The Regional Cleavage in Western Europe: Can Social Composition, Value Orientations and Territorial Identities Explain the Impact of Region on Party Choice?

Abstract

In this article regional differences in party support are analysed by means of comparative survey data from 15 West European countries. The main research question is: How can we explain that people in different regions vote for different political parties? On the basis of the literature on new regionalism and old and new politics, this article formulates two sets of hypotheses which are considered related to new regionalism and new politics, and old regionalism and old politics, respectively. A causal model is formulated which includes three groups of explanatory variables, namely, (other) socio-structural variables, value orientations and territorial identities. The main findings support the old regionalism and old politics notions of how we can explain the regional cleavage. Social structure is the most important explanatory variable, and, among the various value orientations, old politics values are most important. There is, however, some evidence that values and territorial identities are important explanatory variables in three of the countries where the regional cleavage has increased since the 1970s: Belgium, Italy and Spain.

Notes

1. This was also very evident in Rokkan's work on the cleavages in Norway. In some of the most important works he analysed the degree to which religious, linguistic and class differences could explain the regional differences in party choice. Rokkan's main finding was that these differences played a major role in this respect (Rokkan 1967; Rokkan and Valen 1964; Valen and Rokkan 1974).

2. The authors did not find it fruitful to group the regions along a centre–periphery dimension because countries could have many centres and also many peripheries (Rose and Urwin 1975: 7–9). They therefore used region as a nominal-level variable and developed measures for tapping regional differences in party choice which reflected this assumption. This approach is found in many other studies of the regional cleavage: The regional variable is simply treated as a nominal-level variable and the ranking of regions, or an effort to find criteria for ranking the regions along a centre–periphery axis, is absent and not considered important for the empirical analysis.

3. Territorial identities might be correlated with new politics values, but according to the new regionalism perspective which is indicated in the third hypothesis below, the essence in the new regionalism hypothesis regarding territorial identities is that territorial identities have a causal impact on party choice even when (New Politics) value orientations are controlled for.

4. Territorial identities might have been important at the time of the formation of the regional centre–periphery cleavage in European countries, as the discussion of Lipset and Rokkan's analysis of the cleavage above indicated, but such identities became weaker in typical industrial society. Modernisation has been associated with nationalisation and the weakening of regional cultures and identities. This is why, for example, Keating uses the formulation that historical territorial identities might have been revived as a basis for political mobilisation (Keating 2006: 144).

5. In addition, in Iceland, the postal codes were coded for region, which comprised about 100 categories. The region variable in the Norwegian data set was not available in the WVS cumulative file, only in the national file. For Ireland there was no regional variable in EVS 3, and Ireland was not included in this study.

6. NUTS is an acronym for Nomenclature for Territorial Units for Statistics, which has been established by Eurostat in order to provide a single uniform breakdown of territorial units for the production of regional statistics for the EU (NUTS 2003). Iceland, Norway and Switzerland are also classified according to NUTS, although they are not members of the EU. It has been used for collection, development and harmonisation of the EU's regional statistics since 1988. NUTS is a three-level hierarchical classification which classifies regions in three levels. Level 3 (NUTS 3) is the lowest level (with the largest number of regions) and NUTS 1 the highest (with fewest regions). The criteria for belonging to the various levels are that the average population sizes of the various units are within certain limits. This implies that larger countries tend to be classified into all levels while some of the smaller countries are classified into only one or two levels (level 3 or levels 2 and 3). Norway, Sweden and Switzerland are classified at only the two lowest levels, while Denmark is classified only at the lowest level. Iceland is not classified at any of the levels due to its small population.

7. This applied to Denmark and Iceland.

8. The question is formulated as follows: ‘If there was a general election tomorrow, which party would you vote for?’

9. Cramer's V is the most popular of the chi-square-based measures of nominal association because it gives good norming from 0 to 1 regardless of table size, when row marginal equals column marginal (Liebetrau 1983: 14–16).

10. Multinomial logistic regression (MNLR) can be used to perform multi-variate analyses with nominal-level dependent variables like party choice. There is no standardised measure for each independent variable in MNLR, but there are several analogues to the linear regression R². These pseudo R² measures can be used to decompose the ‘pseudo variance’ in multi-variate analyses when the independent variables are entered stepwise. Pseudo R² measures are less easily interpretable than linear R², but they are very useful for this purpose. I perform such stepwise analyses in this article, and I rely on Nagelkerke's pseudo R², which is one of the most frequently used measures (Menard 2002: 24–7). I also use this measure for tapping bi-variate relationships between party choice and various social cleavages. Although Nagelkerke's pseudo R² is not formally a correlation coefficient, I use the notion correlation also for Nagelkerke's pseudo R² for the sake of simplicity.

11. CDU and CSU are kept as separate parties. In many analyses of German data, these parties are treated as one category, but given the research problem here, collapsing these parties would have considerably undervalued the regional dimension of German politics. The existence of CSU in Bavaria is a central component of the regional cleavage in Germany (Hepburn 2008).

12. The two alternative regional divisions in Switzerland show somewhat inconsistent strength compared with the seven-category classification based on NUTS. The correlation coefficients based on the language division are 0.280 (Cramer's V) and 0.158 (Nagelkerke's R²), while the coefficients based on political system are 0.356 (Cramer's V) and 0.210 (Nagelkerke's R²). The Cramer's V coefficients are then larger than for the seven-category division, while the Nagelkerke's R² are smaller. In Table 1 Switzerland is ranked as number 5 according to both measurements. The same applies to three of the coefficients based on the alternative regional divisions. The Cramer's V for the political system would have placed Switzerland at the top, but this division is somewhat problematic since it is constructed on the basis of criteria that are derived from the other variable in the correlation – the party system.

13. The correlations between the two measures from the survey data and two measures used in Caramani's aggregate analyses based on the electoral results from the 1990s are 0.70 and 0.72 for Nagelkerke's R² and 0.65 and 0.62 for Cramer's V.

14. The correlations between the average standard scores based on the three measures from Rose and Urwin's data reported in Table 1C and the two measures in the survey data are 0.48 for Nagelkerke's R² and 0.50 for Cramer's V.

15. Religious denomination is the structural aspect of the religious cleavage; frequency of church attendance is not. Research carried out by Jagodzinski and Dobbelaere (1995: 87–90) has shown very strong correlations between church attendance and more direct measures of religiosity. On the basis of European Value Surveys (I and II), they find that the correlations vary between 0.41 and 0.73 in different countries. This is a magnitude very rare in survey research, and because of these strong correlations with direct measures of religiosity, frequency of church attendance can be considered an indirect measure of religious versus secular values. In addition, it can be argued theoretically that attending church is not a structural characteristic.

16. This category (versus all other classes) can be considered an indicator of the commodity market cleavage, but I include this category nevertheless (as a matter of no choice) in my social class variable as an indicator of the labour market cleavage.

17. Values below 12 were recoded as 12, and values above 26 were recoded as 26.

18. The response alternatives were: 1) under 2,000 inhabitants, 2) 2,000–5,000, 3) 5,000–10,000, 4) 10,000–20,000, 5) 20,000–50,000, 6) 50,000–100,000, 7) 100,000–500,000, 8) 500,000 or more.

19. Given the level of measurement ascribed to the various conflict variables, education, income and urban–rural residence are treated at so-called co-variates, while social class, religious denomination and region are treated as ‘factors’ in the multinomial logistic regressions.

20. The following example which is based on the Austrian data reported in Table 1A, illustrates the decomposition: The explanatory power of region on party choice according to Nagelkerke's R² is 0.113, and the explanatory power of the (other) socio-structural variables is 0.346. The total explanatory power of region and the other socio-structural variables is 0.406. Region then adds 0.060 (0.406–0.346) to the explanatory power of the other socio-structural variables, which then is the unique component of the impact of region. The compounded component is then 0.053 (0.113–0.060) which is 46.9 per cent of the total explanatory power of region. 0.053 and 46.9 per cent are then the absolute and relative compounded components, respectively.

21. The pseudo-variance explained by social structure is very similar for the two other regional divisions in the Swiss case and exactly the same for the division based on party system (53 per cent) and 61 per cent for the language division.

22. Value orientations are not structural cleavages, but may be considered as prescriptive beliefs that may be closely related to a structural cleavage. For example, religious values may be associated with whether individuals belong to different religious denomination or whether they do not belong to any denomination.

23. I use the two sets of orientations instead of Inglehart's (1977a, 1990) materialist/post-materialist value orientations. One reason for this is that the most elaborate measure of materialist/post-materialist values, the 12-item battery, was not available in EVS 3, which included only the 4-item value battery. The two indices tapping the libertarian/authoritarian and Green value orientations are both based on several indicators.

24. In Norway and Switzerland, religious values and libertarian/authoritarian values are tapped by the same indicators, while economic left–right and environmental values are tapped by somewhat different indicators, as can be seen from online Appendix B (see ‘Knutsen Appendices' at http://structureofompetition.pbworks.com and at http://sites.google.com/site/structureofcompetition).

25. The percentage for Switzerland is 27.5 per cent and somewhat less for the two alternative regional divisions, both 21 per cent.

26. For different frameworks for conceptualising and empirical analyses of territorial identities, see Bollen and Medrano (1998), Carey (2002), Inglehart (1971, 1977b), van Kersbergen (2000), Marks (1999), Medrano and Gutiérrez (2001) and Rokkan and Urwin (1983: chapter 3).

27. The percentage transmitted via territorial identities is even somewhat larger for the alternative regional divisions: 25 per cent (language) and 27 per cent (party system).

28. The basis for the figures in Table 6B-D is a series of multinomial logistic regressions. In the first analyses, the four types of variables are included in a hierarchical order: first region, then the other socio-structural variables, then value orientations and finally the territorial identities. In the second analyses, region is dropped, other socio-structural variables are entered first, and then value orientations and finally the territorial identities.

29. Based on the average for the 14 countries, the percentage of the total compounded component for all explanatory variables, social structure accounts for 73 per cent (41.0 of 55.8).

30. The percentages that are transmitted via social structure, value orientations and territorial identities, respectively, based on the seven-category regional division in Switzerland are 53 per cent, 10 per cent and 7 per cent in Table 6C. On the basis of the alternative regional classifications, we find nearly identical percentages, 61 per cent, 4 per cent and 6 per cent, respectively, for the language division, and 53 per cent, 6 per cent and 8 per cent, respectively, for the party system division.