Religious Voting in a Secularised Country: Evidence From Chile’s 2022 Constitutional Referendum

Introduction

Political science has extensively discussed the relevance of social cleavages in explaining the evolution of party systems and voter voting intentions. For some, religious cleavage is no longer relevant (Lane & Ersson 1997; Dalton 2002, 2013; Franklin et al. 2009; Goldberg 2020). For others, however, this cleavage persists as a predictor of voting intention (Brooks & Manza 2004, Olson & Green 2006, Elff 2007, Norris & Inglehart 2011). This helps explain the so-called “conservative backlash” against the gender agenda (Biroli & Caminotti 2020). This discussion is typically based on polls or election results with competing candidates. These candidates summarise political information and facilitate voters’ decision-making, but what happens when voters are confronted with programmatic choices rather than candidates? A constitutional referendum is an excellent opportunity to assess the effect of religion on voting intentions, especially if it involves redesigning a country’s political, economic, and cultural system, pitting conservative and progressive options against each other. In simple terms, a constitutional referendum is a critical election (Key, 1955). Is it possible that religion could become a strong predictor of voting intention under these conditions, competing even with theoretically more robust variables, such as identification on the left-right axis or voter income level?

The cleavage theory is very broad. Lipset and Rokkan (1967), in their classic study of the formation and evolution of party systems, inaugurated an approach that is still current in political science. Part of this approach has been very useful for studying the institutionalisation of party systems, especially in Latin America. The contributions of Mainwaring and Scully (1995), Coppedge (1998), Kitschelt et al. (2010), and Roberts (2013) stand out. This literature focuses on identifying historical conjunctures, institutional variations, and the evolution of the explanatory power of social cleavage as drivers of political change. Regarding the specific analysis of religious cleavages, examples include Boas (2023) and Smith and Boas (2023) in Latin America, Valenzuela et al. (2007), Boas (2014), Bargsted and De la Cerda (2019), Raymond (2021) in Chile, Boas (2016), Layton et al. (2021), Azevedo and Robertson (2022) in Brazil, and Pally (2020) in the United States, analysing the emergence and consolidation of a radical and populist right linked to evangelical religion.

In this study, I measure the effect of religion on voting intention in a constitutional referendum in the context of Chile as a secularising country (Valenzuela et al. 2007; Bargsted & De la Cerda 2019) and where the religious cleavage, as Raymond (2021) shows, has an intermittent effect on voting intention in presidential elections between 1999 and 2017. According to surveys by the Centro de Estudios Públicos, the percentage of Catholics decreased from 72.1% in June 1998 to 43.8% in December 2022, while Evangelicals, without significant changes in this time series, went from 16.1% to 17% in the same period. Trust in the Catholic Church fell from 50% in November 2008 to 21% in December 2022. The decline was more pronounced among young people and exacerbated after cases of sexual abuse by priests became known in the mid-2000s. (Claret 2020). On the other hand, trust in the Evangelical churches declined to a lesser extent, from 32% to 22% in the same period. This leads to questions about the impact of religion in a critical election in which candidates are not competing but instead political options that represent continuity or constitutional change.

Chile held a constitutional referendum to approve or reject a new constitution (see Morales 2020, 2021). The approval option reached 38.1%, and the reject option reached 61.9%. As shown below, religion had two significant effects. First, it was a predictor of voting intention. Second, it was a predictor of support for the five programmatic pillars of the new constitutional text: plurinationality, gender parity, elimination of the senate, abortion, and the creation of a public National Health System. The first effect, on the intention to vote for Approve or Reject, allows us to consolidate the idea that religion is a relevant variable in explaining electoral results. The second effect, on support for the contents of the new constitutional text, allowed us to understand the reasons that led evangelicals to reject the constitutional text to a greater extent, especially the inclusion of an article regulating the voluntary interruption of pregnancy (abortion). Consequently, this critical election helps to identify why and how religious cleavage is essential, even in a secularising country.

This study makes three contributions to the literature. First, I analyse a case of constitutional change and legitimisation by referendum in Latin America that is very different from that implemented in other countries (González-Ocantos & Meléndez 2023). For example, in Peru (1993), Venezuela (1999), Ecuador (2008), and Bolivia (2009), constitutional change was driven by a leader or caudillo. By contrast, in Chile, the constitutional process was an elite response to the social crisis 2019. Second, this study links the issue preferences of some key constitutional clauses with socio-demographic variables. In this case, religion. This approach is uncommon in analysing constitutional referendums, since Latin America’s literature concentrates on the institutional designs promoted by caudillos or leaders, especially on presidential re-election. Third, my study complements the recent literature that explains the outcome of a constitutional referendum, mainly in terms of the degree of popularity of presidents (Alemán & Navia, 2023). While I recognise the importance of presidential approval as a predictor of the Chilean referendum vote in 2022, I suggest that the effect of socio-demographic variables such as religion persists, especially in explaining support for some constitutional content such as abortion. On this point, I complement the work of González-Ocantos and Méléndez (2023), who, although they emphasise the relevance of second-order considerations, such as presidential popularity, over issue preferences, suggest that some key constitutional clauses generated polarisation and that this polarisation was not total but asymmetric. In other words, it was they were concentrated in certain groups. Finally, I relate support for constitutional clauses (including abortion) to socio-demographic variables (such as religion) associated with cleavage theory.

The Puzzle and the Hypotheses

The effect of religion on voting intentions and political preferences has been studied extensively. From the seminal work of Lazarsfeld et al. (1944), through the classic study by Lipset and Rokkan (1967), we know that religious preferences are linked to voting intentions and that these preferences preceded the formation of party systems based on the axis of clerical-anticlerical or church-state competition. This was the case in the European and Latin American party systems in the 19th century. For some, this religious cleavage has lost strength as a result of secularisation (Lane & Ersson 1997, Dalton 2002, 2013; Franklin et al. 2009; Goldberg 2020), being relegated to presumably more relevant cleavages such as class or left-right ideology, among others. However, this conclusion is debatable as religion remains a robust predictor of political preferences, at least in some European democracies (Brooks & Manza 2004; Olson & Green 2006; Elff 2007; Norris & Inglehart 2011). This is even more evident in the constitutional change process. Following Cohen and Laborde (2015) and Bâli and Lerner (2017), constitutional processes in Europe, Asia and Africa have been strongly shaped by religious tensions or conflicts. However, little literature has been devoted to this issue. That is, as Bâli and Lerner (2017, 2) explicitly point out, “the challenge of democratic constitution-drafting under conditions of disagreement over the state’s religious or secular identity.”

More specific studies show that religion is an essential predictor of voting intention in secularised and non-secularised countries. Among others, the analysis of Catholicism, Protestantism, and support for the Nationalsozialistische Deutsche Arbeiterpartei (NSDAP) in Germany (Spenkuch & Tillmann 2018), the effect of religion on support for abortion in the Irish referendum in 2018 (Elkink et al. 2020), the impact of religious education on British voting intention in the European Union referendum of 2016 (Kolpinskaya & Fox 2019; Huang 2020), the incidence of candidates’ religious stereotypes on voters’ voting intention in the United States (McDermott 2009), Brazil (Boas 2014) and Chile (Boas 2016). More recently, as we announced, the literature associates people’s religious preferences with their intention to vote for populist candidates (DeHanas & Shterin 2018; Siles et al. 2021) and other literature on the political behaviour of evangelicals in the United States (Grose 2018, Allen & Olson 2022). Some of these studies were based on the concept of the politicisation of religion. Following Lindberg (2014:567), this implies that “religion, once thought of as a provider of societal cohesion, now becomes a site of difference, contestation, and sometimes also of conflict”. As Foret (2023) underlines, the politicisation of religion often leads to fragmentation, not necessarily polarisation. The latter, however, could occur particularly when radical right-wing or populist right-wing parties confront the governing parties as opposition, defending conservative ideas that reactivate the religious cleavage (Morales & Pérez 2024).

From Latin America, meanwhile, the literature has focused on the effect of the Catholic religion on the emergence and consolidation of religious parties, predominantly Christian Democratic parties (Mainwaring & Scully 2003), highlighting the cases of Mexico (Mantilla 2016) and Chile (Morales & Poveda 2007, Herrera et al. 2019, 2023). In parallel, more specific studies on the electoral behaviour of evangelicals have emerged, with the work of Campos (2012), Boas (2014, 2023), and Boas and Smith (2015) standing out. In general, these studies have focused on voting behaviour for elections between candidates, which contrasts with European literature that has analysed the effect of religion on different types of referendums. The latter is somewhat surprising considering that in Latin America, there have been, at least since 1990, referendums for new constitutions in Bolivia, Ecuador, Venezuela, Peru, and Chile, and constitutional amendments in Uruguay, Colombia, Guatemala, and Panama.

The specific case of Chile has also generated controversy, some of which may be related to the historical period analysed. For some, religion is not a predictor of electoral preferences compared to political cleavages, such as the left-right axis or the democracy-authoritarianism divide derived from General Pinochet’s dictatorship (1973–1990) (Tironi & Agüero 1999, Torcal & Mainwaring, 2003). Valenzuela, Scully, and Somma (2007) quickly confronted this dominant hypothesis in Chilean political science, who noted the effect of religion on voting intentions. Later, Bargsted and De la Cerda (2019) analysed the evolution of political disaffection in Chile as a function of religious identification. They argue that disaffection is stronger in non-religious segments than in Catholics, especially those who regularly attend religious services. Raymond (2021: 101) asserts “that religious-secular divisions in voting behaviour will persist in future elections as long as candidates reflect the divisions between religious and secular in the wider society”. More recently, the literature has stated that religion, particularly evangelicals, has activated the religious cleavage (Meléndez et al. 2021; Boas 2023; Castillo et al. 2023). This conclusion applies not only to Chile but also to several Latin American countries, such as Brazil. Little by little, it has been noticed that the electoral preferences of evangelicals who, without the prejudice of having progressive groups, tend to lean in favour of candidates of the radical right or populist right.

This controversy has become more relevant in this case study. Chile voted for a new constitution in September 2022. As I have pointed out, competition was not between candidates but between political options. Does religion have any effect on voting intention? The constitutional proposal implied significant political, economic, and cultural change. It proposed the elimination of the Senate, a state with autonomous regions and communes, a law on the voluntary interruption of pregnancy, and recognition of different justice systems for indigenous peoples. To a certain extent, these important modifications anticipated a negative response from the most conservative sectors, accustomed to gradual changes. In addition, the body that drafted the constitutional text, the Constitutional Convention, was predominantly composed of left-wing political parties and groups. Of the 155 seats that made up the Convention, right-wing parties won only 37 (23.8% of the total), to which almost a third of independent representatives of left-wing movements and 11% of seats reserved for indigenous peoples were more aligned with the left than with the right (Belmar et al. 2023).

In this article, I suggest four points. First, some of the contents of this constitutional proposal led to its rejection, especially by evangelicals and Catholics, in contrast to agnostics. According to the theoretical approach of the “systematic political reasoning approach” or “political knowledge” (De Angelis et al. 2020), it could be assumed that voters made a rational evaluation of the contents of the proposal, which guided their voting decisions. Second, the effect of religion is significant, even when incorporating short-term variables that are theoretically more powerful in explaining voting intentions. I specifically refer to presidential approval. According to the “heuristic approach” (Bartle 2005; Jacobs 2018), voters’ perception of the President allows them to “save” information costs. A voter may not have read the constitutional text and their vote may be defined based on the president’s evaluation as a type of “informational shortcut”. Third, although voters carry political predispositions such as ideological identification on the left-right axis, party identification, or class voting, following the “socio-psychological” approach (Lazarsfeld et al. 1944; Campbell et al. 1960), religion continues to have an impact on voting intentions, mainly when the discussion deals with value issues. Fourth, although evangelicals may be seen as a more depoliticised group (Fediakova 2012), there is abundant evidence that they are interested in and discuss political issues (Boas 2014; 2023). In addition, they show more conservative positions, although not always homogeneous (Freston 2001; Boas 2023), an issue intensively studied in the recent work of Castillo et al. (2023). There is some consensus that evangelicals tend to be more conservative in matters related to sexuality and gender, but as Castillo et al. (2023) correctly underline, a progressive evangelical identity persists, is more distant from the right, and promotes an expansion of social rights. It is a group that emerged as a response to the conservative predominance and that, for the constitutional referendum of 2022, supported the “Approve” option.

The other discussion refers to the reasons contributing to understanding religion’s impact on voting intentions in a constitutional referendum, specifically in a secularising country like Chile. The question is simple: What constitutional contents are linked to religious cleavage? In the context of secularisation, one might think religion is irrelevant in explaining voting intentions in the constitutional referendum. However, in this article, I argue the opposite: The contents of the constitutional proposal rekindled the religious divide. This constitutional proposal is based on the following five pillars. First, plurinationality, which implied recognising the 11 indigenous peoples, impacted their political representation, access to greater benefits, and even independent justice systems to regulate specific conflicts. Second, gender parity is present in all instances of popular representation, and in some organs and powers of the state. Third, the Senate, an institution that has been in place in Chile since 1812, was eliminated. Fourth, voluntary interruption of pregnancy (abortion) was established. Fifth, the National Health System differed greatly from the current system, which combines public and private agents. Indeed, another group of contents of the constitutional proposal linked to social rights and environmental protection could be added, as in González-Ocantos and Meléndez (2023). However, for this survey, we chose those issues that generated the most conflict and that, in principle, helped distinguish between “Approve” and “Reject” voters. I am aware that these are not the only contents that explain the differences between voters who supported one option or the other; however, from my perspective, they are helpful for this article.

Based on the above, my hypotheses are as follows.

Materials and methods

An online survey database comprising 2,177 participants was used in this study. This survey was conducted by the Millennium Nucleus Center for the Study of Politics, Public Opinion, and Media in Chile (MEPOP, NCS2021_063) between 25 August and 03 September 2022. That is, ten days before the constitutional referendum. The debate on constitutional texts was particularly intense. The political parties that supported the “Reject” option accused the political parties that supported the “Approve” option of promoting a refoundational Constitution. This was due to the changes made to the political system, for example, with the elimination of the Senate and the transition from a unitary state to a regional state characterised by greater autonomy for the regions and greater bureaucracy. Plurinationality was another point of debate. The constitutional proposal recognised the territorial autonomies of the original peoples and their justice systems, which generated intense criticism from the parties that supported the “Rejection” option, which defined the new Constitution as “divisive”.

Table 1 lists the variables used in this study. As is evident from the hypotheses, I used two dependent variables: voting intentions and support for the constitutional proposal. My objectives are twofold: On the one hand, I want to determine the effect of religion on the “Approve” and “Reject” options. On the other hand, I want to identify the contents of the constitutional proposal that generated more and less support according to each person’s religion.

Independent variables were arranged into three groups. First, political variables are theoretically the most robust in explaining voting intentions. This is the case of identification on the ideological scale and presidential approval, which is also a proxy for the general evaluation of the country at political and economic levels. Second, there are religious variables. I included identifying religion as a central variable and added “church attendance” as a statistical control to capture religious intensity. Third are traditional socio-demographic variables such as gender, age, socioeconomic status, and area of residence.

Descriptive and inferential statistical techniques were used to analyse the data. After performing a general bivariate analysis between religion and voting intention and between religion and support for the five axes of the constitutional proposal, I move on to multivariate analysis. The first is a multinomial logit regression model (mlogit), in which the dependent variable is voting intention for the constitutional referendum, and the central independent variable is religion. Second is an Ordinary Least Squares (OLS) model for each of the five axes of the constitutional proposal, in which religion is also the central independent variable.

Results

Figure 1 shows the voting intention for the constitutional referendum according to religious identification. A simple chi-squared test yielded a statistically significant association between the two variables (p < 0.001). These differences are evident. While 26.3% of Catholics leaned towards “Approve” and 49.3% towards “Reject”, in the case of Evangelicals, the differences were greater, with only 13% for “Approve” and 58.9% for “Reject”. Voters of “other religions” showed a voting behaviour very similar to that of Catholics, while agnostics strongly distanced themselves from this trend: 62.4% preferred “Approve”, and only 24.9% opted for “Reject”. Finally, voters with no religion were split almost in thirds: 36.7% marked “Approve”, 29.7% favoured “Reject”, and 33.6% preferred the NV/NA/DK options.

Figure 1

Figure 2 uses support for the five axes of the constitutional proposal as the dependent variable. ANOVA tests for one factor showed statistically significant differences (p < 0.001) between these five issues. Concerning plurinationality, the differences are more eloquent when comparing Catholics, Evangelicals, and other religions with agnostic and nonreligious people. While evangelicals, for example, supported the proposal by 34.5%, agnostic respondents supported it by 67.3%. In the gender parity proposal, the differences are less impressive but statistically significant. Again, evangelicals were least supportive of the measure. In the case of abortion, the differences are enormous: Only 22.8% of evangelicals supported the measure, rising to 42.2% among those identified with other religions, 44.4% among Catholics, 59% among those with no religion and 77.8% among agnostics. It is not surprising that 22.8% of evangelicals supported the constitutional article on abortion, considering the existence of a more progressive group of evangelicals (Castillo et al., 2023). Regarding the proposal to eliminate the Senate, respondents identified with a religion supported the proposal to a lesser extent than agnostics and people with no religion, showing similar behaviour to that recorded for plurinationality. Finally, support for the National Health System (NHS) was higher among agnostic and non-religious groups than among Catholics, evangelicals, and those identified with other religions.

Figure 2

In summary, bivariate analysis showed the following. First, religion is a relevant variable for understanding support for the content of constitutional proposals. Second, there are marked differences between religious and non-religious people, with religious people being least sympathetic to the contents of the constitutional proposal. Third, evangelicals mostly rejected the contents of the constitutional proposal, with the most marked differences from the rest on the issues of abortion and gender parity.

To test my hypotheses, I used two types of inferential analysis. The first is a mlogit model with respondents’ choices for the constitutional referendum (Reject/Approve/NV/NA/DK). Second, a set of multivariate linear regression models was used for each of the five axes of the constitutional proposal.

The first hypothesis (H1) suggests that religion significantly impacts voting intentions for the constitutional referendum, including variables in the statistical model that are theoretically more powerful in explaining voting behaviour. The results of the statistical models support this hypothesis (Table 2). A multinomial logit model was constructed using voting intention as the dependent variable (Model 1). I have placed “Approve” as the reference category. Religion was found to have a significant effect, even alongside other theoretically more robust variables in explaining voting intention, such as left-right identification and presidential approval. The models show that religious voters- Catholic, Evangelical, or of other religions- preferred the “Reject” option more than the “Approve” option. This coefficient was higher in the evangelical group. In the model comparing the NV/NA/DK group with the “Approve” group (Model 2), it is noticeable that agnostic voters – longer-lived than average, with higher incomes and with ideological positioning further to the left – were more likely to “Approve” than to declare themselves as part of the NV/NA/DK group.

There are no statistically significant differences in the frequency of church attendance when comparing the options “Reject” and “Approve” (Model 1). However, differences emerged when comparing the NV/NA/DK group with the Approve group (Model 2). Simply put, more frequent churchgoers tended to choose the NV/NA/DK option over the ‘Approve” choice, consistent with Fediakova’s (2012) central argument regarding the greater depoliticisation of churchgoers. This makes sense and portrays, in part, the consolidation of religious cleavage as a robust predictor of voting intentions. The “Approve” social basis did not capture the religious vote on two dimensions. The first of these is religious identification. Second, there is a religious intensity. The differences between “Approve” and “Reject” were mainly at the level of religious identification, with Catholics, Evangelicals and voters of other religions significantly more supportive of the rejection of the new constitution. Then, if we consider religious intensity, differences emerge between the NV/NA/DK and “Approve” options, with intense religious voters being more likely to support the former than the latter.

Political variables have the expected effect. The ideological scale indicates that the further one moved from left to right, the more support for “Reject” grew, and the more support for “Approve” fell. As I have pointed out, this is explained by the fact that the Constitutional Convention – the political body that drafted the constitutional proposal – was primarily composed of left-wing parties and movements. Then, concerning presidential approval, those who approved of the President were more inclined to “Approve” and less prone to “Reject.” Presidential approval is the variable that has the most robust effect on voting intentions for the referendum, a result that is consistent with the research of Alemán and Navia (2023). This result is not surprising given that the government actively participated in the programmatic debate on the new constitution. Thus, the government acted as an informational shortcut for voters. This finding supports the heuristic approach, but does not imply that presidential approval is the only relevant variable, as the other model results show.

The control variables show, very generally, that voters’ gender did not affect their voting intention. Then, the older the voter, the more significant the support for rejection; the higher the socioeconomic level, the greater the support for rejection; and the area of residence does not explain voting intention.

Based on the above model, a predicted probability plot was constructed for all response categories in the religion question (Figure 3). Here, religious voters’ preference for the “Reject” option is clearly shown. The result is most evident among evangelicals, but the differences in favour of this option are significant across all religions. Then, agnostics and “no religion” voters have different behaviours. Both groups’ preferences for “Approve” and “Reject” are statistically indistinguishable. In summary, there is sufficient evidence to support H1 regarding the impact of religion on voting intentions for the constitutional referendum.

Figure 3

Why was “Rejection” stronger in religious people than in the rest of the population? I have already shown that religion significantly affects voting intentions for constitutional referendums. The next step was to identify the contents of the constitutional proposal that generated the most disenchantment to evaluate Hypothesis 2 (H2). As I pointed out, the survey measured support for five axes of the proposal: plurinationality, gender parity, elimination of the Senate, abortion, and the creation of a public National Health System.

Descriptive analysis revealed statistically significant differences according to religion for these five axes. To test whether this relationship is robust, I build a multivariate linear regression (OLS) model for each content of the constitutional proposal. The independent variables were the same as those used in the model explaining voting intention. I also built a second model with an interaction term between religious identification and ideological scale to evaluate Hypothesis 3 (H3). The results are presented in Table 3.

Five were the most relevant findings. First, evangelicals rejected constitutional content more independently of the model’s configuration. Second, the further right along the left-right axis, the more the content of the constitutional proposal was rejected. Third, the frequency of church attendance had a negative and robust impact on support for abortion and gender parity. Those who were more religious opposed both issues of value-based nature. Fourth, Catholics were more divided, rejecting abortion to a lesser extent than evangelicals, and supporting gender parity. Fifth, the interaction term between religion and ideological identification was significant only in the case of abortion.

To understand the latter, I constructed a graph of predicted values to explain the support for abortion and plurinationality. I chose abortion because it was the value issue that generated the most controversy in constitutional discussions and plurinationality. For both cases, I took the model that included the interaction term between religion and ideological identification as a basis (Models 2 and 8). To show the results more clearly, I only had the three largest groups: Catholics, Evangelicals, and respondents with “no religion”.

The results are shown in Figure 4. As expected, the evangelicals were significantly different from the other groups. Next, evangelicals’ support for abortion was barely sensitive to variations in the ideological scale. This contrasts starkly with other groups. In general, support for abortion was the highest among left-wing respondents, declining as one moved to the right along the ideological axis. However, among evangelicals, right- and left-wing voters showed similar rejection levels of abortion. This evidence supports hypothesis 3 (H3).

Figure 4

In the case of plurinationality, support was lowest among Catholics and Evangelicals and highest among the “no religion” group. In all groups, support was higher among left-wing voters and much lower among right-wing voters. In the case of Evangelicals, the low support for plurinationality is more notorious, which could be explained by a certain “nativism” frequently linked to more conservative voters (Meléndez et al. 2021). Though not shown here, something similar happened with support for eliminating the Senate, NHS, and gender parity. Consequently, everything seems to indicate that for the evangelical segment, abortion was an unacceptable and intolerable proposal, regardless of the ideology of each voter, even though disagreement with this measure was not unanimous. This makes sense if we consider the exciting work of Castillo et al. (2023) on the political behaviour of evangelicals, highlighting the existence of a more progressive group that favoured approving the new constitution despite its opposition to abortion. The evangelical-progressive segment was willing to tolerate abortion in exchange for an economic model further away from neoliberalism and closer to social and democratic rule of law.

I performed the same analysis but with an interaction term between presidential approval and religion. I do not present the results here, but the interpretation is the same. That is, evangelicals who approve and disapprove of the government have similar levels of support for abortions. In the case of support for plurinationality, the effect is not statistically significant.

Discussion and conclusions

What happens to religious cleavage when voters face programmatic choices rather than candidates, how determinant is religion in a critical election such as a constitutional referendum, and is it possible that religion is a relevant factor in explaining electoral outcomes in a secularised or secularising country? In this article, I have provided evidence on the prevalence of religious cleavage in a country with strong indicators of secularisation and in a critical election that rarely occurs: a constitutional referendum. Although constitutional referendums in Latin America have been studied preferably from institutional approaches, examining, among other things, the contents of the proposal concerning variations in presidential powers, there is a lack of literature that analyses them from the perspective of electoral behaviour and, more specifically, from the standpoint of social cleavages, such as religion. Indeed, there have been advances in understanding the electoral behaviour of particular religions, such as evangelicals (Boas 2023), and even their political behaviour in the face of the Chilean constitutional referendum of 2020 (Castillo et al. 2023). This literature provides a new impetus for studying the impact of religion on political decisions.

Consequently, and following the first hypothesis, religion significantly affects voting intention. What is most striking is that this effect survives while controlling for theoretically more robust variables, such as ideological identification and presidential approval. As the theoretical models of social cleavages and heuristics suggest, voting behaviour in constitutional referendums is fundamentally explained by ideological identification and informational shortcuts voters take. The approval of the President or Prime Minister plays a central role in the latter. Voters bring political and ideological predispositions built up in family socialisation processes, but they also update these preferences according to short-term variables. Presidential approval is one of them. The information costs involved in knowing a complex constitutional text may lead voters to use shortcuts. A popular president and a favourable economic situation may incline voters to vote the way that the president suggests, even without reading the constitutional text in detail. This analysis shows that religion has a statistically significant effect on voting intention, including the impact of long-term variables such as ideological identification and short-term variables such as presidential approval.

Why did religious voters reject the constitutional proposal in greater proportion than agnostics and voters with no religion? According to my second hypothesis, and after confirming the validity of the religious cleavage, I show the constitutional content. Among these, the most controversial from a value-based point of view is the voluntary interruption of pregnancy. Here, I have noticed two divisions. The first is the division between people identifying with a religion and agnostics or those without religion. The “religious” rejected this article more broadly than the “non-religious.” The second is the division between Catholics, evangelicals, and people with other religions. Evangelicals rejected the proposal for voluntary termination of pregnancy in a greater proportion than the rest. This group was much more homogeneous on this issue than Catholics and adherents of other religions. This, in no way, hides the ideological diversity of Evangelicals, as Castillo et al. (2023) note. However, on an issue as important as abortion, my results indicate, for example, that left- and right-wing evangelicals were almost equally opposed to this measure.

This finding is relevant because it allows for a more detailed understanding of the electoral behaviour of evangelicals. Additionally, this finding calls into question the great predictive capacity of voters’ self-positioning on the left-right axis to explain electoral behaviour. In the case of evangelicals, my results indicate that, regardless of whether the voter is left or right, they oppose constitutional regulations on abortion (Hypothesis 3). In other words, value issues seem more critical for evangelicals than the traditional division of parties along the left-right axis. Likely, the electoral mobilisation of evangelicals responds more to this type of issue, which is not necessarily done by creating a political party but instead of evangelical organisations with solid social bases and even study centres (Mansilla et al. 2019). However, Chilean Evangelicals founded the Christian Conservative Party in 2020, which disappeared in 2022, and the Christian Social Party (PSC) in the same year. Of the 25 established parties in Chile, the PSC is the thirteenth with the most militants, with 14,930, according to the Chilean Electoral Service records as of 29 February 2024. PSC also has three deputies in the current Congress out of 155. While forming an evangelical party is not necessarily synonymous with politicising religion, its programmatic proposals move in that direction. During the constitutional debate, the defence of conservative ideas, especially abortion, was notorious. This marked a certain continuity with the program of José Antonio Kast, the radical right’s presidential candidate in 2017 and 2021, who showed strong roots in most conservative evangelicals (Morales & Pérez 2024).

In short, to answer the initial question, religious cleavage confirms its relevance even in an election in which options and not candidates compete. Moreover, it does so in a secularised country. Although identification with one of the religions in Chile has declined in recent decades, and its effect on electoral results is intermittent, a critical election such as a constitutional referendum can politicise religious preferences, transforming them into a strong predictor of voting intentions.

Funding Information

This research is sponsored by Fondecyt Project number 1220004 of Conicyt, and Millennium Nucleus Center for the Study of Politics, Public Opinion and Media in Chile (MEPOP, NCS2021_063).

Competing Interests

The author has no competing interests to declare.