Introduction
In January 2018, the Inter-American Court of Human Rights (IACHR) ruled that Costa Rica was obligated to legalize same-sex marriage, immediately transforming the dynamics of its ongoing presidential election campaign. While most candidates equivocated in their response to the ruling, the two who staked out the clearest positions for and against same-sex marriage, Carlos Alvarado and Fabricio Alvarado, went from single-digit standings to become the top two finishers (Zúñiga Ramírez Reference Zúñiga Ramírez, Guadalupe and Grundberger2018). Views on sexuality politics emerged as a major cleavage. In a post-electoral survey that inquired about the reasons for one's presidential vote, the biggest gap between supporters of the two candidates was in the percentage who voted ‘to defend the traditional values of Costa Rica’: 54 per cent for Fabricio Alvarado and 9 per cent for Carlos Alvarado (Alfaro Redondo et al. Reference Alfaro Redondo2018).
Religion also played new roles in the election. Nine days after the court ruling, the Catholic Bishops Conference and the Costa Rican Evangelical Alliance Federation issued a joint statement supporting ‘the family founded on the marriage between a man and a woman’, calling on citizens to meditate on their presidential votes before God (Arroyo Reference Arroyo2018). Such stances might have pushed both conservative Catholics and evangelicals into the arms of Fabricio Alvarado, an evangelical pastor. However, in the second round campaign, media attention to ostensibly anti-Catholic statements from an ally of Fabricio Alvarado prevented an inter-religious alliance (Alfaro Redondo et al. Reference Alfaro Redondo2018). On election day, religious affiliations and attitudes were among the strongest correlates of vote choice (Díaz González and Cordero Cordero Reference Díaz González and Cordero Cordero2020; Pignataro and Treminio Reference Pignataro and Treminio2019).
The example of Costa Rica's 2018 election raises important general questions: (1) how and when do new electoral cleavages emerge in developing democracies and (2) why do we see them in some places but not others?Footnote 1 Throughout Latin America, same-sex marriage, abortion, and the treatment of gender and sexuality in public school curricula are transforming political battles between the left and the right in a way that seemed unthinkable just a decade ago. Simultaneously, electoral gaps have emerged between religious groups, sometimes between evangelicals and Catholics and sometimes along religious-secular lines. Yet these changes are not uniform across the region, nor have they proceeded gradually in a pattern that might be attributable to the growth of evangelicalism or the religiously unaffiliated. Some countries have largely avoided religious cleavages and battles over sexuality politics, with electoral competition still centring on materialist issues such as redistribution or law and order. In others, like Costa Rica, new electoral cleavages seem to burst onto the scene in a spectacular, discontinuous fashion.
We argue that the emergence of new electoral cleavages around sexuality politics and religion is attributable to the specific actions of political elites and social movements. Gradual demographic changes such as the growth of evangelicalism or human development and post-modernization (Inglehart and Welzel Reference Inglehart and Welzel2005) help set the stage for new cleavages to emerge but such changes are insufficient on their own. Sometimes, issues of sexuality and the family are thrust onto the political agenda as the result of a prominent judicial decision, such as the Colombian Supreme Court's legalization of same-sex marriage in 2013 or the IACHR decision bearing on Costa Rica in 2018. In other instances, they arise thanks to alliances between social movements and legislators, as with Mexico City's decriminalization of abortions in 2007 or the movement to do so in Argentina between 2018 and 2021 (Daby and Moseley Reference Daby and Moseley2022; Reuterswärd Reference Reuterswärd2021). Once elites and movements place these issues on the agenda, the public responds, and vote choice becomes tied to attitudes on sexuality politics. Religion enters the story because it affects vote choice in myriad ways via issue attitudes, discussion within congregations and political messages from the pulpit. Because different sexuality politics issues are salient in different religious communities, voters sometimes realign along a Christian-secular cleavage and, at other times, along an evangelical-Catholic one.
This paper seeks to explain how voting behaviour shifts when sexuality politics issues are made salient and when candidates adopt contrasting positions on these issues.Footnote 2 Our analysis draws on multiple data sources and methodologies: panel data on newspaper coverage of same-sex marriage and abortion across Latin America, multilevel analysis of seven waves of the AmericasBarometer and Latinobarómetro surveys between 2004 and 2019, and a conjoint survey experiment in Brazil, Chile, and Peru. First, we show that as newspaper coverage of abortion or same-sex marriage/civil unions increases, typically in response to a legislative proposal or judicial decision, both sexuality politics attitudes and religion become more predictive of left-right voting decisions. Next, we show that in an experimental context where hypothetical candidates differ in their issue stances, voter-candidate agreement on abortion policy has a comparable or greater effect on vote intention than agreement on crime policy or the role of the state in the economy. Finally, we argue that the rising salience of sexuality politics may partially displace materialist voting: increased news coverage of these issues is associated with a declining correlation between support for redistribution and vote choice.
Our findings contrast with and help explain prior findings that positions on sexuality politics weakly predict voting behaviour in Latin America (Carlin, Singer, and Zechmeister Reference Carlin, Singer, Zechmeister, Carlin, Singer and Zechmeister2015; Singer and Tafoya Reference Singer, Tafoya, Vanden and Prevost2021; Zechmeister and Corral Reference Zechmeister and Corral2013) and that religion's correlation with vote choice varies from one election to another (Boas and Smith Reference Boas, Smith, Carlin, Singer and Zechmeister2015). We show that Latin American voters' stances on sexuality politics sometimes strongly affect voting behaviour but only under certain conditions: when heightened media coverage makes these issues salient and when candidates stake out opposing positions. Similarly, religious citizens become religious conservatives at the ballot box when sexuality politics issues arrive on the agenda. Thus, sexuality politics debates at the elite level catalyze the growth of the new religious right.
Our arguments speak to a large literature explaining political polarization and ‘morality’ policies in the US and Europe. Scholars show that polarization on culture war issues in those regions began among political, religious, and social movement elites before spreading to the electorate. We argue that a similar process has taken place in Latin America, albeit in a very different partisan context. Strong partisanship in the US and Europe helps bind attitudes on sexuality politics and religion. In Latin America, sorting occurred despite multiparty systems and weaker mass partisanship. Thus, our research suggests that the rise of sexuality politics on the policy agenda can lead to electoral realignments even absent strong partisan identities.
Theory
What explains the growing importance of sexuality politics, issues, and religion in some Latin American elections? Classic political science scholarship posits that policy stances are a key short-term determinant of election outcomes as voters seek the candidates closest to them on a menu of issues that evolve from one election to the next (Campbell et al. Reference Campbell1964; Downs Reference Downs1957). The questions are how and why that menu evolves. Brody and Page (Reference Brody and Page1972, 455–6) posit that policy voting requires not only an estimate of voter and candidate issue preferences but also an ‘affective link between the voter and the issue’. Some voters just care more about particular issues in certain times and places. This early scholarship, however, leaves unresolved how context shapes such affective links.
One prominent hypothesis to explain the rise of sexuality politics in Latin America is that the explosive growth of evangelicalism and Pentecostalism has influenced voters' issue priorities (Corrales Reference Corrales, Bosia, McEvoy and Rahman2020; Costa, Marcantonio Junior, de Castro Reference Costa, Marcantonio Junior and de Castro2018). The timing seems right. The percentage of Latin Americans adhering to these religious traditions has grown five-fold since 1970 (Pew Research Center 2014), coinciding with the rise of sexuality politics on political agendas. Like their counterparts worldwide, Latin America's evangelicals and Pentecostals adopt conservative theological stances, adhering to strict doctrinal interpretations that might dovetail with conservative policy positions (Robbins Reference Robbins2004). The often implicit analogy to evangelicalism and Pentecostalism in the US and Sub-Saharan Africa (Grossman Reference Grossman2015) primes observers to expect a natural linkage between these traditions and culture war politics.
On further scrutiny, though, this explanation looks incomplete. Prior to the 2000s, growing evangelicalism and Pentecostalism did not automatically translate into right-wing voting or produce a broader societal cleavage over sexuality politics. Evangelicals and Pentecostals have played a prominent electoral role in some Latin American countries since the 1980s (Boas Reference Boas2023) yet, as recently as 2012, they were not consistently voting to the right (Boas and Smith Reference Boas, Smith, Carlin, Singer and Zechmeister2015). Social pressures may induce members of evangelical churches to cohere in vote choice (Smith Reference Smith2019), but those choices have not been uniformly rightist in Latin America. The same is true of issue attitudes. Thanks to their lower-class social origins, Latin American evangelicals and Pentecostals have taken substantially more leftist positions than their American counterparts on policy issues ranging from social welfare to the environment (McAdams and Lance Reference McAdams and Lance2013; Smilde Reference Smilde2007). Frequent analogies to the US may mislead. US evangelicals are substantially more politically conservative than their counterparts in other wealthy English-speaking democracies (Bean Reference Bean2014). In short, evangelical and Pentecostal growth has built a sizeable cohort of conservative Latin Americans on sexuality politics issues, but this growth has not automatically created a consistently conservative voting bloc.
A second hypothesis relates to modernization and human development. While classic modernization theory claimed that development would bring about the ‘death of religion’ (Norris and Inglehart Reference Norris and Inglehart2011, 3), more recent post-modernization arguments envision a growing culture war conflict as the push for progressive policies on sexuality and the family prompts a backlash from the remaining social and religious conservatives (Gaskins, Golder, and Siegel Reference Gaskins, Golder and Siegel2013; Inglehart and Welzel Reference Inglehart and Welzel2005). Scholars thus predict that human development will widen cleavages between religious conservatives and secular voters and intensify battles over sexuality politics issues. This perspective would argue that the growth of evangelicalism does not automatically produce conflict; rather, it does so only when it coincides with growing secularization and demands for policy liberalization.
The human development perspective offers an important insight; there can be no conflict over issues not yet on the political agenda, and liberalizing efforts on sexuality politics have gone furthest in Latin America's wealthier countries. We see major cleavages over abortion and LGBTQ rights in Argentina, Brazil, Chile, Colombia, and Costa Rica, but not in Guatemala, Paraguay, or Venezuela. Change over time also matters. Bargsted and de la Cerda (Reference Bargsted and de la Cerda2019) show that Chile's growing secularism and economic development coincided with increased ideological polarization between observant and non-observant Catholics. We can thus consider broad, development-driven value change as a necessary condition for attitudes on sexuality politics to influence voting behaviour in Latin America.
Yet the gradual value changes associated with post-modernization should not be sufficient for sexuality politics attitudes to influence voting behaviour. First, even if sexuality politics becomes increasingly salient with economic development, particular policy issues within this cluster come and go from the agenda. In Peru, several same-sex civil unions bills were introduced in Congress between 2013 and 2016, but none was successful. More recently, the conflict has centred on the treatment of gender in school curricula (Boas Reference Boas2023). In Argentina, attention to same-sex marriage spiked in 2010 as legislators were debating its legalization, but the issue's salience declined significantly after the policy passed. The more recent battleground was the 2018–2021 struggle to legalize first-trimester abortion, an issue seen as politically untouchable before feminist groups mobilized (Daby and Moseley Reference Daby and Moseley2022). As abortion, same-sex marriage, and related issues rise and fall on political agendas, the degree to which they influence voting behaviour should also fluctuate rather than increase steadily with economic development.
Second, beyond the timing of policy battles, issue attitudes are more likely to influence voting behaviour when candidates differentiate themselves on those issues. If an issue becomes salient due to media coverage or social movement pressure, but major candidates strategically ignore it or adopt similar positions, public attitudes on the issue are unlikely to affect voting behaviour. In Chile, Raymond (Reference Raymond2021) shows that religion influences voting behaviour, but only when candidates are divided by religious affiliation or associated policy stances. Throughout Latin America, leftist and rightist candidates reliably take opposing positions on economic redistribution and security, but they have not consistently done so with sexuality politics. In the Supplementary Materials, we show that Latin American party programmes vary little on ‘traditional morality’ issues such as divorce, abortion, or church-state separation, and they often ignore these issues entirely. This lack of differentiation may result from left-wing parties' ambivalence on these issues (Friedman Reference Friedman2009), born out of a desire to prioritize redistribution and a pragmatic recognition that many low-income voters hold socially conservative attitudes.
In emphasizing issue salience and candidate position-taking, we shift the focus from gradual social change to the specific actors who place sexuality politics on the agenda: social movements and political and judicial elites. In both Europe and the US, ideological polarization and electoral realignment around culture war issues began with political and social elites followed by parallel transformations at the mass levels (Engeli, Green-Pedersen and Larsen Reference Engeli, Green-Pedersen and Larsen2012; Fiorina, Abrams, and Pope Reference Fiorina, Abrams and Pope2005; Hunter Reference Hunter1992; Smith and Tatalovich Reference Smith and Tatalovich2003; Studlar and Burns Reference Studlar and Burns2015). In recent decades, Latin American social movements and elites have led a similar realignment by placing new issues on the agenda.Footnote 3 On average, the region's political class holds more progressive attitudes than the general public on same-sex marriage, abortion, and related issues (Boas and Smith Reference Boas and Smith2019; Corral González Reference Corral González2013). Progressive policy initiatives have typically responded to organizing by feminist and LGBTQ movements rather than widespread public demand for policy change (Daby and Moseley Reference Daby and Moseley2022; Díez Reference Díez2015; Encarnación Reference Encarnación2016; Htun Reference Htun2003). And progressive triumphs have often spawned conservative social movements that further shape public opinion. Examples include ‘Don't Mess with My Children’, founded to oppose progressive sexual education curricula in Peru, and movements to ban abortion in Mexican states following progressives' success in legalizing it in the capital (Corredor Reference Corredor2019; Reuterswärd Reference Reuterswärd2021; Rousseau Reference Rousseau2020). When wedge issues transform political agendas in such a fashion, candidates are more likely to stake out opposing positions and those that remain neutral risk electoral decline (Chhibber Reference Chhibber1999), as in Costa Rica's 2018 election.
When sexuality politics rises in salience, and candidates stake out opposing positions, religious communities intensify individual realignments. As in the US and Europe (Putnam and Campbell Reference Putnam and Campbell2012; Smith and Tatalovich Reference Smith and Tatalovich2003), abortion and same-sex marriage are the two long-standing issues that most strongly cleave the Latin American electorate along religious lines. In the Supplementary Materials, we show that religious attendance strongly depresses support for same-sex marriage and abortion in Latin America. By contrast, religious attendance and Protestant (versus Catholic) affiliation are associated with progressive – not conservative – positions on economic redistribution and crime. Hence, when sexuality politics remains a latent cleavage, religious affiliation and church attendance may actually push Latin Americans to vote for the left. However, when sexuality politics becomes salient, we expect that religious citizens will crystallize as a conservative electoral base. In the process, this transformation may partially crowd out traditional issue cleavages, such as the economy (De la O and Rodden Reference De la O and Rodden2008).
We also expect different religious cleavages around different issues. Through internal dialogue and socialization, Christian theological traditions construct diverse interpretations of biblical doctrine, with implications for members' views on political issues. Pentecostalism and charismatic Christianity, the theological traditions of most Latin American evangelicals, are distinctively associated with ‘moral asceticism’ and conservative positions on LGBTQ rights (Corrales Reference Corrales2017; Grossman Reference Grossman2015; Klinken Reference Klinken2014; Robbins Reference Robbins2004). On some issues, they may even outflank devout Catholics. Recent work shows that Brazilian evangelicals are generally more conservative than Catholics on homosexuality and gender, whereas religiously devout Christians from both traditions are similarly opposed to abortion (Smith Reference Smith2019). In the Supplementary Materials, we show that, after accounting for inter-group differences in church attendance, there is no difference between Catholics and evangelicals in abortion attitudes. However, a gap persists in views of same-sex marriage. As a result, we expect that policy debates over abortion will produce an electoral cleavage between the non-religious and Christians of either tradition, while debates on same-sex marriage will also trigger electoral gaps between Catholics and evangelicals.
If religion influenced voting behaviour simply via its effect on issue attitudes, it might be a relatively unimportant part of the story, located far back in the ‘funnel of causality’ from demographics to vote choice (Campbell et al. Reference Campbell1964). However, we expect that when sexuality politics increases in salience, religion will influence voters' decisions in multiple ways beyond the effect mediated by issue attitudes. First, social pressure within religious communities often leads to high levels of political cohesion as churches orient vote choice through social and identity-based processes (Djupe and Gilbert Reference Djupe and Gilbert2009; Smith Reference Smith2019). Even church members who are personally unconvinced on the issues may end up voting with their fellow congregants. Second, religious leaders may deliberately work to strengthen the linkage between policy attitudes and vote choice through active campaigning or less overtly partisan messages during sermons or other interactions with church members.
Based on the discussion above, we formulate and test the following hypotheses:
H1a. When sexuality politics issues are more salient, views on these issues will be more strongly linked to vote choice.
H1b. When sexuality politics issues are more salient, both religiosity and religious affiliation will be more strongly linked to vote choice.
H2. In a context where candidates adopt contrasting positions, views on sexuality politics issues will be strongly linked to vote choice.
Empirical Analysis
To test our hypotheses, we combine two distinct research designs: (1) multilevel analysis of voting behaviour in multiple waves of the AmericasBarometer and Latinobarómetro surveys, conditioning on coverage of same-sex marriage and abortion in major newspapers during the period surrounding the election and (2) a candidate-choice conjoint experiment conducted via online surveys in Brazil, Chile, and Peru. These two research designs complement one another. The multilevel analysis testing H1a and H1b casts a broad net, examining multiple years of elections across fifteen countries. Meanwhile, the conjoint experiment testing H2 excels in terms of internal validity, allowing for a causal interpretation of the effect of the voter-candidate agreement on abortion policy in three countries.
Multilevel Analysis
Contextual measures
For our public opinion analysis, we develop real-world measures of the salience of abortion and same-sex marriage. Our contextual measures reflect the frequency with which these issues are mentioned in news coverage in each country year, following Grossman (Reference Grossman2015). Using the Factiva database, we identified the major newspaper with the most complete full-text coverage for each Latin American country and searched terms for abortion, same-sex marriage, and civil unions (see Supplementary Materials for specific search terms and the newspapers). Newspapers vary in the amount of coverage they devote to any given topic, and the Factiva database has more complete coverage of some publications in recent years, so we standardize coverage of abortion and same-sex marriage by the number of stories about politics, measured via a full-text search on the term ‘política’. Due to limited full-text newspaper coverage, we have no measures for El Salvador, Honduras, and Nicaragua or for same-sex marriage in the Dominican Republic.
The resulting raw measures, ranging from 0 to 0.173 for abortion and 0 to 0.116 for same-sex marriage, represent the ratio of the number of stories on each issue to the total number of stories about politics. Figure 1 summarizes these measures for each of the fifteen countries for which we have data. The major spikes in coverage are readily identifiable as corresponding to periods of high salience for abortion or same-sex marriage, typically because of a proposed or actual policy change, as summarized in Table 1.
Abortion and same-sex marriage as a share of political coverage in the main newspaper, 2002–2019.
Identifying peaks in coverage of sexuality politics issues
Countries with no identifiable peaks for these issues also serve to validate the measure. For abortion, all countries with low and steady levels of news coverage experienced no legislative change or proposed change during the period examined. In most of them, abortion is either totally prohibited or allowed only to save the life of the mother (Marcus-Delgado Reference Marcus-Delgado2019). Likewise, the three countries with consistently low levels of same-sex marriage coverage, Guatemala, Paraguay, and Venezuela, have seen little progress on LGBTQ rights. In these countries, abortion or same-sex marriage is largely off the political agenda, a fact that is reflected in the absence of newspaper coverage.
For a contextual measure of salience to use in our regressions, we averaged the news coverage ratios from the year of the election in question and the previous year in order to develop a smoothed estimate of the salience of these topics in the lead-up to each political contest. Since the AmericasBarometer asks about votes in prior elections, these contextual measures are lagged to the time of the election. We then rescaled the proportions to run from 0 to 1 and transformed them by taking their square roots, which retains the 0–1 rescaling but reduces the skew. Histograms of the resulting contextual variables are included in the Supplementary Materials.
Survey data
Our multilevel analysis integrates these contextual measures with public opinion data to examine how issue salience modifies the relationship between a series of individual-level variables and voting behaviour. Our primary source is the 2012, 2014, 2016–17, and 2018–19 waves of the AmericasBarometer surveys, all of which inquired about respondents' support for therapeutic abortion as well as same-sex marriage. The former is measured via a yes/no question: ‘Do you believe that the interruption of pregnancy, or an abortion, is justified when the life of the mother is in danger?’ The latter is measured via the question ‘How strongly do you approve or disapprove of same-sex couples having the right to marry’, with responses on a 10-point Likert scale from strongly disapprove (1) to strongly approve (10). We recode abortion views as a dichotomous measure indicating support for the right to a therapeutic abortion and same-sex marriage views on a 0 to 1 scale. For both recoded variables, higher numbers indicate a more progressive policy position.
To test the conditional impact of religion and religiosity, we use two measures. The first is self-identified religious affiliation, recoded as Catholic, evangelical/Pentecostal, no religion, and other religions. The second is church attendance, which we recode to run from 0 (‘never or almost never’) to 1 (‘more than once a week’). Since this latter item was not asked in the 2014 AmericasBarometer, the religion analysis incorporates only the 2012, 2016–17, and 2018–19 waves. In other regressions, we control for religious affiliation but not religious attendance to maximize the number of waves we can include.
To account for potentially confounded ideological and policy views, we include a number of other attitudinal measures. We control for economic views using the only relevant question present in all four survey waves: ‘The (Country) government should implement strong policies to reduce income inequality between the rich and the poor. To what extent do you agree or disagree with this statement?’ Answers are measured on a 7-point Likert scale, recoded 0–1, with higher numbers indicating support for redistribution. In some models, we also include a dichotomous indicator of support for tough-on-crime policies (‘fighting crime by increasing punishment’), which was asked in most but not all country years. To measure ideology, we include indicator variables for those placing themselves on the left (positions 1–4) or right (positions 7–10) of the 1–10 ideological scale. Given high and non-random rates of non-reporting of ideology (Zechmeister and Corral Reference Zechmeister and Corral2013), we include respondents with missing values on these questions, coding them as 0 on our ‘leftist’ and ‘rightist’ dummy variables. Finally, we incorporate relevant demographic variables potentially correlated with religion and policy attitudes, including gender, household wealth, education, age, ethnic identification, and size of place of residence. The latter helps address the concern that anti-urban elite sentiment might be driving a relationship between religion and voting behaviour.
Our dependent variable is the ideological direction of vote choice. The underlying measure involves a question about respondents' votes in the last presidential election (the first round in the case of majority run-off systems). Across all countries and waves, 57 per cent of respondents name a candidate or party in response to this question. To generate a dependent variable that can be compared across countries, we follow Carlin, Singer, and Zechmeister (Reference Carlin, Singer, Zechmeister, Carlin, Singer and Zechmeister2015) in transforming votes for specific candidates/parties into a 1–10 measure of the ideology of vote choice. We merge in left-right estimates for each candidate/party from the Parliamentary Elites in Latin America (PELA) surveys from each country, in which legislators are asked to place major politicians and parties on a 1–10 left-right scale. Our values are drawn from the PELA wave most proximate to the specific election mentioned in the AmericasBarometer survey question. We use the mean ideological rating of the politician if available and party ratings otherwise.Footnote 4 Details about a few difficult-to-score cases are in the Supplementary Materials. As the dependent variable runs from 1 to 10 and is approximately normally distributed, with a mean of 5.7, all our analysis uses standard multilevel linear models, which facilitates the interpretation of effects directly from regression results.
A table in the Supplementary Materials summarizes the countries, elections, and survey waves for which we have valid measures of news coverage, as well as the share of respondents from each wave voting for candidates or parties for whom we lack a valid measure of ideology. For most countries, we are able to use all four AmericasBarometer waves. For the Dominican Republic and Paraguay, we drop the 2012 wave; for Guatemala, we omit 2014. On average, candidates for whom we lack ideology measures garnered only 1.0 per cent of the vote. The only case where this figure rises above 5 per cent is Chile's 2013 election, which featured an unusual number of outsider or small-party candidates.
Results
The Results of our analysis provide strong support for our hypotheses that issue salience modifies the role of sexuality, political attitudes, and religion in shaping vote choice (H1a and H1b). Figures 2 and 3 show our key results based on full multivariate multilevel models. The Supplementary Materials contain tables corresponding to these figures, as well as results from a separate, unconditional analysis.
Vote choice as a function of issue attitudes and news coverage.
Source: AmericasBarometer 2012–2019.
The figure plots the marginal effect of each issue attitude on the ideology of vote choice (higher values = rightist voting). Ninety-five per cent confidence intervals are shown. Estimates are based on the full multivariate model as shown in Supplementary Materials.
Vote choice as a function of religion variables and news coverage.
Source: AmericasBarometer 2012, 2016/17, and 2018/2019.
The top panes plot the gap in the ideology of vote choice between evangelicals and Catholics. The bottom panes plot the marginal effect of church attendance (higher values = rightist voting). Ninety-five per cent confidence intervals are shown. Estimates are based on the full multivariate model as shown in the Supplementary Materials.
Figure 2 supports H1a: the impact of sexuality politics attitudes is strongly contingent on news coverage. In the right panel, we find that the effect of same-sex marriage attitudes on the ideology of vote choice varies with news coverage of the issue. In times and places where same-sex marriage is highly salient, attitudes on this issue are predicted to matter more than economic views and nearly as much as ideological identification. The left panel shows that news coverage of abortion likewise moderates the effect of abortion attitudes on vote choice. However, we also find a counterintuitive result. In times and places where abortion is not widely discussed, people who support therapeutic abortion tend to vote for rightist candidates. We discuss this finding in the following section.
Figure 3 examines the interaction between news coverage of each issue and measures of religion or religiosity: evangelical versus Catholic identification and church attendance. The figures show that when abortion and same-sex marriage are in the news a great deal, religious cleavages emerge, supporting H1b. As expected, abortion coverage more strongly shapes the cleavage between religious and secular citizens (as measured by church attendance), while same-sex marriage coverage affects the cleavage between evangelicals and Catholics but only weakly moderates the effect of church attendance on the ideology of vote choice.
Further analysis of abortion attitudes
In Figure 2, we unexpectedly found that where there is little news coverage of abortion, people who support therapeutic abortion (that is, when the mother's health is in danger) tend to vote for right-leaning candidates. One potential explanation concerns the nature of the abortion question in the AmericasBarometer. Therapeutic abortion is a relatively limited right and is legal in most countries in Latin America (Marcus-Delgado Reference Marcus-Delgado2019). Perhaps when potential legal changes are not on the agenda, those who support the status quo tend to vote for candidates on the centre-right for other reasons.
To examine whether the limited nature of the AmericasBarometer abortion question affects our results, we turn to the 2004, 2007, and 2015 waves of the Latinobarómetro, which include a broader and more sensitive measure of abortion attitudes on the same 1–10 Likert scale as the AmericasBarometer question about same-sex marriage. The question reads, ‘Please use this card to tell me whether you think abortion can always be justified, never be justified, or somewhere in between.’ As before, this variable is recoded to run from 0 to 1, with higher values indicating a more progressive position. Our dependent variable is based on the question, ‘If there were elections this Sunday, what party would you vote for?’ Given the contemporaneous nature of this vote measure, we code the outcome variable using party-based ideology estimates from the PELA wave closest to the survey year. Similarly, our contextual measures are tied to the survey year. Lacking a measure of church attendance, we are not able to assess the role of religiosity. In the Supplementary Materials, we provide further detail on variable coding and the countries, elections, and survey waves included in the Latinobarómetro analysis.
As Figure 4 shows, the effect of abortion attitudes is substantially stronger in the Latinobarómetro analysis than in the AmericasBarometer, perhaps because Latinobarómetro's measure of abortion attitudes is finer-grained and better covers the range of possible views. In this regression, the impact of abortion attitudes rivals the gap between rightists and centrists. Moreover, at the low end of the scale of abortion coverage, people who support abortion rights are neither more nor less likely to vote for rightist candidates.
Vote choice as a function of abortion coverage, Latinobarómetro.
Source: Latinobarómetro 2004, 2007, and 2015.
The figure plots the marginal effect of support for abortion on the ideology of vote choice (higher values = rightist voting). Ninety-five per cent confidence intervals are shown. Estimates are based on the full multivariate model as shown in the Supplementary Materials.
Conjoint Experiment
In the theoretical discussion, we suggested that the relationship between sexuality politics attitudes and voting behaviour should depend not only on the salience of these issues but also on whether candidates adopt contrasting positions. In most of Latin America, abortion and same-sex partnerships were largely off the national political agenda until around 2010, when the first serious liberalization proposals were advanced (Table 1). Presence on the political agenda affects not only the salience of an issue in news coverage but also whether left- and right-wing candidates stake out opposing positions during the campaign. Moreover, even when an issue becomes salient, left- and right-wing candidates may strategically converge, as seen in Brazil's 2010 election campaign where Dilma Rousseff distanced herself from prior statements in favour of abortion liberalization after the issue flared up (Smith Reference Smith2019).
In contrast to the salience of abortion and same-sex marriage in news coverage, we lack a readily available and comparable measure of candidates' campaign stances on these issues over time.Footnote 5 Moreover, as suggested by the example of Dilma Rousseff in 2010, left-wing presidential candidates have been slow to adopt progressive positions on these issues during campaigns, so we would likely obtain less variation in a real-world contextual measure of contrasting campaign stances, limiting statistical power.
In order to test our hypothesis about the effect of issue attitudes when candidates' positions diverge (H2), we turn to conjoint survey experiments in Brazil, Chile, and Peru involving fictional candidates whose issue stances we can manipulate. As highlighted in Table 1, campaign dynamics and progressive legislative initiatives made sexuality politics issues particularly salient in these countries during the 2010s (Boas Reference Boas2023). However, major candidates have not consistently adopted contrasting positions, meaning that voters may have lacked opportunities to act on their issue opinions. Our conjoint experiment gives them this opportunity while also strengthening internal validity, albeit at the expense of external validity.
Research design
The candidate-choice conjoint experiments were embedded in online surveys administered in Brazil, Chile, and Peru from 7–22 May 2019. Respondents were recruited via Facebook advertisements, a common approach to convenience sampling for experimental research in comparative politics. To ensure a diverse sample, advertisements were targeted to Facebook users in distinct strata of age, sex, and region in each country. The valid N was 1,817 respondents in Brazil, 3,732 in Chile, and 3,698 in Peru. The Supplementary Materials offers sampling details, data on representativeness, and a further discussion on using Facebook for survey recruitment.
The conjoint experiment presented subjects with a choice between two hypothetical candidates for Congress and asked which one they would vote for. Subjects read the following introductory text (the name of the office varied across countries) and then were presented with a table of candidate attributes, with the value of each attribute for each candidate randomly chosen from the two options.
Imagine that the legislative elections were this coming Sunday and that you were deciding between two candidates for federal deputy/deputy/congressperson with the following characteristics. Which candidate would you vote for?
As is standard in conjoint experiments, each profile was randomized independently of the other, so while the two profiles could differ in every attribute, as shown here, they almost always involved some common traits. The three policy positions were randomly shown first or last to prevent anchoring biases while allowing for a logical presentation of items. The order of items within the policy and non-policy blocks was fully randomized. To increase statistical power, the conjoint experiment was repeated three times for each respondent, with a new random draw of candidate characteristics each time.
Policy positions in the conjoint experiment were constrained to those that an evangelical representative might reasonably take. On economics and crime policy, evangelical issue positions span the ideological spectrum, so progressive and conservative stances are both plausible. Few evangelical politicians publicly advocate for liberalizing abortion laws, so the leftmost plausible position on this issue was to maintain the status quo, which was roughly similar across countries as of 2019, with abortion permitted in only very limited circumstances (Marcus-Delgado Reference Marcus-Delgado2019). Since laws on same-sex partnerships differed more significantly across countries, this issue was not included.
Prior to the conjoint experiment, separated from it by a block of questions about party identification and voting in previous elections, respondents were asked for their own issue positions on abortion, economics, and crime policy. Choices included the same two positions the candidates could adopt. For abortion, respondents were also offered the option of legalization.
Specification and results
Our hypothesis H2 holds that ‘in a context where candidates adopt contrasting positions, views on sexuality politics issues will be strongly linked to vote choice’. Testing this hypothesis via a conjoint experiment requires a different approach than that of standard surveys and conventional conjoint analyses. With observational survey data, where each respondent chooses among real-world candidates with fixed issue positions, we might examine whether a left- versus right-wing stance on abortion correlates with support for the left- versus right-wing candidate. In the conjoint experiment, where each candidate has a left- or right-wing stance on a given issue that varies randomly, we are interested in the effect of the candidate and respondent adopting the same issue position – just as they would in a conventional survey when, for example, an anti-abortion respondent reports voting for a right-wing candidate who would ban the procedure. This estimand differs from the Average Marginal Component Effect (AMCE) (Hainmueller, Hopkins, and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014) since it depends not on randomly assigned characteristics of the candidate per se but on whether the candidate's randomly assigned characteristics match those of the respondent. For abortion, where candidate issue positions were restricted, respondents who favour full legalization are treated as agreeing with candidates who want to maintain current laws.
Moreover, rather than the conventional effect on the probability of voting for the candidate over a randomly chosen opponent – who, with a binary variable, would be expected to adopt the same issue stance about half the time – we are specifically interested in comparing an opponent who adopts the opposite stance on the issue in question. Hence, we subset on those choice tasks where the candidates differ in their policy stances such that only one of two candidates agrees with the respondent.Footnote 6 Specifically, for respondent i, candidate profile j, and choice task k (such that the candidates in the choice task differ on the policy issue of interest), we estimate the following regression:
Vote takes on the value of 1 if the candidate's profile was chosen and 0 if it was not. For comparison purposes, we estimate separate regressions for abortion, crime, and economic policy, each of which implies a different subset of candidate pairs. Since $\epsilon _{ijk}$
will be correlated within choice tasks (if one candidate is chosen, the opponent is not) as well as respondents, standard errors are clustered on the respondent i (Hainmueller, Hopkins, and Yamamoto Reference Hainmueller, Hopkins and Yamamoto2014).
Results from the conjoint experiment offer strong support for H2. When hypothetical candidates adopt contrasting issue positions, voter-candidate agreement on abortion significantly affects voting behaviour (Figure 5). Averaging across all other candidate characteristics and issue positions, respondents are much more likely to favour a candidate with whom they agree on abortion policy than one with whom they disagree. The size of the effect – 41 percentage points in Brazil, 49 percentage points in Chile, and 32 percentage points in Peru – is of a similar magnitude to voter-candidate agreement on the economy or crime. Hence, these results present a clear contrast with earlier findings that economic and crime policy stances predict left- versus right-wing voting in Latin America while abortion policy stance has no effect (Carlin, Singer, and Zechmeister Reference Carlin, Singer, Zechmeister, Carlin, Singer and Zechmeister2015). Indeed, in Chile and Brazil, we obtained a significantly larger estimate for abortion than for economic policy, reversing these earlier findings.
Effects of policy agreement on vote choice when candidates differ: conjoint experiment.
The dependent variable is an indicator for voting for the candidate. The independent variable indicates policy agreement on each issue. Icons give point estimates, and lines give two-sided 95 per cent confidence intervals. Standard errors are clustered on the respondent. The sample is limited to choice tasks where candidates differ in their policy stances such that only one candidate agrees with the respondent.
As shown in the Supplementary Materials, we obtain slightly smaller coefficient estimates (though they support similar conclusions) when we limit the sample to respondents who favour current laws or a full abortion ban. Thus, part of what drives our results is that proponents of full legalization prefer the lesser of two evils, a candidate who favours the status quo. We expect we would obtain even larger effects for agreement on abortion if these voters could choose a candidate who exactly matched their issue position.
Implications for Redistribution-Oriented Voting
Our theoretical argument and the results of the conjoint analysis suggest that, as sexuality politics attitudes start to matter more for voting behaviour in Latin America, traditional materialist concerns about redistribution might start to matter less. For both voters and candidates, stances on the economy, crime, and sexuality politics issues are certainly correlated with one another, but not perfectly so. Given a finite number of viable candidates in most presidential elections, voters who choose candidates based on sexuality politics issues are less likely to be swayed by their support for redistribution or state intervention in the economy.Footnote 7 In analyses of advanced industrialized countries, De la O and Rodden (Reference De la O and Rodden2008) and Stegmueller (Reference Stegmueller2013) both find evidence of a ‘moral’ dimension of voting behaviour that at least partially displaces materialist voting on the basis of redistributive issue preferences or income. Similarly, we might expect that when same-sex marriage or abortion becomes prominent on the Latin American electoral agenda, they reduce the effect of redistributive preferences on voting behaviour.Footnote 8
To examine this question, we modify our prior AmericasBarometer analysis to examine whether increasing news coverage of same-sex marriage and abortion reduces the association between support for redistribution (as operationalized above) and the ideology of vote choice. The regression specification is the same as in the main analysis.
As shown in Figure 6, salient news coverage of same-sex marriage and abortion significantly modifies the relationship between support for redistribution and the ideology of vote choice. Where there is little coverage of either issue, views on inequality are significantly associated with voting for the left. The magnitude of the coefficient is about the same as that for same-sex marriage attitudes when this issue is most salient. Yet, as coverage of abortion and same-sex marriage increases, support for redistribution becomes a much weaker predictor of left-wing voting. At the highest levels of same-sex marriage coverage, the relationship is insignificant.
Vote choice as a function of redistribution issue attitudes and news coverage.
Source: AmericasBarometer 2012–2019.
The figure plots the marginal effect of each issue attitude on the ideology of vote choice (higher values = rightist voting). Ninety-five per cent confidence intervals are shown. Estimates are based on the full multivariate model as shown in the Supplementary Materials.
Discussion and Conclusion
Over the past several decades, a number of Latin American countries have witnessed dramatic changes to their policies regarding abortion, LGBTQ rights, and other issues related to gender and sexuality (Díez Reference Díez2015; Encarnación Reference Encarnación2016; Marcus-Delgado Reference Marcus-Delgado2019). On same-sex partnerships, some Latin American countries have been regional or even global leaders. Argentina was the tenth country in the world and the second in the Americas (after Canada) to legalize same-sex marriage when it did so in 2010. Changes to abortion legislation have been more limited and hesitant, but there has been a wave of reform since 2007 when Mexico City first decriminalized the procedure. Some of these changes or attempted reforms have come through legislation, while others have resulted from judicial decisions. Especially in the latter case, policy change has often occurred before widespread public support existed. Brazil's High Court legalized same-sex marriage in 2013, approximately three years before a majority of the public approved of the change. Likewise, Mexico's Supreme Court decriminalized abortion in 2021 despite substantial public opposition to the procedure.
Latin America's wave of sexuality politics reforms has had clear effects on mass politics throughout the region. While recent advances in rights have stimulated the rise of social movements opposing ‘gender ideology’ (Corredor Reference Corredor2019), there is evidence that progressive policy change can shift public opinion as a whole towards support for those policies. As Maia, Chiu, and Desposato (Reference Maia, Chiu and Desposato2023) demonstrate, highly educated voters responded to a variety of LGBTQ rights reforms in the 1990s and 2000s by becoming more supportive of same-sex marriage and LGBTQ candidates running for office. They further argue that such policy changes did not drive attitudinal polarization, measured as citizens' distance from mean public opinion in their countries.
Yet, even if policy changes reduce attitudinal polarization and boost support for progressive reforms, they can, paradoxically, increase electoral polarization around these issues by activating cleavages that were previously latent. Prior to the 2010s, differing attitudes on abortion or same-sex marriage had few repercussions for voting behaviour because these issues were largely absent from the political agenda and because candidates did not stake out opposing positions. As policy changes related to gender and sexuality have been proposed or implemented, and as left- and right-wing social movements have mobilized for and against these reforms, they have started to become major topics of debate. Even if public opinion is becoming more supportive and less polarized in the aggregate, the differences that remain now matter more for voters' decisions at the polls.
We argue that the increasing media salience and electoral contestation around sexuality politics in some countries of Latin America is prompting electoral realignments. In a region where materialist issues such as economic redistribution and crime control used to be the major drivers of voting behaviour, opinions on gender and sexuality are emerging as a new cleavage that influences decisions on election day. These changes are also shifting the implications of religion for voting behaviour in the region. When materialist issues dominated the agenda, Latin America's evangelical Christians tended to vote to the left, perhaps thanks to their lower-class social origins. But where sexuality politics has risen in prominence, this religious minority has shifted to the right. When abortion is on the agenda, evangelicals are joined by their Catholic brethren, with both traditions voting more conservatively than the non-religious. Hence, the rise of sexuality politics is prompting the consolidation of a new Christian right – a support base for candidates such as Jair Bolsonaro in Brazil (Layton et al. Reference Layton2021) – even as it may be shifting aggregate public opinion toward support for more liberal policies.
Our argument about the emergence of a new sexuality politics cleavage in Latin America has implications for the literature on ideological polarization and electoral realignment in the US in recent decades. In the US context, partisan sorting or realignment by issue attitudes has been a central driver of polarization. Sometimes, citizens changed their attitudes to match their partisanship while, at other times, they switched parties to match their attitudes on high-profile issues highlighted in the media, such as abortion (Carmines and Stimson Reference Carmines and Stimson1980; Goren and Chapp Reference Goren and Chapp2017). As the Republican Party became publicly identified with conservative religiosity, partisan sorting also increased along religious lines, and partisanship and religious affiliation became mutually endogenous (Margolis Reference Margolis2018; Putnam and Campbell Reference Putnam and Campbell2012). The growing overlap among social identities, as well as social influence within increasingly homogeneous groups (including religious communities), intensified partisan and ideological polarization (Klar Reference Klar2014; Mason Reference Mason2018).
Though there are clear parallels between the sorting processes in Latin America and the US, partisanship constitutes a key difference between the two regions. In the US realignment process, party identification serves as the glue that binds together issue attitudes, religion, and vote choice. By contrast, in the Latin American context, the binding power of partisanship may be closer to that of Scotch Tape. Party systems vary greatly across Latin America but are universally younger and more fragmented than in the US (Levitsky et al., Reference Levitsky, Loxton, Van Dyck and Domínguez2016; Mainwaring, Reference Mainwaring2018). Levels of party identification are substantially lower, parties have weaker brands, and partisanship is more fickle (Lupu Reference Lupu2016; Samuels and Zucco Reference Samuels and Zucco2018). Yet even in this seemingly adverse context, we find that elite and social movement priming of sexuality politics issues can trigger electoral realignments. In countries like Brazil, social conservatives may flit from one party to the next while their emerging tendency to support right-wing candidates remains constant.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0007123423000613
Data availability statement
Replication data for this article can be found in Harvard Dataverse at: https://doi.org/10.7910/DVN/7GIJPI
Acknowledgements
Previous versions of this paper were presented at the 2020 Annual Meetings of the American and Southern Political Science Associations, as well as Columbia University and the Charlemos series at the University of Pittsburgh. Our thanks to Javier Corrales, Mike Hendricks, Bob Kaufmann, Eduardo Moncada, Vicky Murillo, Orçun Selçuk, Metehan Tekinirk, and our anonymous reviewers for comments on previous drafts, and to Miguel Fernández, Breno Hermann, and Julie Weaver for assistance with administration of the online surveys. We thank the Latin American Public Opinion Project (LAPOP) and its major supporters (the United States Agency for International Development, the Inter-American Development Bank, and Vanderbilt University) for making the AmericasBarometer data available. We thank the Latinobarómetro Corporation and its major sponsors for making the Latinobarómetro data available.
Financial support
Amy Erica Smith's work was supported by a Carnegie Fellowship (number G-F-20-57740). Taylor Boas's work was supported by a Boston University faculty research account.
Competing interests
None.
Ethical standards
The online surveys in Brazil, Chile, and Peru were declared exempt by the Institutional Review Board of Boston University (Protocol 5141X).
Open access
