the linking variable.
As every panel survey, BHPS/USoc experienced attrition. This attrition happened for a number of reasons, including residence abroad, death, long-term refusal, and long-term no contact (Knies 2018). Consequently, in our analytic sample the number of respondents declined from 1664 at age 11 to just 497 at age 25, the last age that includes data from all ten cohorts of 11 year olds. Interestingly, this dropout was not selective on political interest and voting intentions, as our two outcomes of interest. Although those who expressed the lowest level of political interest at age 11 showed a slightly higher rate of missing values on political interest at ages 15 and 25 than those who expressed the highest level of political interest at age 11, this difference was not significant. The same pattern applied to voting intentions. We therefore have no reason to suspect that there is less variation in and higher levels of political engagement among the older ages due to attrition. In any case, our modelling strategy is able to retain respondents with missing records on one or more waves (see further below).

Variables
We draw on two items from the BHPS/USoc Youth Survey to measure political interest and intention to vote as outcomes reflecting the motivational dimension of political engagement. These items are continued in all waves of the Main Survey (except BHPS Waves 7-10 regarding political interest). Intention to vote is tapped indirectly by an item asking people which party they would vote for when given the possibility. Although this item asks in first instance about party preference, the response categories also include the options "none" and "don't know" aside from a list of parties, which we considered to indicate relative disengagement. Of course these options could also reflect genuine involvement with politics in combination with dissatisfaction or indecision, but we found respondents choosing these options to have significantly lower levels of political interest (at ages 13 and 18) and to express a significantly lower likelihood to vote (at age 18) than the ones choosing one of the parties. 1 Moreover, one study found that the rise in the number of people not expressing a preference for any party can explain 80 percent of the fall in voter turnout in United Kingdom over the last 25 years (Aidt and Rauh 2019). We therefore assumed both options to be valid indicators of lower engagement.
We should further note that there are minor differences between the two surveys in the wording of the items and the response categories. In the Youth Survey political interest was captured with the item "How interested are you in politics?" with the categories (1 = not interested; 2 = fairly interested; 3 = very interested). In the Main Survey, this became "how interested would you say you are in politics" with four response categories (1 = not at all; 2 = not very interested; 3 = fairly interested; 4 = very interested). We used these original items for the analyses, i.e. the 3 category one of the Youth Survey and the 4 category one for the Main Survey. The voting intentions item in the Youth Survey was phrased as "If you could vote for a political party which would you vote for?" with response categories 1-5 representing the different parties and −1 representing "don't know." The equivalent in the Main Survey is "If there were to be a general election tomorrow, which political party do you think you would be most likely to support?" with the categories 1-7 representing the different parties, 10 "none," 11 "can't vote," and −1 "don't know." We recoded the "don't know" category of the Youth Survey item and the "none" and "don't know" categories of the Main Survey as 0 = "none or don't know" ("can't vote" was coded as missing). Choice for a particular party was coded as 1, indicating higher engagement. Because of these differences in item wording and response categories we run the analyses separately for the Youth and Main Survey data and indicate the split between them in the figures further below.
We relied exclusively on parental education as indicator of social background rather than the more encompassing parental SES. Education not only comes first in the causal chain between education, occupation, and income (as the other two components of SES) (Lahtinen et al. 2019), it has also been argued to be the most important dimension of social background influencing political engagement (Verba et al. 2005;Gidengil et al. 2016). Indeed, preliminary analyses taught us that parental education has a stronger impact on our two outcomes of political engagement than a synthetic measure of parental SES combining all three components. In fact, parental occupation and income did not show significant relationships at all with these outcomes when entered in models as separate variables together with parental education. Following Verba et al. (2005) we averaged mother's and father's education level by household to create a parental education variable. 2 Mother's and father's education are given in the database as ISCED levels: 1 = primary; 2 = incomplete lower secondary; 3 = lower secondary and level 2 vocational; 4 = upper secondary and level 3 vocational; 5 = higher vocational; 6 = bachelor degree; 7 = masters and PhD degree. Parental education thus ranges between 1 and 7 with higher values denoting higher levels of education. Since it is our sole indicator of social background, we will use "parental education" rather than "parental SES" from now on.
We further created two variables to capture parental political engagement. The first one, parental political interest, is based on the earlier mentioned item from the Main Survey on political interest. The second, parental party support, relies on a question in the Main Survey about supporting a particular political party. The responses to this question ("yes," "no," and "don't know") were recoded as 0 = "no or don't know" and 1 = "yes," with the value 0 considered to indicate disengagement (as with intention to vote-see above). These variables also represent the household average of mother's and father's responses.
In the Latent Growth Curve (LGC) models (see further below), we will include several controls relating to other family characteristics. These concern household type (HH type) (0 = living with a single parent; 1 = living with both parents), household size (HH size) (in number of persons), and tenure (0 = owner; 1 = renting from council; 2 = renting from a landlord). We also control for year of birth to take possible cohort effects into account (0 = 1983-1986; 1 = 1987-1989; 2 = 1990-1992) and gender (0 = boy; 1 = girl). We did not include a control for ethnicity as there are less than 5% non-white respondents in the sample. Appendix A presents the basic descriptive statistics for all the variables.

Analytic Strategy
We start by presenting trends in aggregate political interest and voting intentions by level of parental education and politicization to assess how social gaps in political engagement evolve during adolescence and early adulthood until age 25. Of particular interest is to see whether these gaps are already in place at age 11 (as hypothesized earlier) or only emerge later in life. Growing gaps indicate that the effects of parental education and politicization are becoming stronger as young people age. Although we use the term "effects," which suggests a unidirectional impact of parents on children, we recognize that children can also influence their parents' engagement with politics, as compellingly argued and found by Linimon and Joslyn (2002) and Dahlgaard (2018). Beck and Jennings (1982), however, maintain that the causal direction mainly runs from parents to children.
To assess these effects more rigorously we subject them to several controls in a Latent Growth Curve analysis (LGC analysis). LGC analysis permits the modelling of the initial level and subsequent change in some outcome for each individual by estimating a random intercept and slope, which are generated as latent variables. These variables allow researchers to compare the interindividual variation in the change to that of the initial level and thus get a sense of the malleability of this outcome. Subsequently, it can relate covariates to the intercept and the slope, enabling an assessment of the extent to which the interindividual variance in the initial level and in the subsequent change in the outcome can be explained by predictor variables (Bollen and Curran 2006). In an approach similar to Neundorf et al. (2013) we will use it to assess the effects of parental education and politicization on both the age 11 level of political engagement and the post-11 change in this engagement. 3 A positive effect of, say, parental education on the post-11 change in engagement means that engagement has risen faster (or declined less steeply) among young people from privileged backgrounds than among those from disadvantaged backgrounds, indicating divergence and thus growing social inequalities in political engagement. We will run growth models with random intercepts and random slopes. These will include no less than fifteen measurements of the two outcomes of interest ranging between ages 11 and 25 (see further below).
An additional advantage of LGC analysis is that it retains cases showing missing values in one or more waves of the dependent variable. This is a major benefit as attrition is a key feature of panel data, particularly when this involves a large number of waves covering an extended period of time. Mplus, the program we use to run LGC models, is able to retain these cases by offering a maximum likelihood single imputation estimation (Neundorf et al. 2013). In the end, this facility allowed us to preserve between 58 and 71 percent of the respondents of the analytic sample (see the N in Tables 3 and 4). Figure 1 shows the development of political interest between ages 11 and 25 by parental education (in quantiles) and parental political interest (by the most and the least interested groups). The lines in the figure represent the average political interest for these groups. We see that children from the two parental education groups hardly differ in their interest in politics at age 11, with the ones from less educated families even showing a slightly higher level of interest. The mean scores of the two groups are closer to 1 (not interested) than to 2 (fairly interested) indicating broad disinterest. However, differences soon emerge and by age 15 there is a clear relation with social background: the political interest of children from the 50 percent most educated parents has hardly changed, whereas that of the 50 percent least educated has declined. By age 16, which marks the start of the Main Survey, there is a pronounced difference between the two groups in the mean level of political interest with the former showing a higher level. At this age political interest is also at its lowest level for both groups across the 16-25 age range. From age 16 political interest rises steadily for both quantiles. Although hardly perceptible, the political interest of the top quantile rises at a slightly higher rate. Consequently, at age 25 the gap between the two groups is larger than at age 16. Levels of political interest stay quite low: only at age 25 does the political interest of those from the most educated families reach the mid-point of the scale (i.e. 2.5 on a scale from 1 to 4). The political interest of people from less educated families falls well short of this point.

Trends in Political Engagement by Parental Education and Politicization
We can see strikingly similar trends when we replace parental education with parental political interest. The only difference is that parental interest already matters for children's political interest at age 11 as those whose parents are fairly or very interested show higher levels of interest than those from more disengaged families. All the other trends are the same: widening differences until age 15; lowest point for both groups at age 16; steady increase thereafter until the mid-twenties. Although we may add that young people from the most and the least politically interested families do not show the slight divergence in political interest noticeable among the parental education groups for the 16-25 age range. Figure 2 shows trends in the percentage of respondents stating their intention to vote for a particular party (1 = yes; 0 = none/don't know) by parental education (in quantiles) and parental support for a particular party (broken down by "neither parent does so" and "one or both parents do so"). The trends in voting intentions by parental education are different from those in political interest in three respects. First, we see a uniform rise in voting intentions among the 11-15 year olds rather than the differentiated change observed for political interest. Second, from age 16 the trend lines are more fluctuating than for political interest. 4 Third, the divergence between the two quantiles seems to be more pronounced than for political interest, although the fluctuations somewhat obscure this pattern. Nonetheless there are also similarities. As was visible for political interest, there is hardly a difference between children from the most and the least educated families in their voting intentions at age 11, but this difference becomes wider during early adolescence. Second, similar to political interest, young people from the most educated families are always showing higher levels of voting intentions than those from the least educated ones.
In contrast, there is already a pronounced gap between children of politically engaged and those of disengaged parents at age 11 in their voting intentions (much more so than was the case for political interest) and this gap increases only slightly afterwards. Another difference with political interest is that the gap in voting intentions between young people from engaged and those from disengaged families seems to become smaller among the older age group, although again this observation is somewhat disturbed by the volatile pattern. Nonetheless, we can see very clearly that by age 25 the gap in voting intentions between young people of different parental education groups is much larger than that between young people of different parental engagement groups. Finally, as with political interest, overall levels of voting intentions remain quite low, hovering roughly between 25 and 60 percent, which indicate continuing disengagement with mainstream party politics throughout adolescence and early adulthood. The nonexistent social gaps at age 11 in combination with the pronounced widening of them among the youngest age group indicate that the effect of parental education is weak at age 11 but rapidly becomes stronger for both forms of political engagement during early adolescence. The pronounced difference at age 16 and the slight divergence from age 16 to age 25 further suggests that this effect is well established in mid adolescence and might become even more salient thereafter. In contrast, the trend lines by parental political interest and parental support for a political party suggest that there is already a strong parent-child correspondence on political engagement at age 11 and one that may become slightly stronger for political interest and weaker for the intention to vote for a particular party.

The Changing Impact of Family Characteristics on Young People's Political Engagement
Obviously, the trend lines of figures 1 and 2 may hide considerable variation around the mean, and therefore, we run LGC models using the parental characteristics as predictors to explore whether the suggested effects by the trends lines are robust and significant. However, before we turn to the results of these analyses we present unconditional growth models as these allow us to assess the interindividual variation in the initial level and the subsequent change in the two outcomes of political engagement. They also allow us to assess the direction of the change and whether this change reflects divergence or convergence. Table 1 shows the unconditional growth models on political interest for each age group (Youth Survey and Main Survey). In the fixed effects models, the intercept and slope are fixed on their mean values. Looking at the mean slope of political interest we see that political interest declines between the ages 11 and 15 by an average of 0.012 per year and that it rises by 0.031 per year between the ages 16 and 24. As political interest ranges between 1 and 3 for the youngest age group and between 1 and 4 for the oldest one, these are the modest changes. The model fit improves markedly, however, once we allow the intercept and slope to vary (see the much lower root mean square error of approximation (RMSEA)  values in the random effects models). Both the linear and the nonlinear random effects models show a significant interindividual variation in both the intercept and the slope for the youngest age group. Although the variation in the intercept is much larger than in the slope, the variation in the latter is still significant. This means there is every reason to expect predictors, such as family background characteristics, to not only influence initial levels of political interest but also the subsequent change in this outcome. In the older age group, however, the variance in the slope is no longer significant in the nonlinear growth model, which performs slightly better than the linear growth model in terms of model fit. 5 Thus, when we run a more accurate growth model, there are no longer significant differences between individuals in their growth trajectories, suggesting there is little variation in these trajectories to explain for any predictors. We further see a negative correlation between the intercept and the slope for both age groups, indicating that children with lower initial levels show a steeper growth (or less pronounced decline) in political interest than those with higher initial levels. This does not necessarily mean convergence, however, as the inversion of positions may be so pronounced that differences are larger at the end of the time series than at the beginning. Indeed, while the variance in the intercept at age 11 is 0.127, it is 0.168 at age 15 (not shown in the table), which indicates that young people are growing further apart in their political interest during early adolescence. Table 2 presents the unconditional growth models for intention to vote. We omitted ages 16 and 17 from these models because of the high numbers of missing values for these ages (as noted above). The results are remarkably similar to those of political interest. Thus, we see that the variation in the slope remains significant for the youngest age group in the two random effects models and becomes nonsignificant for the oldest age group in the nonlinear growth model, which outperforms the linear one in model fit by a razor margin (see the decline in the log-likelihood). A further similarity with political interest is the negative relation between the intercept and slope, indicating the inversion of initial and ultimate positions outlined above. Unlike political interest, however, voting intentions appear to become stronger between ages 11 and 15 and are stable in the oldest age group. Because the nonlinear growth models score better on model fit than the linear ones for both age groups and for both outcomes, we will continue with estimating the conditional models on a nonlinear growth assumption. Table 3 presents the results of these conditional models for political interest. To begin with the 11-15 age group, we see that both parental education and parental political interest show a significant positive relation to the 11-15 change in political interest. This means that the political interest of young people with educated and engaged parents has risen faster (or declined less) than that of young people from less educated and disengaged families. In other words, young people of different social backgrounds and from families with different levels of engagement have drifted further apart in their levels of political interest, which confirms the growing influence of family background during early adolescence (as already provisionally shown by figure 1). We further see that neither parental education nor parental political interest is significantly related to initial levels of political interest at age 11 (although the positive effect of parental political   interest is bordering on significance), which suggests that parental influences are weak in the years prior to adolescence. The pattern is completely different for the older age group. Here we see that the two indicators of family background show very strong positive links to the initial level of political interest at age 16 but no significant links to the subsequent change in political interest between ages 16 and 25. In other words, by the time young people are in their mid to late teens the impact of family background is well established (with young people from well-educated and engaged families showing much higher levels of political interest) and subsequently remains fairly stable during early adulthood.
Turning now to voting intentions, we see that parental education also has a positive effect on the 11-15 change in voting intentions (albeit at the lowest level of significance) and no effect on its initial level at age 11 (see table 4). In contrast, parental party support, as the second indicator of parental political engagement, shows a strong positive link to the initial level of voting intentions and no relation to the subsequent changes in voting intentions. Thus, unlike parental political interest, it does seem to influence political engagement in early childhood. The pattern of effects for the older age group corresponds more closely to that of political interest. Thus, we also see significant positive effects of the two family background predictors on initial levels of voting intentions at age 18 and no effects on the subsequent change from 18 to 25. The effect of parental party support on the initial level is fairly weak though, and three levels of significance lower than its effect on the age 11 level of voting intentions. This suggests a declining influence of parental party support during adolescence.
Importantly, the patterns on both outcomes indicate that the period before mid-adolescence is crucial for the manifestation of family influences. For political interest, this is clearly restricted to early adolescence while it may well also include early childhood for voting intentions. After mid-adolescence the impact of family background appears to be largely set-at least until age 25. Thus, if one is interested in identifying the causal pathways through which parental education and political engagement increasingly shape the political engagement of their offspring, the most promising life stage to focus on is the early years and particularly early adolescence.
Among the control variables, gender is one of the stronger predictors, with boys developing higher levels of political interest during early adolescence and showing higher levels of voting intentions at age 18. Renting from a council is associated with lower levels of political interest and voting intentions in midadolescence. People born at the end of the 1980s have higher levels of political interest at ages 11 and 16 than other birth cohorts, but their voting intentions decline more rapidly between ages 11 and 15.

Discussion
This paper explored changes in the impact of family background characteristics on young people's political engagement. It focused on two such characteristics, parental education and parental politicization, and drew on two perspectives on the development of political engagement across the life course, the impressionable years and the family socialization perspective, to develop nonrivalling hypotheses about the dynamics of this impact. The findings show mixed support for the hypotheses. Thus, consistent with the hypothesis drawn from the impressionable years perspective that the impact of parental education grows during adolescence and early adulthood and stabilizes thereafter (Hypothesis 1a), we indeed see that the effect of parental education (as indicator of SES) increases between ages 11 and 15 for both political interest and intentions to vote, as our two political engagement outcomes. However, it stabilizes earlier than anticipated as it shows no effect on the change in the two outcomes between mid-adolescence and the mid-twenties. Moreover, defying the hypothesis draws from the family socialization perspective that the impact of parental SES should already be visible in early adolescence (Hypothesis 2), parental education is not significantly related to initial levels of the two outcomes at age 11. Thus, in a relatively short space of time, parental education develops from an irrelevant to a strong predictor of political engagement and becomes a stable influence immediately thereafter. In other words, at the beginning of adolescence there are no social differences in political engagement, but these differences soon start to appear and to grow wider at ages 14 and 15. After age 16 they stabilize with young people from educated families showing consistently higher engagement levels than those from disadvantaged backgrounds. The findings on the transmission of political engagement from parents to children match the hypotheses well for voting intentions. Thus, as anticipated, parental support for a political party (as indicator of parental political engagement) already has a strong effect on voting intentions at age 11 (Hypothesis 2) and sees its influence wane during adolescence (Hypothesis 1b). This is in agreement with the tenet of social learning theory that politically engaged parents already transfer their dispositions in early childhood through socialization and emulation (Verba et al. 2005;Gidengil et al. 2016) and with the expectation that experiences and life cycle choices during adolescence and early adulthood diminish the legacy of these parental influences (Glenn and Grimes 1968;Jennings et al. 2009). However, on political interest the parent-child transmission is not in accordance with these hypotheses. Instead parental political interest, as indicator of parental political engagement, follows the trend of parental education by not being related to initial levels of political interest and quickly becoming a significant predictor during early adolescence.
How can we explain the unexpected patterns on political interest? Possibly the world of politics is still too abstract for children in their preteen years to develop a general sense of curiosity in it. They might only develop this curiosity once they start to understand the terminology and the different party positions on a range of issues. In other words, parental characteristics may only start to influence political interest once children are becoming receptive to political information. Existing research on the development of political interest using panel data has not been able to address this conjecture because it explores this development only from late adolescence onwards (e.g. Jennings and Markus 1984;Shani 2009;Jennings et al. 2009;Prior 2010;Neundorf et al. 2013). Shehata and Amna (2019) come closest to the age range of the current study by investigating the development of political interest among Swedish teenagers from ages 13/14 to 18/19. They find that the change in family political interest is positively related to the change in respondent's political interest across this age range, which suggests that, as our findings do, young people are open to parental political input during early and mid-adolescence. However, they have not explored family influences on the initial level of political interest at age 13. Thus, for now the proposition that young people may only become receptive to parental influences during early adolescence regarding political interest remains at the stage of conjecture.
Let us return to the findings on parental education. The unexpected absence of a link between parental education and age 11 levels of both political engagement outcomes provisionally suggests that parental education influences political engagement mainly through shaping children's educational experiences rather than also in a more direct way as proposed by social learning theory (cf. Verba et al. 2005). Yet the stability of the effect of parental education after age 16 on both political engagement outcomes does not offer much support for the status transmission theory either, which postulated that parents mainly shape the political engagement of their offspring by influencing their children's educational attainment and social status (Verba et al. 2005;Schlozman et al. 2012;Gidengil et al. 2016). After all, if status transmission were the key mechanism, then we should have seen ever widening social differences between young people during late adolescence and early adulthood (and thus a growing effect of parental education), because young people only start to show meaningful differences in educational attainment and social status during these life stages.
The rapidly growing effect of parental education during early adolescence suggests it is through other educational conditions, such as school choice, choice of peers, and choice of subjects, that parental education exerts its effect on political engagement. This tallies with perspectives that de-emphasize the role of parents in socialization processes and accord more value to other agents. Harris' (1995) theory of group socialization, for instance, considers peer groups to be the prime influencers and posits that parents only have an indirect impact on their children's personality by shaping their offspring's choice of friends. However, as the current research has not been able to test this proposition, a prime task of future research is to assess whether the effect of parental education indeed runs through these mechanisms or whether this effect is more direct simply because children are only becoming receptive to parental input during early adolescence. The strong effect of parental political engagement on the initial level of voting intentions, however, is not in accordance with the idea of growing receptivity. Thus, future research would also do well to examine whether parental characteristics influence political engagement in the same way for a variety of outcomes.
If it were indeed established that educational conditions experienced in lower secondary play a key role in amplifying social disparities (i.e. disparities by parental education) in political engagement, then policy makers prioritizing the reduction of such disparities over the freedom of parents to shape the educational experiences of their children have a case to advocate reforms that reduce the impact of parents on these educational conditions. Possibly, comparative research has a useful role to play in this regard by identifying contexts where the impact of parental education on political engagement grows less rapidly during early adolescence than in the United Kingdom and by matching these patterns with policies that are associated with a weaker parental imprint on the said educational conditions. Hoskins and Janmaat (2019: 2) have already found that the effect of parental SES on the voting intentions of 14 year olds varies greatly among European states, with the United Kingdom showing the strongest effect. This does not only suggest that social disparities in political engagement have increased at a much lower rate during adolescence in some of these states (and thus that our findings based on British data cannot just be generalized to these contexts) but also that the United Kingdom may have a lot to learn from these countries in terms of policies and institutional structures that are associated with smaller social gaps in political engagement.