Literature on RRPs and religion

Scholars have found mixed evidence that religion positively affects the likelihood of individuals voting for the far-right. Van der Brug et al. (Reference van der Brug, Hobolt and de Vreese2009) suggest that the role of religious identity on voting is inconsistent and context-dependent. Arzheimer and Carter (Reference Arzheimer and Carter2009), using a structural model of Christian identity on far-right attitudes, find that there is no relationship between RRP party choice and holding orthodox, religious beliefs. Building off of Arzheimer and Carter, Montgomery and Winter (Reference Montgomery and Winter2015) find evidence that religious practice and belief decrease votes for European RRPs. Papastathis and Litina (Reference Papastathis and Litina2018), alternatively, find that religious beliefs enhance both mainstream right and far-right attitudes in voters. Additionally, Schwörer and Fernández-García (Reference Schwörer and Fernández-García2021), when examining the religious language of European election manifestos, find that references to “Christian roots” and exclusionary language around Islam has steadily increased since the 1980s. Furthermore, they find that this increased emphasis on religion has become salient beyond the far-right as more mainstream parties adopt religious language in their own manifestos.

One possible reason for these mixed findings is the different measures used to capture “religiosity” in the literature.Footnote ² Survey questions may capture different aspects of religion and are thus relaying different types of information about religion on the individual level (Dargent, Reference Dargent2019). For example, self-reported church attendance may capture the practice of religious rites, but it does not specify theological differences, even among members of the same sect. Church attendance is also a public and social practice, in contrast to other forms of religious behavior, such as individual time spent in prayer. Along similar lines, Hackett (Reference Hackett2014) suggests that surveys on religion are prone to definitional problems and responses may be subject to social desirability bias. The mixed results may demonstrate a problem in defining indicators of religiosity and religious identity.

While individuals may vary in their reasons for behaving religiously, it is also known that religious behavior adds new dimensions to the political values of voters. De La O and Rodden (Reference De La, Ana and Rodden2008) find evidence that suggests religious identity adds a moral values dimension to political ideology. Religious institutions promote moral rules which can motivate religious voters to support conservative parties. It may be the case that religious individuals vote for more conservative parties in general as a result—leading to a rise in support for both traditional conservative but also far-right movements across Europe.

In summary, religious behavior and belief may play a role in growing support for far-right parties, but measurement issues of individual-level “religiosity” can obfuscate straightforward conclusions. It may be the case that religious individuals, driven by an additional set of moral policy commitments, are more supportive of conservative parties or parties that claim to defend “traditional values,” including RRPs. The literature on the far-right also presents an alternative theory which suggests that religion is not a driving factor in far-right support. Instead, the far-right draws on nativist or anti-immigrant attitudes.

Nativism versus religion: support for Le Pen where Muslims are present but few

It may be the case that religious individuals are more likely to support the far-right because religious voters lean conservative. Another possibility is that religious identity provides an avenue to promote nativist policies while simultaneously appealing to historically democratic values, such as laïcité.

A prominent proponent of the latter argument, Brubaker (Reference Brubaker2017), argues that nationalists use religious symbols to frame international politics as a clash of civilizations where the far-right fights for “Christian civilization.” Molle (Reference Molle2019) finds that religious cultural symbolism affects RRP vote patterns despite the lack of religiosity in Europe. This suggests that RRPs can appeal to religious sentiment to win votes, even if surveys indicate a decline in religious belief. Schwörer and Romero-Vidal (Reference Schwörer and Romero-Vidal2020) find that RRPs use religious language primarily to suggest that Muslim immigrants present a threat to the European in-group defined by “Christian heritage.” These findings offer evidence that the importance of religion itself may be incidental to the politics of RRPs. In the religion-as-civilization framework, far-right nationalist groups that try to signal their religious commitments are not doing so to capture true believers but rather true patriots. In this way, the signaling of Christian culture takes on an ethno-religious, nationalist bent as opposed to a signal that the party represents piety.

There is support for the theory that the RRPs support is related to immigration. In the 1980s, with rising immigration and growing economic inequality, RRPs emerged that promoted cultural nationalism, and ethno-centrist policies (Rydgren, Reference Rydgren2007). Votes for RRPs are higher in states with higher immigration numbers (Knigge, Reference Knigge1998; Lubbers et al., Reference Lubbers, Gijsberts and Scheepers2002; Swank and Betz, Reference Swank and Betz2003; Van der Brug et al. Reference Van der Brug, Fennema and Tillie2005). These national-level findings, however, have the potential to lead researchers down an ecological fallacy, where sub-national votes do not correspond with similar findings (Rydgren, Reference Rydgren2007). It is also unclear what the mechanism for RRP support is if the primary issue for voters is immigration. It could be related to perceived economic disparities between members of the middle class as a result of modernization (Gurr, Reference Gurr2015), a loss of identity as a result of globalization (Koopmans and Statham, Reference Koopmans and Statham1999), or protest against perceived failure of mainstream political parties (Lubbers et al., Reference Lubbers, Gijsberts and Scheepers2002; Norris, Reference Norris2005). The narrative of civilizational nationalism often defines non-nationalists as a threat. As the literature on laïcité suggests, FN may be making a concerted effort to re-frame immigration issues precisely in this way. By turning the conversation into one about defending France's secular culture from religious extremism, the party is able to maintain its general policy position while avoiding explicit discussion of ethnocentric or nativist topics.

In parts of Europe that are already more religiously diverse, such as large urban and commercial centers, it is likely that general support for RRPs will be lower as these areas tend to have a longer history of and exposure to immigration (Warf and Winsberg, Reference Warf and Winsberg2008). It is also likely that RRPs know this and generally target specific areas where long-term integration has not occurred (Ivaldi and Dutozia, Reference Ivaldi and Dutozia2018). While some parts of Europe have grown increasingly urban and globalized, it is also true that this change has not been uniform. Fitzgerald (Reference Fitzgerald2018), looking at the 2002 election campaign of Marine Le Pen's father, Jean-Marie Le Pen, notes similar findings about the appeal of RRPs to Brubaker but refined the analysis at a more local level. While the debates about globalization and civilization are international in nature, Fitzgerald suggests that support for RRPs has more to do with local politics. Specifically, European support for RRPs is higher in homogenous, rural towns that have not seen growth from globalization or that are economically declining. Fitzgerald demonstrates that rootedness to a town, a measure created by indexing long-term homeownership and residential zoning, has a positive association with far-right voting. The localized version of the civilization theory suggests that it is the appearance of losing to forces of globalization that drives local support for RRPs. We combine Brubaker and Fitzgerald's theories to gain a more nuanced understanding as to the relationship between religious in-out groups in elections at the local level. We suggest that local demographics, framed as “civilizational” issues, increase support for RRPs.

Recent work on political parties and political geography supports the theory of local change framed as a greater threat leading to increased RRP support. Hafez (Reference Hafez2014) and Hafez and Heinisch (Reference Hafez and Heinisch2018) find that RRPs have recently sought more mainstream support across Europe by relying on an “anti-Muslim” political agenda. Specifically, they argue that a shared support for Islamaphobic policies has connected disparate RRPs across Europe. Gravelle et al. (Reference Gravelle, Medeiros and Nai2021) examine the relationship between Islamic presence and support for a Dutch RRP, finding that visible diversity, measured by mosques with minarets, is associated with greater support for the RRP among right-leaning voters.

Recent qualitative work conducted in France further supports the civilizational-nationalist theory. Weitzel (Reference Weitzel2020) finds that even the sound of mosques playing prayer music can lead to perceptions of out-group differences for French Muslims. Beaman (Reference Beaman2017) finds that despite French law and society not recognizing ethnicity and race as valid demographic categories, North African immigrants face marginalization along racial lines in France. Furthermore, this marginalization is spatial, tied to specific, local areas where the “otherness” of non-white residents is more apparent. Following this line of reasoning, we suggest that another measure of diversity is the number of religious buildings in a town. Muslim religious centers help approximate the size or presence of Muslim communities as well as highlight areas where Muslim culture is visible in a way that mirrors Weitzel's perceived outgroup theory of sound in French towns. Areas where a small number of mosques exist could, similar to Weitzel's study, be spaces which are culturally contested, prompting reactionary engagement with the far-right.

Under this theory, religion is not a primary factor in driving RRP support—but it may offer an instrumental means for RRPs to gain support without sounding anti-immigrant. Particularly, in the case of the FN, the party has recently started to frame anti-Islamic policy positions and statements as a defense of laïcité.

French secularism and far-right framing of ethnocentrism

While France is a historically Catholic country, the modern state is secular (Dargent, Reference Dargent2019). The French conception of state secularism, laïcité, is rooted in the establishment of separation between the Catholic Church and French rulers who saw each other as threats to political hegemony over the state (Baubérot, Reference Baubérot1998; Belorgey, Reference Belorgey2006). Over time, however, the definition and political salience of laïcité has changed as France faced issues related to religious education and immigration (Baubérot, Reference Baubérot2016; Davis, Reference Davis2020).

The earliest successful, modern RRP was FN, however, it was not until the 2000s that the party became a political contender at the national level (Mudde, Reference Mudde2002, Reference Mudde2013; Rydgren, Reference Rydgren2005, Reference Rydgren2007; von Beyme, Reference von Beyme2013). Beginning in the 2000s, FN, particularly under the leadership of Marine Le Pen, began to promote laïcité. Unlike previous iterations of the concept, which were originally related to Catholic overreach into state politics, FN framed laïcité around Islam and its incompatibility with the ideals of French Republicanism (Almeida, Reference Almeida2017). Marine Le Pen would eventually call for a separation of “mosque and state,” as well as restrictions on immigration (Larquier, Reference Larquier2011).

This is further supported by scholarly work on the FN and the speeches of Le Pen that suggest her leadership has led to a change of the party lingo from an overt Catholic integralist stance to a more secular but anti-Islamic position (Alduy and Wahnich, Reference Alduy and Wahnich2015; Fieschi, Reference Fieschi2020). Thus, the connection between religion and RRP support may be less about Catholic religiosity and more about anti-Islamism. This reframing as to what laïcité means in contemporary French politics ultimately provides an avenue for promoting nativist policy positions while claiming to uphold “traditional” French values. The next section outlines why we have chosen to examine the French 2017 election as a test case for studying whether far-right support is related to the religious or nativist theory of RRPs.

A most-likely crucial case for religious votes

While single-case approaches to political analysis face the challenge of generalizability, we suggest that, per Seawright and Gerring (Reference Gerring, Box-Steffensmeier, Brady and Collier2008), the 2017 election was a most-likely crucial case where an RRP could attract religious voters. Despite the challenge of generalizability, it is possible that the theory of this study is applicable in either non-majority Christian contexts or other regions outside of Western Europe. Research by Jamal (Reference Jamal2005, Reference Jamal2007), Livny (Reference Livny2020), and Mecham (Reference Mecham2017) suggest similar theories are applicable in Islamic contexts, with evidence suggesting that Islamic electoral advantage is related to in-group trust rather than overt religiosity. As a most-likely crucial case, we believe France's 2017 election is one in which the response variable in question is expected to return a specific result (Przeworski and Teune, Reference Przeworski and Teune1970; George et al., Reference George, Bennett and Bennett2005; Gerring Reference Gerring, Box-Steffensmeier, Brady and Collier2008). We suggest that this is a most-likely case for two main reasons. First, the state has an increasingly popular national RRP, The FN, which has steadily increased support in both local and national elections for the past twenty years (Auberger, Reference Auberger2008; Stockemer, Reference Stockemer2014, Reference Stockemer2019; Stockemer and Amengay, Reference Stockemer and Amengay2015; Zaretsky, Reference Zaretsky2016; Almeida, Reference Almeida2017; Buhr, Reference Buhr2017; Dumitrescu, Reference Dumitrescu2017; Bastow, Reference Bastow2018).Footnote ³ Second, the 2017 election was unique for France because, due to scandals involving the major center-right candidate, Francois Fillon, the second round of the election came down to a newcomer party and FN. This scenario left Catholic voters, who were largely mobilized under Fillon, an avowed conservative Catholic, with no clear party choice.

The 2017 French presidential election

French presidential elections are conducted in two rounds. The first round is open to many parties but if there is no single candidate with the majority of the votes then the second round occurs two weeks after the first. In the second round, the top two candidates from the first round compete in a run-off.

France has a presidential–parliamentary system with a number of parties across the political spectrum. In general, the two most powerful parties have been Les Republicains (The Republicans), a moderate, conservative party, and La Partie Socialiste (The Socialist Party or PS).Footnote ⁴ In the 2017 election, the Republican candidate was Francois Fillon, former Prime Minister as well as the Deputy Mayor of Paris. He appealed to conservative Catholic voters because of his conservative social policies (Beardsley, Reference Beardsley2017).

The typical PS-Republicains fight did not occur as usual. The leftist vote was split as former PS member Jon-Luc Melenchon created a new party called La France Insoumise (France Unbowed). This split among the leftist elite led to a near collapse of the traditional left parties and a loss in the first round (Clift and McDaniel, Reference Clift and McDaniel2017).

Additionally, Emmanuel Macron began gaining political momentum with another new party En Marche!. Macron's centrist policies focused on renewing and improving European integration and promoting populist sentiment focused on modernizing French politics (Kuhn, Reference Kuhn2017). Fillon was also facing a major scandal regarding accusations of misuse of public funds. It is clear that these scandals had an effect on the likelihood of his winning (Lees, Reference Lees2017). Fillon ultimately lost, as did Melenchon to the newcomer, Macron.

This election is an ideal case for studying the relationship between religious affiliation on RRP vote share. The two candidates in the second round did not have clear connections to the existing religious electorate. One option was Emmanuel Macron, a non-religious candidate. On the other side was Le Pen, running a nationalist party with a history of Catholic fundamentalism. It was unclear at this time whether Catholics would move to the more conservative candidate with controversial opinions regarding immigration or toward the centrist candidate. In the following section we outline how we combined both voting data and geographic data to help elucidate voting patterns among those who ultimately supported FN.

Number of religious buildings per commune

We use religious VGI as a means of measuring the visibility and presence of religion at the commune level in France. Religious buildings, such as mosques, prayer houses, and churches, exist to support the spiritual needs of a religious community. It follows that areas with more religious buildings for a given denomination have a stronger social, cultural, or historical affiliation with that denomination. Comparatively, areas with no or few religious buildings belonging to a given denomination have less of an overt, visible affiliation. Similar assumptions about the presence of religious buildings are used in examining the link between religion and social outcomes, with recent work arguing that local religious institutions provide public goods and information bridges (Cao et al., Reference Cao, Duan, Liu and Wei2018). While gazetteers of religious buildings are not always available, many religious communities in the digital era rely on online tools for believers to connect with local religious centers. These websites are designed to accurately provide religious individuals with information as to where they can visit, worship, and connect with the local religious community. We leverage the publicly available information on these websites to create novel datasets on the frequency and location of Catholic and Islamic religious buildings across France.

The reason why we collected data on both Catholic buildings and mosques is to examine not only if greater Islamic presence in a commune increases RRP support, but if Catholic presence does as well. If it is true that FN uses Catholic cultural symbols to appeal to religious voters, then we should expect areas with a greater number of Catholic buildings to at least have a non-negative relationship with FN vote share.

Hypotheses

If the theory of religion and support for RRPs is correct, then we can expect that areas which have a higher number of Catholics will have more support for far-right parties.

It is possible that the number of Catholic buildings does not directly correspond with the number of religious voters in a commune. This is partially because Catholic buildings in France are not always active houses of worship. They often represent a cultural and historical presence of Catholicism at the local level but may exist as local, historical monuments rather than active religious buildings.

Alternatively, if it is the case that support for an RRP like FN is related to the anti-immigration/civilizational perspective presented by Brubaker then we should expect that smaller towns with some minority presence should see higher support for FN. In our operationalization, we use the number of religious buildings in a town as an approximate estimate of this cultural presence of Islam. This leads us to our second hypothesis:

We next explain the data gathered to evaluate these hypotheses.

Collecting a dataset of Catholic buildings and mosques across France

This study leverages VGI and web-scraping techniques to build a novel dataset representing Catholic buildings and mosques across the entire nation of France. Information on Catholic buildings was collected from The Catholic Directory (catholicdirectory.com), a website of community-contributed information on Catholic churches, organizations, and businesses worldwide. We scraped The Catholic Directory for religious buildings, which can include churches, abbeys, chapels, and more. Similarly, we turned to another community-contributed website called Trouve ta Mosquée (trouvetamosquee.fr) to gather information exclusively on mosques. The information present on these websites is user-generated, volunteered content generated by members of Catholic and Muslim religious communities to serve as a utility for individuals to find local places for worship (Goodchild, Reference Goodchild2007). As such, we suggest that the data presented in these collections is accurate as it was created to provide information for members of the communities themselves.Footnote ⁵ Both catholicdirectory.com and trouvetamosquee.fr are regularly updated and represent large, free, online repositories of data on Catholic buildings globally and mosques in France. In particular, the latter has been used in several recent publications examining Islam in France (Ivaldi and Dutozia, Reference Ivaldi and Dutozia2018; Drouhot, Reference Drouhot2020).

From these websites we collected the name and address of each building located within the continental borders of France. The address of each building was then geocoded using the Google Maps API to obtain longitude and latitude. We then counted the number of Catholic buildings and mosques, separately, located in each commune. At the time of web-scraping (July 2021), we were able to collect, locate, and map 2445 mosques and 9478 Catholic buildings across France (Figure 1).

Figure 1. Point locations of Catholic (orange) and Islamic (green) buildings across the nation of France. Gray lines indicate department boundaries. Data on Catholic buildings from The Catholic Directory (catholicdirectory.com) and data on mosques from Trouve ta Mosquée (trouvetamosquee.fr)

Response variables: commune election data

Data on French elections are publicly available from the Ministère de l'Intérieur and the open platform for French public data (data.gouv.fr). We downloaded the final results for the first and second Wave of the 2017 French presidential election. For each commune we calculated the percent of votes cast for candidates of interest as the number of votes for a given candidate over the total number of votes expressed. We restrict the analysis to communes located on the continental mainland of Europe, thus excluding overseas regions and Corsica. This results in n = 34,922 communes with election data. Communes can vary in size but are, in general, small geographical regions, with some having populations smaller than 100 individuals.

Control variables

We rely on recent work by Fitzgerald (Reference Fitzgerald2018) to identify control variables relevant to local voting in France. Fitzgerald proposes a number of variables including total population, foreign population (%), unemployment (%), low education (%), high education (%), employment in large professional sectors (agricultural, artisan, high skilled, and intermediary, white collar, and manual labor), and a multidimensional index of local rootedness. The rootedness index is a measure of home ownership, residential zoning, and homes under the same owner since 1990. Fitzgerald provides detailed information on the construction of these variables from the 1999 French census. We identified equivalent variables from the 2017 French census and downloaded publicly available tables from Institut national de la statistique et des études économiques (insee.fr). A detailed sourcing of the census tables and variable construction notes is available in the Supplementary Materials.

As a final step we reduced the combined dataset to instances where data is available across all election and control variables. There were n = 731 communes that did not have demographic data reported for one or more of the demographic control variables, and these removed communes were distributed across the whole of France. In addition, there were n = 67 communes which appeared to have erroneous data (e.g., the range of values exceeded 100%). The final dataset included n = 34,124 communes for analysis, encompassing 95.5% of all possible communes that reported election results in the 2017 French presidential election.

Analysis

Table 1 provides a descriptive overview of the primary response and explanatory variables at the commune-level. The mean number of mosques per commune was 0.07, which was lower than the mean number of Catholic buildings per commune at 0.27. In addition, the maximum number of Catholic buildings in a given commune far exceeded the maximum number of mosques in a given commune. Specifically, we observed the maximum of 79 Catholic buildings in Marcoux, an alpine commune located nearby to the departmental bishopric of Digne. In comparison, we observed the maximum of 26 mosques in Strasbourg, a city that is home to one of the largest mosques in France.

Table 1. Descriptive statistics of 2017 French election data

Moving beyond the counts of religious buildings, we also draw attention to certain commune level control variables. The Fitzgerald multivariate measure of rootedness indicates that the average level of rootedness across communes was 53.16% (std. dev. 6.46). The employment sectors of intermediate, white collar, and manual labor had the highest occupational categories. Lastly, considering education, we observe that communes reported larger proportions of high education (41.68%, std. dev. 9.62) compared to low education (22.75% std. dev. 7.35).

In the first round of voting Le Pen had the highest average commune-level vote share at 26.42%. Macron and Fillon had similar levels of average commune-level vote share at 20.41 and 19.87%, respectively. However, as we expect, Le Pen's highest average does not hold in the second round of voting. Across all communes Le Pen secured, on average, 42.18% (std. dev. 11.93) of the vote, whereas Macron attained a higher average of 57.82% (std. dev. 11.93).

The relationship between religious buildings and election results

Table 2 describes the regression estimates of the fully adjusted models for Wave 1 of the election. Of particular interest are the coefficients relating to the count of Catholic buildings and the count of mosques. We observed a negative association between the number of Catholic buildings and Le Pen vote share in Wave 1. Each additional Catholic building was associated with a −0.33% (SE: 0.04) decrease in Wave 1 Le Pen vote share. This observation is reversed for Macron, wherein each additional Catholic building in a commune was associated with a 0.12% (SE: 0.03) increase in his Wave 1 vote share. Comparatively, these associations change direction when considering the relationship between mosques and election results. Each additional mosque in a commune was associated with an increase in Le Pen vote share by 0.39% (SE: 0.13) and a decrease in Macron vote share by −0.73% (SE: 0.09). The associations observed for Fillon largely reflect Macron, although to a lesser magnitude. We observe a slight positive association with Catholic buildings for Fillon (coefficient 0.09, SE: 0.04) and a moderate negative association with mosques (coefficient −0.57, SE: 0.12).

Table 2. OLS regression results on the relationship between election outcomes and presence of religious buildings for Wave 1 (preliminary) of the 2017 French Presidential election, adjusted for commune-level controls

*p < 0.1, ** p < 0.05, *** p < 0.01.

We can make several observations on the association of control variables with each candidate. There was a strong negative association with commune vote share for Le Pen and areas with higher levels of immigration, high education, and areas with higher rootedness scores, but areas with higher levels of unemployment demonstrated a strong positive association. Comparatively, there were moderate to strong positive associations with commune vote share for Macron and areas with areas with higher population and education levels, although a strong negative association for areas with higher levels of unemployment. Model fit was strongest for Le Pen with an adjusted R ² of 0.256, whereas model fit for Macron and Fillon were lower at 0.147 and 0.121, respectively.

Because Wave 2 of the presidential election contains only the leading candidates of Le Pen and Macron, a commune's expressed vote share will always be the remainder of the other candidate. For example, if a given commune had 60% of the votes go to Macron, the remaining 40% must go to Le Pen. Thus, the regression point estimates for one candidate will be equal and reverse to the other. We present the regression estimates for both candidates for clarity in Table 3. These models use the same specification described previously. We observe similar associations in magnitude and direction for Wave 2 as observed in Wave 1. Catholic buildings are negatively associated with Le Pen vote share (coefficient −0.41%, SE: 0.05) while mosques are positively associated with Le Pen vote share (coefficient 0.48%, SE: 0.17). Wave 2 model fit has an adjusted R ² of 0.252, similar to the Wave 1 models. We present a visual representation of the coefficient estimates and 95% confidence intervals in Figure 2.

Table 3. OLS regression results on the relationship between election outcomes and presence of religious buildings for Wave 2 (final) of the 2017 French Presidential election, adjusted for commune-level controls

*p < 0.1, ** p < 0.05, *** p < 0.01.

Figure 2. Coefficient plot for Wave 2 OLS models estimating candidate vote share. The Intercept and Total Population (per 100,000) coefficients were omitted from the plot as their large values obscured variation between the remaining coefficients

Of the coefficient estimates, the direction and strength of the number of mosques with candidate vote share appeared counter-intuitive because Macron's electoral success in metropolitan areas. In France, cities and metropolitan areas contain higher numbers of mosques, as seen in Figure 1. To provide further insight into the association between religious buildings on vote share, we examined how the estimated Le Pen vote share varies for communes with differing numbers of mosques. Figure 3 shows the average fitted value, 5th percentile of the fitted value and 95th percentile of the fitted value for Le Pen vote share for communes with different numbers of mosques. Other explanatory variables were held at their observed values in the data. Communes with zero or very few mosques tend to have the largest estimated Le Pen vote share. However, this vote share slowly decreases toward zero as the number of mosques in commune increases, with the estimated vote share plateauing for communes with eight or more mosques. The coefficient estimate, which is the average effect holding al else equal, obfuscates this nuance. We contextualize this finding in the discussion section. In general, we observe that areas with more mosques tend to have lower or negligible estimated Le Pen vote share, although there is lower precision for the few communes with more than 15 mosques.Footnote ⁶

Figure 3. Average fitted value for commune-level Le Pen vote share in Wave 2 of the 2017 French presidential election, by the number of mosques in a commune. Fitted values were gathered from fully adjusted regression models, holding other explanatory variables at their observed values.

Sensitivity analysis #1: consideration of Gironde

Although Catholic buildings and mosques are distributed across all of France, Gironde has a unique set of circumstances that requires special consideration. Gironde is the seat of the Roman Catholic Archdiocese of Bordeaux and has the highest number of Catholic buildings among all departments. In addition, Gironde was a strong area of support for Jean Marie Le Pen and FN in the 2002 presidential election. For these reasons we follow Fitzgerald and repeat our analysis on the communes comprising Gironde.

We present the key coefficients of interest for Gironde communes compared to communes across France in Table 4. 534 communes were analyzed, which represents 98.5% of the total communes in Gironde. Comparing these Gironde-specific coefficients to those estimated for all communes of France reveals several nuances. Firstly, the negative association between Le Pen Wave 1 vote share and the number of Catholic buildings changes direction, although the standard errors are much wider and the point estimate is no longer significant. A similar association reversal is observed for Macron. The positive association between Macron Wave 1 vote share and the number of Catholic buildings changes direction, again with wider standard errors. The wider standard errors are likely due to the dramatic reduction in the number of observations considered in the Gironde models (n = 34,124 for all communes of France compared to n = 534 for all communes of Gironde). These reversals for Catholic buildings are more pronounced in the Wave 2 results. However, the Gironde estimates for the number of mosques in a commune have the same direction as the estimates calculated from all communes of France, although the point estimates are smaller in magnitude and have wider standard errors. Le Pen still has a positive association for vote share and the number of mosques, whereas Macron has a negative association.

Table 4. Sensitivity analysis of OLS regression for the association of religious buildings on election share outcomes for the French Presidential election

Models are compared between all communes of France and all communes of Gironde.

Sensitivity analysis #2: consideration of sensitive urban zones (zone urbaine sensible, ZUS)

The analysis of mosque numbers to vote-share thus far has assumed that out-groups living in France would not vote for the FN. However, out-group counterintuitive votes are possible and, in the absence of individual-level voter surveys, may be discernable in local geographic units. A proxy measure for out-group presence at the commune level is immigration. A strong, positive relationship between votes for FN and immigrant composition would challenge our assumptions about voting behavior in France and suggest alternative explanations for the present results. To test the assumption that Muslims and, more broadly, immigrants living in France did not vote for the FN, we conduct an additional sensitivity analysis examining the relationship between votes expressed for Le Pen and immigrant composition. This comparison is made amongst the sensitive urban zones (zone urbaine sensible or ZUS) of France. These areas are a common destination for immigrants coming to France and are profiled as relating to urban terror, although this association has been critiqued (Body-Gendrot, Reference Body-Gendrot2008). While our general models show a negative relationship between immigrant levels and support for FN, we examine this subset of areas as a sensitivity analysis.

A complete atlas of the ZUS is provided by the Systeme d'Information Geographique de la Politique de la ville Footnote ⁷. ZUS are defined at the geographic level of quartier. No comprehensive crosswalk linking quartiers to communes was identifiable from online sources. Therefore, we linked the ZUS atlas to the nationwide election dataset based on department codes and commune names. This method successfully linked n = 480 of the n = 715 mainland ZUS (67% of mainland ZUS or 64% of all ZUS) to the nationwide election dataset. Although we were not able to link all ZUS, tabulations of demographic characteristics of the linked ZUS indicate that the sample is consistent with INSEE reportsFootnote ⁸. For example, the communes containing ZUS in our sample have a high proportion of immigrant residents (mean 15.44%, 95% CI 14.68–16.21) and unemployment (mean 13.38%, 95% CI 13.14–13.63). These levels were much higher than the proportion of immigrant residents (mean 4.21%, 95% CI 4.16–4.26) and unemployment (mean 8.12%, 95% CI 8.09–8.16) for the rest of France.

We begin the sensitivity analysis by exploring the association between immigrant composition and votes expressed for Le Pen in Wave 2 of the 2017 French presidential election amongst communes containing a ZUS (Figure 4). A negative correlation is visible wherein communes with a lower proportion of immigrant residents have a higher proportion of votes expressed for Le Pen and, conversely, areas with a higher proportion of immigrant residents have a lower proportion of votes expressed for Le Pen (Pearson's R of −0.41).

Figure 4. Scatterplot of commune-level proportion of votes expressed for Le Pen in Wave 2 of the 2017 French presidential election and commune-level proportion of immigrant residents. Each point represents one sensitive urban zones (zone urbaine sensible, ZUS). n = 480 ZUS are shown in total, representing 67% of the n = 715 ZUS located in mainland France. Not all ZUS could be included in the analysis due to a lack of comprehensive crosswalk available between ZUS and Communes

We present the key coefficients of interest for communes containing ZUS across mainland France in Table 5. The associations between the number of religious buildings and candidate vote share amongst this sub-sample of the data are consistent with the main findings. A higher number of Catholic buildings are negatively associated with Le Pen vote share, while a higher number of mosques are positively associated with Le Pen vote share. While this relationship could suggest that Muslims are counter-intuitively voting for FN, the correlation between immigration and FN support in Figure 4 suggests this is not likely. Precedence for counter-intuitive voting has been observed in other contexts (Alamillo, Reference Alamillo2019). Nevertheless, considering sub-samples of the national dataset offers a practical method to test conceptual assumptions and model consistency.

Table 5. Sensitivity analysis of OLS regression for the association of religious buildings on election share outcomes for the French Presidential election

Models are compared between all communes of France and n = 480 sensitive urban zone (SUZ) communes across France.

Limitations

Although VGI has been identified as a promising digital innovation, VGI has also been subject to several criticisms (Sui et al., Reference Sui, Elwood and Goodchild2012). One primary criticism arises from the challenges of validating massive amounts of user-contributed content (Fonte et al., Reference Fonte2015; Basiri et al., Reference Basiri, Haklay, Foody and Mooney2019). A potential consequence of this limitation could be an overestimation of the number of religious buildings present in a commune, and subsequent point estimates for the number of Catholic buildings and mosques may be exaggerated. Relatedly, there exists the possibility of underestimation for religious buildings not reported in the online databases or incorrectly geocoded. Other unofficial sources claim that France has more than 70,000 religious buildings, including more than 45,000 Catholic buildings.Footnote ¹⁰ If such figures are true, it is likely that our VGI collection efforts underestimated the number of religious buildings in France for both Catholicism and Islam.

In addition, we assume that the presence of religious buildings is indicative of the importance of religion in an area. However, such a method of measuring religiosity remains an ecological proxy (Shelton et al., Reference Shelton, Zook and Graham2012). We do not weigh the impact of each building with information on congregation size, number of tithings gathered, years of activity, or another institutional measure that might capture levels of religiosity (Zwingmann et al., Reference Zwingmann, Klein and Büssing2011). This additional information was not available for all buildings, would require additional validation, and is likely to vary over time. In addition, there may be substantive differences between Catholic and Islamic buildings in France. For example, mosques are likely to be built more recently compared to Catholic buildings, and, as such, counts of mosques may reflect the prominence of an immigrant community over anything else. We attempt to account for this by adjusting models for both immigration, nationality, and rootedness.

The present study was designed as a cross-sectional analysis. We cannot establish causality between the explanatory and response variables, nor can we comment as to how the association between these variables changes over time. The primary challenge in extending the current study across time is the collection of information on religious buildings across France. Gathering spatio-temporal religious information is of paramount importance to enable future research. For example, resources could be allocated for a building survey that would gather the founding date as well as the previously mentioned, non-identifiable measures of community religiosity. This data could be used in conjunction with individual-level survey data to establish a stronger causal connection while controlling for higher level factors.

Lastly, we did not consider all possible minority religions in France. Recent estimates suggest that these non-Islam minority religious groups make up the smallest proportions of the French population.Footnote ¹¹ However, consideration of these other minority religious groups is important. Historically, FN candidates have drawn intense backlash over discriminatory remarks made toward Judaism, and several scholars have argued that the emergence of Islamophobia alongside existing anti-Semitism reinforces ethnocentric fears of the “other” in Western European states (Bangstad and Bunzl, Reference Bangstad and Bunzl2010; Mayer, Reference Mayer, Salzborn, Davidov and Reinecke2012). Our results are most relevant to discussions of Catholicism and Islam in France, and the direction of the observed associations may change direction or not hold when considering other minority religions.