Political socialization

Starting with the seminal work by Mannheim (Reference Mannheim and Kecskemeti1952) and Hyman (Reference Hyman1959), a large literature in sociology and political science has argued that some fundamental political dispositions develop early on in life and remain relatively stable over individuals’ life course. Early research already identified a variety of agents of this process of ‘political socialization’, ranging from micro-level contexts like the family – which favor the transmission of values across cohorts – to the influence of broader social trends and events (what Hyman Reference Hyman1959, p. 132, calls ‘the Zeitgeist’), which instead may yield generational value change.

Subsequent empirical work has found extensive evidence that individuals who go through the same experiences at the same stage in their lives tend to develop distinct common outlooks and patterns of behavior. Socialization experiences have been linked to a variety of effects on political preferences (Colwell Quarles Reference Colwell Quarles1979; Russel et al. Reference Russel, Johnston and Pattie1992; Gimpel et al. Reference Gimpel, Celeste Lay and Schuknecht2003; Lau and Redlawsk Reference Lau and Redlawsk2018; Madestam and Yanagizawa-Drott Reference Madestam and Yanagizawa-Drott2012; Schuman and Cornig Reference Schuman and Cornig2012; Bartels and Jackman Reference Bartels and Jackman2014) and voting behavior (Firebaugh and Chen Reference Firebaugh and Chen1995; Franklin Reference Franklin2004; Bhatti et al. Reference Bhatti, Hansen and Wass2012; Dinas Reference Dinas2012; Erikson and Stoker Reference Erikson and Stoker2011; Rico and Jennings Reference Rico and Jennings2016; Daniele et al. Reference Daniele, Aassve and Le Moglie2023). A common finding in these studies is that the ‘impressionable years’ (Inglehart Reference Inglehart1967) of adolescence are central to the acquisition of political attitudes and habits that last a lifetime. Indeed, at this stage of psycho-social development, when individuals have less experience and therefore higher margins for cognitive gains, attitudes are more pliable and receptive to external stimuli (Sears and Valentino Reference Sears and Valentino1997; Bartels and Jackman Reference Bartels and Jackman2014).

Political socialization is most often associated with the circumstances in which young people first engage directly with politics in a democracy: during elections (Dinas Reference Dinas2012; Bhatti and Hansen Reference Bhatti and Hansen2012; Dassonneville and McAllister Reference Dassonneville and McAllister2018), when information is more readily available (Campbell and Wollbrecht Reference Campbell and Wollbrecht2006; Lau and Redlawsk Reference Lau and Redlawsk2018; Zeglovits and Aichholzer Reference Zeglovits and Aichholzer2014). Our argument develops two key findings in this literature: first, that party images and attachments formed when one first encounters party politics tend to remain ‘sticky’ thereafter (Tilley Reference Tilley2002; Jennings and Niemi Reference Jennings and Niemi2014; Tyler and Iyengar Reference Tyler and Iyengar2023); second, that the formation of enduring political attitudes depends on the availability and clarity of political information during the ‘impressionable years’ (Sears and Valentino Reference Sears and Valentino1997; Valentino and Sears Reference Valentino and Sears1998; Osborne et al. Reference Osborne, Sears and Valentino2011). These insights are reflected in our model’s distinction between ‘sorting’ and ‘cueing’ mechanisms, respectively.

At the same time, this paper advances our understanding of political socialization processes in two senses. First, while most existing studies focus either on issue or on party preferences, our contribution examines the congruence between the two. Secondly, we focus on the issue of immigration, which has not been central to this literature (with some recent exceptions, e.g., Jeannet and Drazanova Reference Jeannet and Drazanova2023 and Laaker Reference Laaker2024).

Issue salience and political change

The term ‘salience’ is often used but rarely defined (Epstein and Segal, Reference Epstein and Segal2000; Taylor and Fiske, Reference Taylor and Fiske1978; Wlezien, Reference Wlezien2005). On the one hand, it indicates the importance that political actors assign to different issues. On the other hand, it also captures the degree to which an issue is prominent in people’s minds (Taylor and Fiske Reference Taylor and Fiske1978). The concept has long been central to theories of issue voting (Edwards et al. Reference Edwards, Mitchell and Welch1995; Bélanger and Meguid Reference Bélanger and Meguid2008), which posit that individuals tend to choose the political platform that matches their preferences on the most salient issues (Clarke Reference Clarke2009; Lefkofridi et al. Reference Lefkofridi, Wagner and Willmann2014; Gidron Reference Gidron2022; Brader et al. Reference Brader, De Sio, Paparo and Tucker2020; Steiner and Hillen Reference Steiner and Hillen2021). Indeed, a great deal of political competition in democratic polities is not so much about swaying voters’ issue stances as it is a struggle over which issues are made salient to them (Green-Pedersen Reference Green-Pedersen2006; Green-Pedersen Reference Green-Pedersen2007; Baumgartner et al. Reference Baumgartner, Green-Pedersen and Jones2006; Rovny and Edwards Reference Rovny and Edwards2012; Wagner and Meyer Reference Wagner and Meyer2014; Lukes Reference Lukes2021).

Salience has emerged as a prominent conceptual tool to explain major patterns of political change of the last decades in Europe, including the ‘realignment’ of electorates away from the traditional economic left-right axis and the increase in support for new parties. In this perspective, these shifts are not primarily driven by changes in voters’ preferences but rather in their priorities: changes most prominently associated with the rise in salience of immigration and other non-economic issues (Rydgren, Reference Rydgren2008; Dennison and Geddes, Reference Dennison and Geddes2019; Ford and Jennings, Reference Ford and Jennings2020; Kustov et al,. Reference Kustov, Laaker and Reller2021; Dennison and Kriesi, Reference Dennison and Kriesi2023). Crucially for our purposes, recent work suggests that the extent to which immigration preferences are predictive of political and electoral outcomes has not only increased over time but also varies across cohorts (De Vries et al. Reference De Vries, Hakhverdian and Lancee2013; Steiner and Hillen Reference Steiner and Hillen2021; Steiner Reference Steiner2023; Mitteregger Reference Mitteregger2024; Jocker et al. Reference Jocker, Van der Brug and Rekker2025). This paper contributes to this line of research by clarifying the role of salience in this process of ‘generational realignment’: i.e., it is specifically cohorts who grew up in high-immigration salience environments that exhibit particularly high levels of voter-party agreement on immigration. Moreover, our operationalization of ‘issue congruence’ as the extent to which individuals say that they agree with their party of choice’s immigration position is novel in this literature, which instead tends to rely on the correlation between individuals’ and parties’ positions on polytomous scales.

Sorting

Our stylized theory of partisanship formation starts from the assumption that adolescents encounter politics already having developed a set of basic dispositions towards social objects. These may be formed through moral reasoning, a function of self-interest, or otherwise ‘inherited’ from their families and social context – likely, a combination of all three. In this framework, socialization is a distinctive window of time because partisanship is formed. We further posit that adolescents evaluate the available choice set of political platforms, attaching different weights to different issues: the more they believe an issue to be important, the more weight it will carry in determining party preference (RePass Reference RePass1971; Brader et al. Reference Brader, De Sio, Paparo and Tucker2020; Costello et al. Reference Costello, Toshkov, Bos and Krouwel2021). Much has been written, for instance, on the plight of ‘left-authoritarian’ voters in Europe, who tend to find themselves cross-pressured between left-liberal and right-authoritarian parties, and who have increasingly aligned themselves with the right as ‘sociocultural’ issues have come to the fore of political competition (Lefkofridi et al. Reference Lefkofridi, Wagner and Willmann2014; Walgrave et al. Reference Walgrave, van Erkel, Jennart, Lefevere and Baudewyns2020; Steiner and Hillen Reference Steiner and Hillen2021; Gidron Reference Gidron2022).

In this process of platform evaluation, the weights attached to each individual issue are in turn endogenous to the issue salience environment through two channels. The first is straightforward: individual-level salience. The more the public, the media, and parties talk about an issue, the more likely it will be that an individual will think and care about it. The second consists in how high issue salience reduces informational costs. When an issue dominates the agenda, information about the parties’ policy choice set on that issue will be ‘cheaper’, making partisan sorting on it less cognitively demanding than on issues that parties do not feel the need to emphasize (and therefore require more information-seeking ‘work’ from voters).

Finally, in this perspective, partisan choices made during socialization are particularly relevant to partisan identity, as voting reinforces partisan loyalty (Sears and Valentino Reference Sears and Valentino1997; Sears and Funk Reference Sears and Funk1999; Dinas Reference Dinas2014). Thus, even if the process of policy platform evaluation and party choice may be iterated at every election, later in an individual’s life, this will take place downstream from previous choices, which have cemented partisan identities. Thus, salience contexts experienced during political socialization – when, by assumption, partisanship is something akin to a blank slate – should keep casting a shadow on party choice for voters.

A possible fly in the ointment for our argument is that, over an individual life course, the choice set of parties – and even political systems in toto – may change, making past partisan choices and loyalties moot. However, we believe that contextual factors affecting partisan sorting at the time of socialization can still influence long-term voter-party issue congruence. As well as partisan identity, political choice in adolescence may in fact crystallize forms of ‘macro-partisanship’ (such as being ‘on the left’, or ‘on the right’, or ‘pro-independence’, or ‘Peronist’), which are stable in spite of party label churn within these ideological families. Thus, even if parties come and go, individuals’ socialization experience may still have an influence over the choice set of parties one considers (Oscarsson and Rosema Reference Oscarsson and Rosema2019).

Cueing

If the preceding paragraphs outlined a process through which individuals endowed with issue preferences select themselves into a partisan identity conditional on issue salience, the mechanism detailed below is the exact mirror image: the process through which individuals endowed with partisan identities develop their issue preferences conditional on issue salience. The assumptions about adolescents’ political attitudes are correspondingly flipped around: here, individuals ‘inherit’ partisanship from their parents and, more generally, familial background (Campbell and Wollbrecht Reference Campbell and Wollbrecht2006; Rico and Jennings Reference Rico and Jennings2016), and their policy preferences are endogenous to the interaction between partisanship and context. Specifically, people will tend to adopt issue stances that are consistent with their pre-existing partisanship and thus increase party-voter issue congruence, in line with the findings of existing research on polarization and partisan-motivated reasoning (Levendusky Reference Levendusky2010; Lavine et al. Reference Lavine, Johnston and Steenbergen2012; Stanley et al. Reference Stanley, Henne, Yang and De Brigard2020).

A ‘cue’ is ‘a message that people may use to infer other information and, by extension, to make decisions’ (Bullock Reference Bullock2011, p.4). Cues allow individuals to form their stances on different issues in a cognitively cheap way (Vössing Reference Vössing2021) by matching the positions of co-partisan elites (Mondak Reference Mondak1993). The higher the context-level salience of an issue, the easier it will be for voters to acquire information on where parties stand on it. As a consequence, partisanship-consistent position-taking will be easier on issues that are cued more heavily than on issues that are underplayed by parties, resulting in higher voter-party congruence on salient issues than on non-salient ones, as documented empirically across a range of political contexts (Lefkofridi et al. Reference Lefkofridi, Wagner and Willmann2014; RePass Reference RePass1971; Neundorf and Adams Reference Neundorf and Adams2018; Walgrave et al. Reference Walgrave, van Erkel, Jennart, Lefevere and Baudewyns2020; Costello et al. Reference Costello, Toshkov, Bos and Krouwel2021).

The last step left to do to complete the argument is to connect this idea of easier partisanship-consistent position-taking on salient issues to the timing of political socialization. The core of the ‘impressionable years’ thesis is that ‘individuals are most susceptible to political attitude and behavior change at younger ages’ and that ‘political attitudes crystallize as individuals near adulthood, making it improbable that these attitudes will shift during one’s adult years’ (Rosenthal and Farhart Reference Rosenthal and Farhart2023, p. 1936; see also Niemi and Jennings Reference Niemi and Jennings1991; Krosnick and Alwin Reference Krosnick and Alwin1989). Under this assumption, the political attitudes that are more heavily cued during someone’s early experience of politics should remain, over time, the ones that are most consistent with their partisanship. Conversely, attitudes that form and ‘crystallize’ in a relatively cue-free socialization context are likely to be less bundled with partisan or ideological belief systems.Footnote ³

Data and measures

We employ data from Europinions, a web-based survey conducted between 2017 and 2019 on samples of adults from ten European countries (Goldberg et al. Reference Goldberg, van Elsas, Marquart, Brosius, de Boer and de Vreese2021): the Netherlands, Denmark, Germany, Hungary, Spain, Czechia, Greece, France, Poland, and Sweden. Two versions of the dependent variable were recorded from a battery of items asking, ‘How suitable do you think each of the following parties is to deal with the issue of immigration?’ about the six top parties in each country (eight for the Netherlands), with possible responses ranging from 1 (completely unsuitable) to 7 (completely suitable). The item was asked in all samples only once: in April 2019, ahead of the European Parliament elections in May, so the data structure is that of a simple cross-section. First, a congruence score (immigration) integer variable was coded by assigning to each respondent the 1–7 point rating of the competence on immigration of the party the respondent intends to vote for in a national election. Secondly, we coded a binary congruence (immigration) variable, taking the value of 1 if the respondent’s party of choice is the top rated on immigration (inclusive of ties), and taking the value of 0 if the respondent rates a different party as more suitable than their party of choice to deal with the issue. Respondents who did not report a vote intention, did not rate parties on the issue, or intended to vote for a minor party (i.e., one that did not appear on the issue competence rating item) were droppedFootnote ⁶ .

We coded from Europinions also a series of ‘base’ control variables included in all models: age, education, gender, domicile, religiosity, and strength of party identification Footnote ⁷ . Moreover, we code respondents’ left-right self-identification, the party family (of their vote intention), and an immigration attitudes scale: we use these as covariates in model specifications where we assume either exogenous partisanship or exogenous preferences.Footnote ⁸ Finally, for the purposes of conducting placebo tests, ‘congruence score’ and ‘binary congruence’ variables analogous to our main dependent variables were coded for the four other issues that Europinions respondents were asked to rate parties’ suitability over: the EU, social welfare, the economy, and terrorism.

The key independent variable of interest was measured from Comparative Manifesto Project (CMP) data on parties’ thematic focus in their electoral manifestos (Lehmann et al. Reference Lehmann, Franzmann, Al-Gaddooa, Burst, Ivanusch, Regel, Riethmüller, Volkens, Weßels and Zehnter2025). It was computed at the country-election level as the sum of the number of quasi-sentences relating to ‘multiculturalism’ (CMP codes 607 and 608) in each manifesto as a percentage of all policy quasi-sentences, weighted by the party’s share of the vote in that electionFootnote ⁹ . In an election year where no party has a multiculturalism-related pledge, the variable takes the value of 0; in theory, the variable can take values up to 100 in a hypothetical election where all party manifestos include only multiculturalism-related policy quasi-sentences.Footnote ¹⁰ Figure 1 shows country-election estimates of multiculturalism salience from the CMP for the countries in the Europinions sample used for Study 1. The increase in salience of the issue over time varies substantially across countries, reaching the highest values in Denmark, the Netherlands, and Sweden. While recent elections tend to coincide with increased salience of the issue, this upward trend is not monotonic, with earlier ‘spikes’ in the 1990s in countries like France and Germany (as observed also in Study 2).

Figure 1. Multiculturalism salience in manifestos, by country election, from CMP.

Of course, ‘salience of multiculturalism’ is not the same as ‘salience of immigration’; however, this is the closest proxy we can obtain from text data without extensive recoding of the primary sources, which add up to over 1,200 manifestos in over 150 elections. Reassuringly, however, the CMP multiculturalism salience measures for the manifestos of the countries and parties included in the analysis are highly correlated ( $ r=0.78$ ) with those compiled by Dancygier and Margalit (Reference Dancygier and Margalit2019), who undertook a substantial manual classification effort (the ‘Immigration in Party Manifestos’ dataset) for a subset of country elections, limiting their analysis to the main left and right parties in each contest.Footnote ¹¹ Moreover, the literature on multiculturalism highlights how it was a key frame for political mobilization and competition on the issue of immigration in Europe (Ivarsflaten and Sniderman Reference Ivarsflaten and Sniderman2022). In sum, although historically and in a strictly theoretical sense multiculturalism is not the same as immigration, in this particular context it is a very proximate concept.

The main independent variable, multiculturalism salience in the Europinions dataset, takes the value of the CMP country-election-level multiculturalism salience in the first election the respondent was eligible to vote in, using imputed birth year as the linking variable. This follows from the assumption in the political socialization literature (Sears and Valentino Reference Sears and Valentino1997; Bhatti et al. Reference Bhatti, Hansen and Wass2012; Daniele et al. Reference Daniele, Aassve and Le Moglie2023) that the first election after the age of majority is a particularly important catalyst of the development of political attitudes. The value is recorded as ‘missing’ for respondents who were older than 23 in the first election they were eligible to vote in (this is the case for the older Czech, Hungarian, Polish, Greek, Spanish, and East German respondents who came of age prior to the democratic transition). Moreover, we use the same protocol to record, for each respondent, (1) the salience of multiculturalism in the second election they were eligible to vote in, the third election they were eligible to vote in, the fourth, and so on; and (2) the salience of multiculturalism in the election before they became eligible to vote, two elections before they became eligible to vote, three elections before, and so on. Thus, noting $ t$ as the first election the respondent was eligible to vote, we code values of multiculturalism salience from $ t-10$ to $ t+10$ . Finally, we recorded placebo-independent variables, recording the salience of other issues (in the first election after vote eligibility): welfare, economic ideology, national way of life, moral issues, labor unions, foreign policy, the European Union, the environment, and corruption. Table 1 reports descriptive statistics for the main variables of interest.

Table 1. Descriptive statistics (Study 1)

We employ linear regression models to gauge the change in the respondent’s rating of their own party on immigration associated with changes in the salience of multiculturalism in the first election a respondent was eligible to vote in. In all model specifications, we control for a fixed set of base covariates: gender, age, age squared, domicile, education, religiosity, and strength of party identification, and we include country fixed effects. Note that, as the data is cross-sectional, age and cohort are perfectly collinear; thus, controlling for a polynomial of age is the same as controlling for cohort; within the limitations of cross-sectional data, this adjustment should net out some of the cross-cohort heterogeneity due to factors other than socialization contexts. The main model (1) regresses congruence score (immigration) on the independent variable multiculturalism salience and the base covariates. In model (2), we further control for left-right self-positioning; in model (3), we let the slope for left-right self-positioning vary by country; in model (4), we control for the nominal variable party family instead of left-right self-positioning. Models 2–4 thus assume exogenous partisanship. In model (5), we control for immigration attitudes; in model (6), we let the slope for immigration attitudes vary across countries. We employ identical model specifications and control sets in logistic regression models where we regress binary congruence (immigration) on salience plus covariates. Throughout, we cluster standard errors at the country/birth-year level to account for the clustered nature of treatment assignment.

The linear and logistic regression coefficients for both sets of models are presented in Table 2 in Section 5.2. We repeat this exercise, substituting measures of salience of multiculturalism in elections other than the first the respondent was eligible to vote in for the main independent variable. These alternative estimates of the relationship between salience context and congruence serve primarily as a term of comparison for the main estimate and are plotted in Figure 2 in section ‘Results’ alongside average marginal effects from the main models. Finally, in section ‘Further analysis’, we discuss a raft of robustness checks, placebo tests, and additional analysis drawing on different data sources that we employ to probe the validity of the analysis’ substantive inferences.

Table 2. Estimated change in congruence associated with a 1-point increase in salience of multiculturalism in the first election the respondent was eligible to vote in. The top panel reports OLS coefficient estimates, and the bottom panel reports log-odds coefficients. Base controls include gender, education, domicile, religiosity, and strength of party identification. Standard errors are clustered at the birth-year/country level

Note: *p $ \lt$ 0.1; **p $ \lt$ 0.05; ***p $ \lt$ 0.01.

Figure 2. Estimated change in own party’s rating on immigration (top panel) or probability of being ‘congruent’ on immigration (bottom panel) associated with a 1 pt increase in multiculturalism salience in past election ( $ t$ , in red is first election R was eligible). ATEs presented correspond to regression coefficients from model 1 in Table 2; analogous visualizations of results across all 6 model specifications are presented in the Appendix, section A.1.

Results

Table 2 shows the results of a series of regression models where the two versions of the dependent variable – congruence score and binary congruence – are regressed on the salience of multiculturalism in the first election the respondent was eligible to vote in. For ‘congruence score’, we employed linear regression so that the coefficients in the top panel of the table indicate the estimated change in the respondent’s rating of their own party of choice on the issue of immigration associated with a one-point increase in the salience of multiculturalism when they entered the electorate. For ‘binary congruence’, we employed logistic regression so that the coefficients are to be interpreted as the expected change in log-odds of the probability of being ‘congruent’ on immigration (i.e., rating one’s party of choice as the best to handle the issue among those surveyed). The six model specifications all include controls for age, age squared, gender, education, domicile, religiosity, and strength of party identification, as well as country fixed effects. Models 2 through 4 further control for variables capturing partisanship; models 5 and 6 control for attitudes to immigration, in line with the theoretical mechanisms that assume either partisanship or issue preferences as exogenous to political socialization.

Substantively, the OLS models predict that someone who first became eligible to vote in an election that records the highest levels of salience in our data (10.22, representing the percentage of manifesto sentences related to multiculturalism) would rate their party of choice 0.28–0.34 points higher than someone socialized in an election where immigration was not salient at all. While the log-odds coefficients for the ‘multiculturalism salience’ variable in the bottom panel of the table are not easily interpreted substantively, the associated average marginal effects are comprised between 0.008 and 0.01. This corresponds, approximately, to an average 0.8–1 percentage point increase in the probability of ranking one’s party as the best to handle immigration for each unit increase in salience. The results are consistent with the hypothesis that exposure to high migration salience environments in one’s formative years leads to higher voter-party congruence on the issue of immigration. Crucially, we find that socialization effects remain roughly constant in magnitude and significance even when we condition on partisanship (models 2–4) and on attitudes towards immigration (models 5–6). This implies that it is unlikely that the higher degree of congruence associated with the experience of high immigration salience during the ‘impressionable years’ is due to these cohorts being distinctively more likely to side with certain parties or having distinctively pro- or anti-immigration views. Rather, it suggests that individuals belonging to these cohorts are more neatly ‘sorted’ into parties whose immigration positions they share, regardless of the specific type of parties and the polarity of these individuals’ immigration positions.

Figure 2 shows the results of the additional analysis conducted by re-running the same regression model as ‘model 1’ in Table 2, but varying the timing at which the salience of multiculturalism is measured. In both panels, the solid circles in red represent average marginal effect estimates, with 95% confidence intervals, for salience measured in the first election the respondent became eligible to vote in. Specifically, the solid circle in red in the top panel represents the predicted change in the congruence score associated with a one-point increase in multiculturalism salience in the first election the respondent was eligible to vote in, and the solid circle in red in the bottom panel represents the predicted change in the probability of being ‘congruent’ on immigration. Letting the first election the respondent was eligible to vote be $ t$ , the open blue circles show instead the average marginal effects for ‘multiculturalism salience’ measured at $ t-1$ , $ t-2$ , etc. (i.e., when the respondent was younger, or in some cases unborn), as well as $ t+1$ , $ t+2$ , (i.e., when the respondent was older). As discussed, these alternative estimates are not, strictly speaking, placebos: issue salience is likely correlated across elections, and changes in partisanship and attitudes may take place at different times over an individual’s life course, so that we should not necessarily expect a null. However, they are a useful benchmark for understanding to what extent the context experienced in one’s ‘impressionable years’ is distinctive in its consequences for party-voter congruence.

In the top panel of Figure 2, where the dependent variable is ‘congruence score’, the salience of multiculturalism in the first election a respondent was eligible to vote in is the only measure that is a consistently significant positive predictor of congruence: alternative socialization estimates have statistically and substantially insignificant average marginal effects across model specifications (see also Appendix section A.1).Footnote ¹² In the bottom panel of Figure 2, where the dependent variable is ‘congruence score’, we find a comparable estimate for the slope of multiculturalism salience in the last election before respondents became eligible to vote. We have reasons to believe that this can be interpreted substantively: adolescents may be receptive to political cues before they become eligible to vote. In fact, this finding actually chimes with the findings from Study 2, which suggest that the clearest effects of socialization context correspond to immigration salience measured at 16 and 17 years of age, so just before voting age (18 in most countries). Overall, these results are consistent with the established finding that the first election is particularly important for the formation of political habits (Pacheco Reference Pacheco2008; Dinas Reference Dinas2012; Daniele et al. Reference Daniele, Aassve and Le Moglie2023); however, they also suggest that some of these processes may start – at least for some individuals – slightly earlier than legal adulthood, when adolescents first start paying attention to political news.

Further analysis

In part A of the Appendix, we present a series of ancillary analyses of the cross-country analysis: robustness checks, placebo tests, and a new analysis of party-voter alignment on immigration with European Social Survey and Chapel Hill Expert Survey data.

Data triangulation: analysis with the European Social Survey (ESS)

In section A.6 of the Appendix, we further validate our measure of immigration salience at the time of socialization by showing that it also predicts higher levels of alignment between voters’ positions on immigration and the immigration stances of their party of choice. To do so, we use data from the same 10 countries as in Study 1 on voters’ preferences (from the European Social Survey, ESS waves 3–10, 2006–2022) and on parties’ positions (from Chapel Hill Expert Survey, CHES, 2006–2019) (ESS 2024; Jolly et al. Reference Jolly, Bakker, Hooghe, Marks, Polk, Rovny, Steenbergen and Vachudova2022). Extending the ‘Age-Period-Cohort’ modeling approach adopted in Jocker et al. (Reference Jocker, Van der Brug and Rekker2025), we fit a series of OLS models, where the vector of immigration positions of the party each respondent voted for in the past national election is regressed on each respondent’s own immigration position. Crucially, we interact the respondent’s immigration position with both the cohort (a categorical ‘decade of birth’ variable) and our moderator of interest: multiculturalism salience in the first election of eligibility. Conditioning on a number of ‘base’ controls – age group, gender, education, religiosity, social class, and country-year fixed effects – and further political covariates in alternative model specifications – left-right self-placement and party family – we find robust evidence in favor of the testable implications of our argument. That is, the relationship between individuals’ immigration attitudes and the immigration position of their party of choice is stronger among those who grew up at times of higher salience of immigration. This core finding is visualized in the predicted values plot of Figure 3; regression tables, additional results, and a more detailed outline of this analysis are provided in section A6 of the Appendix.

Figure 3. Predicted voter-party alignment on immigration (ESS data): marginal means and 95% confidence intervals. The marginal means are computed from an OLS model regressing the voter’s party of choice’s immigration position on the voter’s immigration position, interacted with salience of immigration in the first election of eligibility; the model controls for age group, cohort, cohort $ \times $ voter’s immigration position, social class, education, religiosity, and country-year fixed effects.

Data and measures

The within-country analysis draws on data from two long-running German surveys: ARD DeutschlandTrend (ARD-Landesrundfunkanstalten and Infratest-dimap 2023; data for 1998–2021) and Politbarometer (Forschungsgruppe Wahlen 2023; data for 1986–2021). The former was used to measure immigration issue congruence and the individual-level covariates, while the latter was used to estimate past immigration issue salience at the federal state/year level. Issue congruence is measured from survey items that read, ‘I will now read you some political objectives. Please tell me which party in Germany you trust the most to achieve them. [Carry out a good policy on immigration and asylum]’. The question is available for 13 years between 1998 and 2021. We code the dependent variable, binary congruence (immigration), by assigning the value of 1 if the respondent’s preferred party to deal with immigration is the same as their vote intention, and 0 if they are different. Respondents without a stated vote intention are excluded from the analysis. We code similar ‘placebo’ dependent variables for other political objectives asked in the same survey years at least as often as the immigration question: advocating for social justice, securing retirement pensions in the long term, implementing a good labor market policy, attracting investments in Germany, implementing a good healthcare policy, implementing a good foreign policy, and implementing a good environmental policy. From ARD DeutschlandTrend data, we also code covariates for age, birth year, education (three-category), income (tercile of survey item bands), federal state, gender, and party.

The independent variable is measured from the Politbarometer data as the weighted percentage of respondents who cite ‘immigration’ or related terms as one of the top two concernsFootnote ¹⁴ at the federal state-year level. Figure 4 visualizes these estimates. Crucially for our purposes, there are discernible cross-state differences in trends of immigration salience. In the first wave of high salience of immigration – which coincided with ‘one of the most contentious political debates in post-war Germany’ (Kirchhoff and Lorenz Reference Kirchhoff, Lorenz, Rosenberger, Stern and Merhaut2018, p. 51) on asylum reform, following a spike in immigration (Münz and Ulrich Reference Münz and Ulrich1998; Green Reference Green2013) and high-profile episodes of anti-migrant violence and mobilization (Karapin Reference Karapin1999) – the issue was much more salient in Western than in Eastern States.

Figure 4. Estimates of state-year-level salience of immigration (data from Politbarometer).

This is likely due to two main factors. First, the dramatic disparity in economic conditions following reunification meant that opportunities for migrant labor were heavily concentrated in the West – which saw its foreign population increase from 4.5m to 7m between 1988 and 1994 (Münz and Ulrich Reference Münz and Ulrich1998, p. 38) – leading to highly geographically unequal migration pressures. In fact, unlike the East, the West had been exposed to such pressures since the 1980s (Thränhardt Reference Thränhardt2002). Secondly, in this phase the East was undergoing massive economic transformations, putting pressing issues of unemployment, labor market disruption, hardship, and displacement to the fore of East Germans’ concerns (Burda and Hunt Reference Burda and Hunt2001; Kitschelt Reference Kitschelt2003).Footnote ¹⁵ The salience environment of the 1990s migration wave contrasts markedly with the context of the second ‘peak’ in migration concerns, which coincided with the 2015–2016 European migrant crisis. This was a heavily ‘nationalized’ phase of migration politics: unlike the 1990s, most newcomers were asylum seekers rather than labor migrants, so that Germany’s asylum quota system distributed migration pressures more evenly over the country’s territory than had been the case in the 1990s (Schaub et al. Reference Schaub, Gereke and Baldassarri2021).

Individual-level survey data from ARD DeutschlandTrend is linked to immigration salience data at the state/year level via the respondent’s state and their age. Specifically, letting $ x$ be any age between 12 and 65, each respondent is associated with an immigration salience at age x variable, which corresponds to the average immigration salience estimate from Politbarometer for that state in the three years when the respondent was between $ x-1$ and $ x+1$ years of age. For instance, a 40-year-old survey respondent from Hamburg polled in 2020 will receive as a value for immigration salience at age 20 the average of estimates for immigration salience in Hamburg in 1999, 2000, and 2001. This ‘three-year window’ approach allows us to smoothen the distribution of the independent variable, as yearly salience estimates are very ‘peaky’, while our assumption is that political socialization takes place over a diffuse period of time. In any case, in part B of the Appendix, we show that we can reproduce our main results with both narrower (one-year) and broader (five-year) windows. Table 3 shows descriptive statistics for some of the key variables used in Study 2.

Table 3. Descriptive statistics (Study 2)

Note: Figures computed from a dataset including only observations for which the dependent variable ‘Binary Congruence immigration’ could be coded.

It is worth noting that the data we use impose some limitations on our measurement. First, because our salience estimates only go back to 1986 for West German states and to 1990 for East German states, we often have to discard a sizable chunk of older respondents in the sample, for whom immigration salience at age $ x$ cannot be computed when $ x$ is relatively low. Secondly, the ARD data does not record the state where the respondent grew up or where she was born, so we have to make do with the state of residence as a proxy, which – while plausible for a majority of respondents – is clearly unideal.

We implement four logistic regression models for each independent variable (immigration salience at age $ x$ , with $ x$ comprised between 12 and 65), with binary congruence (immigration) as the dependent variable. In all cases, we control for gender, income (factor), and education (factor) and introduce federal state and survey-year fixed effects. In models (1) and (2), we control for birth year and birth year squared; in models (3) and (4), we control for age and age squared. In models (2) and (4) we further control for party (i.e., assuming exogenous partisanship). We cluster standard errors by survey year throughout.

Results

Moving now to the analysis conducted on repeated cross-sectional data from Germany, Figure 5 shows the average marginal effects associated with an increase in salience of immigration when the respondent was $ x$ years of age on their probability of supporting the party they deem best to deal with immigration, for values of $ x$ ranging from 12 to 65. Recall that the independent variable estimate corresponds to the percentage of respondents who cite immigration as one of their top two concerns, aggregated at the state-year level, averaged over the three-year window centered in the year when the respondent was $ x$ years old. Models 1 and 2 control for the second-order polynomial of birth year, which means that the estimates are to be interpreted in terms of the predicted variation in congruence associated with exposure to immigration salience among individuals who had otherwise similar cohort-level experiences. Models 3 and 4 swap a quadratic function of age for birth year. Finally, models 2 and 4 control for party of choice, which serves as a proxy for the assumption of exogenous partisanship.

Figure 5. Estimated average marginal effect on probability of being ‘congruent’ with their party on immigration of immigration salience when R was $ x$ years old. All models include controls for gender, income band, education, year (factor), and federal state fixed effects. Models 1 and 2 include controls for birth year and birth-year squared; models 3 and 4 include controls for age and age squared. Models 2 and 4 further control for party voted. Standard errors clustered at the survey level.

As shown in the figure, we find small positive effects for values of $ x$ comprised between 16 and 18 (highlighted in red). Effect estimates are consistently significant at the conventional 95% confidence level for immigration salience measured at age 17 in all four models and for ages 16 and 18 in two out of four. The point estimates for $ x=16$ and $ x=17$ are consistently the highest recovered.Footnote ¹⁶ Substantively, in our sample, the estimated average effect of a one-percent increase in immigration salience when the respondent was 17 is an increase in probability of being ‘congruent’ on immigration just north of one percentage point. The finding that the clearest effects are for estimates centered at 17 years of age is consistent with the finding from Study 1 that the strongest effects on immigration congruence correspond to salience of multiculturalism measured in the first election the respondent was eligible to vote in and – to a lesser extent – in the election just before. Effect magnitude is also comparable: the largest average marginal effect estimates are one order of magnitude smaller than the one recovered in Study 1, which is to be expected given that the independent variable measured from Politbarometer survey data has a standard deviation about 10 times larger than the salience scores coded from CMP data.

Finally, in part B of the Appendix, we present ancillary analysis where we further probe the robustness and assumptions of the models. First, in section B.1.1, we let the size of the ‘socialization window’ vary and find that the results are substantively similar whether we consider immigration salience context measured as the average of the three years the respondent was aged 16–18 (as in the main analysis), the five years between ages 15 and 19, or exactly in the year the respondent was 17. Second, in section B.1.2, we show that we can reproduce the main results of the analysis aggregating the salience variable at East/West rather than at the state level. Finally, in section B.2, we present the results of a series of placebo tests, where we substitute salience of immigration when the respondent was 17 with salience of other issues, and others in which we substitute voter-party congruence on immigration with congruence on other issues.