Introduction

Observers have lamented a “crisis of representation” in Latin America since the 1990s, with established parties losing ground in some countries, and with ongoing party system weaknesses in others (Reference Hagopian, Agüero and StarkHagopian 1998, Reference Hagopian2011; Reference Mainwaring, Bejarano and LeongómezMainwaring, Bejarano, and Pizarro Leongómez 2006; Reference RobertsRoberts 2013). Freed from their partisan moorings, outsider politicians and those from new political parties are seen as liable to deviate from electoral mandates and public opinion, implementing whatever policies they see fit (Reference O’DonnellO’Donnell 1994; Reference StokesStokes 2001). Clientelistic linkages are often thought to exacerbate this trend by displacing programmatic representation (Reference Kitschelt, Hawkins, Luna, Rosas and ZechmeisterKitschelt et al. 2010).

Undoubtedly, partisan weakness poses challenges for democratic representation. Yet political parties are not the only form of interconnection between citizens and elites that can foster interest representation. Nor does territorial representation—politicians prioritizing the interests of constituents in a geographically limited district—exhaust the list. When traditional institutions such as party or territorial constituency fail to channel popular concerns, other forms of citizen-elite interconnection, including civil society organizations, can still provide a basis for effective representation (Reference Hochstetler and FriedmanHochstetler and Friedman 2008).

In this article, we argue that descriptive representation, or demographic similarities between legislators and the public, can provide effective substantive representation of citizens’ concerns, especially for groups that have been traditionally excluded from the political class. Women, those of lower socioeconomic status, and ethnic, racial, and religious minorities were not always well served by older forms of partisan and territorial representation, even when these functioned most effectively (Reference HtunHtun 2016). The increasing diversity of the Latin American political class, a notable trend in many countries, has the potential to ensure that these groups’ preferences are better represented in the policy-making process (Reference HtunHtun 2016; Reference MadridMadrid 2012; Reference Schwindt-BayerSchwindt-Bayer 2010; Reference Van CottVan Cott 2007).

Yet mere demographic similarities between citizens and elected officials do not ensure that the interests of marginalized groups will be effectively represented in the policy-making process (Reference HtunHtun 2016), or even that minority citizens and legislators will hold similar opinions. When party-based representation works well, it does so not only because party members and elected officials share interests and identities, but also because parties socialize masses and elites into a common worldview and provide channels for citizens to hold elected officials accountable. If alternative forms of representation are to take up the slack for political parties, they will need to replicate these functions.

We illustrate our argument about descriptive representation using the case of Brazil. Most of Brazil’s political parties fall into the catchall or clientelistic category, where supporters share few interests and identities and there is minimal capacity to impose a common worldview on masses and elites. In this context, we expect demographic similarities between politicians and the public to provide a stronger basis for substantive representation, especially on issues affecting core group interests, such as racial or class-based preferences in university admissions. Yet Brazil’s politically underrepresented demographic groups also vary in their ability to socialize masses and elites into a common way of thinking, as traditional mass parties once sought to do. Whereas gender and race provide a weak basis for educating group members, evangelical Christians are well equipped to do so, given their comprehensive worldview, organizational capacity, vertical authority structure, and homogeneous social networks. We thus expect evangelical legislators to do a particularly good job of representing the preferences of their descriptive constituents.

Examining a variety of issues, we show that the highest levels of congruence, or alignment between the preferences of masses and elites, are almost always between legislators and citizens in specific demographic categories. By contrast, legislators are often more distant from territorial constituents and co-partisans than from the public as a whole. Among demographic categories, evangelicals generally achieve greater congruence than women, Afro-Brazilians, or those of lower social class—a product, we argue, of leadership from the pulpit. Drawing on original surveys of evangelical and Catholic clergy and congregations, we show that opinions cluster most tightly among evangelical churchgoers, but only on issues that pastors prioritize in their preaching. Our analysis breaks new ground in terms of its dyadic approach to estimating congruence—examining the difference in opinion among all possible pairings of legislator and constituent—as well as the breadth of issues and range of data sources it examines.

Our research underscores that personalistic politics—often decried for weakening party systems, diluting ideology, and displacing programmatic appeals—need not mean that citizens are poorly represented by elected officials. Voting on the basis of candidate characteristics rather than party program may indeed limit opinion congruence—and, by extension, substantive representation—along partisan lines. But personalistic voting does not necessarily imply electing the most charismatic candidate or the local political boss with a strong patronage network. For voters from historically underrepresented social groups, choosing someone who looks (and worships) like them may be a good way to elect someone who thinks similarly on major policy issues.

Opinion Congruence and Descriptive Representation

Opinion congruence, or elite-mass alignment in ideology and issue opinions, is an important factor in democratic representation. Certainly, substantive representation—elected officials acting on behalf of constituent interests (Reference PitkinPitkin 1967)—is possible even if politicians think differently from the people they represent. Legislators’ voting behavior may deviate from their own preferences for any number of reasons, including party discipline, reelection incentives, and a sense of duty. Yet personal opinions also matter, especially for lower-visibility behavior such as bill sponsorship and voting within committees (Reference Carnes and LupuCarnes and Lupu 2015).Footnote ¹ When elections produce governments, legislatures, or parties in which representatives’ preferences match those of voters, principals’ monitoring costs and agents’ temptations to shirk are both reduced. This consideration is particularly important in societies with substantial differences in education between the people and the political class, as well as for disadvantaged groups in any democracy.

The study of opinion congruence in older, wealthy liberal democracies has contributed a rich theoretical foundation, debating the effect of institutions such as proportional representation and direct democracy mechanisms (Reference Bernauer, Giger and RossetBernauer, Giger, and Rosset 2015; Reference Blais and BodetBlais and Bodet 2006; Reference FerlandFerland 2016; Reference Golder and StramskiGolder and Stramski 2011; Reference Huber and Bingham PowellHuber and Powell 1994). Nonetheless, distinct challenges prevail in newer and less wealthy democracies, where volatile party systems, limited party identification, and nonprogrammatic mass-elite linkages may inhibit congruence along party lines. In Latin America, informal institutions, high levels of poverty, and severe class divides between politicians and the public constitute additional constraints (Reference CorralCorral 2013; Reference España Nájera and Martínez-RosónEspaña Nájera and Martínez-Rosón 2012; Reference Luna and ZechmeisterLuna and Zechmeister 2005; Reference Kitschelt, Hawkins, Luna, Rosas and ZechmeisterKitschelt et al. 2010; Reference Otero Felipe and ZepedaOtero Felipe and Rodríguez Zepeda 2010).

Our research contributes to the study of congruence in old as well as new democracies via its focus on descriptive representation and opinion leadership. Few scholars have examined congruence along descriptive lines, and they have done so only in advanced democracies (Reference Kissau, Lutz and RossetKissau, Lutz, and Rosset 2012; Reference Reingold and HarrellReingold and Harrell 2010). Yet in countries that are stratified by race and class and have personalistic, nonprogrammatic party systems, electing members of underrepresented groups could potentially enhance congruence more than party voting. When civil society leaders of these groups effectively orient elite and mass opinion, congruence becomes even stronger.

In thinking about theoretical expectations for congruence, we argue that it is useful to distinguish among three sources of elite-mass opinion alignment. First, congruence might derive from self-selection when preexisting attitudes prompt citizens and/or politicians to join a group that is open to new recruits: for example, an environmentalist choosing to affiliate with the Green Party, or an opponent of same-sex marriage joining an evangelical church. Second, congruence can result from material interests that impinge on all members of an identity group. Examples include affirmative action policies for racial minorities or “pork” provided to residents of a territorial constituency. In both instances, congruence obtains fairly naturally by virtue of group membership.

A third source of congruence is socialization, or mutual influence among group members and their leaders that brings elite and mass opinion closer to one another over time. For example, an evangelical convert initially attracted by the church’s style of worship, but without well-formed attitudes on abortion or same-sex marriage, might come to embrace conservative positions on these issues after exposure to sermons and conversations with other church members. As highlighted by this example, socialization can be horizontal, involving repeated and often informal interaction among friends, family, and associates, as well as vertical, whereby group leaders seek explicitly to shape the opinions of their members.

Because socialized congruence often involves organized effort, it is less common than the other two forms. Elite socialization of the masses is most effective when a group’s leadership embraces a comprehensive worldview—one defining “correct” positions on issues across a variety of policy domains, and ideally derived from Manichaean ideological or theological principles. Socialization is also more effective when group leaders are endowed with high levels of authority; when an organizational structure facilitates regular contact with group members; and when individuals’ social networks contain a high proportion of other group members, who implicitly and explicitly reinforce messages from authority figures.

For any channel of representation, overall expectations for opinion alignment should derive from the combination of these distinct forms of congruence. The European class-mass party of the mid-twentieth century (Reference DuvergerDuverger 1959) scores highly on all three. Given its working-class constituency, shared interests should lead elected officials and members to agree on numerous issues. The fact that joining the party requires an active decision, including paying party dues, allows for congruence via self-selection. Class-mass parties also extensively socialize members and leaders, imposing ideological congruence in areas beyond those at the core of working-class interests (Reference Przeworski and SpraguePrzeworski and Sprague 1988). Vertical authority structures, homogeneous social networks (including those sponsored by the party), and regular contact through party organizations all facilitate socialization.

We expect lower levels of congruence in most contemporary political parties. Modern programmatic parties (Reference Gunther, Diamond, Diamond and GuntherGunther and Diamond 2001; Reference WolinetzWolinetz 1991), such as European Green Parties, may have high levels of congruence on some key issues because of self-selection (Reference DaltonDalton 1985). However, their capacity for socialized congruence will vary widely. Programmatic parties embrace a comprehensive worldview, yet many lack the organizational capacity to socialize members through interpersonal interaction, especially outside of electoral periods (Reference KitscheltKitschelt 1994). For their part, catchall and clientelistic parties are expected to exhibit low congruence. Catchall parties lack the well-defined worldview of programmatic parties, limiting both self-selection and socialization (Reference Gunther, Diamond, Diamond and GuntherGunther and Diamond 2001; Reference Kirchheimer, LaPalombara and WeinerKirchheimer 1966; Reference WolinetzWolinetz 1991). Elites from clientelistic parties have high levels of personal contact with voters and are often inserted into their networks (Reference SzwarcbergSzwarcberg 2015), but such parties focus on the contingent provision of private or local club goods rather than socializing voters with respect to a party program (Reference Kitschelt, Hawkins, Luna, Rosas and ZechmeisterKitschelt et al. 2010, 20).

In the case of territorial representation, expectations for congruence are likely to depend on district size. When districts are small, congruence may be fairly high, given the likelihood of shared interests and identities, the possibility of self-selection through migration, and the potential for socialization through direct contact between elected officials and citizens. As districts grow and the number of representatives per district (i.e., district magnitude) rises, our expectations for territorial congruence fall (Reference Lau, Patel, Fahmy and KaufmanLau et al. 2014). Interests are more diffuse, self-selection implies moving greater distances, and there are too many constituents to allow for effective socialization.

We can also identify expectations for demographic categories. Although full-fledged class-mass parties are largely extinct in present-day democracies (Reference Katz and MairKatz and Mair 1995), religious organizations generate similarly strong expectations of congruence, especially in countries or among denominations that are highly observant. Not only does conversion allow for self-selection, but religious elites and masses have shared interests in policies such as state subsidies for religious groups or religious freedom legislation. Socialization can also align elite and mass opinions. Religions embrace comprehensive, often Manichaean worldviews, and clergy have the organizational capacity—including regular contact with worshippers—to communicate these views to both masses and elites. In a recent special edition of this journal, Wood (Reference Wood2014, 190) rightly notes that the “lived religion” of Latin American citizens “occurs … in a relationship of tension with” religious institutions. Nonetheless, leaders are often imbued with substantial authority, and homogeneous social networks should reinforce socialization.

We have somewhat weaker expectations of congruence among women or racial and ethnic minorities. Congruence due to self-selection is implausible; we expect that policy attitudes rarely affect gender and racial identity. Masses and elites may naturally align on issues related to material interests, such as gender- or race-based preferences. Racial or ethnic minorities may also be congruent on policies benefiting the lower class when their ranks substantially overlap. Yet interest-based congruence also depends on the social construction of identities and interests. Especially in countries with more traditional social structures, women might not support gender-based preferences seen as prejudicial to husbands or sons. And in societies such as Brazil, where racial categorization is fluid and subjective (Reference TellesTelles 2004), dark-skinned citizens might oppose racial preferences if they fail to identify as black or brown.

Achieving congruence through socialization should also be more somewhat more challenging for women and racial and ethnic minorities. Htun (Reference Htun2016, 12) argues that elites become effective representatives of mass interests only through “processes of group mobilization and deliberation in civil society.” Even when women’s and minority rights organizations effectively socialize in-group politicians, they will find socializing citizens more difficult. Unlike clergy, women’s rights activists do not regularly interact with large proportions of their group members. Civil society leaders are typically endowed with less vertical authority, and women are almost always embedded in mixed-gender networks, limiting the potential for group members to reinforce messages. For racial and ethnic minorities, these factors may be somewhat more propitious: more geographically concentrated populations generate homogeneous social networks and facilitate organization and contact with leaders, who may enjoy greater authority. However, we should expect less congruence among more dispersed groups that are assimilated into mainstream society. Moreover, to build support across other lines of cleavage, racial, ethnic, and women’s movements typically eschew comprehensive worldviews, focusing more narrowly on policies directly tied to their constituencies.

Opinion congruence along descriptive versus party and territorial lines should thus depend on the nature of a country’s political institutions and the organization of demographic groups. In established European democracies, with numerous programmatic parties, a weaker capacity for religious socialization, and smaller nonwhite populations, party representation may provide the strongest basis for opinion congruence. In the United States, congruence may obtain from some combination of party and territorial representation, thanks to increasing ideological polarization of parties and a single-member district system with smaller, more homogeneous electoral districts.

In many new democracies, expectations differ. Religious organizations often have more socializing capacity than in Europe, and racial and ethnic diversity is often greater.Footnote ² Proportional representation entails large, heterogeneous electoral districts. Finally, catchall and clientelistic parties prevail, leading to low expectations of congruence along party lines. Descriptive representation may therefore provide a more propitious basis for congruence, particularly on core issues and when identity groups are well organized.

Opinion Congruence in Brazil

Our analysis of opinion congruence in new democracies focuses on the case of Brazil, Latin America’s most populous democracy. Brazil has been identified as a sustainable but low-quality democracy due to persistent inequality in the face of popular demands for redistribution (Reference Weyland, Hagopian and MainwaringWeyland 2005). In such a context, the question of alignment between elite and mass opinions is of both theoretical and normative interest. Brazil’s party and electoral systems have been faulted for the weakness of democratic representation and generate low expectations for partisan or territorial congruence (Reference MainwaringMainwaring 1999; Reference Desposato, Heller and MershonDesposato 2009). Yet descriptive representation generates greater prospects for congruence among politically marginalized groups, including women, Afro-Brazilians, those of lower social class, and especially evangelical Christians. For this reason, it constitutes a brighter counterpoint to a traditionally dim view of the quality of democracy in Brazil.

Brazil is a case in which we would expect little issue congruence along party lines. First, with the partial exception of the Workers’ Party (Partido dos Trabalhadores, PT), no major party has a cleavage-based core constituency, limiting the potential for elite-mass alignment due to shared material interests. Second, while masses and elites regularly self-select into parties, ideology and issues play a minor role in this process. Politicians tend to choose parties on the basis of career ambition and stance vis-à-vis the current government rather than policy positions (Reference Desposato, Heller and MershonDesposato 2009; Reference Samuels, Siavelis and MorgensternSamuels 2008). At the mass level, party identification is comparatively low and is not strongly related to ideology, but rather responds to government performance, scandals, current events, and personalist attachment to individual party leaders (Reference Baker, Sokhey, Ames and RennoBaker et al. 2016; Reference SamuelsSamuels 2006; Reference Winters and Weitz-ShapiroWinters and Weitz-Shapiro 2014).

Most Brazilian parties also have little capacity to socialize members into a comprehensive worldview (Reference MainwaringMainwaring 1999). The Party of the Brazilian Democratic Movement (Partido do Movimento Democrático Brasileiro, PMDB),Footnote ³ a large centrist party, is a collection of state-level political machines that has only a vague ideological profile (Reference HagopianHagopian 1996; Reference Samuels and ZuccoSamuels and Zucco 2015). The Party of Brazilian Social Democracy (Partido da Social Democracia Brasileira, PSDB), which led the opposition from 2003 to 2016, is loosely organized with few ties to its membership base (Reference RomaRoma 2002). Though recent survey experiments show that the PT and PSDB party labels can cue shifts in public opinion (Reference Samuels and ZuccoSamuels and Zucco 2014), this effect is quite distinct from real-life socialization.

We also do not have strong expectations of congruence for the PT, which is often considered an exception among Brazilian parties (Reference KeckKeck 1992). Despite its shift to the center, the PT is still best classified as a programmatic party (Reference HunterHunter 2010). Nonetheless, interests and policy views diverge within its broad array of civil society allies, including labor unions; progressive Catholics; environmentalists; and women’s, Afro-Brazilian, and landless peasant movements (Reference AmaralAmaral 2011; Reference Samuels and ZuccoSamuels and Zucco 2015). Moreover, the substantial numbers of new PT identifiers (petistas) over the past two decades—many drawn in by personal support for President Luiz Inácio Lula da Silva (Reference Baker, Sokhey, Ames and RennoBaker et al. 2016)—have few ties to organized interests. In the 2007–2012 AmericasBarometer surveys, between 81 and 89 percent of petistas reported that they never attend meetings of labor unions, community or professional associations, or women’s groups. New partisans might exhibit congruence on the issue of redistribution—arguably the policy most identified with Lula—but not necessarily on a broader agenda. Finally, though the PT continues to socialize elites into its governing vision, even expelling deviating legislators (Reference Samuels and ZuccoSamuels and Zucco 2015), it has much less capacity to socialize partisans. The “base nuclei” originally charged with political education have long since withered, and today’s municipal directorates do not maintain the same educational mission or intensity of contact (Reference AmaralAmaral 2013; Reference RibeiroRibeiro 2010). In the 2007–2012 AmericasBarometer surveys, 88 percent of petistas never attended party meetings, and 9 percent did so only once or twice a year.

Moreover, Brazil’s legislative electoral system, like many proportional representation systems, is not expected to produce high territorial congruence. Entire states serve as districts, with the number of representatives for the lower house ranging from eight to seventy. Only the three smallest states have populations approximately equal to that of a US congressional district. Most Brazilian states combine large urban centers with vast rural areas. Apart from supporting regional development and broadly targeted pork (e.g., a federal highway through the state), masses and elites from the same state should not automatically share interests and identities. Self-selection is also unlikely to boost congruence along territorial lines. Although a few people might move out of a region because they disagree with its prevailing political attitudes, internal migration in a middle-income country like Brazil is primarily driven by economic concerns (Reference LucasLucas 1997).

While Brazil’s party and electoral systems work to undermine opinion congruence and the quality of democratic representation, we argue that descriptive representation provides an alternate basis for the alignment of elite and mass opinions, especially among marginalized populations. We examine congruence among four demographic groups underrepresented in the Brazilian Congress: women, Afro-Brazilians, evangelical Christians, and those of lower social class.Footnote ⁴ While women and Afro-Brazilians each account for 51 percent of the population, and 89 percent of Brazilians twenty-five or older lack a college degree (our measure of social class), these categories constitute only 8–19 percent of elected legislators (see the appendix). Evangelicals have done better in numerical terms but are still underrepresented; they made up 15–22 percent of the population during the period under study, and 6–12 percent of Congress. As shown in the appendix, there is some correlation among membership in these groups; evangelicals are more likely than other Brazilians to be black and female (on the intersectional politics of evangelicalism and gender, see Reference HallumHallum 2003). However, the four groups are sufficiently distinct that it is meaningful to talk about them separately from one another. In the appendix, we also show that each demographic category cleaves public opinion on a variety of issues, suggesting that they are good candidates for examining the implications of descriptive representation for congruence.

Each of these demographic groups has at least one issue related to material interests on which we might expect high congruence. Women, Afro-Brazilians, and those of lower social class all have a potential interest in supporting nondiscrimination, affirmative action policies, or redistribution (Reference LustigLustig 2017; Reference Valente and BerryValente and Berry 2017). For their part, evangelical elites and masses should both support religious equality laws that grant rights and privileges on par with those of the Catholic Church. Such “natural” opinion congruence may help to explain Afro-Brazilians’ and evangelicals’ observed preference for in-group candidates (Reference Aguilar, Cunow, Desposato and BaroneAguilar et al. 2015; Reference BoasBoas 2014).

Beyond material interest, congruence will depend on potential for self-selection and civil society leaders’ capacity for socialization. Self-selection into gender, race, or class categories is uncommon and unlikely to be based on policy attitudes, especially when class is operationalized in terms of education. As argued above, women’s organizations have only limited capacity for socialization in most societies. Given their geographic dispersion, as well as national myths of racial democracy that inhibit organizing (Reference Layton and SmithLayton and Smith 2017; Reference TellesTelles 2004), we also expect Afro-Brazilians to have limited socialization capacities. Women’s and Afro-Brazilian organizations have contact with only a small share of the population and few ties to in-group legislators (Reference HtunHtun 2016). And while labor unions have long sought to generate class consciousness and socialize workers, their organizational reach pales in comparison to the 89 percent of Brazilians who lack a college degree.

By contrast, both self-selection and socialization are more likely among evangelicals. Evangelicals’ share of the Brazilian population has risen from a few percentage points in 1970 to 22 percent in the 2010 census, largely through conversion. Though most conversions are probably unrelated to politics, some newcomers might be inspired by evangelical denominations’ issue stances.

Evangelical churches are also particularly well equipped to achieve congruence through socialization. Pastors in Brazil are not shy about discussing politics (Reference SmithSmith 2019). Sermons routinely touch on homosexuality, abortion, and other issues in the legislative limelight. Evangelical worshippers, moreover, are inclined to follow clergy cues on political matters (Reference Boas, Smith, Carlin, Singer and ZechmeisterBoas and Smith 2015). Socialization likely affects evangelicals in public office as well as their nonelected coreligionists. Moreover, some prominent denominations exert direct influence over “their” politicians. The Universal Church of the Kingdom of God routinely selects members to run for office, supports their campaigns, and instructs church members on how to vote. Once elected, such politicians tend to consider themselves church rather than party delegates (Reference MachadoMachado 2006, 38–39).

In sum, Brazil generates low expectations of congruence along party and territorial lines and somewhat higher expectations for descriptive representation. We expect the greatest congruence among evangelicals, given the potential for self-selection and a strong capacity for elite and mass socialization. Yet evangelical congruence may also vary by policy area, depending on the issues pastors prioritize in their preaching.

Data and Indicators

In comparison to existing literature, our research exploits an unusually wide array of data sources to highlight congruence across a large number of policy domains. We draw on three waves of the Brazilian Legislative Surveys (BLS) and fifteen distinct surveys from Latinobarómetro (LB), AmericasBarometer (AB), and the 2010 Brazilian Electoral Panel Study (BEPS).Footnote ⁵ The BLS is an opt-in rather than random sample of legislators, but its composition is quite similar to that of the entire Congress, as shown in the appendix. We also show in the appendix that our fifteen mass surveys employ similar sampling procedures and generate comparable population estimates of basic demographics.Footnote ⁶ Our analysis covers the years 2002–2013, since earlier mass surveys from Latinobarómetro did not sample from the entire national population. Each wave of the BLS during this period (2005, 2009, and 2013) is matched to public opinion surveys from the same year and the three previous ones. Hence, we measure mass opinion when these legislators were elected as well as each year that they served in office (excepting senators, who serve eight-year terms).

Measures of partisanship are taken directly from each survey. For elites, we use party affiliation at the time of election. Our mass surveys measure partisanship somewhat differently from one another: LB asks what party the respondent would vote for in a hypothetical election, whereas AB and BEPS ask what party he or she sympathizes with.Footnote ⁷ In the appendix, we show that, for policy issues that draw substantially from both data sources, we obtain similar estimates of party congruence when using only LB or AB/BEPS.

To operationalize descriptive representation, we combine variables from the mass surveys with data external to the BLS. Lists of evangelical legislators were drawn from a variety of secondary sources (Agência DIAP 2010; Folha de São Paulo 2006; Reference GonçalvesGonçalves 2011; Reference PereiraPereira 2010; Reference PriscoPrisco 2010); from the website of the evangelical caucus, or Frente Parlamentar Evangélica; and from data collected by the lower chamber’s Centro de Documentação e Informação (CEDI).Footnote ⁸ For mass survey respondents, our evangelical category includes those self-identifying as evangelical, Pentecostal, Protestant, or any non-Catholic Christian denomination other than Mormon, Jehovah’s Witness, and Seventh Day Adventist. We code legislators’ gender and education—the latter used to operationalize social class—on the basis of candidate registration data from the Superior Electoral Court (Tribunal Superior Eleitoral, TSE). The TSE began collecting data on politicians’ race only in 2014, so for legislators who ran in that election, we identify Afro-Brazilians using self-classification as black (preto) or brown (pardo). The minority who did not run in 2014 are classified as Afro-Brazilian if they appear on the lists contained in Santos (Reference Santos2010) and UNEGRO (2011), both of which also appear to code race on the basis of self-identification. For the mass surveys, we classify as Afro-Brazilian those who describe themselves as black or brown.

To measure congruence, we focus on seven areas of overlap between the elite and mass surveys: economic regime preference (statist versus free market), ideological self-placement, abortion, gay marriage, environmentalism, and university quotas for Afro-Brazilians and those of lower income. For each measure, we use all available years and surveys in which mass and elite data can be paired to one another. Question wordings and applicable years are contained in the appendix. Most issues draw from a single wave of the BLS and several contemporaneous mass surveys. Ideology and economic regime preference draw from all three waves of the BLS and span the entire time period, 2002–2013; in the appendix we show that similar estimates of congruence are obtained when disaggregating by BLS wave. Elite and mass sample sizes vary by issue and group, as shown in the appendix.

For most issue areas, we use a single question from each available survey. For economic regime preference, we pair a single question from the BLS to a variety of mass survey questions measuring free market attitudes. For any survey with three or more such questions, we extract scores on a single factor; when there are two, we sum them to create an index. All measures are scaled from zero to one, with higher numbers indicating rightist positions. Appendix Figure 4 plots the mean differences in elite and mass positions on each measure for the various groups we examine.

Since our analysis focuses on the distance between elite and mass attitudes, it is potentially sensitive to differential item functioning between legislators and the public (Reference Golder and StramskiGolder and Stramski 2011). We effectively assume that a score of 0.5 denotes the same centrist position on both mass and elite surveys, which might not always be the case.Footnote ⁹ Differential item functioning partly results from question wording. For the two items on university quotas, wording is identical for elites and masses; for abortion, gay marriage, ideological self-placement, and environmentalism, it is quite similar. Question wording differs to a greater extent for economic regime preference. However, aggregating across multiple survey questions and years of mass-level opinion ought to mitigate the differential item functioning concern for this item.

We also made several operationalization decisions with an eye to reducing differential item functioning. We exclude from our mass measure of economic regime preference a battery of AB questions (ROS1, ROS2, and ROS3) that asks about respondents’ agreement or disagreement with statist economic positions, rather than free-market ones, as in the majority of questions we use; the concern is that acquiescence bias might affect these items differently. Following Power and Zucco (Reference Power and Zucco2012), we also recode the BLS question on economic regime preference to collapse the two most statist categories, such that a score of 0.5 corresponds to equal state and private sector participation in the economy.

For our multivariate analysis of congruence, we merge into the mass and elite data several variables from other sources. For legislators, these include age (from the TSE candidate declarations), total years of experience in the Senate and Chamber of Deputies (collected from the legislative biography pages), and the log of each candidate’s corporate donations as a percentage of total corporate donations per seat (federal deputy or senator) in that state and election year.Footnote ¹⁰ For the mass data, we merge in the log of the population of the respondent’s municipality in the 2010 census.

Analysis of Elite-Mass Congruence

Operationalizing opinion congruence calls for important decisions about the level of aggregation at which mass and elite opinion are measured. Most existing studies examine what Golder and Stramski (Reference Golder and Stramski2011) term “one-to-many” congruence—comparing, for example, the position of a single legislator to the distribution of attitudes among constituents. This approach corresponds well to the delegate model of territorial representation in single-member district systems. Yet in the case of descriptive or party representation, or of territorial representation in a proportional representation system, citizens almost always have more than one representative. If the female public is well represented by women legislators, for example, the entire distribution of women’s opinion on a given policy issue should be similar for both masses and elites.

Given Brazil’s proportional representation system and our focus on descriptive representation, this study examines “many-to-many” congruence, comparing the full distribution of attitudes at the elite and mass levels. Yet we move beyond most existing studies of many-to-many congruence (Reference Golder and StramskiGolder and Stramski 2011; Reference Bengtsson and WassBengtsson and Wass 2012; Reference Andeweg, Denters, Rosema and AartsAndeweg 2011; Reference España Nájera and Martínez-RosónEspaña Nájera and Martínez-Rosón 2012; Reference CorralCorral 2013) by also examining legislator-voter dyads (see also Reference Lupu, Warner, Joignant, Morales and FuentesLupu and Warner 2017). Though the dyadic approach is most commonly associated with one-to-many congruence (Reference Hill and HurleyHill and Hurley 1999), it can also be adapted to the many-to-many context by examining all possible pairings of elites and masses from a particular category. By shifting the analysis to the dyadic level, we examine how congruence depends on shared partisanship, state of residence, or demographics while also controlling for characteristics of individual legislators or citizens, such as age and political experience.

Raw distributions

As a first step toward analyzing many-to-many opinion congruence, we calculate the area between the cumulative distribution functions for elites and masses, as recommended by Golder and Stramski (Reference Golder and Stramski2011).Footnote ¹¹ Mathematically, this quantity can be expressed as

${\int }_0^1\left|F_e\left(x\right)-F_m\left(x\right)\right|\mathrm{dx}$

where F_e(x) and F_m(x) are the cumulative distribution functions for elites and masses, respectively. Since issue opinion is scaled 0–1, the area between cumulative distribution functions has a theoretical range from 0, when the distribution of mass and elite opinions are identical, to 1, when masses and elites are all located at separate ends of the scale. For economic regime preference and ideology, which span more than one wave of the BLS, we apply weights to correct for cross-wave differences in BLS response rates and in the number of mass respondents paired to each BLS wave (discussed further in the appendix). In addition to descriptive categories, we calculate this measure of congruence for elites and masses from the three largest parties (PT, PSDB, and PMDB) and two largest states (São Paulo and Minas Gerais).

A first conclusion from examining differences in attitude distribution is that legislators are generally further from their partisan and territorial constituents than they are from their descriptive constituents. Figure 1 plots the elite-mass differences in distributions for each group and area. In most cases, the most congruent category—that is, the one with the smallest area between cumulative distribution functions—is a demographic one. The major exception is gay marriage, where PSDB partisans and legislators are closer than any demographic category. This particular issue notwithstanding, the least congruent category is most often a partisan one. These results corroborate a study showing low levels of programmatic party structuration in Brazil in the late 1990s (Reference Kitschelt, Hawkins, Luna, Rosas and ZechmeisterKitschelt et al. 2010), though Luna and Zechmeister (Reference Luna and Zechmeister2005) find moderately high levels of party-based congruence across the entire Brazilian political system over a similar time period.

Figure 1 Opinion congruence: Differences in distributions.

Among political parties, the PT stands out as least congruent. In several areas, including abortion, gay marriage, and ideological self-placement, petistas in the legislature cluster on the left, while the opinion of their partisans is either diffuse (ideology) or leans toward the right (abortion and gay marriage).

Among descriptive representation categories, evangelicals most often stand out in terms of congruence. In three areas—economic regime preference, ideological self-placement, and abortion—evangelicals are the closest to their descriptive constituents; in a fourth, gay marriage, they are closely behind those with no college education. The exceptions concern environmentalism, where evangelicals lie in the middle, and both forms of affirmative action, where they are the most distant demographic group. Evangelicals achieve high congruence in several areas despite this group having a relatively small number of legislators in our data set. In the appendix, we use a simulation to show that smaller elite samples tend to generate slightly lower levels of congruence because with fewer observations it is harder to match the full distribution of mass opinion.

Elite-mass dyads

To formally test for differences in congruence along demographic versus partisan and territorial lines, we disaggregate our data into elite-mass dyads. We analyze these dyadic data using weighted least squares regression, applying weights to correct for disproportionality in the number of mass respondents paired to each BLS wave, the partisan affiliation of our elite sample, and the regional distribution of several mass surveys (discussed further in the appendix). We use two-way cluster-robust standard errors to account for the same mass respondent being paired to multiple elites and the same elite respondent being paired to multiple mass respondents. The dependent variable is the absolute value of the difference in opinion for each elite-mass dyad.

Our regressions use the following specification (with i indexing masses and j indexing elites):

$\begin{array}{c}\left|\mathrm{DIF}{F}_{\mathrm{ij}}\right|= {\beta }_0+ {\beta }_1\hspace{0.17em}\mathrm{Evangelica}{l}_{\mathrm{ij}}+ {\beta }_2\hspace{0.17em}\mathrm{Femal}{e}_{\mathrm{ij}}+ {\beta }_3\hspace{0.17em}\mathrm{NoColleg}{e}_{\mathrm{ij}}\\ \hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}+{\beta }_{4\hspace{0.17em}}\mathrm{AfroBrazilia}{n}_{\mathrm{ij}}+{\beta }_5\hspace{0.17em}\hspace{0.17em}\mathrm{SamePart}{y}_{\mathrm{ij}}+ {\beta }_6\hspace{0.17em}\mathrm{SameStat}{e}_{\mathrm{ij}}\\ \hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}+{\beta }_{7-9}\hspace{0.17em}\mathrm{Control}{s}_i+ {\beta }_{10-12}\hspace{0.17em}\mathrm{Control}{s}_j+ \hspace{0.17em}{ϵ}_{\mathrm{ij}}\end{array}$

The first four variables in the equation are indicators for respondent pairs in which both belong to the corresponding category, so the reference group includes all other combinations. For economic regime preference and ideological self-placement, which span years in which respondent race was and was not asked, we estimate one specification that includes the indicator for Afro-Brazilian descriptive representation (analyzing years 2007–2013) and another that drops it (analyzing years 2002–2013). The fifth and sixth variables in the equation are indicators for elites and masses being from the same party and state, respectively. At the dyadic level, we can group together all partisan and all territorial pairings—something that was not possible when examining differences in distributions, where we were limited to categories with sufficient numbers of elite respondents. Hence, these regressions allow us to compare each form of descriptive representation to partisan and territorial representation as a whole, rather than to particular parties and states that, though large, might be unusual.

In terms of controls, we include several variables at the elite or mass level that ought to affect congruence. If citizens support politicians because of clientelistic ties, they might rely less on programmatic linkages and, hence, be less congruent on policy attitudes (Reference Kitschelt, Hawkins, Luna, Rosas and ZechmeisterKitschelt et al. 2010). Thus, we include a BLS variable measuring the degree to which legislators think that voters demand clientelism.Footnote ¹² In addition, we control for normalized corporate campaign contributions, as defined above, since legislators who rely heavily on corporate campaign donations may deviate from public opinion due to commitment to particularistic interests (Reference Boas, Daniel Hidalgo and RichardsonBoas, Hidalgo, and Richardson 2014). Finally, political experience may make legislators more familiar with their constituents’ opinions, facilitating congruence, so we also control for years in office.

At the mass level, a variety of factors should affect congruence. We expect greater congruence among citizens who are more interested in politics, as they should be more likely to voice their opinions, vote on the basis of programmatic criteria, and follow elite cues (Reference ZallerZaller 1992). Political interest was asked inconsistently across our mass surveys, so we substitute news consumption (self-reported frequency from all available media sources, including television, radio, newspapers, and the internet).Footnote ¹³ We also expect that legislators who aim to mirror public opinion will pay more attention to urban areas where votes are concentrated and polls are conducted, as well as to younger voters who constitute the bulk of the electorate; both dynamics should affect congruence. Therefore, we include controls for age and for population (logged) of the respondent’s municipality. Finally, for issue areas involving more than one mass survey, we employ survey fixed effects.

In interpreting the results of these regressions, we focus on the difference in coefficients between each descriptive representation category and the indicators for elites and masses being from the same party or same state. Figure 2 plots point estimates and 95 percent confidence intervals for these differences (a full table of coefficients can be found in the appendix). A negative value indicates that elites and masses in the descriptive representation category are less distant from one another—hence, more congruent—than those from the same party or state.

Figure 2 Opinion congruence: Effect comparisons.

Note: Negative values mean that the first group is more congruent than the second. Solid dots indicate differences that are statistically significant at p < 0.05.

Results from these regressions are largely consistent with our theoretical expectations. First, descriptive representation often leads to greater levels of congruence than does party or territorial representation. Admittedly, there are a number of insignificant differences between coefficients for demographic and state or party pairings. For some issues—such as abortion and the environment, each of which is covered by only one mass survey—low statistical power may be to blame. However, where coefficient differences are significant, these are mostly negative, meaning that the demographic pairing generates more congruence than a party or territorial match. Race- and class-based quotas in higher education are exceptions, in which party representation fares comparatively well.

Second, evangelicals stand out most frequently among the descriptive categories. They are significantly more congruent than both state and party matches with respect to ideology, abortion, and gay marriage, though they are significantly less congruent on the two affirmative action issues. To put these differences in perspective, we can scale, by the distribution of the dependent variable, the absolute value of the difference in elite and mass opinions. For both abortion and gay marriage, evangelical legislators and citizens are about half a standard deviation closer to one another than elites and masses of the same party, and about one-third of a standard deviation closer than those from the same state. For ideology, they are about one-quarter and one-seventh of a standard deviation closer, respectively.

With respect to control variables (results reported in the appendix), we find support for several of our hypotheses. On multiple issues, older citizens are significantly further from legislators, and those from larger towns are significantly closer. News consumption is associated with being less congruent on a variety of issues, contradicting our expectations. Elite controls are mostly insignificant, likely due to the smaller samples of the BLS. To show that our main estimates do not depend on the inclusion of control variables, we also estimate a version without them. Results, reported in the appendix, are essentially unchanged.

Mechanisms of evangelical congruence: Conversion or socialization?

We hypothesized that evangelicals would be the most congruent category of descriptive representation for two key reasons: the potential for self-selection via conversion, and churches’ ability to socialize masses and elites, which should align opinions on other issues. In this section, we examine both mechanisms. We argue that socialization plays a major role in evangelical congruence, whereas there is little evidence for self-selection.

If self-selection leads to a group’s high levels of congruence, it could be because agreeing with elites influences the decision to join the group or because disagreeing affects the decision to leave. In Brazil, evangelical self-selection almost exclusively involves conversions into rather than out of the faith. In a 2014 survey by the Pew Forum on Religion and Public Life, only 3 percent of Brazilian nonevangelicals reported being raised evangelical and later leaving the religion. However, conversions to evangelicalism are routine. In Pew’s surveys in 2006 and 2014, around two-thirds of Brazilian evangelicals reported that they were raised in or previously practiced a different faith, primarily Catholicism.

It is clear that few evangelicals are leaving the faith, whether because they disagree with elites or for other reasons. Thus, if self-selection is the mechanism underlying evangelical congruence, it would have to take the form of newcomers being closer to elite opinion than lifelong evangelicals and, by their act of conversion, making the faith as a whole more congruent. However, as Table 1 shows, the issue attitudes of both groups are indistinguishable from one another in the 2006 and 2014 Pew surveys. Across all relevant issue areas—economic regime preference, ideology, abortion, and gay marriage—there are no significant differences in attitude between lifelong evangelicals and converts. Consequently, elite-mass congruence, operationalized through our difference in distributions measures, is nearly identical, as shown in the appendix. There is also no consistent pattern to the small differences that do exist; for some issues and some years, converts are more congruent, whereas for others, lifelong evangelicals are closer to elites. We conclude that self-selection via conversion does not explain evangelicals’ greater opinion congruence in Brazil.

Table 1 Issue opinion: Lifelong evangelicals versus converts.

Note: All variables scaled 0 (left) to 1 (right). P-values are from two-sided difference-in-means t-tests assuming equal variances. Data are from surveys by the Pew Forum on Religion and Public Life: “Spirit and Power: A 10-Country Survey of Pentecostals” (2006) and “Religion in Latin America” (2014).

To test whether churches’ socialization efforts impose congruence on the evangelical public, we turn to surveys of Catholic and evangelical clergy and congregations conducted as part of the 2014 Churches North and South Project (Reference SmithSmith 2019). The primary location of the project was Juiz de Fora, Minas Gerais, a medium-sized Brazilian city with a religious composition essentially identical to that of the country as a whole. The congregation-level study surveyed attendees at the worship services of eight Catholic and evangelical churches, as well as members of the general public at nearby health centers and a shopping mall. The survey of clergy interviewed ninety-seven evangelical and Catholic priests. More details on the sampling and interview procedures for both surveys are contained in the appendix.Footnote ¹⁴

If churches’ socialization efforts explain evangelicals’ higher-than-average levels of congruence, we have several expectations. First, we should find that the issues where evangelicals are highly congruent are those that pastors emphasize more in their preaching. Second, if evangelical pastors are successfully orienting the flock, we should see less variance in attitudes towards prioritized issues within particular evangelical congregations than within comparable Catholic congregations or groups drawn from the general public. On issues that are not the focus of socialization efforts, we would not expect evangelical opinion to vary less than that of other groups.

As expected, the clergy survey shows that evangelicals preach more frequently about homosexuality and abortion, where congruence is high, than about protecting the environment, where congruence is lower (Figure 3). Especially with respect to homosexuality, evangelical churchgoers receive much more socialization than Catholics. Meanwhile, they hear significantly less about racism and the environment than their Catholic counterparts.

Figure 3 Frequency of discussion of selected topics among Juiz de Fora clergy.

To analyze the variance of attitudes among congregations, we conduct a variance function regression (Reference Western and BloomeWestern and Bloome 2009). For each issue area, we first regress opinion (scaled 0–1) on indicators for each church or community site. We expect congregations to follow their pastors, but pastors themselves may convey different messages from one another. We then regress the squared residuals from this first model on indicators for type of site—Catholic congregation and general community, with evangelical as the excluded category.Footnote ¹⁵

On the basis of these models, we can calculate the mean predicted squared residual for each group and policy area. Figure 4 plots point estimates and 95 percent confidence intervals for these quantities; full tables are in the appendix. As expected, the variance of attitudes on gay marriage is much lower in evangelical congregations than within Catholic congregations or the general public. Evangelicals also have the lowest variance on abortion attitudes, though the difference is only significant with respect to the general public. Meanwhile, there are no differences in the variance of attitudes on the environment, and for racism, evangelical attitudes are significantly more dispersed than those of the general public.

Figure 4 Variance in support for policy issues, by type of community site.

Our analysis of the clergy and congregation surveys suggests that evangelical churches achieve congruence through socialization on issues they prioritize in their preaching. We find no evidence that congruence is merely an artifact of particular issue opinions influencing the decision to convert.

Conclusions

Whether and under what circumstances elites effectively represent citizens’ views has been a core empirical and normative concern for scholars of Latin America. Electoral democracies across the region have often been characterized by great divergence between citizens’ preferences and the policies adopted and implemented by elites. This phenomenon affects democratic quality and, at the extreme, may actually endanger democratic institutions themselves. The collapse of established party systems in some countries, and their continued weakness in others, has been seen as a worrisome contributor to this crisis of democratic representation. Moreover, the high-magnitude proportional representation systems used in much of the region provide a limited basis for substantive representation along territorial lines.

In this article, we argue that demographic similarities between citizens and elected officials provide an alternative basis for representation, especially for politically marginalized groups. Our analysis of mass and elite survey data from Brazil shows that, across a variety of issue areas, congruence is almost always higher for legislators and citizens from one or more historically underrepresented demographic groups than for those of the same state or party. In some cases, core interests may be at play, as with class- or race-based preferences in university admission. Yet socialization matters for congruence across a broader array of issues, and evangelical Christians stand out. Examining original surveys of clergy and congregations, we argue that evangelical congruence depends on whether pastors prioritize an issue in their preaching, which effectively orients the masses, or discuss it less often, which allows heterogeneous opinions to persist.

Our analysis also suggests that when ideologically focused, programmatic parties shift to the center and gain supporters over time, they may become substantially less congruent. For several issue areas, particularly abortion, gay marriage, and ideological self-placement, PT legislators cluster on the left, whereas the opinion of their partisans is more highly dispersed. We suspect that petistas in Congress were more congruent with their smaller, more left-leaning base of voters and identifiers in the 1980s. Yet as the party has shifted to the center and gained new supporters, many PT legislators have retained strong noncentrist attitudes on a variety of issues, leaving them further, on average, from their expanded partisan base. The same might be true of the moderate Left elsewhere in the region, such as Uruguay’s Frente Amplio and Chile’s Socialist Party, as well as with right-wing parties that have moved in a catchall direction in recent years, including Mexico’s Partido Acción Nacional and Chile’s Renovación Nacional.

Beyond the case of Brazil, our results suggest a new research agenda for scholars of Latin America. Do demographic categories provide the strongest basis for representation across the entire region? What party system and institutional variables affect congruence along party, territorial, and descriptive lines? If socialization increases congruence over time, can we test this hypothesis using panel studies of issue attitudes and exposure to opinion leaders? These and other questions should be important to the study of democratic representation in Latin America in the years to come.