When the chickens come home to roost: The long-term impact of party positions on religious voting

. Despite the widespread secularisation of West European societies, research has only found mixed evidence of a decline in the inﬂuence of religion on people’s electoral preferences. A relatively recent line of inquiry has adopted a ‘top-down’ approach to this problem, arguing that the impact of religion not only depends on structural social changes, but also on parties’ convergence on moral issues. Drawing upon this ‘top-down’ approach and the ‘impressionable years’ model, this article argues that parties’ political strategies aimed at (de-)mobilising social cleavages have a lasting effect on voters’ party preferences. Using nine rounds of the European Social Survey for 19 West European countries, I ﬁnd the impact of religiosity on voting for the centre-right (Conservative and Christian Democratic parties) to be signiﬁcantly smaller for voters who were exposed during emerging adulthood (aged 15– 25) to a centre-right party that adopted similar positions on moral issues to those of its main competitors. These ﬁndings have important implications because they highlight the role of generational replacement in bringing about electoral change, even when this is prompted by parties’ strategic choices.


Introduction
Unlike other parts of the world, in Western Europe the evidence of society's increasingly low reliance on supranational beliefs, matched by a dwindling proportion of churchgoers, is so strong that current academic debates do not revolve around the existence of secularisation but, rather, around its causes and actual significance (Norris & Inglehart 2004: 83−110). Given that the politics of most advanced democracies has been shaped by the religious/secular cleavage, it has often been assumed that religious identity and beliefs would soon cease to impact West Europeans' voting behaviour. This assumption rests upon two mechanisms that have not always received clear support in empirical research. First, it is argued that, as the number of religious citizens decreases, parties that used to rely on their support will be forced to moderate some of their policy stances in order to attract new constituencies (Dalton 2002, p. 201). Second, as religion loses much of its social and political significance, the values and political preferences of religious and non-religious voters are often expected to become more alike (Evans & Northmore-Ball 2018). Although there is little doubt that religious voters have become a smaller (and, therefore, also a less influential) group, research into the electoral impact of religion has failed to find consistent evidence of a decline in religious voting. Conservative and Christian Democratic parties continue to perform significantly better than other parties among religious voters in contemporary West European democracies, even though the size of this effect varies substantially across countries (Knutsen 2018). Many studies have shown that the differences between the party preferences of religious and non-religious voters have not consistently decreased across most of Western Europe -and, in some cases, they have even increased over time (Knutsen 2004(Knutsen , 2018Elff 2007;Van Der Brug et al. 2009;Raymond 2011;Tilley 2015). In contrast, other studies analysing a longer (and more recent) time period have found a certain decline in the impact of religion in countries such as Germany, Switzerland and the Netherlands, but not in other countries such as Britain (Lachat 2007;Goldberg 2020). The picture is, therefore, more complicated than it seems. This begs the question, why has secularisation not blurred the differences between the political preferences of religious and non-religious voters?
A recent approach to this issue questions the simplistic view that the extent or timing of secularisation can, on their own, explain the variation in religious voting across elections and countries. On the contrary, political elites are deemed to be essential for understanding the impact of religion on vote choice. In a nutshell, when differences between religious and non-religious voters have shrunk it has been because Conservative and Christian Democratic parties have actively sought to converge with other parties on issues, such as traditional morality, that religious voters care about (Elff 2009;Jansen et al. 2012;de Graaf et al. 2013;Evans & de Graaf 2013a). Although this 'top-down' approach is genuinely promising and more sophisticated than previous approaches, it has so far focused on exploring the role of short-term changes in parties' positions between elections rather than their long-term impact, which seems relevant for explaining the influence of structural factors such as religion. The goal of this piece is therefore not to question the role of parties in activating or deactivating cleavages between elections, but rather to expand this approach by exploring the potential long-term impact of parties' past strategic choices on individuals' electoral behaviour.
This article aims to contribute to the current debates on parties' ability to shape the effects of structural factors by focusing on the role of generational dynamics. Drawing on the impressionable years literature, which has been widely applied to electoral behaviour theory since at least Campbell et al.'s (1960) lifelong persistence model, I argue that the relative positions on moral issues held by political parties when voters are aged between 15 and 25 years have a lasting impact on how religiosity influences the latter's vote choices. Evidence supporting this hypothesis is shown using survey data from 19 West European democracies from 2002 to 2018 in combination with party manifesto data on past party positions. The findings have important implications because, besides contributing to our understanding of the role of political parties in activating or deactivating political cleavages, they also help to explain why the effect of religion on vote choice differs across cohorts and its overall impact changes ever so slowly even in contexts of widespread or increasing secularisation.
The first section of this article reviews the literature on the effect of religion on voting. The second section lays out the theoretical background, which integrates the 'top-down' approach to religious voting with the 'impressionable years' model of political learning. After describing the dataset and explaining the case selection, the operationalisation of the main variables and the method employed in the third section, the fourth section presents the results. As is customary, the last section then discusses the implications of the findings. parties were among the first parties to appeal to religious voters in Europe because, while they are not confessional, they played a major role in resisting Liberal anticlerical attacks seeking to deprive organised religion from its control over issues such as family and education. In some countries (particularly in those where the Right was not able to stop Liberal reforms), Christian Democratic movements emerged later on as separate organisations seeking the support of religious voters. Although Conservative and Christian Democratic parties competed for some time with each other for the anti-Liberal vote, in those countries where Christian Democracy was electorally successful and established itself as the main party of the conservative political space, it eventually ended up co-opting conservative political elites and replacing Conservative parties altogether (Kalyvas 1996, pp. 222−226). The opposite was nevertheless the case in countries such as France, Spain or Greece, among others, where successful Conservative parties prevented the survival of explicitly confessional platforms.
In the context of the religious cleavage, many researchers make a conceptual distinction between denominational identities (which are associated with denominational voting) and the secular-religious conflict (which is associated with what tends to be called religious voting) (Wolf 1996;Brooks et al. 2006;Elff & Roßteutscher 2011). Even though the main differences in contemporary voting behaviour are found between religious and non-religious voters, significant denominational differences are still present in many European countries (Knutsen 2018, pp. 131−141). 1 Denominational voting rests upon identarian bonds between particular religious groups and political parties that are contingent on countries' historical developments. Conservative and Christian Democratic parties usually forged long-term alliances with the major religious groups that stood in opposition to Liberal anticlericalism (Lipset & Rokkan 1967, pp. 33−41). In countries dominated by Catholicism this was the Catholic Church, whereas in majority-Protestant countries, such as Britain and the Nordic countries, conservative parties' interests became intertwined with those of the dominant national church, and therefore smaller groups of voters such as non-conformist Protestants (and Catholics in Britain) became aligned with nonconservative forces (Lipset & Rokkan 1967, p. 38;Madeley 1982). 2 In religiously mixed countries, where Protestants often allied with the nation-builders to limit the influence of the Catholic Church, the interests of conservative forces aligned closely with those of Catholics (and other anti-Liberal groups like Calvinists in countries such as the Netherlands).
Although patterns of denominational voting vary across countries, the differences between religious and non-religious voters are much more consistent. Religious voting has an important identity component, but it is also guided by individuals' wider religious orientations and values (Langsaether 2019). Religious voters, regardless of their denomination, are not only more likely than their less religious counterparts to have morally conservative attitudes, but their opinions on moral issues also have a stronger influence on their vote, making them more likely to vote for Conservative and Christian Democratic parties (Stegmueller et al. 2012;Stegmueller 2013). Religious orientations have a separate impact on vote choice that tends to affect individuals across all major Christian denominations, and this can be seen in countries such as Britain, where Tilley (2015, p. 919) finds religious practice to decrease support for the Labour party among all three main denominations (Church of England, Catholic and non-conformist), or in Germany, where again church attendance increases voters' probability to vote for the Christian Democratic Union/Christian Social Union (CDU/CSU) regardless of their denominational affiliation (Elff & Roßteutscher 2011).

Bottom-up and top-down approaches
Religious voting has received much attention in the last decades in the context of the secularisation process that has affected virtually all West European democracies (Norris & Inglehart 2004, pp. 84−92;Voas 2009). Secularisation entails a steady decrease in religious beliefs and church attendance for all religious groups, but also the individualisation of religious orientations and, therefore, the separation between individuals' religious beliefs and their political preferences (Knutsen 2018, pp. 73−75). In this context, it is somewhat surprising that religious orientations continue to be an important predictor of support for mainstream conservatism (often simply referred to as centre-right) 3 in contemporary West European democracies (Duncan 2015;Tilley 2015;Knutsen 2018). Yet, as Lachat (2007) argues, political dealignment can take place through two different mechanisms: structural change and behavioural change. In the context of secularisation, structural change refers to the reduction in size of the group of religious voters. Scholars such as Best (2011) contend that the numerical decline of religious voters has significant political implications because it entails a more or less irreversible reduction in their contribution to party vote shares and, therefore, pushes parties to compensate for that loss by attempting to attract non-religious voters. Of course, religious voters have not disappeared from the face of Earth and, in fact, there is evidence that religious decline has already bottomed out in early-declining countries, where there is a baseline of between 40 and 50 per cent of the population who consider themselves members of an organised religion (Kaufmann et al. 2012). In many proportional systems, securing the support of the majority of a group of this size could still guarantee a party's victory in a general election. However, there is little doubt that this group has become less decisive than it used to be, and, as a consequence, religion may have also become a less important factor for electoral competition. The second mechanism of dealignment is behavioural change. The political preferences of religious and non-religious voters have been argued to become more alike as a result of secularisation, partly because of the decreasing social and political influence of religious groups (Dalton 1984). The loss of relevance of religion in the public sphere is often expected to entail wider societal acceptance of previously contentious issues such as sexual morality, even among religious individuals (Finke & Adamczyk 2008). However, it has also been argued that secularisation may have increased the distinctiveness of those who remain religious (Evans 2010, p. 643), and could even be associated with greater religious polarisation (Ribberink et al. 2018). Overall, the evidence that secularisation has brought about behavioural dealignment (or realignment) in Western Europe is mixed. Knutsen (2004) finds no consistent trend in the effect of both religious identity and religious orientations on vote choice in eight West European countries between 1970 and 1997, with statistically significant declines in Denmark and Italy but fairly stable patterns in Belgium, Britain, France, Germany, the Netherlands and Ireland (see also Raymond 2011, who finds no decline in religious voting between the 1960s and the 1990s in Germany and Britain). Similarly, Van der Brug et al. (2009) find no clear downward trend in the explanatory power of religion on party preferences in Europe between 1989 and 2004, with early decreases followed by an increase in the 2000s. In contrast, other studies that analyse longer (and more recent) time series have found a decreasing effect of religion in Germany, Switzerland and the Netherlands, although not in Britain (Lachat 2007;Goldberg 2020).
One of the reasons why there is no unequivocal evidence of behavioural dealignment in Western Europe may lie in the role of parties' strategic choices. Religious voters' greater degree of moral conservatism is key for understanding their support for centre-right parties (Stegmueller 2013).
However, as Evans and de Graaf (2013a) argue, even though voters' political preferences are shaped by socio-demographic variables, their electoral choices are constrained by the existing party platforms. Although social cues are particularly relevant in contexts where political parties differ greatly on the issues that are associated with individuals' social position, when party platforms converge around those issues voters are compelled to base their voting decisions on other matters. One of the expectations of the secularisation process and the decreased relevance of religious groups was precisely greater convergence between parties on moral issues. Yet, even though mentions of traditional morality are now less frequent than they used to be for many centre-right parties (Euchner 2019, pp. 28−30), linear declines are not found everywhere (de Graaf, Jansen & Need 2013). This could therefore explain the variation in the impact of religiosity on voting across Western Europe. 4 This supply-side (or top-down) approach to religious voting has found empirical support in several studies. De Graaf et al. (2013) and Jansen et al. (2012) analysed religious voting in the Netherlands between 1976 and 2001, and found that the effect of religion increased when Christian Democratic parties emphasised moral traditionalism in their manifestos. In a similar vein, Elff's (2009) study, which focuses on Belgium, Denmark, France, Great Britain, Italy and the Netherlands over a period of 28 years , finds that moral traditionalism increases parties' electoral support among regular churchgoers and decreases it among non-churchgoers. Top-down patterns of change in religious voting have also been found in case studies of countries such as Italy and France, although the evidence is either modest or inexistent for countries such as Germany or Spain, among others (Evans & de Graaf 2013c).

Long-term change and the top-down approaches to religious voting
Part of the appeal of the top-down approach to religious voting is that, contrary to the bottomup approach, it can also explain short-term changes in the electoral strength of religiosity. Nonetheless, it would be wrong to conclude that long-term changes are always fully explained by bottom-up forces. As explained earlier, one of the reasons why secularisation is thought to lead to behavioural changes is because parties that drew much of their electoral support from religious individuals are expected to moderate their stances on moral issues to make their policies more palatable to the growing group of secular voters. Yet we know that some voters respond more swiftly than others to political changes -especially when these are not abrupt and radical. Therefore, it may take years before such changes fully bear fruit.
The decline of cleavage voting has been a very gradual process, not least because older generations of voters have continued to be strongly influenced by social loyalties such as class and religion (Franklin 2009). Research has shown that certain aspects of human central beliefs and dispositions, including values, identities and political orientations such as ideology and party identification, are most susceptible to external influence when individuals are young but crystallise and become fairly stable once people reach a certain age (Jennings 1990;Alwin & Krosnick 1991). This empirical regularity provides support for what social psychologists have labelled the 'impressionable years' model of political learning (Alwin & McCammon 2003). The 'impressionable years' model poses that people's core political beliefs are acquired through a learning process that is influenced by a wide range of experiences (interactions with family, friends, personal events, but also the general political and social context). Political learning seems to start early on in life, but is accelerated (at least in Western societies) during emerging adulthood (Delli Carpini 1989). This period is so important that adults tend to disproportionately recall memories of events that occurred during early adolescence and the early adult years (Rubin et al. 1998;Schroots et al. 2004). During this time, which Alwin and McCammon (2003, p. 38) characterise as a period of 'sorting and sifting', individuals gather a large amount of new political information, and sometimes change their views quite dramatically. However, as individuals learn more about politics, they form political predispositions and become less susceptible to changing their core political attitudes. This is consistent with Zaller's (1992) 'resistance axiom', which states that individuals assess every new bit of information against their predispositions, and explains why people's political views do not change all the time. Under normal circumstances, political predispositions are formed during the period of emerging adulthood and become highly persistent thereafter (Sears & Funk 1999).
Even though individuals are never completely 'set in their ways' and changes do occur later in life, the lasting influence of the 'impressionable years' (a loosely defined period which seems to peak at around 15-16 and 21-22 years of age) on people's voting behaviour is remarkable (Neundorf & Smets 2017). It also explains why generational replacement is one of the most important driving forces of long-term political change, as older voters are less likely to respond to normal political events by switching political allegiances (Hooghe 2004). Accounting for the political context during people's 'impressionable years' is essential for understanding, at least partly, the differences in voting behaviour across individuals born at different points in time.
Research has demonstrated that political scandals and wars can have a lasting effect on the political preferences of very young voters (Erikson & Stoker 2011;Dinas 2013), but other more mundane short-term factors, such as the structure of political supply, can also have distinct long-lasting effects on voters (Smets & Neundorf 2014). This has important implications for the study of the effect of religion on party choice. As political preferences are shaped by the political context that individuals experience during emerging adulthood, parties' stances on moral issues could leave a significant imprint on the way in which religious beliefs inform young voters' electoral choices. We can therefore speculate that religious differences in voting will be weaker for voters who are socialised into politics at times when parties converge on moral issues, and vice versa when parties' platforms are very different from each other. So, while religiosity increases the probability of voting for centre-right parties, its effect should be even greater for those voters who are exposed during their impressionable years to centre-right policy platforms that are distinctively more traditionalist on moral issues than those of their competitors.
The idea that structural factors could be a more influential factor for some generations of voters than for others has found empirical support in several studies. Van der Brug (2010), for instance, found that religion and social class play a more important role for understanding the party preferences of Europeans socialised before the 1970s (i.e., during the 'heyday' of cleavage politics). Similarly, Van der Brug et al. (2009) found (non-linear) cohort differences in the impact of religion on party preferences in Europe. Although differences between age cohorts have not always been corroborated by studies of specific countries, this may largely be due to the different trajectories of political parties' efforts to mobilise the religious issue. Thus, while there is some evidence that support for the German CDU/CSU has deteriorated among younger churchgoers in recent elections (Elff & Roßteutscher 2017), in the Netherlands, where Christian Democratic parties' traditionalism has only been subject to timid changes and there is no evidence of a decline over time (de Graaf, Jansen & Need 2013), differences in the effect of religion across cohorts are rather small (van der Brug & Rekker 2020). Note: In Ireland, both Fianna Fáil and Fine Gael can be categorised as centre-right parties, although the former has a more populist outlook, favours more economic interventionism and some of its leaders have sometimes referred to the party's fuzzy ideology as 'left of centre' (Puirséil 2017). It will be tested if results remain robust when Fianna Fáil rather than Fine Gael is considered as the main centre-right party.

Data, operationalisation and methods
To test the hypothesis outlined above, this study employs data from waves 1 to 9 of the European Social Survey (ESS), fielded biennially between 2002 and 2018 (European Social Survey Cumulative File, EES 1-9 (2020)). The analysis will focus on 19 West European democracies. Postcommunist countries, Turkey and Israel are excluded from the analysis to maximise comparability across widely similar political and social contexts. 5 The operationalisation of the main variables is as follows. The dependent variable is a dichotomous indicator measuring whether respondents voted for the main Conservative/Christian Democratic party in the most recent legislative election (1) or for any other party (0). 6 A list of the parties included in each country can be found in Table 1.
The main independent variable is a measure of the centre-right's divergence on moral issues (in relation to other parties) when individuals were aged between 15 and 25 years. This variable was created using information from the Manifesto Project (Volkens et al. 2019) measuring positive and negative mentions of traditional moral issues in party manifestos. The Manifesto Project codes as 'traditional morality' policy statements referring to traditional and/or religious moral values, including the stability and maintenance of the traditional family (as opposed to supporting modern family structures, same-sex marriage, divorce, abortion, and so on), the suppression of immorality and unseemly behaviour, and the role of religion and churches in state and society (Volkens et al. 2019, p. 19). To the best of my knowledge, this is the only available data source that tracks parties' positions on moral issues over a sufficiently long period of time. It also has the advantage that it is an objective indicator based on parties' own documents (Bakker & Hobolt 2013).
The centre-right's divergence on moral issues (from here onwards also referred to as 'moral divergence') is measured as the relative position of the main Conservative/Christian Democratic party (which is the party that the dependent variable refers to) in relation to all other parties represented in parliament. This indicator was constructed for every election since the end of WW2 in the following manner: 7 • First, each party's position on moral issues was estimated by using the log ratio of positive to negative mentions of traditional morality in their manifestos (see Lowe et al. 2011 for more details about the procedure). 8 This indicator correlates highly (Pearson's r = 0.7) with the 'social lifestyle' variable from the Chapel Hill Expert Survey (Bakker et al. 2020), which measures parties' positions on issues such as rights for homosexuals and gender equality. • Next, I calculated the difference between the position of the main Conservative/Christian Democratic party and the mean position of all the other parties represented in parliament (weighted by their relative electoral strength). 9 Thus, the indicator takes on positive values when Conservative/Christian Democratic parties are more traditionalist than their average competitor, whereas a value of 0 indicates that they all share a similar position (a 'convergence' scenario). Conversely, negative values signify that Conservative/Christian parties adopt less traditionalist stances on moral issues than their average competitor (either because the former adopt liberal views on these issues or because their competitors emphasise traditional moral values even more strongly than they do). Overall, the data show different trajectories across countries rather than uniform party-system convergence on moral issues over time (see Figure 1a of the Appendix in the Supporting Information).
After computing the centre-right's 'moral divergence' indicator for each election, individuals in the sample were assigned the average value of those elections that took place when they were aged between 15 and 25 years. 10 Although the period between the ages of 15 and 25 is very often used in the literature, there is no consensus about the precise age range that constitutes the impressionable years, with some (but not all) authors claiming that it starts earlier than 18 and others showing that it can continue until individuals are at least 30 years old (Fox et al. 2019, p. 384). Therefore, the robustness of results will also be checked against alternative age ranges (18-25 and 15-30).
The next key independent variable, religiosity, is an index that comprises religious orientations and church integration -two concepts that tend to be strongly correlated (Jagodzinski & Dobbelaere 1995). The index was created using principal component factor analysis with three variables: frequency of church attendance (a 7-point scale ranging from never to weekly), frequency of prayer (a 7-point scale ranging from never to every day) and subjective religiosity (an 11-point scale ranging from not at all religious to very religious). 11 A potential problem is that religiosity is measured at the time that the survey was conducted rather than during voters' impressionable years. This problem is, nevertheless, minimised by two factors. First, although individuals' religious beliefs could potentially change later on in life, there is evidence that these, too, tend to settle and remain fairly stable after people reach their mid-20s (Voas & Doebler 2011). Second, the main mechanism described in this article assumes that the political context during individuals' impressionable years moderates the effect of religion on voting, and that this effect will remain visible for decades because political preferences tend to crystallise shortly thereafter. This moderating effect of the political context on religiosity, however, will not be observable with cross-sectional data if religious beliefs change later on in life in a substantial manner but party preferences do not change accordingly (or vice versa). Therefore, measuring religiosity at the time when the survey was conducted can only lead to an underestimation of the influence of the political context during individuals' formative years.
With regard to other independent variables, the models control for cohort-level religiosity (measured as the mean of the religiosity index for each electoral cohort within each country) and the centre-right's moral divergence at the last election. The purpose of the first control is capturing structural differences between generations (e.g., because religiosity might play a less important role in the electoral behaviour of individuals belonging to a very secular generation). The second control is introduced to capture the moderation effect of parties' contemporary positions on moral issues on people's religiosity. Although the data do not allow for a proper test of short-term changes (policy shifts) on moral issue policy positions, the introduction of this control will make sure that the effect of past policy positions during individuals' impressionable years is not fully explained by parties' current stances on moral issues. Other controls include individuals' age (divided by 10, as otherwise coefficients were too small), left-right self-placement (a 11-point scale ranging from extreme left, 0, to extreme right, 10), gender (male = 0, female = 1), educational attainment (4-point scale including primary education or less, lower secondary education, higher secondary education and tertiary education), urbanisation of respondents' place of residence (village, town, city/suburbs) and a five-category indicator of social class based on Oesch's (2008) schema. Finally, denominational differences are captured by three dichotomous variables measuring whether individuals identify themselves as Roman Catholic, Protestant, Christian Orthodox, or none (the reference category). 12 Given the uneven distribution of denominations across countries and the different levels of religiosity across denominations, denominational effects are problematic to analyse in comparative research. Denomination is not an accurate indicator of religiosity in countries that are virtually monoreligious, and its use can lead to over-estimating religiosity in countries with high levels of nominal affiliation (Evans & Northmore-Ball 2018). Moreover, as Stegmueller (2013) shows, once the effect of religiosity and other variables is taken into account, the differences in party and issue preferences between denominations are much smaller than the differences between Christians and the non-affiliated. Nevertheless, a control needs to be introduced to account for denominational differences in voting behaviour that are unrelated to individuals' degree of religiosity. 13 In the ESS, individuals are nested within years (country-year combinations) and countries. 14 In addition, as the article focuses on a variable that varies with age, it must be taken into account that members of the same electoral cohort are more alike as they experience a similar political context. 15 Therefore, I estimate multilevel logistic regression models with random intercepts at four levels: individuals, cohorts, country-years and countries. 16 Modelling random effects in this way -a strategy adopted by other recent comparative research on the effect of the political context during individuals' impressionable years (e.g., Dassonneville & McAllister 2018) -is more flexible than using a cross-classified model because it allows accounting for the fact that cohorts belonging to different national contexts cannot be assumed to be similar to each other. Nevertheless, I verified the robustness of results when (a) cross-classified models are used instead, and (b) fixed effects, rather than random effects, are introduced to account for differences between periods and countries. Table 2 shows the results of the main analysis. The first model (Model 1) introduces only variables measured at the individual level without any interactions and can be used as a benchmark against which the size of the effect of religiosity can be assessed. Consistent with previous research, results confirm the importance of religiosity for explaining electoral support for the centre-right in Western Europe. The positive and statistically significant coefficient for religiosity reflects higher electoral support for centre-right parties among more religious individuals. On average, a one-standard deviation increase in religiosity leads to a 4-percentage-point increase in the probability to vote for centre-right parties. With regard to control variables, the probability to vote for the centre-right increases with educational attainment and, rather unsurprisingly, with right-wing ideology, but decreases with urbanisation. Moreover, centre-right parties are relatively more successful among small business owners and the services classes than among skilled and (particularly) unskilled workers. Also, there is no significant gender gap in voting.

Findings
Model 2 introduces an interaction between religiosity and the centre-right's divergence on traditional morality when individuals were aged between 15 and 25 years. As expected, the interaction is positive and significant, providing support for the main hypothesis of this article, which states that the impact of religiosity on vote choice will be stronger for those individuals who were socialised into politics at a time when the centre-right adopted a more traditionalist position on moral issues than its average competitor. To better illustrate this finding, Figure 1 shows the effect of increasing religiosity by one standard deviation from the mean on the probability to vote for the main centre-right party (y-axis) across different levels of centre-right parties' moral divergence during individuals' formative years (x-axis). 17 As a reminder, 'moral divergence' is a relative measure that takes on positive values when the main centre-right party was more traditionalist on moral issues than its average competitor, and negative values when the opposite was the case. Overall, religiosity increases the probability to vote for the centre-right for every cohort -except for those socialised in contexts where the centre-right was very liberal (negative values below −0.24 on the x-axis), but these cases are extremely rare, as can be seen in the rug plot at the bottom of Figure 1. The effect of religiosity, however, is clearly stronger for cohorts who experienced a traditionalist centre-right party when they were aged between 15 and 25 years. A religious individual (one standard deviation above the mean) socialised in a context of party convergence on moral issues (a value of 0 on the x-axis) is about 1.3 percentage points more likely to vote for the centre-right than an individual with an average level of religiosity socialised in the same context. For those socialised in a scenario of greater divergence on moral issues (a value of 0.73 on the x-axis, which represents one standard deviation from the mean), the difference is 7.2 percentage points. 18 Considering that this is a long-term mechanism, the effect is not small. 19 Model 3 in Table 2 adds three controls: (a) an interaction between religiosity and the age of respondents to control for any possible age effects, (b) an interaction between respondents' religiosity and the religiosity of their cohort in order to control for the impact of cohort-level secularisation and (c) an interaction between respondents' religiosity and the centre-right's moral distinctiveness at the last election to control for contemporaneous effects of party positions. Results only show a small and statistically non-significant moderating effect for age entailing   Note: Coefficients and standard errors (in parentheses) of mixed-effects logistic regression models shown. *** p < 0.001; ** p < 0.01; * p < 0.05. a weaker impact of religiosity for younger individuals. In contrast, there is strong evidence that the effect of individuals' own religiosity is even stronger for those belonging to more religious generations (as reflected by the significant interaction between religiosity and cohort-level religiosity) and also in contexts where the centre-right currently holds relatively more traditionalist positions on moral issues (as reflected by the significant interaction between religiosity and centre-right's moral distinctiveness at the last election). More importantly, though, the coefficient for the interaction between religiosity and the centre-right's divergence on moral issues during individuals' impressionable years continues to be positive and significant when these controls are added. Results do not change either when denomination is added as a control (Model 4). As expected, Catholics, Protestants and Orthodox Christians are (in this order) more likely than nonaffiliated individuals to vote for the centre-right. Overall, results demonstrate the importance of considering parties' histories, as strategic decisions made in the past can have a lasting effect on individuals' current vote choices.
Findings are robust to alternative model specifications (see the Appendix in the Supporting Information). In particular, the interaction between religiosity and parties' stances on moral issues when individuals were aged between 15 and 25 years continues to be positive and statistically significant when (a) fixed effects rather than random effects are used to account for Figure 1. Effect of religiosity on the probability to vote for the mainstream right across different levels of the centre-right's divergence on moral issues during individuals' impressionable years. Note: Shaded areas are 95 per cent confidence intervals. The rug plot at the bottom of the graph displays the distribution of the centre-right's divergence on traditional morality at age 15-25. contextual variation (Model 3a in Table 2a, Appendix in the Supporting Information), (b) random effects are estimated using a cross-classified random-effects model with cohorts cross-classified across surveys (country-year) and countries (Model 4a in Table 2a, Appendix in the Supporting Information) and (c) a random effects are estimated using a cross-classified random-effects model with separate cohorts by country (Model 5a in Table 2a, Appendix in the Supporting Information). Results are also robust to the introduction of cohort random effects for religiosity (Model 6a in Table 2a, Appendix in the Supporting Information), and to the use of alternative time windows for the 'impressionable years' such as [18][19][20][21][22][23][24][25] in Table 4a, Appendix in the Supporting Information) and [15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30] Table 4a, Appendix in the Supporting Information). Conclusions also remain the same when Fianna Fáil rather than Fine Gael is defined as the main centre-right party in Ireland (Model 11a in Table 5a, Appendix in the Supporting Information).

Conclusions
Support for West European centre-right parties continues to be higher among religious voters than among their non-religious counterparts in spite of secularisation, but research has shown the strength of this effect to be a function of parties' strategic choices (Elff 2009;Evans & de Graaf 2013b). When centre-right parties have converged with other parties on issues related to traditional morality, the role of religiosity has diminished significantly. This article has now provided evidence that party convergence on moral issues, or the lack thereof, has a lasting influence on the way in which individuals use religious orientations to inform their vote choices. In particular, religious differences in party choice are significantly smaller among individuals who are exposed to a context of party convergence on traditional morality at 15-25 years of age.
The implications of this finding are manifold. In the first place, it lends further support to the idea that generational replacement is one of the main driving forces of electoral change, as the impact of the political context in which individuals are socialised during emerging adulthood is still visible several decades later. Furthermore, the mechanisms identified in this piece can help us to better understand the political consequences of the secularisation process in Western Europe. The shrinking number of religious voters in West European democracies led many to expect centreright parties to confront a dilemma similar to the one faced by the Social Democrats when the industrial working class waned in size as a result of the emergence of the service classes (Kalyvas 1996, p. 241). In particular, the centre-right was expected to adopt more liberal stances on moral issues in order to attract non-religious voters, which would in turn lead to the decline of religious voting. Thus, the assumption was that structural changes would also entail behavioural changes in voters, but the picture has proved to be more complicated than that. This article sheds more light on this complicated picture because, while it emphasises the role played by political agency in shaping the strength of political cleavages, it also contributes to explaining why it sometimes takes time for the full effect of parties' actions to be realised. Indeed, the explanatory potential of 'topdown' approaches need not be limited to short-term changes. We know that individuals respond to short-term policy shifts, but electoral change cannot be fully understood without reference to longer term forces and generational replacement.
Even though research has found support for the centre-right to have waned among younger religious voters in some elections and countries (Elff & Roßteutscher 2017), overall the moderating effect of age found across West European countries in this article was very small once cohort differences were accounted for. This means that at least part of the effect of age may be related to relatively recent strategies enacted by centre-right parties, which may have affected new religious voters but is unlikely to have had an enormous impact on those who have been supporting such parties for decades.
As is customary, it is worth mentioning some of the limitations of this study. An obvious limitation is that the analysis relies on repeated cross-sectional data. Although this is the research strategy adopted by the overwhelming majority of cohort studies in the field of electoral behaviour (partly because longitudinal data going back several decades is only available for extremely few countries), future research is necessary to dig more into within-individual differences, even if that entails focusing on a single country. Second, this article has focused on West European countries. Although the mechanisms explained here could equally apply to similar post-industrial democracies, the dynamics of religious voting might nevertheless have followed a different path in most post-communist countries, not least because traditional morality and religious issues in many of such countries have been actively used, and sometimes even 'owned', by parties other than the centre-right (Raymond 2014). The particularities of these cases on both the demand-and the supply-side make them particularly worthy of further exploration.

Online Appendix
Additional supporting information may be found in the Online Appendix section at the end of the article: Figure 1a. Centre-right's moral divergence across countries and over time. Figure 2a. Effect of religiosity on the probability to vote for the main centre-right party across different levels of centre-right's divergence on moral issues during individuals' impressionable years (estimation based on Model 4 in Table 2).  Table 1 using Fianna Fail rather than Fine Gail as the main centre-right party in Ireland Supplementary material Notes 1. There is some debate over which of these aspects of the religious cleavage is more important, with scholars such as Brooks et al. (2006) and Wolf (1996) claiming that the religious-secular divide is now more important than denominational identities for explaining voting behaviour. Similarly, Stegmueller (2013) finds that, while religiosity and affiliation (being a member of a religion or not) have a significant impact on both individuals' stances on moral issues and their likelihood to support centre-right parties, denomination (measured as Catholic or Protestant) does not.
2. This was partly reversed in some Nordic countries during the twentieth century with the emergence of religious parties that appealed strongly to non-conformist voters (Karvonen, 1994). 3. In this article, the term centre-right will be used to refer to the mainstream conservative right, which comprises mainstream Conservative and Christian Democratic parties. Although many European liberal parties may also qualify as centre-right, these have historically been secular (or even openly anti-clerical) and are therefore not considered to be part of the conservative right. 4. Another aspect of party strategies that should not be disregarded is issue salience, because policy divergence could lead to even greater religious voting in contexts where moral issues play a key role in parties' discourses (Langsaether, 2019). However, as Wagner (2012) shows, salience and positions are not separable characteristics of positional issues. Parties are significantly more likely to emphasise issues on which their position is distinctive from those of other parties, suggesting that both position and salience tend to respond to the same strategic aims. 5. Post-communist countries deserve a careful, separate analysis that is beyond the scope of this article.
Mainstream Conservative and Christian Democratic parties in post-communist countries lack the long historical links with organised religion that we find in many West European democracies, and in some cases have struggled to 'own' the religious issue, given that clericalism is one of the most distinct features of the populist radical right in Central and Eastern Europe (Pirro, 2014) and that left-wing parties have sometimes adopted moral traditionalism in order to attract religious and culturally conservative voters (Raymond, 2014). Moreover, patterns of secularisation in many post-communist countries differ from those found in Western Europe due to decades of state-forced secularisation (Müller, 2011), and that is likely to have distorted the link between religiosity and party choice found elsewhere in Europe and in advanced democracies more generally. 6. In Germany, the dependent variable refers to the party list vote ('second vote'). In Ireland, the dependent variable refers to voters' first preference. Non-voters are excluded from the analysis. 7. For Greece, Portugal and Spain, party manifestos are available from the moment they became democracies in the 1970s. The indicator was calculated separately for the two Belgian linguistic communities (Flanders and Wallonia) as they have different party systems. Parties that stand candidates in both regions (currently only the Workers' Party, but this was a much more common feature before the 1970s) were included in the calculations for both regions. The Christian Social Party was the main centre-right party in both regions before 1968, when it split along linguistic lines. 8. Using only positive mentions of traditional morality (rather than the log ratio of positive to negative mentions) produces the same substantive findings. 9. For the Netherlands, the weighted average of the three predecessors of the main Christian Democratic party (CDA) was used for all elections held before 1973. 10. In Greece, Portugal and Spain, individuals who came of age before the transition to democracy were assigned the value corresponding to the first cohort that experienced democratic elections. Similarly, in Italy older individuals were set to 'restart' their learning process in 1994, where the party system changed completely. Research has demonstrated that older voters were forced to (re-)learn their way around the Italian party system after it collapsed (Franklin and van Spanje, 2012). East Germany is a special case because, although Eastern Germans have been exposed to West Germany's political parties since 1990, the communist state implemented a policy of forced secularisation that might have affected the link between religion and party choice there. Research has shown that patterns of religious voting among Eastern Germans began to emerge as early as 1994 and were already similar to those found in West Germany by 1998 (Elff and Roßteutscher, 2009). Therefore, Eastern Germans who came of age before 1990 were assigned the value corresponding to the first cohort who experienced democratic elections. Nevertheless, results are robust to the exclusion of (a) Eastern Germans born before 1972 (see Model 1a in Table 1a, Appendix in the Supporting Information), and (b) all pre-democratic cohorts in Greece, Portugal, Spain and East Germany, as well as the pre-1994 cohort in Italy (see Model 2a in Table 1a, Appendix in the Supporting Information). 11. There is a single factor solution with eigenvalue = 2.3. 12. For consistency with the theory, individuals affiliated with a non-Christian religion and those born abroad who arrived in their country of residence after age 15 are excluded from the analysis. 13. The sample size is somewhat smaller for models containing this variable because information on individuals' denomination is not available for Finland 2004, Cyprus 2006 and Great Britain 2006. 14. The number of country-level units is 20, as the Belgian sample is split into Flanders and Wallonia.