Introduction
Populist radical right (PRR) parties are often more successful in some regions, municipalities, and neighbourhoods than in others, overperforming in, for example, rural and/or peripheral areas, suburban towns, or working-class neighbourhoods in large cities. At the same time, there are many rural, peripheral, suburban and working-class areas where few people support PRR parties. The relationship between local context and PRR support is thus not straightforward. Scholars have sometimes found that economic and/or cultural contextual factors, such as economic decline or the presence of immigrants, matter for PRR support, but these relationships are often contingent (Berning Reference Berning2016; Bowyer Reference Bowyer2008; De Blok and Van der Meer Reference De Blok and van der Meer2018; Dinas et al. Reference Dinas2019; Dustmann, Vasiljeva, and Piil Damm Reference Dustmann, Vasiljeva and Piil Damm2019; Hangartner et al. Reference Hangartner2019; Janssen et al. Reference Janssen2019; Savelkoul, Laméris, and Tolsma Reference Savelkoul, Laméris and Tolsma2017).
In this paper, we argue that the role of contextual factors can be well understood by assessing how their effects are mediated and moderated. Instead of a one-size-fits-all model, we argue that the relationship between context and political outcomes is more complex than sometimes assumed. First, different contextual factors – cultural and economic – can influence citizens through similar individual-level mechanisms (mediation), producing populist and nativist attitudes and PRR voting. Second, some citizens are more strongly influenced by the context in which they live than others (moderation).
Regarding the former – mediation – we argue that we must examine the causal mechanisms linking economic and cultural contextual factors to PRR voting. We expect that different kinds of local experiences create similar grievances, resulting in a backlash against immigrants and elites (Harteveld et al. Reference Harteveld2022). Thus, we hypothesize that three different contextual developments – immigration, economic hardship, and demographic decline – can contribute to a generalized sense of local decline, translating into populist and nativist attitudes and, other things being equal, PRR voting. For instance, PRR support rooted in a sense of nativism might emerge in areas that do or do not feature many citizens with an immigration background. Including perceived local decline as a mediator in our models explains why PRR voters have similar attitudes despite living in areas that experience different kinds of societal trends.
Regarding moderation, we do not expect that contextual factors will influence all residents of a given area to the same extent. It has been demonstrated that some citizens – especially the higher educated and those with their main ties outside their neighbourhood – are less likely to be affected by the context in which they live than others (Fitzgerald Reference Fitzgerald2018; Small and Feldman Reference Small, Feldman, van Ham, Manley, Bailey, Simpson and Maclennan2012; Steenvoorden and Van der Meer Reference Steenvoorden and van der Meer2021). The presence of immigrants, economic hardship, or demographic decline does not necessarily foster PRR support if citizens are ‘shielded’ from the impact of their environment. Thus, we hypothesize that citizens' resources and their embeddedness function as key moderators of the relationship between contextual factors on the one hand and perceived local decline, populist and nativist attitudes, and PRR voting on the other hand. In other words, we expect that citizens with more resources and a weaker place-based identity are less likely to respond to the contextual features that, among those with less resources and a stronger place-based identity, would lead to PRR support.
To test these hypotheses, we use a unique dataset that combines fine-grained contextual and survey data from four West European countries: France, Germany, Great Britain, and the Netherlands. We conducted geocoded surveys in each of these countries on large geo-stratified samples of around 20,000 respondents in the three larger countries and 8,000 respondents in the Netherlands. These unique data enable us to test the causal mechanisms linking local contexts to individual attitudes and political behaviour. Moreover, the data allow us to test our hypotheses based on contextual data at the local neighbourhood level rather than at higher levels that are comprised of larger geographical units, such as regions or provinces (Georgiadou, Rori, and Roumanias Reference Georgiadou, Rori and Roumanias2018; Kestilä and Söderlund Reference Kestilä and Söderlund2007; Lubbers and Scheepers Reference Lubbers and Scheepers2000) or municipalities (Berning Reference Berning2016; Bowyer Reference Bowyer2008; Coffé, Heyndels, and Vermeir Reference Coffé, Heyndels and Vermeir2007; Rink, Phalet, and Swyngedouw Reference Rink, Phalet and Swyngedouw2008) that are normally analysed in comparative studies. We rely on considerably smaller areas with between 1,000 and 6,000 inhabitants. Although context effects may also play out on larger scales (Van Wijk, Bolt, and Johnston Reference Van Wijk, Bolt and Johnston2019), especially in more segregated areas, a lower level of analysis is particularly fruitful (Biggs and Knauss Reference Biggs and Knauss2012). We move the analyses closer to the so-called ‘locales’ – our respondents' immediate surroundings – that shape their everyday experiences and interactions (Johnston and Pattie Reference Johnston and Pattie2006, 43–44; Harteveld and Van der Brug Reference Harteveld and van der Brug2023).
We test our hypotheses using a harmonized design spanning four countries rather than the single countries that previous studies on small geographical units (such as voting districts or neighbourhoods) examined (De Blok and Van der Meer Reference De Blok and van der Meer2018; Evans and Ivaldi Reference Evans and Ivaldi2021; Janssen et al. Reference Janssen2019; Rydgren and Ruth Reference Rydgren and Ruth2013; Savelkoul, Laméris, and Tolsma Reference Savelkoul, Laméris and Tolsma2017; Van Wijk, Bolt, and Johnston Reference Van Wijk, Bolt and Johnston2019). The same survey items were fielded in all four countries, and contextual indicators were harmonized across these contexts. Our design allows us to test our theories in typical post-industrial Western European democracies that have seen a surge in PRR mobilization but vary considerably with respect to their electoral and party systems, migration patterns, and citizenship regimes. We aim to assess whether similar mechanisms underlie PRR support across these contexts.
Our analyses confirm that perceived local decline mediates the effects of context on populist and nativist attitudes and PRR voting. They also demonstrate that, while immigrant presence is the most universal predictor of PRR support, its effect is moderated by education levels. While citizens with lower levels of education are affected by the presence of immigrants, those with higher education levels are not. This latter finding fits the broader literature on context effects, which shows that such effects are highly conditional.
By showing that the relationship between contextual factors and PRR support is mediated and moderated, we make two main contributions to the literature on PRR success. First of all, our study shows that to understand PRR support, we need to understand its equifinality: PRR parties can thrive in different contexts because different contextual developments can generate the same types of grievances in the form of perceived local decline. These contextual factors might be orthogonal, so focusing on only one contextual indicator would paint a biased picture. Second, our model highlights that some areas are highly resistant to developing PRR support, even though the contextual conditions conducive to it are present. If citizens with high levels of education populate areas, PRR parties are unlikely to flourish, even when conditions are favourable. Thus, when studying geographical patterns of PRR support, composition and context effects should be considered.
Theory
The role of context in fostering PRR support has been widely studied, but no consensus has been reached about whether or how contextual factors impact the PRR vote. We briefly discuss the literature on context effects, followed by a discussion of the two remaining inconsistencies and how we intend to shed light on these.
Contextual Explanations for PRR Support
The Populist Radical Right (PRR) is a relatively new party family that emerged in the 1980s. Following Mudde's (Reference Mudde2007) widely accepted definition, three ideological elements set them apart from their mainstream competitors: they believe that non-native elements (persons and ideas) are a threat to the homogeneous nation-state (nativism), they are aggressive towards outgroups and expect submission to the ingroup while espousing a highly conventional view of society (authoritarianism), and they embrace a Manichean view of society that pits ‘pure people’ against the corrupt elite (populism). Unlike parties of the Extreme Right, they do not openly campaign for an authoritarian regime type, but they do reject and undermine key elements of liberal democracy.
There is growing evidence that PRR support is fostered, under certain conditions, by the presence of immigrants and economic hardship (Arzheimer Reference Arzheimer2009; Dinas et al. Reference Dinas2019; Dustmann, Vasiljeva, and Piil Damm Reference Dustmann, Vasiljeva and Piil Damm2019; Georgiadou, Rori, and Roumanias Reference Georgiadou, Rori and Roumanias2018; Golder Reference Golder2003; Hangartner et al. Reference Hangartner2019; Rydgren and Ruth Reference Rydgren and Ruth2013; Savelkoul, Laméris, and Tolsma Reference Savelkoul, Laméris and Tolsma2017; Van Wijk, Bolt, and Johnston Reference Van Wijk, Bolt and Johnston2019). More recently, studies have suggested that demographic decline might also play a role (Dijkstra, Poelman, and Rodríguez-Pose Reference Dijkstra, Poelman and Rodríguez-Pose2020; Harteveld et al. Reference Harteveld2022). The presence of immigrants is expected to increase PRR support by fostering nativism, while the latter two factors supposedly breed political discontent, which resonates with PRR parties' populist message. Below, we briefly discuss each contextual factor in turn.
The reasoning for why and how the presence of immigrants or other perceived ethnic or religious outgroups would matter for PRR support is straightforward. Generally, PRR support follows individual anti-immigrant sentiments or nativism more broadly (Arzheimer Reference Arzheimer and Rydgren2018; Lubbers, Gijsberts, and Scheepers Reference Lubbers, Gijsberts and Scheepers2002; Rydgren Reference Rydgren2007; Van der Brug, Fennema, and Tillie Reference Van der Brug, Fennema and Tillie2000), which in turn can be expected to respond to actual patterns of immigration. Group threat theory suggests migration leads to more perceived threats (Blalock Reference Blalock1967). In this perspective, the mere presence of migrants initiates anticipation of negative consequences for host society members’ economic and social well-being (Quillian Reference Quillian1995; Scheepers, Gijsberts, and Coenders Reference Scheepers, Gijsberts and Coenders2002). The contact theory contrasts this perspective: the presence of immigrants in a neighbourhood increases inter-ethnic contacts, which might reduce prejudices and perceptions of ethnic threat (Allport Reference Allport1954; Pettigrew and Tropp Reference Pettigrew and Tropp2006). Regardless, studies tend to find a positive relationship between the presence of immigrants or ethnic minority members and nativist attitudes and, by extension, PRR support (for recent studies with strong inferential designs, see Dinas et al. Reference Dinas2019; Dustmann, Vasiljeva, and Piil Damm Reference Dustmann, Vasiljeva and Piil Damm2019; Hangartner et al. Reference Hangartner2019).
In the case of economic hardship, the causal mechanism is more complex (Bowyer Reference Bowyer2008). PRR parties do not generally mobilize on clear-cut socio-economic policies, nor do they have outspoken views on such issues, nor do they ‘own’ them (Mudde Reference Mudde2007). PRR parties regularly blame unemployment on immigrants and the elites who are allegedly responsible for their entry into the country. These parties amplify feelings of (fraternal) deprivation by pitting ‘deserving’ in-group members against alleged abuse of welfare arrangements by members of outgroups (Abts et al. Reference Abts2021).
Even in areas with few immigrants, economic hardship can be expected to render voters susceptible to this specific appeal of PRR parties. Ongoing industrial transformations are reducing the financial returns and the social status associated with a below-college education, and PRR parties capitalise on these developments by appealing to workers (Gidron and Hall Reference Gidron and Hall2017; Kurer Reference Kurer2020) by blaming elites and immigrants for the real or perceived decline. Even those who live in areas facing economic hardship and are still employed can be susceptible to PRR parties’ messages because of the uncertainty caused by the developments they see around them. Indeed, there is evidence for a positive relationship between a range of indicators of economic hardship and PRR support (Adler and Ansell Reference Adler and Ansell2019; Arzheimer and Carter Reference Arzheimer and Carter2006; Colantone and Stanig Reference Colantone and Stanig2018; Dancygier Reference Dancygier2010; De Blok and Van der Meer Reference De Blok and van der Meer2018; Golder Reference Golder2003; Hainmueller and Hopkins Reference Hainmueller and Hopkins2014; Lubbers and Scheepers Reference Lubbers and Scheepers2000; Norris and Inglehart Reference Norris and Inglehart2019; Van der Brug and Fennema Reference Van der Brug and Fennema2009).
More recently, scholars have started to point out the role of demographic decline in fostering PRR support (Dijkstra, Poelman, and Rodríguez-Pose Reference Dijkstra, Poelman and Rodríguez-Pose2020; Harteveld et al. Reference Harteveld2022). This refers to the demographic exodus that threatens the livelihood of particular regions. While this can follow in the wake of economic hardship, it also affects moderately prosperous but peripheral ageing regions. A decline in population, or the share of the young, highly educated, and economically active, threatens ‘not only economic prosperity but also potentially the reservoir of social and cultural capital’ (Bock Reference Bock2016, 557). This deteriorates ‘the carrying capacity of current models of business, public and private services’ (Bock Reference Bock2016, 556). The feeling that the community is disappearing fosters political discontent (Cramer Reference Cramer2016; Guilluy Reference Guilluy2019; Woods et al. Reference Woods2012), rendering PRR support more likely. Moreover, the government may respond to the shrinking and ageing populations in affected areas by scaling back the infrastructure, resulting in even more support for the PRR (Cremaschi et al. Reference Cremaschi2022). However, and somewhat paradoxically, Lahdelma (Reference Lahdelma2023) shows that establishing an asylum processing centre in areas that experience demographic decline may help to reduce feelings of local decline because of the positive consequences of bringing in new workers and economic activities.
Expanding the Model by Introducing Mediators and Moderators
All in all, there are good theoretical and empirical reasons to expect context to matter for PRR support by fostering either nativist attitudes (through the presence of immigrants) or populist attitudes (through economic hardship and demographic decline). Still, two important inconsistencies remain. First, we can observe similar outcomes in otherwise different contexts. If nativist attitudes are often prominent in areas with and without immigrants, and populist attitudes are prominent in areas with and without economic hardship or demographic decline, how decisive are these local contexts? Second, looking at any electoral map reveals different outcomes in otherwise similar contexts. For an example of an ethnically diverse, economically struggling, or demographically declining area with strong PRR support, there is another that does not produce such support. The lack of PRR support in many large and diverse cities – such as London or Paris – seems to confirm the words of Alba and Foner (Reference Alba and Foner2017, 239), that ‘the regions and places with the largest immigrant populations are often those where the native majority holds the most positive attitudes toward diversity’. Again, the role of context appears to be complex.
We argue that such complexities can be explained by considering two crossed factors (summarized in Figure 1). First, a crucial mediator – a generalized feeling of local decline – can have diverse origins (immigration, economic hardship, or demographic decline), but they always generate similar political attitudes (nativist and populist attitudes). Second, context may matter in different ways for different people. We develop these two in turn below.
Figure 1. The causal chain between local context and PRR support
Mediation: How Different Local Contexts Yield the Same Outcome
In tracking the contextual origins of PRR views and support, we argue that feelings of local decline play a mediating role. Contextual theories about PRR support imply that citizens take cues from their immediate surroundings. If true, support should be rooted in certain ‘objective’ local conditions (not merely perceptions endogenous to a pre-existing worldview) that citizens perceive and reflect upon. The different contextual conditions discussed above do not automatically translate into PRR support. We argue that a plausible intermediate step is that each factor fosters feelings of local decline, generating negative reactions towards elites and immigrants.
Feelings of local decline imply that citizens perceive things in their immediate area as going in the wrong direction. What this ‘immediate area’ entails will differ between individuals and contexts, and (as discussed below) not everybody interprets or is affected by local conditions to the same extent. Still, a growing body of work shows that localized concerns have substantive political implications (Cramer Reference Cramer2016; Fitzgerald Reference Fitzgerald2018; Hochschild Reference Hochschild2016; Rodríguez-Pose Reference Rodríguez-Pose2018). Ziblatt, Hilbig, and Bischof (Reference Ziblatt, Hilbig and Bischof2020, 5) argue that PRR support can be rooted in ‘a sense that a location and its way of life suffers some form of distributive injustice in terms of power, wealth, and prestige’, and this in turn ‘[shapes] residents’ perception of themselves, elites, and outsiders’. This plausibly relates to all three contextual factors being studied. For those susceptible to ethnic threat mechanisms, the presence of immigrants will be perceived (regardless of accuracy) as a form of local decline in comparison to an earlier state when there were fewer or no immigrants at all. Similarly, economic and demographic decline appear to many as threats to the way of life in their immediate area. These phenomena have one thing in common: they upset existing local social structures and hierarchies.
In contrast to concepts such as ‘place resentment’ (Munis Reference Munis2020), ‘rural resentment’ (Cramer Reference Cramer2016), or ‘regional resentment’ (De Lange, van der Brug, and Harteveld Reference De Lange, van der Brug and Harteveld2023), the perceived local decline does not in itself attribute blame, nor does it point to geographically, ethnically, or socially defined culprits. It merely captures the perception that things are going in the wrong direction locally. In that sense, feelings of local decline are a localized equivalent of pessimism regarding the direction the nation is going, which has been theorized (most famously in Taggart's (Reference Taggart2004) concept of the ‘heartland’) to be a core ‘linking’ feature in PRR ideology, weaving its various ideological components in a compelling picture of the homogeneous and sovereign past slipping away (Steenvoorden and Harteveld Reference Steenvoorden and Harteveld2018).
While perceptions of local decline can originate in distinct local conditions, we expect they will find a common political expression in a package of grievances about elites and immigrants. First, perceptions of a decline are often accompanied by a strong feeling that out-groups are ‘skipping the queue’ (Gest Reference Gest2016; Hochschild Reference Hochschild2016). Second, anti-immigrant and anti-elite rhetoric has consistently been linked together on the supply side, especially by PRR actors themselves (Mudde Reference Mudde2007; Rydgren Reference Rydgren2005). Indeed, Harteveld et al. (Reference Harteveld2022) show that contextual conditions can foster both types of attitudes but do not test any mediating mechanisms. The co-emergence of these two attitudes due to perceived local decline might explain why citizens in different areas can have similar attitudes. We therefore hypothesize that:
H1: perceived local decline is rooted in three different contextual conditions: (a) immigration, (b) economic hardship, and (c) demographic decline
H2: perceived local decline leads to both nativist attitudes and populist attitudes
By positing these hypotheses, our framework assumes a causal cascade from context through perceptions to attitudes and vote choice. We realize, however, that reality is, in some ways, more complex than this. First, the perceived local decline is, to some extent, endogenous to citizens' broader political worldview and their party preferences. Voters pick up cues from their party, including the notion that many areas are in decline and that immigrants and elites are to blame. Second, we cannot rule out that geographical sorting may also play a role. Citizens with nativist attitudes may be most likely to perceive local decline when they live in an area with many immigrants and may be the most likely to move to a neighbourhood with fewer immigrants. However, to the extent that geographical sorting biases our results, we are most likely to err on the conservative side because geographical sorting will depress the observed relationships between local context and perceptions of local decline. So, it is fair to assume that perceptions of local decline are at least partially rooted in actual experiences. These experiences are relevant for understanding contextual effects because they link variations in ‘objective’ context conditions to political outcomes.
While objective decline predicts perceptions and evaluations, subjective perceptions of decline are more relevant to political behaviour (Michener Reference Michener2013). Whether or not the perception of local decline and disorder is (fully) accurate, the perception is expected to lead to attitudinal or behavioural change. To the extent that objective context matters, perceived decline is a likely mediator.
Moderation: How People Respond Differently to the Same Local Context
If an area experiences one or more of the above developments, not all its residents are equally likely to translate these developments into feelings of local decline and, by extension, political grievances. In political geography, more generally, it has long been known that neighbourhood effects are often heterogeneous across residents (Small and Feldman Reference Small, Feldman, van Ham, Manley, Bailey, Simpson and Maclennan2012), but this is rarely acknowledged in the empirical study of context effects on PRR support. Being sensitive to the conditionality of context effects helps transcend the ‘dichotomous perspective’ in which neighbourhood effects exist or do not exist (Sharkey and Faber Reference Sharkey and Faber2014, 560). Therefore, the more pertinent question is this: which voters are most likely to be affected by the conditions in their local area? We focus on two moderators that may help to address this question: resources and embeddedness (for a related argument in the context of political support, see Steenvoorden and Van der Meer Reference Steenvoorden and van der Meer2021).
Our first potential moderator of context effects concerns the resources citizens have to remain unaffected by their local conditions. In particular, we study the role of education, which we consider to be a resource in several respects (for example, transferable skills, human capital, and cognitive capabilities). Although extensive literature documents the strong main effects of education on progressive and cosmopolitan attitudes and vote choices, either as a result of socialization (Surridge Reference Surridge2016) or of selection (Lancee and Sarrasin Reference Lancee and Sarrasin2015; Maxwell Reference Maxwell2019), the role of education as a potential moderator of contextual effects is under-theorized and under-researched. The literature suggests several compatible and mutually reinforcing resource-based mechanisms for moderation through education.
First, voters with higher levels of education are less likely to be affected by their immediate surroundings: they are more mobile, and their outlook is more cosmopolitan (Sobolewska and Ford Reference Sobolewska and Ford2020). They are also more politically sophisticated and more attuned to national discourses (Elevestad Reference Elevestad2009), which reduces the value and importance of local information. Second, and more importantly, voters with higher levels of education respond differently to local conditions. This is most obvious in the case of immigration. Here, education can be both a psychological and economic resource that allows us to better navigate the impact of immigration.Footnote 1 Highly educated voters are less likely to perceive immigrants as threatening (Hainmueller and Hiscox Reference Hainmueller and Hiscox2007). Their cognitive resources and ideological leanings make it easier for them to navigate ethnic and cultural change (Stubager Reference Stubager2008), and they might even value the diversity that immigration brings (Maxwell Reference Maxwell2019). Their education is also an economic resource that makes them less sensitive to competition by immigrants in the labour market. Similarly, local unemployment is less impactful for those with higher levels of education because they possess convertible skills that are less vulnerable to local economic downturns. With demographic decline, the situation is less clear-cut, but because the more educated are also less dependent on public infrastructure and resources, it is at least conceivable that its effect is moderated, too. While our study cannot precisely disentangle these different mechanisms, they all point toward the following expectation.
H3a: contextual factors have a weaker effect on perceived local decline and PRR support among citizens with higher levels of education
The second potential moderator is the extent to which voters identify with, care about, and spend time in their neighbourhood – which we denote here as embeddedness. Embeddedness is comprised of two (reinforcing) components. One is behavioural, which is reflected in the spatial distribution of citizens' social networks and their strong or weak ties to other persons in the local area. The second component our analyses focus on is attitudinal; that is, citizens' affective identification with the area. If citizens have little interaction with others in their local area and do not feel attached to it, they are less likely to notice or care about local conditions enough to let it affect their political views. Conversely, context can be expected to shape feelings of local decline and PRR support more among those strongly attached to their area. Such ‘place-based identities’ are important in shaping citizens' reactions to their environment (Cramer Reference Cramer2016). We therefore hypothesize that:
H3b: contextual factors have a stronger effect on perceived local decline and PRR support among those strongly embedded in their neighbourhood
In short, it is fruitful to think about PRR support as blossoming in one of four types of situations, depending on whether local conditions are favourable and whether individuals lack the resources and wider attachments to navigate this. PRR parties are likely to be most popular in situations where local conditions are not favourable, particularly among those individuals who lack the resources and connections to deal with these adverse local conditions.
Taken together, these factors might explain why areas that are similar in terms of contextual conditions – but not in terms of the levels of education and embeddedness of its residents – still vary widely in their support for PRR parties. For instance, the relatively high levels of education and lower levels of local embeddedness among residents of Central London or Berlin-Mitte might allow them to remain relatively unaffected by local conditions.
Design, Data and Methods
In most research, PRR voting and context effects have been studied at the level of large subnational entities such as provinces or regions (Kestilä and Söderlund Reference Kestilä and Söderlund2007; Lubbers and Scheepers Reference Lubbers and Scheepers2000). A smaller number of studies looked at context effects at the level of municipalities (Berning Reference Berning2016; Bowyer Reference Bowyer2008; Coffé, Heyndels, and Vermeir Reference Coffé, Heyndels and Vermeir2007) or – mostly in single-country studies – electoral wards or neighbourhoods (De Blok and Van der Meer Reference De Blok and van der Meer2018; Evans and Ivaldi Reference Evans and Ivaldi2021; Harteveld et al. Reference Harteveld2022; Janssen et al. Reference Janssen2019; Savelkoul, Laméris, and Tolsma Reference Savelkoul, Laméris and Tolsma2017; Van Wijk, Bolt, and Johnston Reference Van Wijk, Bolt and Johnston2019).
Given the theoretical mechanisms, which imply everyday contact with immigrants, personal experience of decline, and direct competition over resources (Hewstone Reference Hewstone2015; Oliver and Wong Reference Oliver and Wong2003), large geographical units such as countries or even municipalities are not a priori the theoretically relevant scale to study the effects we are interested in. Of course, larger areas shape citizens’ views, but there is evidence they do so on smaller scales (Biggs and Knauss Reference Biggs and Knauss2012; Van Wijk, Bolt, and Johnston Reference Van Wijk, Bolt and Johnston2019). More importantly, the mechanisms we are particularly interested in – those pertaining to perceived local decline and the way citizens relate to their immediate environment – play out locally. Therefore, we employ much finer spatial classifications, as discussed below.
Data
To test our hypotheses, we conducted large-N surveys in France, Germany, Great Britain (specifically England and WalesFootnote 2), and the Netherlands and combined them with contextual data from national statistical offices. Thousands of respondents are required to have enough power to detect the effects of context on attitudes and behaviours. Since we aim to establish the effects of context conditions at the individual level, the surveys must cover a wide range of small-scale geographical areas in the four countries.Footnote 3 The effective sample size (the number of respondents who filled out the survey and could be linked to ecological data) is around 20,000 in France and Great Britain and 7,000–8,000 in Germany and the Netherlands. The online panels through which these respondents were recruited used quota samples crossed by region (as the main aim was to obtain a sample representative of different areas of these countries), as well as age, gender, and education.Footnote 4 Fieldwork for the surveys was carried out close to the first-order elections in the four countries in 2017 (see Table 1).
Table 1. Information about the geo-coded surveys and contextual data
The core questions asked in the surveys were identical across the four countries whenever possible. Exceptions to this rule are discussed below when the operationalizations are presented. The ecological data on the areas where the respondents lived were derived from the main national statistics offices and, where necessary, from other secondary socio-economic and demographic data sources in each country. These ecological data were linked to survey respondents on the basis of the geocodes assigned to them. In each of the countries, the geocodes were developed in line with how contextual data is measured by the National Statistics Office (for details of the geocodes, see Table 1). As the table testifies, the contextual areas range in average population. With the possible exception of the Netherlands, these statistical areas may be somewhat larger than the respondent's image of their neighbourhood. While this is clearly not ideal, it would dilute the correlations between objective conditions and subjective perceptions, resulting in more conservative estimates for these effects.
Operationalization
The biggest challenge in a comparative study of contextual effects is the operationalization of key independent variables, especially at the contextual level. In this study, we endeavoured to use common measures where possible. However, there is inevitably some variation across some measures due to national specificities. Tables T1 and T2 in the Appendix provide information about the exact measurements used in each country. We use static measures in our main models for practical and theoretical reasons. Practically, studying change is hampered by limitations, including shifts in administrative boundaries and privacy regulations. We replicate the regression models using change variables in three of the four countries (omitting Germany), which was to some degree feasibleFootnote 5 and report this in Appendix F2 and throughout the text where relevant. However, prior studies show a positive correlation between levels and change measures of macro indicators (Harteveld et al. Reference Harteveld2022).
We measure the presence of immigrants by the percentage of first-generation immigrants (those foreign-born) in each area. However, in the Netherlands, we rely on the percentage of first- and second-generation immigrants, as this is the figure reported by Statistics Netherlands. While these are two different classifications, we assume they strongly correlate on the contextual level. It should be noted that citizens' perceptions of who constitutes an ‘immigrant’ might differ from official classifications. In particular, it is plausible that citizens distinguish between types of immigrants based on perceived cultural distance. Unfortunately, no finer-grained operationalization was available across all countries. If anything, this too would make any correlation we do find with attitudes a conservative estimate of the true relationship.
We measure economic hardship using the share of the total population that is unemployed (France, Germany), the share of the active population that is unemployed (Great Britain), or the share of the total population receiving unemployment benefits (the Netherlands). Again, while these differences in operationalization produce different absolute levels, they should plausibly correlate to a similar degree with the dependent variables.
We measure demographic decline by the percentage of inhabitants of each area aged between 15 and 25 (‘leaving the area age’) and the percentage of inhabitants of each area aged between 25 and 45 (‘(not) returning to start families age’), as proposed by Harteveld et al. (Reference Harteveld2022), interpreting lower numbers as a sign of demographic decline. The reason is as follows. A low percentage of young people is likely to reflect young citizens or those starting families having left. Thus, fewer young people indicate that a community sees many of its younger residents leave or is not attractive for young people to settle. In addition, their absence will trigger the perceived threats to the demographic, social, and economic sustainability of communities that might predict PRR support.
Table T3 in the Appendix reports the correlations between these context variables, showing they are sufficiently distinct to assess their effects independently of each other. For instance, the correlation between unemployment and immigration in an area ranges between 0.23 and 0.40. Further calculations using quintiles (calculated by country) show that, of the areas that score in the highest quintile of unemployment, 37 per cent are also in the highest quintile for immigration. This ranges from 31 per cent in the Netherlands to 42 per cent in Germany. So, the degree to which the two are independent differs by country, but there is sufficiently independent variation to calculate the effects of the two indicators separately.
Our models control for population density, which we measure as the number of inhabitants per square kilometre of each area.
In our study, we use the following individual-level variables. We measure perceived local decline using a single question that prompts respondents as follows: ‘In the past 5 years, has your neighbourhood …’, with the answer options being ‘Improved’, ‘Declined’, and ‘Stayed the same’. We recoded perceived decline to 1 and all other answers to 0. We acknowledge that communication by local campaigns and media, everyday political talk, and individual factors may shape perceptions of local decline and, therefore, estimate the strength of the links between objective conditions and subjective perceptions.
We measure (attitudinal) embeddedness using a single question that asks respondents, ‘To what extent do you feel attached to your neighbourhood or town?’ with the answer categories forming a 7-point rating scale ranging from ‘Not attached at all’ to ‘Strongly attached’.
We measure education on a 4-point scale ranging from ‘low’ to ‘high’, with Table T2 in the Appendix summarizing the national categorizations of education used.
We measure populist attitudes using the average response to a scale based on the following four statements, which are derived from Akkerman, Mudde, and Zaslove's (Reference Akkerman, Mudde and Zaslove2014) battery of items: (1) ‘The politicians in Parliament need to follow the will of the people’; (2) ‘The people, and not politicians, should make our most important policy decisions’; (3) ‘The political differences between the elite and the people are larger than the differences among the people’; and (4) ‘Elected officials talk too much and take too little action’. The answer options form a 7-point rating scale ranging from ‘strongly disagree’ to ‘strongly agree’.
We measure nativist attitudes through perceived ethnic threat, relying on a scale of four items from the European Social Survey that tap into cultural and economic fears. More specifically, we calculated the average answer on a 7-point rating scale to the following statements (recoded so that a higher value indicates a higher level of perceived ethnic threat): (1) Is it better for a country if almost everyone shares the same customs and traditions? (2) Would you say that [country]'s cultural life is generally undermined or enriched by people coming to live here from other countries? (3) Would you say it is generally bad or good for [country]'s economy that people come to live here from other countries? (4) Now thinking about taxes and welfare, would you say that people who come to live here from other countries generally take out more than they put in, or put in more than they take out?
We measure PRR support with a single item. The formulation of the item slightly varied between countries, taking into consideration differences in election timing and the nature of the elections. In Germany and the Netherlands, respondents were asked: ‘If there were a general election in [country], which party would you vote for?’ In Great Britain and France, a vote recall question was used to refer to the 2015 parliamentary elections (before the collapse of UKIP in 2017) and the first round of the 2017 presidential election. We coded a vote intention or recall for the country's main PRR party (the AfD in Germany, the FN/RN in France, the PVV in the Netherlands, and UKIP in Great Britain) as 1 and vote intention for a different party as 0.
In our regression models, we include several socio-demographic variables that have been shown to predict PRR support at the individual level (Stockemer, Lentz, and Mayer Reference Stockemer, Lentz and Mayer2018): age in years, education measured as discussed above, and gender as a dummy with female as the reference category. In addition, in two of the four countries (the UK and the Netherlands), we could have included controls for economic class (as measured by the NS-SEC five-fold classification). However, for comparability, we do not include these in the main models. But we report them in Appendix F3 with a particular purpose: to assess if the moderating effect of education is predominantly channelled through economic conditions. If so, the moderating effect of education should be reduced after controlling for class. We discuss this issue below when reporting moderation by education.
Methods
To test our expectations, we proceeded in two steps. In the first step, we focus on mediation and employ multilevel Structural Equation Models (SEM). Path analysis through SEM allows us to simultaneously estimate the various hypothesized paths and investigate whether our theoretical model – from context to perceived local decline, through nativist and populist attitudes, to PRR support – presents a plausible structure generating the correlations between variables. We model individuals nested in local areas, separately estimated for each country. This methodology generates more appropriate standard errors accounting for the nested data structure (Hox, Maas, and Brinkhuis Reference Hox, Maas and Brinkhuis2010). All SEM models were estimated in Mplus 8 using weighted least squares estimation with mean- and variance-adjusted chi-square tests. After establishing the relationships between neighbourhood conditions, perception of local decline, and PRR support, we then use multilevel logistic regressions to assess the moderating role of individual levels of education and local embeddedness. These models also allow us to include control variables.
Findings
We start by describing PRR support and the contexts where it can be found. Figure 2 shows that PRR voters generally live in somewhat less densely populated areas, although the differences are not substantial. Figure 2 also shows that PRR voters tend to live in areas with somewhat fewer, rather than more, immigrants. Levels of unemployment seem a little higher in areas inhabited by PRR supporters, while there is no discernible correlation with demographic decline.Footnote 7 All this resonates with the mixed or weak findings in earlier work. However, it is important to stress that we expect the effects of context to become more clear-cut when we take into account the fact that some citizens are more likely to be affected by context than others. We will explore this in the main analysis.
Figure 2. Average scores on context variables among non-PRR and PRR voters
What is the role of the hypothesized mediator (perceived local decline)? Figure 3 shows the average scores on (standardized) attitudinal variables, as well as the fraction that votes for the respective PRR party among those who perceive decline, improvement, or neither. We find that those who feel their area is declining have considerably different views compared to those who think it is improving. These differences exist across all countries and are of similar magnitude. Those who experience local decline score 0.5 to 0.7 standard deviations higher on the immigrant threat perception measure. On the populist attitudes measure, there is more variation between countries, but the differences are sizable everywhere, ranging from almost 0.2 in France to more than 0.4 standard deviations in the Netherlands. The association with vote choice is also sizable. In all countries, support for the PRR party at least doubles among those who perceive local decline. Crucially, these correlations are not confounded by individual characteristics. The size of the effect remains virtually identical when controlling for age, level of education, gender, and left-right position. Figure F1 in the Appendix confirms that perceived local decline is higher in areas with more immigrants and unemployment in all four countries, while there is no robust association with demographic indicators (again, regardless of controls). This makes it plausible that perceived local decline mediates the relation between contextual conditions and PRR support, but establishing this requires a more formal test.
Figure 3. Nativist attitudes, populist attitudes, and PRR support by perceptions about the local area
Mediation (Multilevel Path Model)
We start by mapping the hypothesized causal chains using multilevel path models described in the theory section. The goal is to establish the extent to which feelings of local decline mediate the context effects and whether such feelings result in a negative evaluation of immigrants and politicians. We present the coefficients for all four countries below in Figure 4. For readability purposes, the first and second half of the causal path are visualized separately, even though they are modelled simultaneously. The numbers next to the arrows in the figure are the standardized effect coefficients by country (as indicated by the flag). All models have a good fit (RMSEA between 0.017 and 0.023).
Figure 4. Multilevel path model
Note: Standardized coefficients. There are no sampling weights.
Consistent across all four nations, we find a statistically significant positive relationship between the presence of immigrants and the perception that the neighbourhood is declining, as predicted by H1a. Furthermore, in all countries except the Netherlands, unemployment is associated with increased perceptions of local decline (H1b). Contrary to H1c, evidence for a link between demographic decline and perceived local decline is weaker, with a positive effect in France, a negative in Germany, and no significant relationship in the Netherlands and Great Britain.
Regarding the individual-level paths, we find consistent positive relationships between perceived local decline and nativist attitudes, as well as between perceived local decline and populist attitudes. The fact that both attitudes are affected confirms H2. In all countries, the effect on nativist attitudes is stronger than on populist attitudes, but both are consistently affected. Both nativist and populist attitudes are associated with support for PRR parties. These patterns can potentially account for the fact that anti-immigrant sentiment exists in areas without immigrants but with sizable unemployment, as well as political discontent in areas that experience little economic hardship but feature many immigrants.
It's important that the effects of contextual conditions are not exclusively mediated through perceptions of local decline. Tables T4 to T6 in the Appendix present the results for all remaining paths per country. Besides the mechanisms discussed above, the contextual variables emanate a significant direct effect – that is, without being mediated by any attitude – on PRR support in Germany, Great Britain, and (less robustly) in France, either positively or negatively. We believe that contextual influences should be mediated by some attitudes other than perceptions of decline and that remaining direct effects can mostly be attributed to confounding variables such as socio-demographics. There are also direct influences, of varying degrees, of contextual variables on nativist and populist attitudes. On the one hand, this confirms that contextual factors can affect attitudes even without explicit perceptions of local decline. On the other hand, this also reflects the fact that our measure does not capture such feelings perfectly and that socio-demographics may (again) act as confounders. We aim to isolate the latter in our regression models with the control variables below. However, given that our model includes alternative paths that bypass perceptions of local decline, its roots in objective contextual conditions and its impact on nativism and populism appear remarkably strong.
Moderation (Multilevel Regression)
As discussed in the theory section, we expect context to have a stronger impact on perceived local decline among citizens with a lower level of education (H3a) or stronger attachment to their local area (H3b). Since the perceived decline, in turn, predicts attitudes and PRR support, the same moderators can be expected to determine the correlation between context and PRR support. Therefore, we present the results of models predicting feelings of local decline (directly moderated) and models predicting PRR support (indirectly moderated).
We start with feelings of local decline. How are these affected by contextual conditions, depending on citizens' level of education and neighbourhood embeddedness? To start with the first moderator, Figure 5 below shows the marginal effects of the four context variables on feelings of local decline, split out by citizens' level of education. These average marginal effects are based on models that control for all other context characteristics, as well as age and gender, but not for embeddedness, as this might (partly) mediate the effect of education. Positive values indicate that higher scores on the context variable increase the likelihood of perceiving local decline. Negative values indicate that higher scores decrease this likelihood. We opted for this presentation because we are particularly interested in learning which educational subgroups have the most significant effects.
Figure 5. Marginal effects of context on perceived local decline by country and level of education
Note: Effects of standardized variables (for example, one standard deviation increase). No sampling weights were applied. Includes controls for age, age squared, gender, and population density. With 95% confidence intervals.
The presence of immigrants in a local area increases feelings of local decline in all countries. In line with the hypotheses, this effect is more pronounced among respondents with a lower level of education, except for France, where immigration matters equally for citizens with different levels of education. In the Netherlands, the effect is halved for respondents with the highest level of education compared to those with the lowest. In Great Britain and Germany, the effect is not significantly different from zero for respondents with high or even medium levels of education.
Higher levels of unemployment are also associated with perceptions of local decline in all countries (including, this time, the Netherlands). Again, the effect appears moderated by education. Regarding effect size, unemployment has the largest overall effect on perceived local decline. Conversely, the effects of demographic decline are weak and inconsistent. We would expect the presence of more young people to be associated with less PRR support (even while controlling for the age of the respondent), but we find mostly null results or even positive effects.
Overall, we find evidence for H3a to the extent that context matters; it does so mainly among those with lower levels of education. Our replication analyses using change variables strengthen this conclusion, as reported in Appendix F2. These models suggest that the change in immigration, unemployment, and demographic outlook are less consistently related to perceived local decline than the actual level, except for the case of immigration in Great Britain (in line with previous UK studies such as Kaufman [Reference Kaufmann2017]). Given the weaker main effects, it is to be expected that few interactions are present, but in the case of the one clear main effect (that is, Great Britain and immigration), our central conclusion is confirmed: the lowest educated respondents are affected most. The additional analysis in Appendix F3, which controls for class in the Netherlands and Great Britain, suggests that this is not primarily for purely economic reasons. Controlling for respondents' class position hardly impacts the moderating effect of education, which we take as a sign that education works mainly through the cultural attitudes it tends to instil.
The second mechanism we investigate is moderation by level of neighbourhood embeddedness. Figure 6 shows the marginal effects for different levels of embeddedness. Contrary to our expectations formulated in H3b, it does not suggest that immigration, unemployment, or demographic composition matters most for those with the strongest place-based identities. The effects are either similar across attachment levels or sometimes stronger for weak identifiers. This conclusion is similar to the models using change variables in Appendix F2. We conclude that context effects depend more consistently on people's education (and thus their cognitive, economic, and cultural resources) than their attachment.
Figure 6. Marginal effects of context on perceived local decline by country and embeddedness
Note: Effects of standardized variables (for example, one standard deviation increase). No sampling weights were applied. Includes controls for age, age squared, gender, level of education, and population density. With 95% confidence intervals.
While the core focus of our study is on the mediating role of perceptions of local decline, the ultimate variable in the causal chain presented in Figure 1 is PRR support. Because local decline increases such support (as the SEM model confirmed), it would follow that our context variables matter for this particular outcome, too, especially among the lower educated. At the same time, the patterns should be much weaker because the relationship will weaken with each additional step in the causal chain. This is what we find in our analyses. Figures 7 and 8 show the marginal effects of the context variables on PRR support across the levels of the two moderators. In most cases, the patterns are similar but weaker, so the effects are often less statistically significant.
Figure 7. Marginal effects of context on PRR support by country and level of education
Note: Effects of standardized variables (for example, one standard deviation increase). No sampling weights were applied. Includes controls for age, age squared, gender, and population density. With 95% confidence intervals.
Figure 8. Marginal effects of context on PRR support by country and embeddedness
Note: Effects of standardized variables (for example, one standard deviation increase). No sampling weights were applied. Includes controls for age, age squared, gender, level of education, and population density. With 95% confidence intervals.
In particular, the strongest interactions with education are found in the Netherlands, where immigration and demographic decline exert the strongest effects among the lowest educated citizens. The contextual effects in Germany, France, and Great Britain differ much less between educational groups. On the other hand, local embeddedness is an important moderator in Great Britain and France but not in Germany and the Netherlands.
One of the most striking differences between the patterns observed in Figures 5 and 7 is the negative effect of immigration among the highest-educated citizens in Great Britain. Further analysis shows that this negative relationship is driven by respondents from the London area, which suggests highly educated pro-immigrant (and UKIP-averse) Londoners are also most likely to live among immigrants.Footnote 8 At any rate, the demographic composition is not a consistent predictor. We conclude that immigration, unemployment, and demographic decline shape feelings of local decline, especially among those with few resources and those who feel embedded in their communities. This matters downstream for patterns of PRR support in all four countries. However, the strength of these relationships differs between these countries.
Conclusions
In this article, we investigate the geographical patterns of PRR support using unique geo-referenced data collected in France, Germany, Great Britain, and the Netherlands. We contribute to the literature by studying the role of local conditions – measured at the neighbourhood level – in a comparative way and by theorizing how contextual conditions influence PRR support in four different countries. We propose that local conditions affect PRR support through the perceptions of local decline. Moreover, we test to what extent individual resources or embeddedness moderate this influence.
Our results confirm that feelings of local decline mediate the impact of ‘objective’ contextual features on attitudes and PRR support. We show that immigrant presence positively affects the perception of local decline across all four countries. High levels of unemployment also increase the view that the local area is worse off than before (except for the Netherlands). By contrast, demographic decline, measured as a low percentage of youth in the neighbourhood, is not consistently linked to perceptions of local decline. We note that some recent case studies operationalized the erosion of community life (which speaks to the same mechanisms) more directly (for example, in disappearing services such as pubs; Bolet Reference Bolet2021; Harteveld et al. Reference Harteveld2022) and found a relation with PRR support. We therefore argue that more research is needed to assess the role of such mechanisms and how to best operationalize these. We also find that perceived local decline is associated with increased nativist and populist attitudes, both of which, in turn, consistently predict PRR vote intention. Our findings further show that the contextual effects, with some exceptions, tend to be moderated by education. That is, citizens with higher levels of education do not link immigrant presence and economic hardship to perceived local decline, whereas citizens with lower levels of education make this connection. Indirectly, this translates into PRR support.
With our observational design, we cannot completely rule out the possibility that nativist and populist attitudes make citizens more prone to perceiving local decline, which would lead us to overestimate its importance. Clearly, more research, preferably using experimental designs (for example, priming studies), is required to disentangle the relationship between local perceptions and ideology. However, the links between objective contextual conditions and perceptions of decline, on the one hand, and the moderation of these links by education, consistent with theoretical expectations, on the other, cannot be explained by reverse effects. Therefore, we remain confident that perceptions of decline, grounded in real-world developments, can contribute to a better understanding of the spatial patterns supporting the PRR.
Our findings have important implications for our understanding of the PRR and the geographical distribution of its support but, at the same time, point to some unanswered questions. First, our findings confirm that PRR support is indeed affected by local conditions. While many previous studies have uncovered correlations between context, right-wing attitudes, or PRR vote choice, the mechanisms involved have not been investigated explicitly. Our study finds that these effects are partly due to explicit perceptions of local circumstances. Concerns over local decline, in turn, translate into a package of grievances towards both political elites and immigrants, which fuels PRR support. It seems plausible that this ‘package deal’ emerges because of elite cues, but further research is needed to understand exactly how citizens attribute ‘blame’ in the face of (perceived) local decline. This model can also be employed to understand the roots of PRR success in other types of local conditions, such as crime (Amengay and Stockemer Reference Amengay and Stockemer2019).
Second, our study confirms that context does not shape PRR support to the same extent for everyone. While almost self-evident, this has hardly ever been modelled explicitly in context and political behaviour studies. The higher educated are more likely to remain unaffected by contextual conditions, which we attribute to the various resources that allow them to escape, ignore, or embrace immigration, economic hardship, or demographic decline. Contrary to expectations, we did not find that those most embedded in their local environment were impacted most strongly by it. Perhaps behavioural measures of networks and social capital could provide a more ‘objective’ measure of embeddedness that moderates the effect of context (Fitzgerald Reference Fitzgerald2018; Fitzgerald and Lawrence Reference Fitzgerald and Lawrence2011). At any rate, future studies of context and political behaviour should be sensitive to the possible potential moderation because failing to do so can lead to a substantial underestimation of the role played by context. It also means that a glance at election outcome maps is not enough to understand the role of context because these do not convey which proportion of citizens is ‘at risk’.
Taken together, these implications complicate and enrich our understanding of what PRR ‘strongholds’ are. Rather than inhabiting a single ‘heartland’, PRR support is geographically fragmented, drawing supporters in different types of areas for different reasons. At the same time, it is important not to overstate the role of context. Citizens' views of the world are shaped by a multitude of factors, many of which are not specific to the local context, such as the (national) political debate and media environment. Where one lives does not entirely determine one's political views, the PRR is more prevalent in some areas rather than others, and this cannot be fully understood without being sensitive to the interplay between the characteristics of the individuals and their context. Future work should expand and refine which interactions are important in this respect.
Of course, our study has some limitations. While the country-comparative element and the finer spatial scale provide unique insights not available to previous studies, it also limits the availability of context indicators to common denominators. For instance, our operationalization of immigrant presence might not correspond fully to categories employed by majority-group citizens. Using a broader set of ecological data, country case studies could pick up the thread and study micro-macro interactions and the role of perceived local decline on a small scale. Furthermore, our case selection was restricted to four relatively urbanized Western European countries with a long history of immigration. It is likely that other factors play a role in countries with, for instance, lower population density and less immigration.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007123424000085.
Data availability statement
Replication data for this article can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/YETU0O.
Acknowledgements
The authors want to thank the editors and reviewers of the British Journal of Political Science for their helpful and constructive comments. We also thank everyone who gave us feedback at various stages of this research project.
Financial support
This research was made possible by the Agence Nationale de la Recherche (ANR-15-ORAR-0006), the Economic and Social Research Council (ES/N018745/1), the Deutsche Forschungsgemeinschaft (AR369/4-1) and the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (464-15-225).
Competing interests
None.
Open access
