Religious Voting and Class Voting in. 24 European Countries. A Comparative Study

Transcription

1 0 Religious Voting and Class Voting in 24 European Countries A Comparative Study Oddbjørn Knutsen Department of Political Science, University of Oslo Paper prepared for presentation at the XVII International Sociological Association (ISA) World Congress of Sociology in Gothenburg, July, RC18 Political Sociology, Panel RC Comparative Class and Religious Voting

2 1 Introduction The study of the relationship between social structure and party choice is a classic topic within political science and political sociology. The relationship between social structure and party choice is part of most reports from election studies in many countries, but there are not many works that make systematic comparative studies of the relationship. Most of the books that are supposed to be comparative are organised with country-specific chapters and present some comparative patterns in the introductory and/or concluding chapters (Lipset & Rokkan 1967a; Rose 1974a; Dalton, Flanagan & Beck 1984; Franklin et al. 1992). However, several more recent books or book chapters have tried to analyse the relationship in a more systematic comparative setting (Nieuwbeerta 1995; Knutsen 2004; 2006, Oskarson 2005). In their seminal essay on the development of the conflict structure in Western democracies, Lipset and Rokkan (1967b: 15-23) focused on the historical origins and the major conflicts between the political parties. They identified four central cleavages which had their anchorage in the social structure: 1) The centre periphery cleavage was anchored in geographical regions and related to different ethnic and linguistic groups as well as religious minorities. 2) The conflict between the Church and the State pitted the secular state against the historical privileges of the churches and over control of the important educational institutions. This cleavage has more specifically polarised the religious section against the secular section of the population. 3) The conflict in the labour market involved owners and employers versus tenants, labourers and workers. 4) Finally, the conflict in the commodity market was between buyers and sellers of agricultural products, or more generally, between the urban and the rural population. Previous research has shown that religion and social class have been the most important structural variables for explaining party choice. Rose and Urwin (1969) conducted one of the first comparative analyses of the topic, examining the social basis of party support in 16 western democracies. They found that religion and social class were more important than the other cleavages in the Lipset-Rokkan model, and that, contrary to conventional wisdom, religious divisions, not class, are the main social basis of parties in the Western world today (Rose and Urwin 1969: 12). In a comparative study that included most West European countries (Rose 1974b: 16-18) compared the impact of religion, social class and region on left-right voting on the basis of data from mainly the 1960s, and found that religion and social class was much more important than region, but that religion was much more important than social class in all the Catholic and religiously mixed countries. Only in Britain and the Scandinavian countries was social class the most important predictor for left-right party choice. These findings are confirmed in numerous single-country studies. My own comparative longitudinal study of eight West European countries from the early 1970s to the late 1990s showed that the strength of the religious and class cleavages confirmed previous analyses in the sense that the religious cleavage was more significant in the Catholic countries and the religiously-mixed countries while the class cleavage was of more significant than religion in Britain and the Nordic countries. There were few changes in the relative importance of these main cleavages over time although there was a decline in many countries. The impact of the religious and class cleavages were, however, approaching in most countries due to the fact that the cleavages that traditionally had the largest impact showed the clearest sign of decline. Religious variables were, nevertheless most important for explaining party choice in most countries also in the late 1990s (Knutsen 2004: chap. 7). In a comparative longitudinal study of six Western democracies Brooks, Nieuwbeerta and Manza (2006) found, however, that class voting was larger than religious voting in most

3 2 countries. This study included, however, only one of the two aspects of religious voting (see below), namely religious denominations. In this chapter I examine the impact of the two most important social cleavages according to many researchers, namely religion and social class. It is a comparative study of 24 countries based on a common comparative data set, namely the European Social Survey. Three variables tap aspects of the religious cleavage, namely religious denomination, church attendance and a direct measure of religiosity. I consider these variables to tap the two faces of the religious cleavage which will be explained below. The conflict in the labour market is traditionally associated with class cleavage, and in a broad sense, social class, education and income can be considered as class variables or variables that tap the class cleavage. However, it is important to emphasis that in advanced industrial societies such variables can also tap other aspects of societal conflict. This is explained in details in the section on the class variables below. Research problems and data This paper is a comparative study of the impact of religious and class variables on party choice in 24 countries in the 2000s. The overall research problems are how the religious and class variables vary in their capacity to explain voting behaviour in 24 European countries. The more concrete research problems are as follows: 1. What is the comparative strength of religious and class voting in European countries in the first part of the 2000s ( )? 2. Which party families contribute to religious and class voting (polarisation)? Are there significant cross-country differences in this respect? 3. Do the religious and class variables polarise the parties along the traditional left-right division or do we see patterns of cross-cutting in the sense that the religious and class variables group the parties in ways that do not follow the traditional left-right division? The data source for this paper is a cumulative file based on the three first rounds of the European Social Survey carried out in 2002, 2004 and All countries that have participated in at least two waves are included apart from Ukraine. The data for the various waves are weighted for the following reason. The sample size for the various waves varies, for some countries considerably. The intention with the weighting is to balance the sample size between the various ESS rounds. The principle for the weighting is to take the total N in each country and divide it by two or three depending on the number of rounds the countries have participated in. The number of cases in the various waves in then the same in each country, and the total N for each country is the same as in the data set before the weighting. This study does not examine trends in the impact of the religious cleavage over time. Table 1 shows countries that are included in the study and the number of respondents in the cumulative file. The table also shows that a few of the countries only have participated in two of the three relevant waves. Since the former East and West Germany have fairly different religious structures and different recent historical experiences which are relevant for the social cleavages, I have decided to keep them as different units/countries in this study. 1 1 This decision is influenced by cooperation with German scholars on a book project on Society and Democracy in Europe where different researchers should analyse the relationship between social structure and different political dependent variables. Due to the fact that social structures still are different in the two parts of Germany

4 3 < Table 1 about here > The countries are grouped into five regions which deserves a few comments. The category - the Islands comprise two countries which are located is the same region, but have very different party systems and religious structures, although they sometimes are grouped together under the umbrella Anglo-Saxon. I have kept Britain and Ireland in a separate category partly due to their geographical location and partly because this category is a residual category which cannot reasonably to be grouped among the other categories. France is grouped among the Central European countries although it is a borderline case to Southern Europe. I partly used empirical criteria to decide between the Central and Southern European category. On the variables tapping religious identification and religiosity, France has for example a very different location than the pure Southern European countries. Finally I have grouped East Germany among the Eastern European countries 2 due to the historical experiences with the Communist legacy. The notion Western Europe will be used below to describe the four regions that do not comprise the Eastern Europe category. I will use the region of countries below to sum up some main trends for the patterns for the 24 countries by presenting average figures for each of the regions and then compare the regions. Research strategy and methodology For studying the relationship between the religious and class variables and party choice in a comparative setting the political parties are grouped into 10 party families. These party families are the Conservative, Christian, Liberal, Agrarian, Ethnic-regional, Radical Rightist, Green, Social Democrats, Left Socialist and Communist party families. The distribution of the party choice variable across the 24 countries is shown in Appendix Table 1. 3 The largest party family according to the data is the Social Democrats followed by the Conservative, Christian and Liberal party families. The other party families are much smaller. The average size has partly to do with the fact that several party families are present in only some of the countries. Only the Social Democrats are present in all countries. The Liberal and Christian parties are present in 20 and 18 countries, respectively, while the Conservative, Radical Right and Green parties are present in 14 to 16 countries. The other party families are present in 6 to 10 countries. For details in this respect, see the Appendix table. Due to the large number of countries and the complexity of the research topic, I start many of the empirical analyses by presenting some general findings which are based on the main patterns in the data. The analyses which represent these main patterns are based on a pooled data analysis. The data from the 24 countries are pooled into one analysis where the party choice variable is party families. In this pooled analysis each country is assigned the same weight since the focus is on the average pattern across countries. Each country represents one unit and all countries are weighted equally. 4 and influence political phenomena in different ways, it was decided that the two Germanys should be considered as separate units. 2 I use the notion Eastern Europe instead of Central and Eastern Europe because the former is shorter and since another region is labelled Central Europe. 3 The classification of the parties into party families can be obtained from the author. 4 The number of cases with a party choice for all 24 countries were after the weight which set the number of cases equal for the (two or) three surveys. The number of cases in each country was then set to 71538/24 which, after rounding, is 2980.

5 4 The party choice variable is based on the respondents reported party choice during the last national election in their country. The division between parties to the left and those to the right is central in much literature on predictors of party choice. This study examines both the impact of religious variables and class variables on all parties in the party systems, and the impact on the left right division of parties. It aims to compare the impact on these two ways of treating the party choice variable. The traditional way of determining which parties to include in the socialist or leftist group is that they should belong to the Communist, Left Socialist and Social Democrat party families. (Bartolini and Mair 1990: 42 43; Franklin et al. 1992; Nieuwbeerta 1995: 36). 5 The Green parties were not part of the industrial conflict structure, and therefore did not belong to the established left. The Green parties have been characterised as the post-industrial left. Many of the issues and values that the Greens inject into the political arena are new, challenging the established conflict structure and the established left. Therefore, I present and compare the results of two analyses in which the parties in the Green party family are placed within the non-socialist and the socialist party group respectively. 6 The first grouping of the leftist parties is referred to as the traditional left right division below. In the surveys the Green party family is present in 14 of the 24 countries. The portion of the respondents reporting that they voted for the Greens varies from % in Austria and West Germany to 2-4% in Britain, the Czech Republic, Italy and Ireland. The countries where the Greens are not present in the data set, are five East European countries (all apart from the Czech Republic and East Germany), Denmark, Norway, Greece and Portugal. The division between leftist and rightist parties is often used in analyses of the relationship between social structure and party choice, but it is seldom asked whether this division really taps all the impact of social structure on the entire set of parties. Indeed, researchers seldom develop measures that can be used for this purpose. A major research problem in the present work is to examine the extent to which the impact of the religious variables tends to overlap or cut across the traditional left right division of parties. In Table 2, I have illustrated my approach for analysing the impact of the various sociostructural variables on party choice. The table shows an imaginary cross-tabulation of the relationship between religious denomination and party choice in a party system with five parties where parties A and B are the leftist parties, and C, D and E are the non-leftist parties. The illustration is based on a situation with only one dominant denomination where the values on the denomination variables are no denomination and Roman Catholic denomination. < Table 2 about here > In this imaginary example 75% of the sample indicates that they belong to the Roman Catholic denomination. This can be seen from the marginal of the table. The first part of the table shows the support for each of the five parties, percentage difference (hereafter PDI) between those who belong to the Roman Catholic denomination and those who do not belong to any denomination who vote for each party, the log-odds ratios (hereafter lor) for the differences between no denomination and Roman Catholic denomination and the absolute magnitude of the percentage differences PDI. 5 It is the socialist (or leftist) parties that are defined positively; the non-socialist parties are those not defined as socialist, according to the conventional rules for the socialist/non-socialist division. 6 It should be emphasised that this applies only to the Green parties that are grouped into the Green party family, not those that are grouped among the Left Socialist parties. Both of these party families could be characterised as New Left parties, but they have different history and partly also political profile regarding location on the economic left-right dimension.

6 5 The lor-scores are not affected by the size of the parties in contrast to percentage difference measures and can be used for comparing the impact of religious denomination on the various parties in the party system. In the table we see that the absolute magnitudes of the lor-scores for party C and D are larger than for party B although the latter has a larger PDIscore (in absolute magnitude) than the two other parties. Thus, religious denomination has a larger impact on parties C and D than on party B. While lor is relevant for comparing individual parties, PDI is highly relevant for analysing the impact of religious denomination on the whole party system; that is on party choice. For the impact on party choice larger parties normally contribute most. The larger percentage differences are the most important for the overall impact of the structural variable on the party system, and a larger percentage difference is more easily obtained for a large party. A relevant PDI measure for the impact of religious denomination on party choice is obtained by summing the absolute values of the various PDIs and dividing by two. This is done in the column to the right in the table. In the last part of the Table 1 I have summed up the support from for the leftist and rightist parties and calculated similar measures for left right party choice. In this case the impact of religious denomination on left right party choice is considerably smaller than the total impact from the first part of the table according to the PDI. The reason for this is easy to see from the first part of the table. The impact of religious denomination seems to cut across the left right division to a certain degree. The leftist party A gets stronger support from the Roman Catholics, while the opposite occurs for party B. We find a similar division among the rightist parties. The left right division taps only 11 of the total 26 percentage points (hereafter pp.) which is the total impact of religious denomination on party choice. The portion 11/26 = 0.42 is the degree of overlap, that is the degree to which the impact of religious denomination on left right party choice taps the whole impact of religious denomination s impact on party choice. The inverse concept in this respect is cross-cut, which in the example is 15/26 = If the impact of religious denomination on left right party choice is the same as the total impact (26 pp.) in this case, the overlap is 1.00 and if religious denomination has no impact on left right party choice the overlap is The focus on overlap and crosscut is also the main reason why I use PDI as a statistical measure in the bivariate analyses of the relationship between the socio-structural variables and party choice. As shown above, there is a close relationship between this measure and the degree of overlap and crosscut since this measure is the basis for these calculations. In the example above, the religious denomination variable has only two values and it is fairly easy to calculate the degree of overlap. If there is more than one significant denomination, the degree of overlap can be calculated by estimating the degree of overlap between no denomination and each of the denominations and then by weighting the figures on the basis of the sizes of the denominations in the sample (see under religious denominations and explanations under Table 4 below). When the religious or class variable is a variable with several values and has a higher level of measurement, there might be a problem where to set the cutting points for calculating the overlap. This might be complicated because the distributions of the religious and class variables can be very different cross-nationally (see the discussion under the analysis of Religiosity and church attendance below). I use the eta-coefficient and Cramer s V (for religious denomination and social class) as my main statistical measures for tapping the relationship between the structural variables and party choice. It should be underscored that for these measures all values for the religious 7 15 are calculated by subtracting the sum of the PDI from the second table (11 pp.) from the sum from the first table (26 pp.). 8 For details and additional illustrative examples, see Knutsen (2004: 34-41).

7 6 and class variables are treated as separate values. They are not collapsed as for the PDI measure. For measuring class voting several other measures are used. The two ways of treating the party choice variable are sometimes referred to as total party choice (based on all parties as separate values) and left-right party choice (based on a dichotomous variable). The multivariate analyses are based on logistic regression for left-right party choice and multinomial logistic regressions for total party choice. Logistic regression are preferred to be used for analyses of dichotomized dependent variables, and multinomial logistic regression (MNLR) can be used to perform multivariate analyses with nominal-level dependent variables like total party choice. I use the pseudo R squared measures of model fit from these analyses to compare the impact of religious variables and class variables, respectively, in different countries and to compare the relative impact of religion and social class. I use one of the measures, namely Nagelkerke s R 2 which has the property that it varies between 0 and 1. 9 It should first be emphasized that even thought these measures are considered as equivalent to explained variance, they are not tapping explained variance. For multinomial logistic regression with a nominal level dependent variable, explained variance is not meaningful. Instead, these measures tap the improvement in the log likelihood model from a baseline model with parameters for the independent variables equalling zero (Pampel 2000: 48-52). There is no standardised measure for each independent variable in MNLR or in logistic regression. I therefore use the pseudo R squared measure also for comparing bivariate correlations between total and left-right party choice, and different religious and class variables. An additional reason for using pseudo R squared as a measure of bivariate relationships is that these statistical methods can treat both nominal and interval variables as independent variables. It is then possible to compare the impact of variables with different levels of measurement in a simple way. Pseudo R squared measures are then less easily interpretable than linear R squared, and the various measures show different figures. These measures are then not the same as explained variance, but they are useful for examining the relative importance of various groups of variables and to compare the impact of the same variables across countries as in the present case. Instead of the notion explained variance I use the notion explanatory power frequently below. It is unproblematic to compare the fit of different models with the same dependent variable. It is somewhat controversial whether one exactly can compare the explanatory power for the same set of variables from different data sets which in this paper means different countries. In this paper I assume that the problems with comparing such models are small. I also try to compare the results based on pseudo R squared with other statistical correlation measures in order to examine if they show the same relative strength of different variables. Organisation of the paper I have organised the article in the following way: I first examine religious voting in Europe, then class voting and finally I compare the impact of religious variables and class variables on party choice. This part also contains multivariate analyses of comparative variations in the impact of both religious and class variables. 9 For an overview of different measures tapping the goodness of fit in logistic regression analyses, see Menard (2002: 17-27)

8 7 For each section I first examine the relationship between the structural variables and party families based on the pooled data. The analysis based on the party families for all the 24 countries combined will be called the pooled analysis below. I then focus on comparative patterns by examining cross-national variations in the strength of the relationship between the religious variable and party choice, and the class variables and party choice, where party choice is treated as a nominal level variable. Due to lack of space I do not report in detail where the parties within the various countries are located on the religious and class variables, but when the pattern deviates considerably from the pooled analysis of party families, this is reported. Then I examine the comparative patterns for the strength of the correlations for left right party choice based on the PDI measure, and the degree of overlap. Finally, I examine the total explanatory power of the three religious variables on total party choice and left-right party choice and compare the impact of these variables. The same is done in the section on the class variables.

9 8 Religious voting 10 Introduction In their seminal article on the development of the party cleavages in western democracies, Lipset and Rokkan (1967b) were impressively detailed about the development of the religious cleavage. The religious cleavage was first shaped by the Protestant Reformation, which created divisions between Catholics and Protestants. These divisions had political consequences because the control of the nation-building process often became intermixed with the religious cleavage. Protestants frequently found themselves allied with nationalist forces in the struggle for national autonomy. In Anglican England and the Calvinist Netherlands, the Protestant church supported national independence and became a central element of the emerging national political identity. In other nations, religious conflicts also ran deep, but these differences side-tracked the nation-building process (Dalton 1990: 66; Martin 1993: ). Gradually the political systems of Europe accommodated themselves to the changes wrought by the Reformation. The French Revolution renewed religious conflicts in the nineteenth century. Religious forces both Catholic and Protestant mobilised to defend church interests against the Liberal, secular movement spawned by the events in France. Conflicts over church/state control, the legislation of mandatory state education and disestablishment of state religion occurred across the face of Europe. These conflicts often were intense, as in the Kulturkämpfe in Germany and Switzerland. In reaction to these liberal attacks, new religious political parties formed in Germany, the Netherlands, Switzerland, Austria, Italy and Belgium. These parties ranged from the Calvinist Anti-Revolutionary Party in the Netherlands (named in reaction to the French Revolution) to the Vatican-allied Catholic Partito Populare in Italy (Dalton 1990: 66-67). The party alignments developed at the start of the twentieth century institutionalised the religious cleavage in politics, and many basic features of these party systems have endured to the present. The religious cleavage has two aspects: the various religious communities of which people are members, including a category for those who are not a member of any religious community (religious denomination); and how religious they are independent of the religious community they belong to (Bean 1999: 552; Dalton 2008: ). This latter aspect is normally measured by frequency of church attendance. Many researchers have noted that there is a somewhat paradoxical situation related to the importance of the religious cleavage. Only a small number of political issues clearly follow the religious/secular conflict line. By the same token, there are very few issues that are completely divorced from them. Despite the paucity of explicitly religious issues and the lack of religious themes in most campaigns, religious beliefs have proven to be a strong predictor of party choice in many West European democracies. Smith (1989: 20) has therefore characterised the religious cleavage as a passive rather than an active force in shaping political behaviour. Perhaps the most important reason why religion continues to play an influential role for voter choice is that religious conflicts helped determine the structure of the modern party system and therefore still affect the electoral choices open to the voter. The religious cleavage is also important because it reflects deeply held human values, which have a great potential for influencing behaviour. Although religious issues are not very prominent on the political agenda, religious values are related to a wide range of social and political beliefs: work ethics, achievement aspirations, life-style norms, parent-child relations, morality, social relations, attitudes toward authority and acceptance of the state. Religion signifies a 10 This section is based on Knutsen (2010).

10 9 Weltanshauung that extends into the political area (Dalton 1990: 86). Religious faith is strongly connected not only to party choice. The connection encompasses political ideology, issue outlook, and attitudes towards a wide range of political objects (Wald 1987: chap. 3). Several studies have examined the impact of the religious cleavage (the two faces of it or only one) over time and in a comparative setting (Dalton 1990: 82-88; Dalton 2008: , Elff 2007: ; Inglehart 1977: , ), and numerous studies have focused on trends within a single country. The main findings from these studies are that although there has been a considerable change in the distribution on the religious cleavage variables in the direction of a more secular mass public, the correlation with party choice has shown a surprising persistence at a high level. For example, Dalton (2008: 159) compares the impact of religion on voting with the impact of social class in a comparative longitudinal study and concludes that the trends for religious voting do not show the sharp drop-off found for class voting. My own longitudinal study of eight West European countries from the early 1970s to the late 1990s based on Eurobarometer data showed, however, considerable decline in the impact of religion on party choice in the countries where the religious cleavage has been most pronounced in the 1970s, Belgium, France, Italy and the Netherlands. Due to these declines there was a trend towards convergence in the impact of the religious variables on party choice at a somewhat lower average level than in the 1970s. There were, however, also signs of a considerable persistence in the impact of religion in the other countries (Knutsen 2004: chap. 2, 3, ). Similar findings are reported in van der Brug, Hobolt and de Vreese (2009: ) on the basis of the European Election Studies. They found that there was a significant decline in the impact of religious variables on party choice from 1989 to 1999, but then a small increase from 1999 to 2004 based on data from the countries that were EU-members for the whole period they examined. Operationalisation of the religious variables. Comparative patterns regarding religious structures and beliefs. I start the empirical analysis with presenting the three variables that tap the religious cleavage and a comparative analysis of the distributions of these variables. The three variables tapping religious affiliation and religiosity in the European Social Survey are based on the following questions: Religious denomination Do you consider yourself as belonging to any particular religion or denomination? Those who answer yes are then asked Which one and the interviewer should code the reply into one of several response alternatives. It is evident from the annotations provided to aid translation that the first of these questions is supposed to tap identification with a religious community, not official membership. 11 To the extent that religious denomination can be considered as a structural variable (see the discussion below), the question is then not ideal for tapping religious structure since the notion consider yourself is used. Self-declared religiosity 11 In the source questionnaire the annotation in a footnote is as follows: Identification is meant, not official membership.

11 10 The second question tapping self-declared religiosity is asked in the following way: Regardless of whether you belong to a particular religion, how religious would you say that you are? The respondents are then shown a card with a scale from 0 (Not at all religious) to 10 (Very religious). This question is strictly self-declared religiosity since the respondents are asked to place themselves on a scale. It is not a question about concrete religious beliefs. Church attendance The question on church attendance was asked to all respondents in the following way: Apart from special occasions such as weddings and funerals, about how often do you attend religious services nowadays? The respondents were then shows a card with the following alternatives: Every day (10.00) More than once a week (8.33) Once a week (6.77) At least once a month (5.00) Only on special holy days (3.33) Less often (1.67) Never (0.00) The variable is in this paper transformed into a ratio-scale with values from 0 to 10 in accordance with the figures in parentheses to the right-hand side. These figures were not on the card which the respondents got. Table 3 shows the percentage of the samples in the various countries which replied that they considered themselves as belonging to any religious denomination and the means on the two other variables, namely the self-declared religious and the church attendance variable. A high degree of religiosity and religious identification have the highest values on all these three variables. < Table 3 about here > There are large comparative differences for all the religious variables and the ranking of the countries are very similar. This is also indicated when the scores on the three variables for the 24 countries are correlated. They are all strongly correlated ( ) indicating that all three variables tap a religious/secular cross-national dimension. The highest level of religiosity and religious identification are found in the Southern European countries and in Ireland, Poland and Slovakia, while the lowest level is found in The Nordic countries and in several East European countries. France also has a low level of religiosity and religious identification. These differences are also reflected in the average scores in the latter party of the table (see Table 3D). On the religious denomination variable we can examine two different aspects, namely the portion who indicates that they do (not) consider themselves to belong to any religious denomination, and the dominant denomination among those who declare that they do adhere to a denomination. Since the portion who considers they not to belong to any religious denomination varies so largely between countries, I delimit the analysis of the comparative variations in religious denominations to group the countries according to the dominant denominations in the various countries. This corresponds to the historical differences between the countries in this respect.

12 11 Protestant The Nordic countries, (Britain (even when Northern Ireland is excluded) is a borderline case to the religiouslymixed countries and can also be grouped there). Roman Catholic Austria, Belgium, Ireland, Luxembourg, France, Italy, Spain, Portugal Poland, Slovenia, Slovakia. Orthodox Greece Religiously mixed The Netherlands, Switzerland, West Germany, East Germany, Hungary (Roman Catholics and Protestants) Britain (Established Protestant churches, non-conformists and Roman Catholics) Estonia (Protestant and Orthodox) Several other Christian denominations and also non-christian religions were also coded in the surveys, but they comprised each only a few percentages of the samples. For example, the sum of the various non-christian religions (including Jewish, Islam, Eastern religions and Other non-christian religions ) comprise less than 4% in all samples in all countries. Below I only examine the voting pattern or these categories in the pooled data.

13 12 The impact of the religious variables on party choice Religious denominations The religious denominations that people belong to can be considered as the structural aspect of the religious cleavage (Knutsen and Scarbrough 1995: ). The denomination that families belong to, is frequently transferred from generation to generation. The religious structures are very different cross-nationally as we have seen above. Based on the pooled data I will mainly comment on the voting behaviour of the Roman Catholics and the Protestants in relation to those who do not consider themselves to belong to any denomination. The first comment, however, is related to the voting behaviour of those who belong to the Islamic and other non-christian religions. They are inclined to support the leftist parties, in particular the Social Democrats and the Greens. The comparison of those who belong to the Roman Catholic denomination and those who do not belong to any denomination, shows that in addition to the Christian parties, the Conservative and Ethnic-regional parties have a larger portion of the vote from the Roman Catholics while all other party families (apart from the Radical Right where there is no difference) get stronger support from those who do not belong to any denomination. Among the parties that get strongest support among the non-affiliated the differences are largest for the Greens and the Left Socialists according to the lor scores and among the Social Democrats, Left-Socialists and Liberals according to PDI. Some of the same patterns are found based on a comparison between the non-affiliated and the Protestants. The Radical Right, however, is inclined to get stronger support from those who belong to a Protestant denomination, and there is no difference for the Conservative parties. For the comparison of both the Catholics and the Protestants with the unaffiliated we find that all leftist party families get stronger support from those who are not affiliated with any religious denomination, while in particular the Liberal parties among the non-leftist parties do get stronger support from those who are not affiliated. Since the Green parties get stronger support from the unaffiliated, the overlap is largest when these parties are grouped among the leftist parties. The degree of overlap is somewhat smaller for the Protestants (0.49 and 0.60 when the Greens are placed in the leftist group) than for the Roman Catholics (0.57 and 0.72, respectively) based on the pooled data. Tables 4 A and B shows that the strength of the denominational cleavage in a comparative perspective by means of the Cramer s V coefficient and the PDI measure. < Table 4 about here > The impact of religious denomination on party choice is largest in the Netherlands according to both Table A and B, but there are several differences regarding the ranking of countries. One particularly large deviation is Poland which has a comparatively low correlation according to Cramer s V, but is ranked second according to PDI in Table B. This difference is probably caused by the skewed distribution on the religious denomination variable (see Table 3A). Such a distribution does not course a large correlation for correlation coefficients like CV, but the PDI measure indicate that the Catholics and the affiliated nevertheless vote very differently. The impact of religions denomination on party choice is - according to the average etacorrelations in Table E - largest in Central Europe while there are only small differences between the other regions (Cramer s V ) apart from the Island category where the impact is small. The ranking of the regions is somewhat different based on the PDImeasure.

14 13 Table C shows the PDI measure which is based on the left right division of parties, and where those who do not belong to any religious denomination are compared with those who belong to the major denomination(s). 12 In all countries those who do not belong to any religious denomination are more likely to support the leftist parties than those who belong to one of the major religious denominations although the differences are small in Estonia and Slovakia. We note that the correlation with left right party choice is not considerably larger in the Netherlands than in the many of the other countries with a strong denominational voting differences according to Table A and B. The impact of religious denomination on left-right party choice is according to the average scores in Table E largest in the Central European countries, followed by Southern, the Islands, Northern Europe and finally Eastern Europe. Table D shows that the degree of overlap is very high in most countries. Only in five countries is the degree of overlap less than In average level of overlap in the various regions is fairly similar for all regions apart from Eastern Europe. While the average varies between 0.70 and 0.83 for the former regions, the corresponding figure for the Eastern European countries is We note in particular the low degree of overlap in Estonia, Slovakia, Slovenia and the Czech Republic. The low degree of overlap in several East European countries is mainly caused by Liberal parties who do get stronger support from those who do not belong to any religious denomination. The same applies to the Netherlands where the degree of overlap also is smaller than in most West European countries. The Netherlands has two liberal parties (VVD and D66) which both get strongest support from the non-affiliated. The lower degree of overlap in the Netherlands explains why the left right correlation is comparatively smaller than the correlation with party choice in Tables A and B. In Denmark and Norway where the overlap is fairly low, the most important component explaining this is that the radical rightist parties get stronger support from the non-affiliated. The same applies to Belgium. In Greece the voting pattern of those who belong to the Orthodox Church and those who do not consider themselves to belong to any denomination is small. The degree of overlap is also relatively small. The major reason for the fact that religious voting in Greece cut across the left-right division of parties is that the social democratic PASOK get strongest support from those who belong to the Orthodox Church. Religiosity and church attendance One may argue that frequency of church attendance and the eleven-point scale tapping (self-declared) religiosity, taps much of the same aspect of the religious cleavage. Research carried out by Jagodzinski and Dobbelaere (1995: 87 90) has shown very strong correlation between church attendance and more direct measures of religiosity at the individual level. On the basis of European Value Surveys (I and II), they found that the correlations vary between 0.41 and 0.73 in different countries. Jagodzinski and Dobbelaere concluded that these two variables tap the same aspect of the religious cleavage. Church attendance can then be considered as an indirect measure of religiosity, and is not a structural aspect of the religious cleavage in the same way as religious denomination. 12 For the religiously mixed countries the PDI is calculated by weighting the figures for the various denominations, see note to Table 4.

15 14 This is confirmed in analyses based on the cumulative ESS file. The correlation between church attendance and religiosity is on average 0.58 for the 24 countries. The correlations vary between 0.51 and 0.68 apart from Greece where it is lower (0.45). 13 Based on the pooled data for the 24 countries the correlation with party choice is somewhat larger for church attendance (0.273) than for (self-declared) religiosity (0.218). The average location of the party families on the two scales shows that they are ranked completely identically. In addition to the Christian party family, religious voters support Conservative, Agrarian and Ethnic-regional parties, while secular voters support all the leftist party families (including the Greens) and also the Liberal parties. The average for the Radical Rightist parties is close to the average for all party voters for both variables. The degree of overlap is very high, 0.83 when the Greens are grouped among the leftist parties for church attendance, and 0.82 for religiosity based on the pooled data. 14 For the comparative analysis the PDI measure is calculated by collapsing the three categories for those who report that they attend church 1) every day, 2) more than once a week and 3) once a week. This collapsed category is then compared with those who never attend church. The distribution on the church attendance variable varies considerably between countries as we have seen above and in most countries a small percentage report that they attend church more than once a week (categories 1 and 2 above). However, for a few countries the portion is (much) larger than in the other countries. This applies to Poland, Ireland, Slovakia, Italy, Portugal and Greece (see Table 3B). I have as an alternative collapsed only two categories (1 and 2 above) for these countries. This reduces the religious portion of the samples considerably and increases the PDI measure somewhat. I have nevertheless chosen to rely on a consistent coding for all countries for calculating the PDI and the degree of overlap. The PDI measures for the 24 countries (based on the consistent coding and the alternative coding) are both strongly correlated with the strength of the etacoefficients reported in Table A (0.90 and 0.91, respectively). The strength of the correlations between church attendance and party choice and religiosity and party choice, respectively, are similar although the correlations for church attendance are somewhat larger. 15 The difference is, however, somewhat smaller than for the pooled data set: The averages correlations with party choice for the 24 countries are and for church attendance and religiosity, respectively. The correlation between church attendance and party choice is larger than the correlation for religiosity for 20 of the 24 countries. Here, I therefore focus on church attendance. < Table 5 about here > There are strong variations in the correlation between party choice and church attendance. The correlations are strongest in the Netherlands, Norway, Slovenia and Finland and weakest in Britain and Estonia. Somewhat surprisingly the average strength of the correlation is strongest in the Nordic countries and then Central Europe, followed by Eastern and Southern Europe, while the average correlation is the Islands is smallest due to the extremely low correlation in Britain (see Table E). 13 These correlations at the individual level should not be mixed with the correlations between the comparative means based on the 24 countries reported earlier, which are considerably stronger, The degree of overlap is 0.69 and 0.70 respectively when the Greens are not grouped among the leftist parties. The degree of overlap for church attendance was calculated by comparing those who attend church at least once a week with those who never attend church (see text), while for religiosity overlap this was calculated by collapsing the two extreme categories on the 11-point scale and then using these collapsed categories as a basis for comparison. 15 The similarity is expressed when the eta-correlations for church attendance and religiosity are correlated for the 24 countries. This correlation is 0.90.

16 15 The Green parties get consistently across the countries where they are significant in the data, stronger support from the secular segment of the population. They are therefore grouped among the leftist parties in the analyses of the left-right division below. The ranking of the countries based on the left right division of the parties shows a different ranking on the top of the list (see Table C). Three countries are now ranked ahead of Norway and an additional fourth country ahead of the Netherlands and Slovenia is among the countries where the correlation with left right party choice is lowest. The strength of the correlation between church attendance and left-right party choice is largest in Central Europe according to the average correlation, followed by the Nordic countries and Southern Europe, while the average correlation is considerably smaller in Eastern Europe and the Islands as can be seen from Table E. The degree of overlap is remarkably high for most countries. In the seven countries where the overlap is lowest (less than 0.60) we find the same four East European countries that had the lowest degree of overlap on the religious denomination variable (see Table 4D) and Denmark, the Netherlands and Norway. In Southern and Central Europe and the Islands the average level of overlap is remarkable high according to the average score while the figures for the Nordic countries and Eastern Europe are smaller (see Table E). For the East European countries it is again the Liberal parties that contribute to almost all crosscut by gaining considerably more support among the secular segment of the voters. The same applies to The Netherlands and voters for the radical rightist List Pim Fortuyn also recruited more voters from the secular segment. In Belgium where overlap also is comparatively small (see Table 5D) the Liberal and Radical Rightist parties contribute equally is this respect. Both party families 16 get strongest support from the secular segment. In Norway, both the Conservative and Radical Rightist parties gain strongest support from the secular segment. The same applies to the Agrarian Liberals and the radical rightist Danish People s Party in Denmark. In Norway the components attached to the two mentioned parties are fairly similar, while the pattern for the Agrarian Liberals is dominant in Denmark. The Conservative parties in the Nordic countries are interesting cases regarding the religious cleavage. Apart from the Danish Conservative Party, these parties gain considerably stronger support from the secular segment: The support from those who attend church once a week and more frequently and those who never attend church, respectively are 14% and 20% in Finland, 6% and 20% in Norway, and 9% and 23% in Sweden. In Denmark we do not find the same pattern for the Conservative party (11% and 6%) but for the considerably more successful Agrarian Liberal Party (19% and 32%) which has taken many voters from the Conservatives during the last years. We do not find equivalent tendencies among the leftist parties: All leftist parties get strongest support from the secular segment of the population. There is, however, one important exception, the social democratic PASOK in Greece get stronger support from those who attend church frequently. This is the main explanation for why the overlap in Greece is fairly low I use the notion party families since there are two parties within each family, one Flemish and one Francophone. 17 All the mentioned causes of low degree of overlap for frequency of church attendance are also found for the self-declared religiosity variable.