A societal perspective on cleavage theory

Departing from a societal version of cleavage theory (Rokkan, Reference Rokkan, Flora, Kuhnle and Urwin1999, also Bartolini, Reference Bartolini2000; Bornschier, Reference Bornschier2010; Caramani, Reference Caramani2004; Kriesi et al., Rogowski, Reference Rogowski1990), we are interested in the structural conditions of political mobilization. Consequently, according to this perspective, the degree to which political elites are able to politically mobilize different constituencies depends on the salience of conflicts that are structurally determined and have been the result of the most important “social revolutions.” More specifically, we aim to explain how deindustrialization conditions the success of nationalist elites in the peripheries to mobilize their ethnonational constituency in the context of democratic elections.

While the societal version of the cleavage theory is structural, it follows historical institutionalism to some extent by taking path dependency into account. It does so by acknowledging the long-lasting impact of the institutionalization of party systems in the critical juncture of mass enfranchisement. According to the “freezing hypothesis,” the cleavage constellation after World War I has stabilized Western European party systems through positive feedback effects until at least the 1960s (Lipset and Rokkan, Reference Lipset, Rokkan, Rokkan and Lipset1967). However, in contrast to what would be typical for a historical institutionalist approach (see Hall and Taylor, Reference Hall and Taylor1996), the emphasis in cleavage theory is not on the contingency of paths taken in critical junctures through decisions of political elites but on the structural conditions under which these elites take their decisions.

From the perspective of cleavage theory, in order to explain party system change, we need to identify the most important societal forces that change conflict constellations and the degree to which these conflicts are cross-cutting. As Rokkan (Reference Rokkan, Flora, Kuhnle and Urwin1999) has famously shown, the configuration of party systems in Western Europe can to an important extent be explained with the degree to which conflict structures are reinforcing or cross-cutting. Furthermore, in the meantime it has been demonstrated that the concept of cross-cutting cleavages can even be successfully applied to ethnic voting in contexts of weakly institutionalized party systems beyond Western Europe (e.g., Dunning and Harrison, Reference Dunning and Harrison2010). Hence, it is the more surprising that this perspective has not been systematically applied to explain variance in ethnonational mobilization in Western Europe over time.

A two-dimensional model of electoral mobilization

In our application of the cleavage theory, we maintain that for the context of Western Europe in the 20^th century the electoral strength of the ethnonational parties – parties that primarily promote the interests of “their” ethnonational groups – is to an important extent a residual to the class cleavage. Accordingly, the degree to which ethnonational parties can mobilize their ethnonational constituency depends on the extent to which the class conflict dominates (electoral) politics. At the source of the strength of the class cleavage is industrialization while deindustrialization can, to an important extent, explain its decline (Bartolini, Reference Bartolini2000).

Deindustrialization has been preceded by a decline in the agrarian sector, which has led to the decline or reconfiguration of the urban–rural cleavage (for a review see Knutsen, Reference Knutsen2013, 189–192). The decline of the manufacturing industry, more importantly, has decreased the vote share of the Social Democrats (Best, Reference Best2011). Hence, together with secularization, deindustrialization has allowed for the emergence of often-discussed new cleavages such as the post-materialism (Inglehart, Reference Inglehart1990) or globalization cleavage (Kriesi et al., Reference Kriesi, Grande, Lachat, Dolezal, Bornschier and Frey2008).

With regard to ethnic and national identities, there has been no process similar to deindustrialization that would have undermined the center–periphery cleavage. Consequently, there has been no universal decline of the center–periphery cleavage. Rather, as this article argues, deindustrialization and its weakening of the class cleavage have strengthened the center–periphery cleavage.

While the cross-cutting cleavage hypothesis has intuitive appeal, its underlying assumptions often remain implicit. And although our historical macro-sociological approach does not allow to test the micro-level assumptions, making them explicit is important to gauge the plausibility of the hypotheses and its empirical testing in subsequent analyses. We make five main conceptual decisions to arrive at a parsimonious and testable theory that is – although simplifying – rooted in the empirical literature.

First, in line with previous work (Alonso, Reference Alonso2012; Elias et al., Reference Elias, Szöcsik and Zuber2015; Massetti and Schakel, Reference Massetti and Schakel2015) we conceptualize the class cleavage as the most important cleavage that cross-cuts the center–periphery cleavage. We focus on the class cleavage rather than the state–church or urban–rural cleavage because the class cleavage has clearly been the most important such division for party politics in Western Europe (Bartolini, Reference Bartolini2000; Caramani, Reference Caramani2015). This is not to say that in all contexts, the class cleavage has been dominant. Arguably, in some states the state–church cleavage has been more consequential (Manow, Reference Manow2015). However, while the state–church and the urban–rural cleavage have only been decisive to the configuration of party systems in some European countries, the class cleavage has left its mark on virtually all party systems in Western Europe. Also, while we focus on the class–cleavage our argument can be generalized to the state–church and the urban–rural cleavage when they are cross-cutting to the center–periphery cleavage.

Second, we conceptualize state-wide parties as parties that, on average, have a stronger class character than the ethnonational parties (and hence are cross-cutting). More specifically, with regard to the state-wide parties we assume the party system to be divided into a Socialist party representing the workers and a Conservative and/or Liberal party representing the bourgeoisie. With varying degrees, all of these parties share a preference for the consolidation of the nation-state with the majority culture at the core of the imagination of this nation. On the side of the ethnonational minorities, in contrast, we assume the existence of a single interclass ethnonationalist party or several ethnonationalist parties, which are less segmented by class alliances than the state-wide parties. To be clear, we are not claiming that state-wide parties are not nationalistic, but merely that the linkages with their electorate are more based on class than on ethnonational identity relative to ethnonational minority parties where the linkages are more based on ethnonational identity than on class. Also, while our argument that state-wide parties have a stronger class character than the ethnonational parties might not be equally correct in all contexts (see below), there is ample evidence that class is a much weaker predictor for the choice of ethnonational parties than variables related to ethnonational identity such as ancestry and language and, hence, that the class cleavage is cross-cutting the ethnonational cleavage.Footnote ¹

Third, we want to argue that members of a core group of the antagonistic classes (e.g., industrial working class vs. owners and managers) tend more to vote for a state-wide Socialist or Conservative/Liberal party than other voters. The reason is that it is those groups that have been mobilized most strongly by these parties.Footnote ² The decisive process for the size of the antagonistic groups and hence the strength of the class cleavage is industrialization (Bartolini, Reference Bartolini2000). The change in the size of the antagonistic class groups can be due to generational change or to occupational mobility within the voters’ lifecycles. The stronger the degree of industrialization, the larger the core groups of the worker and the bourgeois political movements and the more the different classes deviate in their political preferences and political organization. Hence, when the degree of industrialization is particularly high and the class division consequently strong, the class conflict might be so salient and the class movements organizationally strong that they tend to co-opt the workers and the bourgeois of the ethnonational minority. This explains why Caramani (Reference Caramani2004) finds that the super-imposition of the economic left–right cleavage indicated a strong universal trend towards territorially homogenous electoral behavior in Western Europe since 1830.

Fourth, while the process of deindustrialization leads to a dealignment of the electorate from the class-based parties, we do not assume such a process at the level of the ethnonational groups. The reason is that those variables that are linked to ethnonational identity change more slowly than occupation (which is at the core of the class cleavage).Footnote ³ This is particularly true for (perception of) ancestry and for language, which are very strongly socialized and hence transmitted over generations. The dealignment of the electorate from the class-based party can again be due to generational change or due to changes in identity during the voters’ lifecycles. From the historical perspective taken here, generational change is probably more relevant. But also identity change during the voters’ lifecycles is important to explain changes in voting patterns. According to a growing literature on Right Radical Parties (e.g., Engler and Weisstanner, Reference Engler, Weisstanner, Careja, Emmenegger and Giger2020; Gidron and Hall, Reference Gidron and Hall2017), deindustrialization leads large segments to embrace culturally rather than economically dominated identities due to status threat. This argument might also apply to voters that identify with an ethnonational minority and hence switch their party preference from a state-wide party to an ethnonational party.

To sum up, for the core segments of the class antagonism, the class imposes itself over ethnonational identity. Consequently, the core segments of the class conflict vote for state-wide class parties independent of their ethnonational identities. Because industrialism is the main determinant of the sizes of the core groups of the class conflict, levels of industrialism impact on the degree to which the class cleavage imposes itself through its composition of the electorate.

We have argued that state-wide parties have a stronger class character than the ethnonational parties. The argument is that ethnonational parties have to be of an interclass character in order to maximize their vote share (Elias et al., Reference Elias, Szöcsik and Zuber2015, 843–846). This is especially true for small groups in national elections where ethnonational parties might not gain representation if they are fragmented. Consequently, on the economic left–right dimension ethnonational parties might favor centrist positions over extreme positions. More precisely, they tend to position close to the preferences of the median voter in the region of their targeted electorate, which depending on the region’s economic structure is moderate to the left or the right of the center (Massetti, Reference Massetti2009; Massetti and Schakel, Reference Massetti and Schakel2015). However, if institutional incentives are not strong enough to hinder the fragmentation of ethnonational parties, the two-dimensional model developed above applies less. Especially since different ethnonational parties from the same ethnonational party bloc tend to have different positions on the left–right dimension (Massetti and Schakel, Reference Massetti and Schakel2015), it cannot be excluded that they compete on the secondary economic dimension with state-wide parties. In this case, state-wide parties were probably never fully able to co-opt the working class and bourgeois voters and are hence less negatively affected by the deindustrialization process.

High levels of industrialization should have a particular strong impact on the degree to which the party system is dominated by the class antagonism. Hence, in these highly industrialized cases, there is a higher potential that the theoretical model applies than in initially more agrarian societies. Assuming, furthermore, that there is some degree of path dependency (see above), we might also think that what matters most is the level of industrialization at the time point when the mobilization of the masses took off.

Case selection

While studying ethnonational mobilization, we focus on parties that target what is referred to as “ethnonational group,” “ethnic group,” “national minority,” or “nation without state,” among others. Since these concepts are “essentially contested” (Gallie, Reference Gallie1955), albeit there is little disagreement on substance, there is no consensus on its use, which also varies strongly across scientific fields (e.g., conflict scholars use the concept “ethnic group” in a broader sense than nationalism scholars). The common denominator to these concepts when used in a similar way is that they refer to a category of individuals with a common subnational identity that claims some degree of self-government. The group “is the entity for which ethnonationalism claims autonomy or independence” (Lecours, Reference Lecours2000). Hence, it typically includes ethnic groups or such that have a self-perception of a nation because of former territorial independence (also see Massetti, Reference Massetti2009, 508–509).

For this study, a group is selected as a case if it fulfills four criteria. First, it must be a rather prototypical case of an ethnonational minority. In conflict studies, the fuzziness of ethnic groups has led to problems of selection bias (Fearon, Reference Fearon2002; Hug, Reference Hug2003). This selection bias can be minimized if one studies the most prototypical cases, that is, by studying those cases which clearly belong to the category of interest (Strijbis, Reference Strijbis2013). In order to include only rather prototypical cases, we consulted three datasets on ethnic and national minorities covering Western Europe and selected only those cases that appeared in at least two instances. The three data sources are the Encyclopedia of Stateless Nations: Ethnic and National Groups Around the World (Minahan, Reference Minahan2002), Pan and Pfeil’s Volksgruppen in Europa (Reference Pan and Pfeil2000), and the Geo-referencing of Ethnic Groups (GREG) data (Weidmann et al., Reference Weidmann, Rød and Cederman2010). These datasets are well suited because they apply a low numerical threshold for the inclusion of a group in the dataset and share an understanding of ethnonational minorities as self-identifying groups consistent with the conceptualization above.

The Encyclopedia of Stateless Nations covers all groups that self-identify as a distinctive group, display “outward trappings” of national consciousness, and have formed a specifically nationalist organization or political grouping that reflects a claim to self-determination (Minahan, Reference Minahan2002, xii). In fact, also very minor organizations are considered to fulfill the last criterion, putting more weight on the first two criteria.

Volksgruppen in Europa defines a “Volksgruppe” as a community that is numerically minor to the rest of a country, whose members are citizens of this country, who are distinct from the majority due to ethnic, linguistic, or cultural markers, and who are willing to maintain this idiosyncrasy (Pan and Pfeil, Reference Pan and Pfeil2000, xiii).

The Geo-referencing of Ethnonational Groups (GREG) data (Weidmann et al., Reference Weidmann, Rød and Cederman2010) contain the data of the Atlas narodov mira (Miklukho-Maklai Ethnological Institute 1964). The Atlas narodov mira has no explicit definition of “ethnonational group,” but relies heavily on linguistic differentiation for its categorization. Also, it has been shown that the categorization is related to an understanding of ethnonational groups as groups whose members tend to marry within the group (endogamy) (Bridgmann, Reference Bridgmann2008, 1). It can therefore be assumed that this data set, too, puts common identity at the center of its (implicit) definition. In order to assure that divergence in the coding of ethnonational minorities is not due to the fact that the Atlas was compiled in the 1960s, we have taken a more recently compiled version of it from Weidman et al.’s Geo-referencing of Ethnonational Groups (GREG) data (Weidmann et al., Reference Weidmann, Rød and Cederman2010).

Second, since the article tries to understand levels of ethnonational mobilization, we only include groups for which the ethnonational identity is politically salient. Here we follow the distinction by Chandra and Wilkinson (Reference Chandra and Wilkinson2008, 524) between ethnicity activated in politics and its mobilization in institutionalized politics. Hence, the question is not when ethnonational categories actually become salient, but when politically salient ethnonational identity is mobilized. In order to make sure that only ethnonational minorities with salient ethnonational identities are included, we restrict the sample to minorities that have been rated as politically salient in the Ethnic Power Relations Dataset (EPR) (Cederman et al., Reference Cederman, Wimmer and Min2010). To correct for the fact that the EPR data do not include the period from 1918 to 1944, we also selected all those with electoral support by an ethnonational party that cannot be found in the EPR dataset.

Third, we restrict the sample to cases of ethnonational minorities that count 50,000 group members or make for a minimum of 0.1% of the total population. This selection criterion is necessary because the strength of ethnonational minority mobilization will be measured in a way, which, with regard to the sources of the data, makes the indicators somewhat sensitive to ethnonational group size (see below).

Finally, since we are interested in levels of mass mobilization and not the question of whether elites from an ethnonational minority movement found parties which contest (national) elections, we restrict our sample to movements which achieved minimal levels of mobilization at least once since 1918. The electoral data allow to identify all ethnonational parties receiving at least 5% of the votes in one electoral constituency in at least one state-wide parliamentary election (Caramani, Reference Caramani2000). Applying this criterion leaves us with 15 ethnonational minority groups for which we analyze electoral mobilization in national parliamentary elections.

As argued in the article, we restrict our sample to movements which achieved minimal levels of mobilization at least once since 1918. This leads to the exclusion of some ethnonational groups like the Basques in France, Frisians in the Netherlands, the Rhaetians in Switzerland, and the Slovenians in Austria, because they never successfully mobilized with an ethnonational party in national elections. Two cases where ethnonational parties were formed and competed in elections concern the Basques in France and the Frisians in the Netherlands. In the case of the former, the attempt to enter national electoral competition was made but was unsuccessful. The Frisian Nationalist Party (Fryske Nasjonale Partij) only contested regional elections where it gained a maximum of 13.2% of the votes (Hemminga, Reference Hemminga, De Winter and Gomez-Reino Cachafeiro2006). The cases of the Faroe Islanders and Greenlanders were excluded because of lack of data. Online Appendix A gives a detailed account on how the different steps led to the sample of this study.

Ethnonational party blocs

This article provides, to my knowledge, the first attempt to describe and explain the performance of ethnonational parties since World War I. Studying the political mobilization already at this early stage is important because this has been the first period of mass mobilization with a strong path-dependent effect (Lipset and Rokkan, Reference Lipset, Rokkan, Rokkan and Lipset1967). In line with previous literature, we define ethnonational parties as those that primarily promote the interests of “their” ethnonational groups and/or the exclusion of others (see Chandra, Reference Chandra2011 for a review). As Strijbis and Kotnarowski (Reference Strijbis and Kotnarowski2015) demonstrate, in line with their political ideology, ethnonationals overwhelmingly attract the vote of their targeted ethnic group members. While strongly overlapping, this separates them from regionalist parties, which promote the interests of the entire population residing within a certain delimited territory and implies parties that represent ethnonational groups that are not regionally concentrated.Footnote ⁴

As our interest is in the political mobilization of ethnonational minorities and not in the electoral success of specific parties, we confine the study to the performance of ethnonational party blocs. As most difficulties in deciding whether parties are ethnonational or not boil down to the question of whether or not a group should be considered an ethnonational group, only a few problems concerning the classification of parties remain. Often, a party’s name provides a clue as to whether it should be considered ethnonational or not. Among the ethnonational parties considered here, however, a few cannot be identified as ethnonational by their name alone. One such case is the Social Democratic and Labour Party (SDLP) in Northern Ireland. In the case of the SDLP, there have indeed been attempts to amend its identity to that of an interethnic socialist party. As the SDLP, however, clearly moved towards becoming a primarily ethnically defined party (Jenkins, Reference Jenkins1997, 152–153; Webb, Reference Webb2000, 20–25), other socialist parties based on interethnic alliances have been founded.

Most difficult is the classification of Catalan parties before the Spanish civil war. Although the ethnonational character at that time of the Regionalist League (Lliga Regionalista) and the Republican Left (Esquerra Republicana) is clear, we have decided with Medrano (Reference Medrano1995) to exclude the Catalan Union (Uniò Català) and the Radical Party (Partit Radical). Also, in the post-Franco era some communist parties, such as the Basque wing of the United Left (Izquierda Unida), had no clear program relating to ethnonational minorities; we have excluded them from the ethnonational party family. We also exclude the Valencians from the dataset because this group showed no strong signs of ethnonational identity in the Second Republic, which is also demonstrated by the fact that group identity was still very weak in the early days after Franco (Coller and Castelló, Reference Coller and Castelló1999).

Table 1 lists all ethnonational groups and ethnonational parties that are included in the analysis. Most ethnonational minorities witnessed splits, dissolutions, and/or (re-)foundations of ethnonational parties during their democratic periods, presenting them with a more or less fragmented ethnonational party bloc. This resulted in 53 ethnonational parties representing 15 ethnonational minorities in seven West European countries in the period since WWI.

Table 1. Ethnonational party blocs in Western Europe, 1918–2018

Notes: ¹participation with ≥ 5% in a constituency; ²without the voter alliance with the Flemish Christian Democrats in 2007; ³includes voter alliances with Francophone Democratic Front of Brussels; ⁴Corsican nationalist candidates officially run as regionalists; ⁵coalition of Tyrolean People’s Party und Free Liberals; ⁶data for 1931 only available for Galicia, votes for nationalist parties could not be calculated for Catalonia and Galicia in 1936; ⁷single nationalistic candidates; ⁸Three candidates, not to confound with the Scottish section of the British Labour Party.

Operationalizations

Since the electoral mobilization of ethnonational parties should be measured as the mobilization of the ethnonational parties of its targeted ethnonational group, and since ethnonational group sizes are not always stable over time, valid comparisons of ethnonational mobilization are enhanced by knowing the size of each group in question. Estimating the group size of the 15 native ethnonational minorities in this sample is, however, far from simple. Until now, studies that have used longitudinal data have made fairly imprecise estimates, assume very constant group sizes, do not contain data on all 15 groups analyzed here, and contain no information before World War II.

In estimating the size of ethnonational groups, we have tried to approach self-classification as narrowly as possible. This has two implications. First, we have tried to rely on information on ethnonational categorization whenever possible. Second, we have tried to rely on census or register data whenever possible (see part A.2 in the SM).

In order to measure the electoral mobilization of the ethnonational movements, we focus on electoral mobilization of ethnonational parties. Although it has been demonstrated that regional elections are more than “second order” (Schakel and Dandoy, Reference Schakel, Dandoy, Dandoy and Schakel2013), for reasons of data availability, we restrict our analysis to vote choice in national elections for the lower parliamentary chamber. Since ethnonational parties primarily want to mobilize their co-ethnics (Chandra, Reference Chandra2004, 5–6; De Winter, Reference De Winter, De Winter and Türsan1998, 210; Diamond and Gunther, Reference Diamond and Gunther2001, 198), their levels of electoral mobilization should be measured as the share of ethnonational group members at voting age, voting for “their” ethnonational party or ethnonational party bloc if several ethnonational parties mobilize the same ethnonational group. Hence, we divide the share of the ethnonational minority parties’ vote by the state-wide share of ethnonational group members (Strijbis and Kotnarowski, Reference Strijbis and Kotnarowski2015, 5–8). The values for this Relative Index of Ethnic Party Mobilization (RIEPM) typically vary between 0 and 1. They can be higher than 1, however, in cases where members of an ethnonational minority vote homogenously for “their” ethnonational parties and have a higher turnout than the members of other ethnonational groups.

The Relative Index of Ethnic Party Mobilization (RIEPM) is based on the assumption that the size of the ethnonational groups is exogenous to the parties’ electoral success. One might, however, argue that ethnonational parties can impact the ethnonational group size. Hence, in this case, the electoral success of parties that are able to increase the size of their targeted ethnonational group would be underestimated with the previously developed measures. In order to test for such a bias in this measure, we will replicate the analyses by taking the mean ethnonational group size as the denominator and include the group size at each election as an independent variable.

For the degree of industrialization, we use as an indicator, the share of industry workers in the region in which the ethnonational minority is territorially concentrated. The regional level is used because this is the territorial level at which ethnonational parties can gain representation. Through interpolation, the share of industry for most regions at most points in time could be estimated. In some cases, however, there were no data for the interwar period. Instead of extrapolating the data for these cases – continuous growth/decrease could not be assumed – the cases were categorized according to the description in historical sources, and the variable was expressed as estimates of the share of industry workers in 10% intervals. In order to test the robustness of this indicator we also report the analyses calculated without our estimations, relying on primary sources alone.

We have argued that the effect of deindustrialization on ethnonational party mobilization should be stronger among those groups that are concentrated in regions that have historically been highly industrialized. In order to test this implication of the theory, we will split our sample between those cases that had at least 30% of workers in the industry in the interwar period and those who were dominated by the agrarian sector.

Finally, we include several variables that have been put forward in the literature: We use a lagged value of the Polity5 index (Marshall and Gurr, Reference Marshall and Gurr2020) in order to control for ethnonational mobilization as a reaction to state repression, the electoral system at the level of the territorial concentration of the ethnonational group, the degree of territorial autonomy making use of the Regional Autonomy Index (RAI) (Hooghe et al., Reference Hooghe, Marks and Schakel2010), level of European integration (Brancati, Reference Brancati2014), and national GDP per capita. The descriptive statistics for all variables are given in Table 2 (for operationalizations and sources see the SM part A).

Table 2. Descriptive statistics

Sources: See Supplemental material

Method

The data are collapsed into five-year periods, producing a largely balanced panel. In those few cases when two elections were held within the same period, the mean was taken. Since the data are non-stationary and autoregressive, we make use of time series cross-sectional regression analysis. More specifically, we use fixed effects for spatial units and panel correct standard errors (PCSE). The fixed effects model reads as (Beck, Reference Beck2001, 283):

$${y_{i,t}} = {x_{i,t}}\beta + {f_i} + {\varepsilon _{i,t}},$$

where f _i is a dummy variable marking region i. Since average levels of the dependent variable over spatial units (regions) are fixed, the model predicts variance over time and is consequently suited to analyze the effect of change (to which the term deindustrialization points). This specification is preferable over a difference model where we would lose degrees of freedom. Also, Table B.2 in the supplemental material shows that the results are robust to inclusion of time fixed effects in a two-way fixed effects model. Finally, we find very similar effects if we lag the independent variable by up to about 20 years, which is approximately one generation (Table B.3 in the SM).

Please note that changes in the boundaries of the electoral constituencies or regions do not bias our analysis. This is because we measure the ethnonational party vote share not at the constituency or regional, but at the national level. Insofar as changes in variables that are related to boundary shifts have a causal effect on electoral mobilization, this should be accounted for by the control variables. However, the measurement of the ethnonational group sizes might be prone to measurement error, which could bias the estimates. As a robustness test, we will replicate the analysis while calculating the Relative Index of Ethnic Party Mobilization (RIEPM) with each ethnonational group’s average relative population size. If the estimates of the group size would have strong impact on its value of ethnonational mobilization, the results would differ heavily from other analyses.