The Impact of Religion on Voting Behaviour – A Multilevel Approach for Switzerland


  • I would like to thank Peter Selb for the help in developing the research idea and Pascal Sciarini plus the two anonymous reviewers for their helpful comments and suggestions. A special thanks to Stéphanie Plante and Cassidy Travis for editing the different versions of the article.

Address for correspondence : University of Geneva, Departement de science politique et relations internationales, 40 bd du Pont-d'Arve - 1211 Genève 4;


The traditional cleavages like social class or religion are often reported to have lost explanatory power for the voting decision. Regarding the religious cleavage, the evidence is ambiguous depending on the choice of cases and indicators used. The present study tests the impact of religion for the preference to vote the Christian Democratic Party of Switzerland (CVP) using data from the 2007 and 2011 Swiss national elections. Additional to the inclusion of individual variables, a special focus lies on contextual effects. The estimated hierarchical linear models confirm a prevailing influence of the simple individual factors and the presence of a significant contextual effect. Statistical evidence is also presented for some of the supposed interaction effects between individual and contextual religious variables.


During the last few decades the focus of electoral research has changed quite significantly. In the 1960s and 70s, the traditional cleavages according to Lipset and Rokkan (1967) were in the centre of scholarly interest. At this time religion was one of the most influential factors for the voting decision. More recent studies claim that new cleavages, or rather divisions, are weakening or even replacing the former influential cleavages (e.g. Bornschier 2010; Brunner and Sciarini 2002; Inglehart 1977, 1989; Kriesi et al. 2006). As a consequence, religion is repeatedly disregarded as a prevailing factor. Broughton and ten Napel (2000: 4) state that religion is often either mentioned only for a historical overview or to highlight its declining impact.

Besides the prominent counterexample of the United States, where the religious dealignment between Catholics and Christian Democratic parties is not observable (Norris and Inglehart 2004), recently several scholars in Europe state that religion is still (or yet again) a significant factor worth to analyse. Even though some acknowledge that the religious cleavage might be weaker compared to previous times, they argue that it is still relevant in electoral decision-making (e.g. Brooks et al. 2006; van der Brug et al. 2009; Dalton 2002; Enyedi 2008; Esmer and Pettersson 2007). Additionally, there are studies that speak of a possible religious revival in post-industrial Europe (Geser 1997; Knutsen 2004; Norris and Inglehart 2004; Wolf 1996). With respect to the religious cleavage, the Swiss case is particularly convenient as the country was built along denominational lines. Consequently, religious differences have always been one of the most important factors in defining how individuals will vote (Hug and Trechsel 2002; Lijphart 1979; Trechsel 1995).

Whereas the individual influence of religious characteristics has been studied extensively, there is less known about contextual effects. Although socio-structural variables like religion are “better conceived as indicators of a collective social experience” (Huckfeldt and Sprague 1993: 293), in most studies they are only treated as individual factors. This is all the more surprising, since the socio-structural approach emphasizes the importance of the social environment (Bornschier and Helbling 2005; Carmines and Huckfeldt 1998). The authors of the Columbia School (see Lazarsfeld et al. 1944; Berelson et al. 1954) have already considered how much individual behaviour depends on the social context. However, most studies since then concentrate on the analysis of individual characteristics. In Switzerland there are significant differences in the religious composition of the cantons. This offers the opportunity to identify and examine contextual effects. The aim of the paper is thus to analyse the impact of the religious environment regarding the individual and contextual level, with a specific focus on the relationship between both levels. The religious context may show a simple direct effect, but it may also interact with individual factors and condition their influence. These possible interactive contextual impacts are understudied in electoral research. To analyse them, the focus is not on all the Swiss parties, but is restricted to the most important party in the religious context, the Christian Democratic Party (CVP) (cf. Kriesi and Trechsel 2008).1

The structure of this article is as follows: The next chapter presents the theoretical significance of the religious cleavage and its empirical importance for Switzerland. Next, I develop the hypotheses of the religious effects. This is followed by the data section, which specifies the operationalisations and the used method. I then test the hypotheses using hierarchical models. A brief discussion summarizing the most important findings concludes the study.

Theory of cleavages

Today the concept of cleavage is sometimes used in inappropriate ways. It is therefore important to properly define the meaning of the concept when discussing its impact.


In political science literature, the term “cleavage” is widely used, however often with different meanings. Frequently it describes any form of political division, although its actual definition is much more restrictive. For Pappi (1983: 185) a cleavage (Konfliktlinie) is a lasting potential of conflict, which is rooted in social-structural groups of a society and expresses itself at elections and votes due to its politicization. This definition already contains the empirical element out of the common conceptualisation by Bartolini and Mair (1990), yet the normative and organisational element is missing. This three-part concept is the only one really defining the notion of cleavage and has become widely accepted in scholarly literature (e.g. Bornschier 2007; Enyedi 2008; Hug and Sciarini 2002; Kriesi 1998; Lachat 2007).

The first element is the empirical one, defined in social-structural terms. A division between opposing social groups may be based on class, religion, status, etc. These groups hold a common set of values and beliefs through which they develop a sense of collective identity, the normative element. The last component is the organisational/behavioural element. It comprises the articulation of the group's interest through institutions or organizations, such as political parties (Bartolini and Mair 1990: 199; Bornschier 2007: 5; Lachat 2007: 27).

When defining a religious cleavage two aspects are central (e.g. Dalton 2002). On the one hand is the religious community a person belongs to (religious denomination), and on the other hand, it is also important how religious a person is independent of the denomination (religious attendance and belief). The latter is normally measured by church attendance. These two aspects stand for different elements of the above presented definition. Religious denomination denotes the empirical or structural element (Knutsen and Scarbrough 1995: 499). Lane and Ersson (1999) make a slightly different distinction between a latent and manifest part of the cleavage. Denomination represents the latent aspect, which is equal to the structural part of the definition used here. The authors argue that “(r)eligious structure is by definition a latent, or unconscious, structure which can be transformed into a manifest structure through a process in which religious cleavages become tied to some kind of religious awareness” (ibid.: 51). This manifest aspect or “value orientation” labelled in Knutsen and Scarbrough (1995: 500) is church attendance which represents the normative element.

A central point in the literature is the reservation of the notion of cleavage for relationships with a certain durability. The links between social groups and political parties persist over several generations of voters, even if the conflict which led to the formation of a given party is no longer present. This might be especially true for the religious cleavage, since lately no major religious issues have entered into political discussion. In the definition of Knutsen and Scarbrough (1995) the structural component (denomination) represents the long-standing ties to a certain party. These voter-party links can possess a direct impact on voting behaviour, but also an indirect one through the normative aspect (church attendance). The latter therefore has a mediating and/or reinforcing impact, but also a simple direct effect. As Lane and Ersson (1999: 52) show, there is a strong relationship between religious denomination and religious awareness (church attendance). This is especially pronounced when church attendance is considered in relation to the size of the Catholic population, which will be examined in the following analysis.

The religious cleavage in Switzerland

Till the late 1960s, Switzerland was a classic example of the “freezing hypothesis”, as the party system had remained unchanged since the 1920s (Kriesi et al. 2005: 3). Like in most other European countries, the religious and the class cleavage were highly important during this time. They continue to show considerable effects nowadays, even if a certain decline has been observed since the 1970s (Lijphart 1979; Linder and Steffen 2006; Trechsel 1995). Earlier studies showed that in comparison to the other important cleavages of class and language, religion was the main determinant of electoral behaviour in Switzerland (Lijphart 1979; Kerr 1987). On the one hand, this can be seen as a result of the nation's history. The formation of the Swiss Confederation was heavily influenced by and loaded with religious tensions (Geser 1997: 6; Kriesi and Trechsel 2008: 6). On the other hand, the significance of the religious cleavage remained higher than in many other countries in Europe due to a comparatively low intensity of the class conflict (Geser 1997). More recent studies accept a certain decline of the religious cleavage, but argue that it is still strong enough to compete with the linguistic or class division (Trechsel 1995: 32).

A development which the early study by Lijphart (1979) already revealed is the growing influence of religiosity. Traditionally, the denominational divide between Catholics and Protestants was the most important factor. Nowadays, it is rather the difference in religiosity (measured by church attendance) that matters, especially between religious and non-religious people (Geissbühler 1999; Lachat 2007, 2008; Trechsel 1995). The latter group is consistently rising in the last years.2 This speaks of a significant change in the structure of the religious cleavage. A recent study by Nicolet and Tresch (2009) reveals additionally that the substantial decline in people attending and belonging to a church (institutionalized religion) does not necessarily lead to a decline in religious beliefs. It is just a different so-called post-traditional form of religiosity. The authors report that around one quarter of the Swiss population in 2007 belongs to this group of believers without belonging to a church (ibid.: 91).

In addition to evidence about the individual religious impact, there are also some studies regarding the religious cleavage in a contextual perspective. Following the classification of Switzerland in three types of cantons by Klöti (1998) and Kriesi (1998), the latter argues that the religious cleavage is especially important for traditional Catholic cantons. In the religiously mixed German- and French-speaking ones it is largely pacified (see also Lachat 2008). To further this argument, Kriesi and Trechsel (2008) and Lachat (2008) add that in Catholic regions the conflict prevails between (practising) Catholics and secularized people, whereas in the other cantons a denominational division between Catholics and Protestants is present. Two studies present evidence of significant effects by the religious environment. Geser (1997) reports a strong impact of the confessional culture on the political ideology and issue positions. Geissbühler (1999) finds evidence of a direct positive influence on the sympathy for the Christian Democratic Party (CVP) depending on the number of Catholics living in a canton.

In terms of parties, the religious cleavage has traditionally opposed the Christian Democratic Party (CVP) with the Radical Party (FDP) (Ladner 2006). This opposition was mainly based on the historical division of the state vs. church conflict. Another competing party today is the Evangelical Party (EVP) (Kriesi and Trechsel 2008). Compared to the CVP, however, this Protestant party is rather of minor importance. A second party which especially represented the Protestants is the Swiss People's Party (SVP). However, this party is not specifically centred on the religious cleavage. In addition, the impact of religion on the vote for the SVP has strongly decreased during the last decade (Lachat 2008; Lutz 2012).

The strong Catholic background of the CVP is on the one hand observable at the demand side. Compared to the overall population, the Catholics show stronger attachment to the CVP and simultaneously have rather hostile feelings towards non-Catholic institutions (Geissbühler 1999).3 This strong Catholic base, however, is also a danger in times of religious decline. In the last decades the party tried to position itself in the centre of the political spectrum with a modernistic and Christian, not solely Catholic, perspective (Linder 2009). It has opened up to attract also Protestants, however only with moderate success, so that the CVP still shows signs of a Catholic-confessional party (Kriesi and Trechsel 2008; Linder 2009).4 On the other hand, the Catholic tradition of the party is also observable at the “supply” side. Geser (1997: 25) reports that the elites of the CVP are still surprisingly homogeneous when it comes to denominational composition. The strong link of the CVP with and dependency on religion makes the party the perfect object for studying the supposed contextual and interaction effects.

The influence of religion

In this part the theoretical approaches of religious voting are linked with the socio-structural influence of the environment. First, I will briefly review the well-established theory regarding the individual effects. In a second step the focus lies on the insufficiently examined contextual effect and the interaction with individual impacts.

Individual effect

As many scholars have already theorized and tested the individual religious effect (e.g. Dalton 1996, 2002; Elff 2007; Knutsen 2004; Lijphart 1979; Norris and Inglehart 2004; van der Brug et al. 2009), it shall be sufficient to present the main arguments and results before discussing the contextual level. Formerly, political topics directly related to specific religious concerns and the strong tensions between Catholics and Protestants have been responsible for the religious impact. Today there are other reasons why religion still plays a significant role in explaining voting behaviour. One explanation is the strong link between religious values and a wide array of general political and social beliefs, which in turn are connected to party choice (Knutsen 2004: 99).5 A second explanation highlights the background of a person to determine its political standpoint. The church has lost its significance in directly influencing the voting decision. Nowadays internalized value systems, the identification with Catholicism as a social group or simply the customization to vote Christian Democratic parties are responsible for the continuing effect of religion (Schmitt 1984: 49). The latter fact is also mentioned by Broughton and ten Napel (2000), who argue that historical links rival and influence the present behaviour of the voters by producing cross-pressures.6

In the past, the difference between Catholics and Protestants was at the centre of the conflict. Nowadays the difference between religious and non-religious people is becoming more important (Brooks et al. 2006; Wolf 1996). Many people are identifying less and less with the institution church. However, the ones who still do and especially those who frequently attend church services differ strongly to atheists or other secularized persons. In this context, the more religious people could take offence at the modern and secular lifestyles, which deepens the division (Elff 2007: 279).

One development responsible for the growing importance of religiosity compared to denomination is certainly the process of secularization.7 This can be observed in most countries throughout Europe. The result is the observable shrinking of the Catholic and Protestant population. On the one hand, this opened a gap between the two denominational and the growing secular group. On the other hand, it narrowed the gap between Catholics and Protestants, as they realize that in the end both of them stand for many equal religious values. People who actively participate in religious life differ in their voting behaviour compared to non-religious people, even though the latter may “on paper” also belong to a church. The difference which influences the political attitudes is their active religious involvement.

Contextual effect

In comparison to the other two schools of electoral research,8 the socio-structural approach, known as Columbia School (Lazarsfeld et al. 1944), emphasizes the importance of the social environment (Bornschier and Helbling 2005; Carmines and Huckfeldt 1998). The environment shows influence on voting behaviour through basic societal conflicts, class consciousness, membership in certain groups and the corresponding background variables like social class, urban rural differences or denomination. The authors of the Columbia School (see Lazarsfeld et al. 1944; Berelson et al. 1954) started to use individual instead of aggregate data to draw inferences about social processes and also considered how much the individual behaviour depends on the social context. In doing so they laid the foundation of contextual analysis and the multilevel understanding of politics (Huckfeldt and Sprague 1993). Their famous quote “[…] a person thinks, politically, as he is, socially” (Lazarsfeld et al. 1944: 27) highlights the notion that in addition to individual factors, the social context also affects political behaviour.9

When defining a contextual effect, one very important point is the consideration of it as exogenous to the individual. Huckfeldt and Sprague (1993: 286) define a contextual effect to be operating when the behaviour depends on some external factor after all individual level determinants have been considered. Something not intrinsic to the individual is responsible for systematic variations across contexts. Disregarding the problem of possible self-selection into a certain social environment, the context lies beyond the reach of the individual control (Carmines and Huckfeldt 1998: 230). Although a context is not necessarily to be defined in geographical terms, most studies follow this example. In the literature normally characteristics of local geographical areas, ranging from neighbourhoods up to whole countries, are used to explain individual political behaviour (Books and Prysby 1988, 1991). For the Swiss case the cantons serve as the contextual level.

The characteristics of a context can include a wide range of aspects. It can be a “real” external factor or an aggregate of some individual characteristics (Przeworski and Teune 1970: 6). Huckfeldt and Sprague (1993: 293) even argue that in general many individual factors are better understood as indicators of a collective social characteristic. Arguing along the same line Stark and Bainbridge (1996: 72) suggest to “discard the assumption that religiousness is primarily an individual trait, a set of beliefs and practices of individuals, and substitute the assumption that religiousness is, first and foremost, a group property”.

In the literature we find two main explanations as to how the compositional effect influences individual electoral decisions. Social interaction being the most common one. It posits that communication with another person influences and modifies individual attitudes through the transmission of the (political) opinions of the interaction partner (Books and Prysby 1988).10 The size of the context and the intimacy of the person with whom the interaction takes place should be irrelevant. Huckfeldt and Sprague (1993) argue that voluntary interaction among friends is politically not more relevant than involuntary contact. The same is true for the contextual size, where smaller units do not have to be more important than larger ones (ibid.). A second explanation for contextual effects is the conformity reaction. This idea appears in the early Columbia studies in the concept of a “breakage effect” (Berelson et al. 1954). The mechanism behind this assumes that a person perceives the composition of its environment and reacts according to it. This reaction is normally a desire to conform to one's community and to agree with its dominant political norms. According to Books and Prysby (1991: 63) religion belongs to the properties of an environment most people are aware of and thus are in some way sensitive to.

For religion, the contextual effect in general and the social interaction in particular seems to play an especially important role. The individual (non-)religiousness is of little relevance for the political behaviour when regarded isolated from their environment. It is, however, very decisive how the individual religiousness is ratified by the social environment (Stark and Bainbridge 1996). Stark and Bainbridge (1996: 72) suppose that “religion is empowered to produce conformity to the norms only as it is sustained through interaction and accepted by the majority as a valid basis for action.” These norms may be politically interpreted as a form of model or rule to vote for certain parties, traditionally Christian Democratic ones. A similar idea is shown in Olson (2012). She argues that the political influence of religious adherence is rather a by-product of being active in this social milieu and receiving indirect but regular political messages through social interaction.

In Switzerland the religious context might be especially crucial. The early denominational divisions are still observable today, particularly in terms of Catholic strongholds in the centre of the country. Geser (1997: 6) reports that the intranational migration between Catholic and Protestant locations has been quite low. Consequently, a lot of communities and cantons are still rather homogeneous in terms of denominational belonging. Besides the Catholic strongholds, there are also mainly Protestant cantons. A different voting behaviour of the local population might then depend on the denominational structure. Combining the individual effect of religion and the influence of the social context, leads to the assumption of a direct effect of the amount of Catholics living in a canton.11 Catholics, which have a traditionally strong link to the CVP, strengthen or confirm their voting pattern not only by talking to other like-minded Catholics, but influence also non-Catholics. The latter is more likely the bigger the group of Catholics in a context is. The resulting hypothesis thus assumes that as more Catholics live in a context, their influence on fellow citizens to vote in conformity with them (namely for the CVP) becomes stronger.

The larger the Catholic proportion in a context, the higher the individual propensity to vote for the CVP. (H1)

Interaction effects

The idea of a contextual influence dates back to the researchers of the Columbia School (Lazarsfeld et al. 1944). Nevertheless it took several years until the effects of the social context on individual voting behaviour were modelled explicitly. One of the first studies, on the “search of a curve”, is by Przeworski and Soares (1971). They criticized that earlier research dealt with variables related to voting behaviour only in additive and thus linear relationships. In contrast to this, the authors argue to focus not only on the substance of a theory. It is equally important to place emphasis on the form of a relationship to derive valid statements regarding voting behaviour. Consequently, their formal model includes interaction terms between individual and contextual variables. A successive study by Huckfeldt (1980) delivers empirical evidence of these interactive models. The argument is a change of the contextual effect depending on the individual characteristics. My argument in contrast is a change of the individual effects according to the social context.

Figure 1 summarizes graphically the link between the contextual and the individual level. In addition to direct effects on voting behaviour from both levels, the respondent and the environment, I assume that the context moderates/conditions the influence of the individual determinants (dotted line). The context as a moderator should affect “the direction and/or strength of the relation between an independent or predictor variable and a dependent or criterion variable” (Baron and Kenny 1986: 1174). More specifically, the religious environment is expected to change the strength of the two individual religious variables (denomination and church attendance).

Figure 1.

Moderating effect of religious context

According to Kriesi and Trechsel (2008) in Catholic regions the religious conflict opposes practising Catholics to secularized people, whereas in non-religious cantons it opposes Catholics to Protestants. Going one step further, I assume that depending on the religious composition of a canton either the effect of denomination or religiosity prevails. In religiously homogeneous contexts with a significant majority of Catholics or Protestants, the difference between the two denominations should be less important. This is due to the absence of a significant opposing group. Geser (1997) reports evidence of this fact for the Catholic population, which lacks the stimulus of forming a denominational culture once they are in an overwhelming majority. Instead of focusing on denominational differences, in religiously homogeneous cantons the difference in religiosity should have significant impact. Exactly the opposite would then be true for religiously heterogeneous cantons where Catholics/Protestants face a significant opposing group. In these cantons the denominational difference should still play a major role.

The incentive to focus on one of the two religious variables is also expected to be displayed in terms of party competition to mobilize the voters. In a canton, where almost solely Catholics live, all competing parties have to represent the attitudes and opinions of the Catholic population in a certain way. However, mobilizing purely around Catholic values and issues (in a positive or negative way) will not give them an advantage in winning parliamentary seats. In contrast, the main focus of the parties lies on other factors like the level of active religious participation. To distinguish itself from other parties, the CVP might specifically concentrate on trying to get the votes from the religiously active population. This includes the Catholic citizens, but also religiously active Protestants. In homogeneous Protestant cantons a slightly different effect may be at work. The Protestant cantons are not as religiously homogeneous as the Catholic ones. Hence, the exclusive concentration on Catholic values can lead to the gain of a small number of seats by receiving the Catholic votes, however this depends on the amount of seats to be allocated. Therefore, especially in small cantons, the CVP might leave the competition empty- handed when only trying to mobilize their Catholic electorate. To gain additional votes from Protestants, the party should again rather focus on general religious values which are shared by the two denominational groups. Consequently, a reasonable strategy of the CVP would be to highlight topics to get the votes from their traditional electorate, the Catholic minority, but also from a certain part of the Protestant majority, namely religious citizens. Combining these lines of argument leads to the expectation of a U-shaped relationship. At both ends of the x-axis representing the density of the Catholic population a stronger effect of religiosity is expected compared to a lower impact in religiously-mixed cantons.

In homogeneous contexts (mainly Catholic or Protestant) religiosity has a stronger impact on the voting probability for the CVP than in religiously-mixed contexts. (H2)

The expectation for an interactive effect of the denominational difference is the exact opposite relation. Here, an inverse U-shaped relationship seems plausible. The difference between Catholics and Protestants is especially important in religiously-mixed cantons.12 As argued in the former hypothesis, in homogeneous cantons the denominational difference is less important than other (religious) factors. In cantons where the population is separated in almost equal parts, meaning that there is a significant ingroup-outgroup antagonism, however, the parties can still focus on denominational differences and thus gain a majority of the votes. The CVP representing primarily Catholics in these mixed contexts are then expected to predominantly receive votes from the Catholic population, independent of their religiosity.

In religiously-mixed contexts denomination has a stronger impact on the voting probability for the CVP than in homogeneous contexts. (H3)

Data and Method

Data from the Swiss Electoral Studies (SELECTS) 2007 and 2011 are used for this analysis. In the first rounds of the SELECTS project a cantonal representative sample was drawn only for a part of the 26 cantons. The last two rounds aimed to include at least 100 respondents for each canton. These numbers are sufficiently large to allow the later presented method of multilevel modelling, as for each canton enough respondents are included.13 Due to some missings on the dependent and the religious independent variables, the number of interviews finally used for this paper is 4125 in 2007 and 4189 in 2011.

Additional information required for the contextual factor is derived from the Swiss Federal Statistical Office (BfS). Contextual data is available for all 26 Swiss cantons, however in 2007 the canton of Nidwalden had to be excluded as no SELECTS data was collected there. This is due to the fact that just one candidate ran for the single available seat, so that no election (also called “tacit” election) took place (Lachat 2011: 649). Thus, only 25 cantons are included in the final sample for 2007.


Dependent variable

Voting probability

Traditionally, studies analysing voting behaviour regard discrete party choice of the electorate as the dependent variable. However, in the 1990s another method that used voting preferences emerged. These improved survey measurements enable more sophisticated analyses of current electoral behaviour (cf. van der Brug et al. 2009; van der Eijk et al. 2006; Lachat and Selb 2005; Lachat 2011; Tillie 1995). Van der Eijk et al. (2006: 425-426) argue that older studies using nominal scaled dependent variables fail to distinguish between choice and the attractiveness of a party. This leads to difficulties in analysing voter choice. According to the authors, the classic works of the social-structural tradition (e.g. Lazarsfeld et al. 1944) are no exception to this. These studies focus on the determinants of utilities for parties and at the same time imply that the electorate selects the party with the highest utility.

Empirical evidence of the strong relationship between the voting probability and party choice has been provided in the literature. For the Dutch case van der Eijk et al. (2006: 435) show that 93 per cent of the respondents voted for the party they gave the highest utility score. The identical number was found by Lachat (2011: 650) for Switzerland using the SELECTS data from 2007. Sciarini (2010: 122) found for the same data the slightly different number of 88 per cent, which still stands for a very strong link. Therefore, the authors come to the conclusion that the relationship between voting propensities and actual choice is almost deterministic.

The concrete operationalization of the voting probability for the CVP is measured on an 11 point scale. Respondents were asked to indicate the probability that they would ever vote for the party. Answers range from a very low (value of 0) to a very high probability (value of 10).

Independent variables

The measurement of religious characteristics follows the mainstream in electoral research. It is argued that religion consists of two central aspects, religious denomination and religious attendance or belief (here referred to as religiosity) (e.g. Dalton 2002).14

Religious denomination

In Switzerland the two main groups are Catholics and Protestants. Additionally, there are some minor Christian groups, Jews and a growing community of Muslims. Today there is also a large third group, which consists of people who do not belong to any religious community. For the statistical models only the difference between Catholics and non-Catholics is considered. A dummy (Catholic) is used with the value of 1 for respondents who indicated they belong to the Catholic church.


The second religious variable measures religiosity in terms of church attendance. Here, respondents have been asked how often they attend church services or other events of their church. Answers are measured on a 7-point scale ranging from “several times a week” to “never”. This scale was reversed, so that high numbers represent regular attendance. The scale then fits to the supposed positive relationship between church attendance and probability to vote for the CVP. In the original variable, around 1000 respondents have been coded as missings. The majority of them do not belong to any church and therefore have not been asked about their church attendance. If this coding would be maintained, almost a quarter of the overall sample would be lost. Since the lowest category of religiosity classifies people who “never” attend church services, all missings that have been coded so due to their prior indication that they do not belong to any church have been grouped in that category as well.


The contextual effect of religion is measured by the number of Catholics living in a canton15. Corresponding numbers are retrieved from the Swiss Federal Statistical Office and originate from the year 2000 (see table A2 in the appendix). The SELECTS data used in the analysis are more recent, thus some minor differences can be expected. A further decrease in people belonging to the Catholic church is very likely, however the differences should be rather small and therefore negligible.

Control variables

In addition to the three independent variables, two control variables are included in the analysis. These regard specific issues of the Swiss electoral system and historical developments of the party competition.

Threshold of exclusion

The National Council in Switzerland consists of 200 seats, which are distributed according to proportional representation. Constituencies are considered at the cantonal level and not on the national level. Thus, in smaller cantons there is a de facto majoritarian voting system.16 This is due to the distribution of seats to the cantons according to their population. More precisely the number of eligible voters is crucial, with a minimum representation of one seat per (half-)canton (Linder 2009: 574). This fact can have consequences for the individual voter, because some of them might vote strategically for their second or third preference in order not to waste their vote. The same is true for the reported voting probabilities.17 Some respondents may realize that their first preference will never be able to win a seat, so they score the party lower then actually desired. In return they score their second or third preference highest or at least equal to the first, because these parties have a real chance of winning.

The effective threshold of exclusion is derived from Lijphart (1997). It reports the proportion of votes a party can receive without winning a seat. The formula is math formula with m as district magnitude. This magnitude ranges from 1 representing the six smallest cantons to 34 in Zürich. Resulting values do not represent the highest possible thresholds, but shall be more realistic representing the average between high and low values. As the distribution of the variable is heavily skewed, following Lachat (2011: 652) the natural logarithm of it is used.


In the context of Swiss referendums or elections, voting behaviour sometimes differs between the German- and French-speaking part of the country. These differences are said to be based on cultural issues and became known as the so called “Röstigraben”. In his analysis comparing the influence of religious, linguistic and class characteristics, Lijphart (1979) found that support of religious parties is stronger in the German-speaking part controlled for denominational belonging.18 In case of existing regional differences in CVP preference independent of the religious composition, the actual religious contextual effect might be hidden. Consequently, a dummy for the six French-speaking cantons Fribourg, Genève Neuchâtel, Jura, Valais and Vaud is included to control for a possible cultural effect.


In order to analyse the simultaneous impact of individual and contextual variables, hierarchical multilevel models are calculated. As the dependent variable is linear, regressions are calculated by the method of OLS. This multilevel method enables the estimation of the effects in appropriate statistical ways. At the same time it allows the inclusion of cross-level interactions (Steenbergen and Jones 2002). The basic random-slope model without interaction terms has the following form:

display math

The voting probability for the CVP (Y) of an individual i in canton j is explained by several factors. In detail, this is a global mean for the propensity of CVP voting (β0), the two individual religious factors (their estimates β respectively), which as random slopes are allowed to vary between cantons (math formula) and math formula) and an individual variation (εijN(0,σ2)). Additionally, the contextual factors are included. These are the religious context (γ1), the two control variables (their estimates γ respectively) and a cantonal variation (math formula).

In the next step cross-level interactions are included to check the varying influence of the two individual religious variables out of hypotheses 2 and 3. In order to see a possible curvilinear relationship, the individual variables do not only interact with the simple Catholic proportion, but also with the quadratic term of it. The resulting formula has the following form:

display math

with X1 = catholic, X2 = church.attendance, Z=catholic.proportion, C1 = threshold, C2 = west.swi

Several scholars argue and demonstrate that in multiplicative interaction models the (in)significance of the single coefficients is not the crucial point and might be also misleading sometimes (for a detailed discussion see, e.g., Brambor et al. (2006), Friedrich (1982) and Kam and Franzese (2007)). According to the authors, a more appropriate handling and interpretation is possible by looking at marginal effects of the independent variables and especially a graphical presentation of them. Since the interest of this paper is the testing of the expected patterns of an (inverse) U-shape, the proposed visual presentation of marginal effects is particularly suitable. Consequently, to ensure a proper analysis, the marginal effects for both individual religious variables are calculated.19


The results of the estimated models are displayed in table 1.20 To check the proposed hypotheses, five different models have been calculated. Additionally, a so called “empty” model (not displayed) was estimated to determine the portion of the variance which is due to the contextual level. The resulting intra-class-correlations are approximately 0.05 and 0.07 respectively, meaning that the contextual level explains only little of the variance. Individual differences are responsible for the majority of the variance.

Table 1. Hierarchical multilevel models for the voting propensity for the CVP
(a) 2007(b) 2011
  1. Standard errors in parentheses

  2. *p < 0.10, **p < 0.05, ***p < 0.01

church attendance0.31***0.31***0.31***0.27***0.240.31***0.30***0.30***0.20*0.28
catholic proportion 1.32***1.24***1.46***0.21 1.55***0.95**0.551.29
 (0.34)(0.40)(0.54)(3.60) (0.45)(0.44)(0.54)(3.44)
catholic proportion2    1.26    0.69
    (3.42)    (3.31)
threshold of exclusion0.14*** 0.28**0.28***0.29***
(0.10) (0.09)(0.09)(0.09)(0.10) (0.11)(0.11)(0.11)
west switzerland-0.23-0.37**-0.35*-0.34**-0.32*-0.23-0.31-0.25-0.25-0.25
catholic X   -1.22**1.70   0.153.34
catholic proportion   (0.57)(3.93)   (0.68)(4.78)
church attendance X   0.100.23   0.22-0.19
catholic proportion   (0.15)(1.10)   (0.20)(1.37)
catholic X    -2.78    -3.00
catholic proportion2    (3.70)    (4.50)
church attendance X    -0.12    0.39
catholic proportion2    (1.04)    (1.29)
sd (catholic)0.410.380.410.280.280.400.430.400.440.46
sd (church attendance)
sd (constant)0.410.300.350.360.380.270.450.340.320.34
corr (catholic, church attendance)-0.96-1.0-0.81-1.0-1.0-0.13-0.09-0.11-0.15-0.14
corr (catholic, constant)0.270.910.550.940.920.04-0.090.00-0.04-0.14
corr (church attendance, constant)-0.53-0.90-0.77-0.93-0.88-0.27-0.51-0.51-0.43-0.40
Observations (groups)  4125 (25)    4189 (26)  

The first three models only include the simple religious variables and the control variables, but not the interaction terms. As expected, in both years the denominational difference between Catholics and non-Catholics and religiosity measured by church attendance show a strong positive significant effect. When considering the contextual level it becomes clear why a separate display of the two election years is necessary. In both years the Catholic proportion shows the hypothesized positive impact. In 2011, however, the strength of the effect is heavily influenced by the inclusion of the threshold of exclusion. The threshold itself shows a significantly positive effect in 2011, in contrast to an insignificant effect in 2007. A change took place leading to a significantly higher probability to vote CVP in smaller cantons. At the same time it lowered the impact of the Catholic proportion in each canton. One possible explanation for this phenomenon is the additional competition by the two youngest parties BDP and GLP.21 Just like the CVP they are centre-parties and compete for a similar electorate. This could lower the attractiveness of the CVP, as these new parties might represent parts of the electorate in the political centre better than the CVP. The significant effect of the threshold may then result out of the fact that one or both parties only compete in PR cantons with at least three seats and not in the very small cantons.22 Since CVP probability remains on a constant high level in the smallest cantons, but drops throughout the bigger cantons, the significant effect of the threshold might be explained (see figure S1 in the appendix). Despite the difference in the contextual religious impact when including the threshold of exclusion in 2011, the first hypothesis assuming a positive contextual impact is confirmed. Additionally, the control variable for West Swiss cantons shows a negative influence, which is significant in 2007, but lost in strength and is no longer significant four years later.

Models 4 and 5 analyse the relationship in an even more sophisticated way now including additionally interactions with the contextual variable(s).23 In the fourth model the two individual religious variables are interacted only with the simple Catholic population. The actual interesting model, though, is the fifth one including also the quadratic term of the contextual variable and its interactions. At first glance, the results are disappointing because none of the religious variables or their interaction terms are statistically significant.24 However, as discussed in the method section, a simple consideration of the single coefficients does not allow an appropriate interpretation of the results. A visual illustration of the results is helpful in this case.

The graphical evidence in figure 2 displays significantly positive effects for both religious variables in both election years.25 Especially interesting though is not the positive impact, but the form of it. According to hypothesis 2 the pattern for church attendance should be a U-shape indicating a stronger effect in religiously homogeneous contexts. This relationship (figures 2a and 2c) is not supported by the graphs. Instead, the point estimates of the marginal effect display a slightly positive tendency. Consequently, the effect of church attendance is, at best, influenced somewhat positively by the higher Catholic proportion living in a canton. Due to the wide confidence intervals, however, there might be also no influence of the context on individual religiosity. Hence, the null hypothesis of no context effect cannot be rejected.

Figure 2.

Marginal effects of religious variables

Regarding the denominational effect in figures 2b and 2d support the expected inverse U-shape out of hypothesis 3. Still, the confidence intervals are too big to verify the observable pattern of an inverse U-shape.26 The pattern itself though is much more striking. In 2007 a prominent downward trend was observable indicating the almost non-impact of denomination in mainly Catholic cantons. Additionally one can observe a slight upwards trend for strongly Protestant cantons towards more mixed ones. The 2011 data shows exactly the expected pattern. A rising impact up to Catholic proportions of around 50 to 60 per cent is observable before a decrease of the effect begins. In cantons with a majority denomination, either Catholic or Protestant, the denominational influence is smaller compared to religiously heterogeneous cantons. Consequently, hypothesis 3 seems to be confirmed, albeit not very strongly.

When interpreting the graphs one minor limitation of the data has to be mentioned. Due to the limited number of cantons, the estimations of marginal effects sometimes heavily depend on just a few cases. This is especially true for non-Catholic cantons. Table A2 in the appendix shows descriptive numbers of the Catholic population per canton. For instance, Bern is a clear outlier as a canton with a very small Catholic population, so that the marginal effects at the extreme left side of the x-axis heavily depend on that case. This is less a problem on the other side. Unfortunately, this issue is often present when working with rather small numbers of cases, but should still be mentioned in the interpretation of the results.


The purpose of the study was to determine the religious impact on voting behaviour in Switzerland, specifically for the Christian Democrats (CVP). In contrast to many other studies, the focus was not purely on the individual religious variables of denomination and religiosity. The main interest was rather the influence of the religious context and its interactions with the individual characteristics. The idea for a contextual influence led first to the expectation of a simple positive effect of a stronger Catholic environment on the propensity to vote for the CVP. Furthermore, I assumed that the two factors of denomination and religiosity differ in their importance depending on the religious context. Whereas religiously homogeneous contexts should favour the strength of an effect of religiosity, denomination was expected to be especially important in heterogeneous cantons. This idea was not completely new, as authors such as Kriesi and Trechsel (2008) assume a similar relationship for the Swiss case. Innovative for the present study was the detailed statistical analysis of these effects.

So far contextual influences have been conducted in a very rudimentary manner with only a few studies assessing the possible effects. The results of the present study lead to a partly confirmation of the hypothesized influences. The first hypothesis is confirmed by saying that citizens who live in a stronger Catholic context prefer the CVP significantly more than citizens in less Catholic cantons. The hypothesized varying strength of the individual religious aspects according to the context could be (weakly) confirmed only for the denominational impact. In the case of religiosity, the expected U-shape is not observable. In contrast, the denominational difference clearly shows the expected inverse U-shape. In particular, the pattern in the 2011 data shows a much more prominent effect of being Catholic in religiously mixed cantons compared to homogeneous (non-)Catholic cantons. Only the relatively wide confidence intervals prevent a stronger confirmation of the expected pattern.

The empirical findings assume that religious values exert a significant positive influence on the preference to vote for the CVP. In addition to simple linear effects for the individual and contextual factors, some evidence for an interplay between both levels could be presented. Besides the already mentioned limitations of the data, the results are furthermore limited as the effects were not controlled for other cleavages. To get a complete picture one needs an expanded analysis that accounts for traditional cleavages such as social class or the rural-urban difference. The additional analysis of other cleavages also provides an opportunity to examine the contextual effect in further detail. In some cases, there may also be relationships between individual and contextual variables from two different cleavages. An example of such a cross-cleavage relationship could be a weaker impact of social class among the religious population. For the latter, religious values still represent the main determinants for their electoral behaviour. In contrast, the secular population's voting decisions tend to be influenced by non-religious values like their social class. Looking at cross-cleavage relationships thus appears as a promising avenue for further research.


  1. 1

    Instead of providing evidence of the total impact of religion for the electoral behaviour, the method used is particularly suitable for examining the relationship between the contextual and individual level.

  2. 2

    This can be seen when regarding the figures of citizens officially belonging to a church retrieved from the Federal Statistical Office.

  3. 3

    In general, Christian Democratic Parties profit from a large Catholic population, but also from a distinct religious heterogeneity (Frey 2009).

  4. 4

    Evidence of the religious divide between the CVP and the other parties is shown in table A1 in the appendix. The voters of the CVP are overwhelmingly Catholic (around 76 and 82 per cent in both election years). In the other parties the proportion of Catholic voters is significantly smaller. The same holds for the religiosity measured by church attendance. Voters of the CVP are clearly more religious compared to the electorate of other parties.

  5. 5

    Among these beliefs are attitudes like work ethics, morality, lifestyle norms, achievement aspirations, parent-child relations or the acceptance of the state (Knutsen 2004: 99).

  6. 6

    Cross-pressure can be understood as a situation in which the individual is exposed to inconsistencies and conflicts, which drive it into opposite directions (Lazarsfeld et al. 1944: 53).

  7. 7

    Secularization changes the focus from older values, especially religious ones, to new rather secular issues (cf. Altermatt 1989; Berger 1973; Inglehart and Baker 2000).

  8. 8

    The rational-choice (Downs 1957) and the socio-psychological approach, known as Michigan School (Campbell et al. 1960).

  9. 9

    In many studies examining the religious cleavage, however, the analysis of individual characteristics is prevailing. Only in a few cases contextual factors are regarded as well, but mostly just in rather basic terms (e.g Geissbühler 1999; Pappi 1985).

  10. 10

    Contrary to the more individualized view of the modern citizen, this is still true today as Levine (2005) reports.

  11. 11

    Geissbühler (1999) analysed this effect of the religious environment (Catholic population) and found that the sympathy for the CVP indeed grows with the number of Catholics. However, his database only consisted of 13 instead of 26 cantons, which weakens the reliability of his results.

  12. 12

    On the European level, the study by van der Brug et al. (2009) provides evidence for a similar reasoning. The authors can show that religion plays a more important role in influencing party choice, the higher the degree of religious fractionalisation. The latter measure is different to the here used dichotomy between Catholics and Protestants. However, it argues in the same direction of a bigger importance of individual religious factors the tighter the religious competition is.

  13. 13

    The absolute smallest number of respondents is 64 in Glarus 2011. Especially in the whole dataset 2007, however, the numbers are much closer to the desired 100 cases or even above.

  14. 14

    There are also voices critical of that rather simple operationalization. Authors such as Wolf (1996) and Broughton and ten Napel (2000) claim that for the many dimensions of religious life more specific measurements are necessary (e.g. the subjective rating of the own religiosity, the importance of religion as one area of life or specific current religious issues, such as abortion or genetic questions).

  15. 15

    In the literature the canton is the common used level for contextual effects. According to Steenbergen (2010: 403) it is normal to consider variations between cantons when regarding vote choice in Switzerland, as the electoral system is mainly defined at the cantonal level.

  16. 16

    Exactly speaking it is a plurality vote, where the candidate with the most votes wins the seat without the requirement of an absolute majority.

  17. 17

    The use of voting probabilities as the dependent variable is particularly convenient when strategic considerations may influence the voters' preferences (see for example van der Brug et al. (2007), van der Eijk and Franklin (1996) or van der Eijk et al. (2006)).

  18. 18

    This result is mainly due to differences for practising Catholics, whereas for Protestants and non-practising Catholics almost no difference between both regions is observable (see Lijphart 1979: 449).

  19. 19

    The exact way of calculation is displayed in the appendix.

  20. 20

    Since some of the coefficients differ in a significant way between 2007 and 2011 (at least to some extent due to a changing party landscape), separate models for both elections have been estimated. In a combined analysis some of the effects would have been blurred.

  21. 21

    Bürgerlich-Demokratische Partei and Grünliberale Partei.

  22. 22

    Exceptions are Ticino, where none of the parties competes and Glarus, the “home canton” of the BDP.

  23. 23

    According to the AIC, in 2007 model 2 is the best in explaining voting preference, slightly better than the fifth model. In 2011 the latter is the best model, pointing towards a stronger interaction effect in 2011.

  24. 24

    This is a general problem of models including interaction terms. Since these often lead to collinearities, the corresponding confidence intervals become bigger, which then result in insignificant coefficients.

  25. 25

    Only at both ends of the x-axis significance can not be ensured in all cases. This is due to the few cases left with corresponding numbers of Catholic citizens.

  26. 26

    A combined analysis of the marginal effect using both data sets confirms the inverse U-shape. It also narrows the confidence intervals, but still not to the extent that no other pattern would be possible.

  27. 27

    All the following equations are retrieved from by Brambor, Clark and Golder.


BfS (Bundesamt für Statistik) - Swiss Federal Statistical Office

SELECTS 2007 and 2011 - The Swiss Electoral Studies


Marginal Effect

Model equation to calculate the marginal effect of the variable “catholic”.27 For reasons of simplicity, the coefficients of the other covariates and interaction terms not necessary for the calculation are not displayed. The used values, however, are the ones estimated in the fifth model of table 1, where the coefficients are controlled by all religious and control variables.

display math

Marginal effect of being Catholic:

display math

Variance of the interaction model to calculate the corresponding standard errors and confidence intervals (with Z = catholic.proportion):

display math
Table A1. Location of voters on two aspects of religious cleavage
Denomination (in %)
Church attendance 4.363.473.283.423.01


  1. The party variable used is the reported vote in the election. The denominational variable shows the proportion of Catholics, Protestants and other/no denomination per indicated party vote. The church attendance variable shows the mean location of parties' voters on a scale from 1 (never) to 6 (several times a week).

  2. Source: SELECTS 2007, 2011

Denomination (in %)
Church attendance 4.243.323.143.312.952.993.25
Table A2. Catholic population per canton 2000
CantonProportion of Catholics in %
  1. Source: Federal Statistical Office (BfS)

Appenzell AR30.5
St. Gallen52.3
Appenzell IR81.3
Figure S1.

Scatterplot for CVP probability and threshold of exclusion

Source: SELECTS 2007/2011, own illustration


  • Andreas Goldberg is Ph.D. candidate and research assistant at the Department of Political Science and International Relations, University of Geneva. He is also a collaborator of the Swiss Electoral Studies (SELECTS) at FORS, Lausanne. His research focuses on electoral behaviour in Switzerland, particularly linked to cleavages. Other fields of interest include survey research (misreporting) and e-voting.