World Politics

I. Introduction

THE most basic assumption of redistributive politics is that the poor favor social welfare spending and the rich resist it. It follows that income predicts support for redistribution, and that the poor vote for politicians who champion it. But this theory of redistributive demand flops where it should operate most seamlessly—in Latin America, one of the most unequal regions of the world. Public opinion surveys show that the poor in Latin America are no more likely to support government efforts to reduce inequality than the nonpoor.¹ Although inequalities in political power may explain why the poor are unable to enact welfare state expansions, such distortions can’t explain the underlying puzzle about social demands: Why don’t the poor want to soak the rich? Or, conversely, why do the rich support spending on the poor?

My argument is that the poor only have an economic interest in supporting social expenditures in contexts where they expect policies to redistribute resources or risks in their favor. This condition is usually fulfilled (to some degree) in advanced industrial democracies, but less so in much of the developing world. In Latin America, social expenditures [End Page 555] historically have done little to aid the poor. Although governments have made significant progress in providing a basic social safety net for their citizens, legacies of truncated welfare states persist. I define truncated welfare states as those that invest heavily in employment-based social insurance; provide flat or regressive transfers; and maintain informal access barriers that limit the benefits provided to the poor.² Social policy exclusion reinforces skepticism about state redistributive efforts among poor voters. Rationally, the poor show less support for redistribution when they do not expect to benefit from it. The rich express limited opposition when they themselves receive benefits from it and foot less of the bill.

Empirically, then, the relationship between income and support for redistribution is contingent on how benefits and access are structured in a country and policy area. Using public opinion data from across Latin America and an original survey conducted in Colombia, I test the impact of welfare truncation in four ways. First, I demonstrate that the receipt of welfare benefits is associated with stronger support for redistribution across Latin America. Second, I show that countries with welfare programs that cover the poor have more polarized redistributive preferences: the poor support benefit expansions that the rich oppose. Third, I hold constant broad features of the political environment that could confound the relationship between welfare state structure and preferences by looking across different social policy areas in Colombia. Income strongly predicts attitudes when respondents are asked about policies that target the poor; it weakly predicts attitudes when coverage of the poor is more limited. Fourth, I probe expectations about welfare benefits. The majority of poor respondents don’t think that they benefit from welfare programs writ large or that the rich pay for them. Poor voters receive less, expect less, and demand less from the welfare state.

This study, while intuitive, reverses a critical mistake in how political economy models have been applied to developing countries. Canonical material-interest models assume that redistribution and popular demand are substitutes.³ In other words, the poor mobilize to demand greater expenditures in countries with inadequate welfare states and the rich strongly resist expansions. The empirical implication is that inequality is associated with more polarized preferences and more popular support for redistribution. And yet there is almost no relationship between inequality and the polarization of preferences across Latin [End Page 556] America. Instead, as I demonstrate, welfare state truncation and the limited polarization of redistributive attitudes are political complements. In many unequal societies, important welfare programs exclude the poor, which dampens the poor’s support for redistribution.

I don’t make a causal claim, but I stress the reinforcing equilibrium created by truncated welfare policies. Limited coverage and access barriers lead the poor to doubt that redistribution will serve their interests (the welfare state structure causing preferences), and tepid demand for redistribution can reinforce the exclusionary nature of social policy (preferences causing the welfare state structure). One implication to which I return in the conclusion is that political parties that include outsiders in the welfare state can stimulate societal demands and thereby strengthen redistributive coalitions.

A second contribution of this article is to offer a bridge between the often-divorced literatures on welfare regimes in developing and advanced economies. On the one hand, behavioral research on advanced industrial democracies recognizes that attitudes are shaped by welfare program design.⁴ In a similar spirit to this article, Pablo Beramendi and Philipp Rehm show that variation in fiscal progressivity explains why income is a better predictor of attitudes toward redistribution in some European countries than in others.⁵ On the other hand, a growing literature on social policy in developing countries documents the more limited scope and depth of formal welfare benefits.⁶ My work explicitly connects differences in welfare state structure to their behavioral consequences in a context in which welfare state incorporation shows far greater variation. I stress the pivotal role of welfare truncation, operationalized by whether individuals receive any benefits from the welfare state. I also move beyond the focus on formal program design to consider informal access barriers that shape real and expected benefits in much of the developing world. The takeaway is that even unequal democracies characterized by clientelistic relations, informal labor markets, and weak political parties have the standard income-based redistributive attitudes when social policies downwardly redistribute resources to the poor. The snag is that many developing welfare states do far less than advanced democracies to aid the poor. [End Page 557]

II. The Puzzle

The materialist model of preferences, as formulated by Allan Meltzer and Scott Richard, assumes that the gap between the median voter’s preferences and mean income determines individual support for redistribution.⁷ Scholars extend this intuition to the cross-national level to predict a larger welfare state in unequal democracies⁸ and greater redistributive conflict in unequal societies.⁹ Synthesizing, the model has three predictions: (1) the poor support more welfare spending than the nonpoor, (2) preferences are more polarized by income in unequal societies, and, more tentatively, (3) unequal societies support more redistribution on average than equal ones. These predictions should find clear support in Latin America, where the income distribution is among the most unequal in the world, but they do not.

First, the poor in Latin America don’t support more redistribution than the nonpoor. The usual expectation is that regressing redistributive attitudes on income (plus a small set of controls like gender, education, and age) results in a negative coefficient, that is, individuals with higher incomes have less support for redistribution. I refer to this quantity of interest as the income coefficient, and think of it as a proxy for the polarization of attitudes between the rich and poor. While some scholars find the expected negative income coefficients in Latin America,¹⁰ others find no relationship between income and preferences.¹¹ The inconsistency of these results contrasts with advanced democracies, where a robust negative relationship (although varying in magnitude) between income and support for redistribution has been found.¹² Beyond statistical significance, the substantive effects of income are usually tiny in Latin America. Merike Blofield and Juan Pablo Luna conclude that while income may predict attitudes in some models and data sets on Latin America, its “significance is less consistent across countries and over time, and the predictive power of the models is weaker overall” than in advanced industrial democracies.¹³ Indeed, as I explain below, I find that the income coefficient is positive in a fifth of Latin American cases (nineteen of ninety country-years), meaning that the rich actually support more redistribution than the poor. [End Page 558]

Second, the Meltzer-Richard model predicts that the gap between the preferences of the poor and nonpoor should be larger in unequal societies. Quite simply, the poor want to soak the rich even more forcefully when income is concentrated at the top. Empirically, then, the income coefficient should be more negative in countries with high levels of inequality. Panel (a) of Figure 1 plots level of inequality measured by the Gini coefficient based on market income (before government taxes and transfers) on the horizontal axis. The vertical axis shows the income coefficient for the most standard measure of redistributive attitudes, namely whether an individual agrees that the government should take action to reduce inequality. In contrast to standard expectations, there is no relationship between inequality and the income coefficient across countries in Latin America.

Third, the standard material-interest model can be extended to predict that a larger share of the population supports redistribution in unequal societies. A mean-preserving spread in market-income distribution implies that a larger poor majority stands to benefit from income redistribution. As Karl Moene and Michael Wallerstein put it, “Welfare policy is expected to ‘lean against the wind’ in the sense that the greater the inequality of pre-tax and transfer income, the greater the electoral support for government policies that redistribute from rich to poor.”¹⁴ Some studies find that a larger share of the public in Latin America supports redistribution than it does in Europe.¹⁵ But the results do not hold within Latin America. I measure the level of support for redistribution by looking at the share of a country’s public that agrees or strongly agrees that the government should take firm measures to reduce inequality. Against expectations, panel (b) shows that when the outlier Venezuela is excluded, there is a negative correlation between market inequality and the share of the population that strongly supports redistribution. Some of the most equal countries, such as Argentina and Uruguay, have the highest levels of support for redistribution, while the most unequal societies, such as Guatemala and Honduras, have the lowest.

These stylized facts lead many scholars to jettison material-interest models. On the one hand, some scholars emphasize that the poor are uninformed about their material interests. Kenneth Roberts, for example, calls Latin America a region of “classless inequality.”¹⁶ Without the mobilizing power of unions and strong left parties, the poor, and [End Page 559]

Click for larger view
View full resolution

Figure 1.

The Weak Explanatory Power of Inequality for Redistributive Preferences^a

^a Predicted income coefficient (a) and share of respondents that strongly support redistribution (b) are based on pooled data from AmericasBarometer 2008–16. Controls are education, gender, and age. Market inequality comes from the Standardized World Income Inequality Database (swiid 5.1), averaged over 2008–14 (or available years), described in Solt 2016. Linear relationships are shown excluding Venezuela, which is an outlier due to the unreliable nature of inequality statistics.

[End Page 560]

particularly those with informal labor contracts, struggle to understand their class interests.¹⁷ Informal insurance systems through religious, clientelistic, or family networks may also dilute more programmatic welfare demands among poor voters.¹⁸ Moreover, the poor in unequal societies can become naturalized to high levels of inequality.¹⁹ Other scholars focus on the ways that ethnic divisions cleave the poor, leading poor whites to oppose redistribution to uphold their status.²⁰ If any of these explanations hold, then poor voters voice limited support for social expenditures of all types.

On the other hand, scholars have proposed that the rich have instrumental reasons to support redistribution, especially in unequal societies. The upper class may want to minimize societal conflict or crime, dislike living in unequal societies, or simply feel altruistic toward the poor.²¹ In weaker states, the rich may also be taxed less, boosting their support for expenditures for which they pay little.²² Many developing countries fund the state partly through commodity rents that reduce the direct costs of social spending.²³ From this perspective, the elevated support of the rich explains the weak relationship between income and redistributive preferences in many developing countries.

I don’t discount that these other factors shape redistributive politics. Rather, the goal of this article is to see how far material-interest models can take us when properly applied. I return to a very simple point: material-interest models assume that tax and spending policies are redistributive. Yet most Latin American countries failed to live up to this assumption for most of the twentieth century and still do much less to redistribute resources and risks than their counterparts in advanced economies. Informal barriers due to uneven state reach, documentation requirements, and bureaucratic obstacles further restrict the poor’s access. Modifying the material-interest model to allow for differences in the expected impacts of redistribution leads to a very different set of behavioral implications in developing countries. [End Page 561]

III. A Theory of Diminished Expectations

Political economists make what Torben Iversen and David Soskice label a “nonregressivity assumption”: the rich always lose from welfare policies and the middle class does less well than the poor, but better than the rich.²⁴ Indeed, advanced democracies substantially favor the poor in their tax and transfer policies.²⁵ This logic holds for both redistributive transfers and social-insurance spending. Moene and Wallerstein emphasize that social-insurance programs draw stronger support from the middle class because those people stand to lose more income in the event of a health shock or job loss.²⁶ But Rehm, Jacob Hacker, and Mark Schlesinger demonstrate that nonregressivity holds even when considering social-insurance expenditures in advanced capitalist societies because low-income households face greater risks and pay less in taxes.²⁷

The nonregressivity assumption is more tenuous in Latin America and most of the developing world. Figure 2 plots inequality against a standard measure of redistributive impact—the change in the Gini coefficient through taxes and transfers (absolute redistribution).²⁸ Redistribution reflects both how much money governments spend (the size of the welfare state) and who pays and who receives the benefits (the progressivity of the welfare state). What stands out is that Latin American welfare states do very little to reduce the gap between rich and poor. On average, taxes and transfers reduce the Gini coefficient by eighteen points in Europe and North America but only by three points in Latin America. Contrary to expectations, the relationship between inequality and redistribution is relatively flat in Latin America.

Regional averages hide substantial variation in the redistributive nature of spending within Latin America. Government spending leaves inequality basically unchanged in Colombia and Guatemala, but improves the income distribution by nine points in Uruguay. The redistributive effect also differs by policy area: public education and health provision play the greatest role in reducing inequality in Latin America, whereas pensions increase it.²⁹ [End Page 562]

Click for larger view
View full resolution

Figure 2.

Redistributive Impact Compared across Latin America, the OECD, and Other Non-OECD Countries^a

Source: swiid 5.1

^a Chile and Mexico are in the oecd, but are plotted in the Latin American regional category to highlight the regional differences.

A related second assumption of material-interest models is that welfare states begin with a commitment to the absolute poorest and differ in how they extend benefits up the income ladder. Jonas Pontusson captures the idea: “Means-tested social assistance constitutes the minimalist core of the modern welfare state—even the least welfare-oriented societies must somehow take care of the indigent—and it is the extent to which they have gone beyond the minimalist core that distinguishes the social market economies from the liberal market economies.”³⁰ Put otherwise, the “minimalist core” assumption is that welfare states cover the poor and differ in their inclusion of the middle class, or what’s often thought of as universalism in the European context.

Latin American welfare states didn’t originate with a minimalist core to provide a safety net for the poor. Quite the opposite, they began in the early or mid-twentieth century with social-insurance programs for public sector and organized workers. The origins of social spending in occupation-based protective schemes aren’t unique to Latin America. [End Page 563] As Gøsta Esping-Andersen underscores, several European welfare states (most notably, Otto von Bismarck’s Germany) began as conservative welfare regimes that linked welfare benefits to labor-market status and created stratified benefits.³¹ But the small size of the industrial working class in Latin America meant that tying benefits to labor contracts left most of the population in the informal and rural sector without social protection.³² In contrast to other parts of the developing world, as they endeavored to promote domestic industry, Latin American governments provided substantial entitlements and labor protections to the industrial working class. Welfare states grew to be “deep but not wide,” to quote Stephan Haggard and Robert Kaufman.³³ Historically, Latin American governments invested in social policies that provided benefits to those above a minimum level of income and risk, rather than below it.

Many Latin American countries still struggle to provide the most basic income and insurance to the poor. Countries differ in how far down the income ladder they’ve moved to extend benefits. Contrast the description of Latin America’s welfare states offered by Alberto Díaz-Cayeros and Beatriz Magaloni to the minimalist core assumption made about Europe: “Social policy in Latin America has traditionally failed to benefit the poor. . . . Latin America’s biggest social policy-challenge is to extend benefits to those who are now excluded from social-insurance programs—or in other words, to reach the poor.”³⁴

Particularly since the 1990s, Latin American countries have reformed and expanded welfare programs to cover poor majorities. Almost every government in the region now runs some form of means-tested income transfer program. Some governments have also moved to delink social-insurance benefits from labor status and have extended coverage through noncontributory pensions and healthcare.³⁵ Nevertheless, some countries and areas of social policy changed far less than others. Coverage and generosity vary dramatically. Consider the case of noncontributory pensions: while Brazil now covers one third of elderly households with monthly transfers of $328, the comparable program in Peru includes just 1.5 percent of the elderly and makes monthly transfers of only $46.³⁶ [End Page 564]

Latin American welfare states remain truncated for three main reasons. First, the lion’s share of the budget goes to contributory social-insurance programs. The portion of social-insurance spending has remained stable over time, even through Latin America’s economic liberalization, state cuts, and full democratization.³⁷ Contributory social-insurance policies concentrate benefits among relatively wealthy labor-market insiders. Noncontributory social-insurance programs that target the poor were layered on top of (and in some cases were even accompanied by expansions in) expensive contributory programs.³⁸

Second, redistributive transfers do less to help the poor in Latin American welfare states. Subsidies for energy costs, higher education, and home mortgages constitute important parts of social-assistance budgets in Latin America. Because richer households spend more in absolute value on these goods, these subsidies disproportionately benefit better-off households. For instance, subsidies for urban transport, gas, and electricity accounted for 10 percent of the total Argentine budget in 2010 and resulted in little downward redistribution.³⁹ Cash transfer programs for the poor are comparatively small, at just 2 percent of the annual budget.⁴⁰ Thus, a “conflicted mix” of transfers that help both the rich and the poor limits the progressivity of welfare regimes.⁴¹

A third challenge comes in what I call informal access barriers. Guillermo O’Donnell first drew attention to the uneven nature of Latin American states, identifying “brown areas” where the state is territorially or functionally absent.⁴² Rural residents must often travel prohibitive distances to reach social services, leading to divergent life prospects within countries.⁴³ The poor also struggle to apply for benefits for which they’re eligible. In their innovative work, Wendy Hunter and Robert Brill show that millions of Latin Americans lack the birth certificates needed to access social programs. Only with the extension of social protection since the 1990s have the poor started to seek such documentation.⁴⁴ Uptake can also lag due to difficult application processes. Each welfare benefit may require a trip to a different bureaucratic agency, complicated paperwork, and unclear qualifying conditions. Corruption and discretion in the application process can further reduce access. [End Page 565]

Social work systems can overcome access barriers by making it easier for the poor to learn about and apply for benefits. To take a classic example, the American welfare state expanded its welfare coverage in the 1960s and 1970s through the establishment of public social work services. Frances Piven and Richard Cloward argue that direct outreach by social workers increased successful application rates and program coverage by helping individuals gain access to multiple programs at the same time.⁴⁵ As Latin American governments have moved to include labor-market outsiders in the welfare state, some have come to realize how important social workers are to reaching the poor. In particular, Chile’s poverty-relief program (Chile Solidario) assigns households a “support worker” to provide integrated welfare assistance.⁴⁶ A creative field experiment shows that providing rural Mexicans with a “facilitator” trained to help individuals apply for social programs increased the number of benefit claims through nonclientelistic avenues.⁴⁷

The absence of social workers opens space for local politicians to act as welfare brokers. In contrast to the dominant view of political brokers providing cash or handouts in exchange for votes, welfare brokers often provide information and bureaucratic assistance.⁴⁸ Brokers help voters organize paperwork, get it to the right office, and follow up on it; they may also accompany applicants who feel stigmatized or overwhelmed when interacting with bureaucrats. For instance, local politicians in Bogotá frequently describe their job as the guides (orientadores) and navigators (tramitadores) of welfare bureaucracies. One city councillor captures a common view:

People don’t know what exists for them so you have to help …. These programs also are difficult to apply for, and so you need to help people figure out what paperwork they need to bring, why they were denied benefits, which office can fix it, and how to get the benefits they’re assigned. Many people, and especially the poorest, blame themselves if they hit an obstacle and just give up.⁴⁹

Whether due to formal policy design or informal access barriers, experiences of welfare state truncation have consequences for public opinion, as illustrated in Figure 3. It’s a common argument that social policies unleash path-dependent dynamics in which beneficiaries organize to protect and defend the tangible benefits they receive.⁵⁰ In the context [End Page 566] of benefit inclusion, this argument is familiar. But it’s less clear what happens when individuals do not receive benefits. I contend that experiences of welfare exclusion diminish the poor’s material expectations about social spending. The poor doubt that the government will change the targeting or access to allow them to benefit from future expenditures. Redistribution is understood as a distant policy for “someone else,” rather than as a visible and tangible way to reduce inequality.⁵¹ To be clear, skepticism of welfare programs doesn’t preclude the poor from making other demands, such as claims for local benefits,⁵² discretionary transfers,⁵³ or forbearance toward legal violations.⁵⁴ My claim is that individuals who perceive a more limited stake in state redistribution are less motivated to support it. The most basic empirical implication is that individuals who access welfare benefits should be more supportive of redistribution than those who do not, all else being equal.

Click for larger view
View full resolution

Figure 3.

How Welfare State Truncation Affects Redistributive Demands

The coverage of welfare benefits should also affect the polarization of redistributive attitudes at the national level. As described above, the standard prediction of the Meltzer-Richard model is that income predicts less support for redistribution, leading to a negative income coefficient. Truncated welfare states have different implications. If the poor are excluded from coverage and the upper class captures more of the benefits, then the link between income and preferences should be more tenuous. Put otherwise, income is more strongly associated [End Page 567] with preferences (the coefficient on income is more negative) in welfare states where social policies cover the poor.

In stressing the material basis of welfare expectations, I depart from a literature that emphasizes generalized distrust of the state or confusion about material interests. If the poor believe the state can’t run social programs that reach them or they struggle to understand their material interests, then they should be less supportive of all types of social expenditures. Yet social policies vary dramatically in coverage within the same country. If experiences of welfare state truncation shape attitudes, then income is more strongly associated with preferences (the coefficient on income is more negative) in social policy areas that include the poor.

I test the mechanism underlying my theory by looking at expectations of social policy benefits. There’s little direct evidence of the causal pathway that leads from experiences of welfare state benefits (or lack thereof) to expectations about benefits and positions on redistribution. My underlying claim is that the poor, and especially those excluded from benefits, do not perceive themselves as the main beneficiaries of redistribution in truncated welfare regimes. I also probe whether expectations about who benefits are associated with support for redistribution. If the poor have diminished expectations, it follows that concerns that the poor do not receive benefits should be associated with less support for redistribution, and identified as the main risk to increased social expenditures.

In short, the poor of Latin America face variable—and sometimes very weak—material incentives to support redistribution, based on the welfare state structure in which they live. Differences in the extent of welfare state truncation should shape the poor’s expectations about future benefits and redistributive demands.

IV. Patterns of Redistributive Preferences across Latin America

In this section, I test my predictions about how variation in welfare coverage at the individual and national level affects redistributive demands. To do so, I draw on five waves of survey data from the AmericasBarometer of the Latin American Public Opinion Project (lapop): 2008, 2010, 2012, 2014, and 2016.⁵⁵

lapop uses the most common operationalization of redistributive demand, namely whether an individual agrees or disagrees that the [End Page 568] government should take firm actions to reduce inequality. Responses are measured on a Likert scale that ranges from “strongly disagree” (1) to “strongly agree” (7) (redistribution). A drawback of this question is that it doesn’t mention the policy’s potential costs. Support likely would be lower if, as is often the case, inequality reduction required some costly tradeoff (in the form of either higher taxes or cuts in other parts of the budget). Ceiling effects, in which most respondents agree to some extent with costless inequality reduction, can also occur. Many scholars therefore recode this question to differentiate between respondents who strongly support redistribution (6 or above) and all others. I use the full coding and show in the supplementary material that the results are robust to a binary specification.⁵⁶ I examine alternative questions that make the costs explicit in my original survey.

Given that questions about inequality reduction are framed in abstract terms, it may seem puzzling to argue that social policy inclusion shapes responses. Wouldn’t individuals express their preferences about the policies they’d like to see enacted? I draw on John Zaller and Stanley Feldman to suggest that the way in which the public responds to survey questions—even quite abstract ones—is shaped by the existing policy environment.⁵⁷ More concretely, when answering survey questions, respondents consider whether inequality reduction involves policies that benefit them or cost them money. Current policy inclusion matters because it anchors these interpretations.

The first implication of my theory is that respondents who receive social policy benefits should have higher levels of support for redistribution than those who do not. Each survey wave includes different measures of social welfare access, such as whether a household member receives a cash transfer (cct); has health insurance (health); or contributes to a pension plan, regardless of whether the respondent draws on the benefits (pension). In 2014 and 2016, another question asks whether “you or anyone in your household receives regular assistance in money, food, or products from the government, without including pensions” (subsidy). I collapse these questions into a single measure of benefit access (benefits), which takes on a value of 1 if the respondent receives some government benefit and 0 otherwise. The intuition is that inclusion in any welfare program is likely to increase support for redistribution, so the coefficient should be positive.⁵⁸

Redistributive attitudes are shaped by present benefit receipt, as well [End Page 569] as expectations about access to future benefits. No question directly captures expectations about who will benefit, but one item in 2016 does measure the territorial reach of the social services. It asks how long it takes to reach medical attention (state reach).⁵⁹ I expect that individuals with greater access to services should be more supportive of redistribution, resulting in a positive coefficient.

Benefit inclusion should matter, holding constant an individual’s socioeconomic position. In low-income countries, wealth indicators are more accurate than self-reported income because recall of volatile income flows can be inaccurate, households tend to smooth consumption patterns, and response rates are higher. I therefore use a wealth-based measure, constructed by forming deciles from a principal-component analysis of durable items, as a proxy for income (income).⁶⁰ I also include level of education (education) as an additional measure of socioeconomic status and as a proxy for skill specificity (there are no occupation questions).

I include a standard battery of demographic controls. I control for gender (female), given that some studies find that women are more supportive of inequality reduction than men.⁶¹ I also include a measure of municipal size, ranging between rural areas (0), small cities, medium cities, large cities, and capital cities (1) (size). If cities have greater welfare access, then support for redistribution may increase with size.⁶² Alternatively, individuals who live in small communities may be more supportive of redistribution because the beneficiaries are proximate.⁶³ Older respondents (age) are more likely to draw on pension and health insurance policies, and therefore should be more supportive of redistributive spending. I also consider whether an individual self-identifies as indigenous, mulatto, or black (nonwhite). Ethnic minorities may be more supportive of redistribution due to weaker status concerns.

Figure 4 visualizes the results from an ordinary least squares regression model, including year and country fixed effects to account for the differences across waves.⁶⁴ All independent variables are standardized (except for indicator variables) so that the coefficients can be interpreted as the estimated change in redistributive support for a standard [End Page 570]

Click for larger view
View full resolution

Figure 4.

Relationship between Benefit Receipt and Redistributive Preferences^a

Source: AmericasBarometer 2008–16

^a Change in redistributive support for a standard deviation change in a covariate estimated from an ols model. Year and country fixed effects, as well as political controls in model 2, are included but not shown.

deviation covariate change. Model 1 reports the baseline results with standard demographic control variables. Consistent with my first hypothesis, benefit receipt is associated with more support for redistribution. Income is associated with less support for redistribution in the pooled data, as predicted by standard models. Education is strongly associated with more support. Model 2 adds political controls for ideology, religiosity, unemployment, vote-buying offers, corruption, and crime concerns. I’m interested in whether the impact of benefit receipt persists after accounting for possible confounders, so I don’t plot the coefficients. The supplementary material describes the variables and confirms much of what’s known from the literature: identifying with the political left and unemployment are both associated with more redistributive support.⁶⁵ Exposure to vote buying, crime concerns, and corruption are correctly signed, but are only statistically significant in some specifications. Model 3 adds the measure of state reach, which is [End Page 571] included in 2016. Individuals with greater access to health services are more supportive of redistribution.

To make these effects more concrete, consider the predicted probability of a typical Guatemalan woman strongly supporting redistribution.⁶⁶ If she receives no welfare benefits, there’s a 38 percent chance that she will strongly support redistribution. The predicted probability jumps to 43 percent if she receives some welfare benefit but doesn’t live near medical care. If she lives ten minutes from medical care, there’s a 50 percent chance that she will strongly support redistribution. The effects of benefit receipt and access are modest compared to differences based on the country in which a person lives. For example, there’s a 71 percent chance that the average Uruguayan woman who’s covered by some welfare program will strongly support redistribution.

Next, I test the relationship between national welfare state structure and redistributive support more systematically. The extent of welfare state truncation is my key national-level explanatory variable. Since there’s no consensus on how to measure truncation, I consider several approaches and concentrate on the robustness of the results across specifications. As in the case of individual benefits, I want to capture a notion of coverage, or roughly the share of the population that receives some benefit from the welfare state. To do so, I use the percentage of the population covered by social assistance or social insurance (coverage) from the World Bank’s Atlas of Social Protection Indicators of Resilience and Equity (aspire) database.⁶⁷ Efforts to extend health care, noncontributory pensions, and cash transfers all indicate greater effort to include the poor in the welfare state, and therefore should make redistribution more salient and the income coefficient more negative.

A second way to capture truncation is through the redistributive impact of taxes and transfers. Progressivity is an important proxy for the extent to which countries downwardly redistribute resources.⁶⁸ I consider the total inequality reduction through taxes and transfers from the Standardized World Income Inequality Database (swiid) (absolute redistribution), as well as a measure of the extent to which social assistance [End Page 572] and social insurance reduce inequality for the poorest quintile from aspire (benefit redistribution). I expect these measures to capture the extent to which the poor benefit from the welfare state.

At the cross-national level, I expect that the coefficient on income will be more negative in countries that do more to include the poor and engage in more downward redistribution. To test this hypothesis, I estimate a hierarchical linear model predicting redistributive attitudes with income and a small set of controls (gender, education, and age), with random intercepts and random slopes. I then recover the country-specific income coefficients (and their standard errors) from best linear unbiased predictions. These income coefficients become my dependent variable; more negative coefficients indicate greater class polarization in a country. I use the inverse of the standard errors as weights in a second-stage regression to account for the fact that the dependent variable is measured with error.

Before turning to the statistical results, I verify that the basic hypothesized patterns exist in the data. Figure 5 plots the estimated income coefficients against the coverage, absolute redistribution, and benefit redistribution measures, revealing strong negative relationships in all cases. Countries with greater coverage and progressivity have a closer relationship between income and redistributive support, as expected.

Many things cluster together at the national level, so it’s important to control for possible additional and alternative explanations. Table 1 displays the regression results, including several possible confounders. Model 1 presents the most basic specification. First, inequality is thought to result in greater conflict between poor and rich. I use the market Gini coefficient to capture inequality (inequality). Second, a larger welfare state may make redistribution more relevant to the lives of the poor (and more costly to the rich), so it should be associated with greater income polarization, just like coverage. I include a measure of social expenditures as a percentage of gdp (social exp.). But because spending levels say little about who benefits, I expect coverage and progressivity to retain independent impacts. Third, national wealth is often associated with stronger states and greater redistributive demands, so I include log gdp per capita (gdp).

Model 2 adds additional measures of state capacity. State capacity will only affect the income coefficient if it has differential effects on the attitudes of the poor and the rich. While state weakness can reduce the benefits expected by the poor, it may also dampen support among the rich. Social welfare in weak states may be less likely to resolve externalities like crime, social unrest, and indigence. I use an index of administrative [End Page 573]

Click for larger view
View full resolution

Figure 5.

Welfare State Truncation and Income Coefficients^a

Source: AmericasBarometer 2008–16, aspire, and swiid 5.1

^a Income coefficients derived from multilevel models (with controls for education, age, and gender). Measure and coverage benefit redistribution are missing for Venezuela; measures of absolute redistribution are missing for Bolivia, Ecuador, and Nicaragua.

[End Page 574]

Click for larger view
View full resolution

Table 1.

Predicting Income Coefficients with Coverage and Progressivity^a

[End Page 575]

quality from the World Bank’s Worldwide Governance Indicators (effectiveness) as a crude measure of state capacity.⁶⁹

Model 3 considers the role of ethnic heterogeneity. Societies that are more heterogeneous may divide along ethnic lines, reducing class polarization. I use the ethnolinguistic fractionalization index as a proxy for ethnic divisions (ethnicity), despite its known shortcomings, and consider an alternative measure of between-group ethnic inequality in the supplementary material.⁷⁰

Model 4 shows the role of taxation. Differences in redistributive polarization can result from the attitudes of the rich or the poor. I emphasize the poor’s attitudes because the relative importance of receiving benefits is much greater for the poor than the tax burden is for the rich.⁷¹ But the fact that the rich pay less in taxes in countries with weak welfare states could strengthen their support and flatten the income gradient. For example, Kimuli Kasara and Pavithra Suryanarayan show that the political preferences of the rich and poor—measured in terms of support for different political parties—diverge less in places with weak fiscal capacity.⁷² In a similar spirit, Isabela Mares argues that doubt that governments will collect tax contributions reduces support for social-insurance schemes even among workers who would benefit from them.⁷³ I consider the possibility that ineffectual taxation explains redistributive polarization through a measure of income taxation as a percent of gdp (income tax). A related concern, which I analyze in the supplementary material, is that some countries draw on commodity rents to fund social programs without taxing the rich.⁷⁴ Both tax measures come from the International Monetary Fund’s World Revenue Longitudinal Data.⁷⁵

A final threat to inference is the fact that welfare state structure has no independent relationship with preferences once we account for differences in left power. Left parties, as well as organized labor allies, can help frame redistributive choices and mobilize class identities. At the same time, left parties have programmatic commitments to expand welfare state coverage.⁷⁶ To capture a power resources channel, model 5 looks at the portion of years between 1945 and 2008 in which the [End Page 576] executive has been from the left (left rule), drawing on the Latin America and the Caribbean Political Dataset.⁷⁷ The supplementary material tests alternative measures, such as union density and programmatic party structuration.⁷⁸

For ease of presentation, Table 1 uses the estimated income coefficients from the multilevel model as the endogenous dependent variable. My key explanatory variables—coverage, fiscal progressivity, and benefit progressivity—are associated with more negative income coefficients in all models. Countries that spend more on social welfare also have more polarized redistributive attitudes. Greater social expenditures may allow countries to expand coverage, much in line with my theory. The supplementary material shows that the benefit redistribution results are very similar.⁷⁹ It also includes cross-level regression models in which coverage and progressivity (and their interaction with income) are used as independent variables to predict redistributive attitudes, and confirms the two-stage results.

Moving to the alternative explanations, the supplementary material shows that countries with more effective bureaucracies and greater income tax collection do have more polarized redistributive preferences.⁸⁰ But the effects lose significance in most models once coverage and progressivity measures are included. Nontax revenues have no clear association with the income coefficient. Ethnic cleavages are associated with a lower overall level of support for redistribution, but have no significant relationship with the income coefficient. Left rule, union density, and programmatic party structuration are all correctly signed, predicting greater class polarization as suggested by power resource theory—but they fall short of statistical significance.

To summarize, countries with greater welfare state coverage and progressivity have more polarized redistributive preferences, even accounting for possible confounders. But the small number of countries and imprecise measures make it difficult to rule out other explanations at the macrolevel. In the remainder of this article, therefore, I look within a single country. This approach allows me to hold constant features of the political environment and to probe the mechanisms that link benefit exclusion to redistributive attitudes. [End Page 577]

V. Cross-Policy Preferences and Diminished Expectations in Colombia

I ran an original public opinion survey in Bogotá, Colombia, because no existing survey provided suitably fine-grained measures of attitudes toward a range of social policies. Colombia decentralized governance in the early 1990s. Its social welfare programs vary in regional coverage, so I focused on the capital to maximize the comparability of responses. Although program quality and coverage can differ substantially across urban space, the assumption is that Bogotá residents have more similar experiences of welfare programs than residents in other parts of the country (and especially those affected by the civil war).

A respected polling firm administered the survey face to face to nine hundred voters. It was essential to conduct a household survey to reach the poorest segments of the city’s population, who are often excluded from online samples, and the richest, who are difficult to contact because many live in gated condominiums. The survey used a stratified sample with equal representation (three hundred respondents) of each major class group. This approach entailed oversampling upper-class groups, which are a small fraction of the residential base. In the supplementary material, I summarize the demographic characteristics of the survey respondents compared to the Bogotá population as a whole, and provide additional details on sampling procedures and response rates.⁸¹

Government class stratifications provide a way to circumvent problems with measuring income data in Colombia. The Colombian government divides the population into six socioeconomic strata (1 is the lowest and 6 is the highest) based on household features. Statistical agencies consider strata 1 and 2 lower class, stratum 3 lower-middle class, and strata 4 through 6 upper-middle class. Because these strata are used to determine eligibility for service prices, households are very aware of their class.

Colombia exemplifies a truncated welfare state with formal and informal access barriers. While the country spends 14 percent of its gdp on social expenditures (about average for the region), it only improves the income distribution by two points through tax and spending policies.⁸² But Colombia also illustrates the challenges of classifying welfare states as a whole. Social programs vary widely in their coverage and progressivity, as Table 2 demonstrates. I therefore asked respondents the standard question on inequality reduction, as well as questions about [End Page 578] their willingness to pay to fund policies that differ in their beneficiaries, such as cash transfers, health insurance, unemployment insurance, and housing subsidies. I selected social policies that differ in their coverage and their insurance or income role. My expectation is that class attitudes should be more polarized with respect to policy areas that do more to include the poor, regardless of whether they serve insurance or income purposes.

Click for larger view
View full resolution

Table 2.

Tax and Benefit Structure in Colombia, Select Policies^a

At the progressive end, Colombia implemented noncontributory health and cash transfer programs. In 1993, the country extended health insurance to the poor through a subsidized public option (Régimen Subsidiado). At the time, observers considered it “one of the most ambitious social reforms ever undertaken in Latin America.”⁸³ Mayors initially retained discretion to select beneficiaries, which politicized the targeting.⁸⁴ Nevertheless, health coverage increased from 6 percent of poor households to 70 percent, and health outcomes improved.⁸⁵ In 2000, the government launched a conditional cash transfer program (Familias en Acción) that rates highly on means-tested implementation.⁸⁶ It reaches 23 percent of households nationwide, but benefits are low in amount, only $33 per month.⁸⁷

Income taxes are also highly progressive in Colombia. But they affect [End Page 579] few individuals as less than 2 percent of Colombia’s adult population (640,000 of 37 million) pays income taxes. Colombia collects most of its revenue through value-added taxes, which are regressive in relative terms because generally speaking, the poor spend a larger portion of their income than the rich.⁸⁸

Progressive benefits have been layered on the existing contributory system. As Table 2 shows, pension payments overwhelmingly benefit upper-income groups. Less than one-quarter of Colombians receive a pension, and the top quintile of the income distribution receives more than 86 percent of government pension subsidies. Pension expenditures consume 4 percent of gdp, compared to the 0.22 percent of gdp that goes to cash transfer programs.⁸⁹ Unemployment insurance similarly favors formal-sector workers—three-quarters of the benefits go to workers who are enrolled in a contributory pension fund.⁹⁰

Transfer programs, such as housing subsidies, provide flat benefits. Colombia has a generous mortgage-interest deduction to subsidize middle-class housing purchases. To help the poor, Colombia adopted a demand-subsidy model in the 1990s. Qualifying households receive a government subsidy, which they then complement with personal savings and bank loans to purchase a house. Yet half of Colombian households have no capacity to save. Among the income bracket eligible for housing subsidies, 70 percent don’t have formal (or stable) labor contracts, and thus struggle to access commercial bank loans.⁹¹ One woman living in a squatter settlement in Bogotá captured a common frustration when she said, “Those programs aren’t for people like me . . . they’re for people with good jobs and savings.”⁹² Thus, truncation can occur through more subtle access barriers; only creditworthy employees, who are generally in the formal sector, can access housing benefits.

My expectation is that class should be more predictive of social policy attitudes for types of expenditures and taxes that aid the poor, such as cash transfers, subsidized health insurance, and income taxes. Class should be less predictive of attitudes toward unemployment insurance and housing benefits, and of answers to general questions on redistribution. To test this cross-policy hypothesis, Figure 6 plots support for spending on each social policy by class. Panel (a) shows the relationship between class strata and support for targeted expenditures. There [End Page 580]

Click for larger view
View full resolution

Figure 6.

Support for Social Policy Across Domains by Class^a

Source: Bogotá Survey

^a Fine dotted lines represent 95% confidence intervals.

[End Page 581]

are relatively sharp differences across class groups, with roughly 80 percent of the bottom two class strata in support of increased expenditures on programs like noncontributory health care, compared to 55 percent of the top strata. Preferences over the tax code are the most polarized by class, with the rich strongly opposed to “paying more taxes to fund social expenditures on the poor” and the poor in strong support.

Panel (b) compares the structure of preferences for less clearly progressive policy areas, including unemployment insurance, housing benefits, and the standard question about support for redistributive policies. Consistent with my theory, the slope on preferences is much flatter.⁹³ Figure 6 also suggests that the largest variation in attitudes comes from the poor. The nonpoor’s attitudes change little by type of expenditure, clustering around 50 percent supportive of additional spending regardless of type. The poor’s support varies widely, from 53 percent supportive of unemployment benefits to 79 percent supportive of public health subsidies.

The way in which spending is targeted seems to be a more important determinant of attitudes than whether a policy insures against risks or provides income. Moene and Wallerstein emphasize that demand for insurance increases with income because wealthier households have more to lose in the event of a job loss or health event.⁹⁴ Thus, social-insurance spending tends to gain greater middle-class support than redistributive transfers. But consistent with my theory, the slopes differ depending on whether the poor are included in insurance payments. The income coefficient is more than double (–0.17 compared to −0.06) for health insurance compared to unemployment insurance. Likewise, attitudes toward cash transfers are much more polarized between rich and poor (an income coefficient of −0.19) compared to housing subsidies (–0.09). These results are hardly dispositive, but they’re consistent with the idea that exclusion of the poor helps to explain the weak salience of income for social policy preferences.

To test the mechanism behind these results, I asked questions about the perceived beneficiaries of welfare expenditures. I didn’t want to induce a consistency bias by asking who benefits from a social policy and then asking about policy support. I therefore asked about perceived beneficiaries of social spending as a whole, probing which group gets “the majority of benefits from social expenditures, such as health, cash transfers, housing, unemployment insurance, and family benefits.” Given the [End Page 582] mixed incidence of social welfare programs, my expectation is that poor respondents don’t perceive themselves as the primary beneficiaries. Respondents selected whether they believe that most social benefits go to the middle class or to the lower class. I also framed the item differently to make the personal stakes in social programs even more explicit, asking respondents whether they believe that welfare expenditures benefit “people like you.” To generalize beyond Bogotá, I compared these questions to a similar one that was included only in Colombia on the 2016 lapop wave. It asked half the respondents “who receives the most state benefits” and the other half “who pays the most in taxes.” Individuals selected from (1) lower class, (2) middle class, (3) upper class, and (4) all classes equally. To simplify, I recoded these questions as “1” if an individual selected that the lower class benefits or pays the most, and “0” otherwise.

Figure 7 shows popular perceptions of progressivity plotted by socioeconomic class. The solid lines indicate the portion of respondents who answered that the poor benefit the most by class (strata and income deciles are rescaled from 0 to 1, with 1 being the highest). Overall, the most striking finding is how few respondents see the poor as the main beneficiaries of social welfare spending. Just 39 percent of the respondents in my Bogotá survey (panel (a)) think that the lower class benefits the most from social expenditures. The results are similar on lapop’s national sample: only 37 percent of Colombians say that the lower class benefits the most from social welfare expenditures and 41 percent believes that the lower class also pays the most in taxes. There are also clear class differences in views. Just 31 percent of the bottom quintile of the income distribution believes that the poor primarily benefit from social spending, while 46 percent of the top quintile has the same belief. These judgments clearly diverge from the assumptions of political economy models that taxes and transfers are understood to be down-wardly progressive.

Encouragingly, linkages exist between objective benefit receipt, expectations about who benefits, and redistributive attitudes. Benefit inclusion positively shapes expectations of benefits: individuals who receive cash transfers or health subsidies or who contribute to pensions are more likely to believe that redistribution benefits people like them. Furthermore, individuals who believe that they benefit are more likely to support redistribution, controlling for the same socioeconomic variables used in past models (see the supplementary material).⁹⁵ Thus [End Page 583]

Click for larger view
View full resolution

Figure 7.

Perceptions of Beneficiaries and Taxpayers by Class^a

^a Fine dotted lines represent 95% confidence intervals.

[End Page 584]

the analysis is consistent with the theory that welfare state truncation dampens expectations of future benefits and redistributive support.

To further distinguish possible mechanisms, I included a question in a follow-up survey to probe popular concerns regarding social spending. The relevant question read: “Many people want to improve the lives of the poor, but they see risks when the government tries to reduce inequality. What is the biggest risk that you see when governments reduce inequality?” Respondents selected from the following choices (given in an order that was randomized across questionnaires): social programs teach the poor to live off the hand of the state, politicians manipulate social programs to win votes, social programs do not reach the poor, government increases taxes on the middle and upper classes, and other.

Figure 8 presents the responses by class. Consistent with my theory, more than half of low-income respondents worry that welfare benefits don’t reach the poor. For the upper class, worries that benefits don’t reach the poor (34 percent) were closely followed by concerns that welfare benefits make the poor dependent on the state (28 percent) and that politicians manipulate welfare programs to buy votes (26 percent). Upper-income respondents were somewhat more concerned than the poor that taxes would increase, but this was a relatively uncommon response, confirming that tax fears are minimal. Admittedly, concerns that benefits don’t reach the poor could extend beyond issues of targeting and access barriers. For instance, respondents may be most concerned about administrative corruption or fraud by welfare applicants. Although the survey can’t distinguish these concerns, several respondents elaborated on their responses and underscored problems of informal access barriers. For example, one upper-income respondent noted: “State help is not taken advantage of by the poor due to lack of communication and bureaucracy that prevents these people from getting it. There’s no initiative to teach these people to take state assistance.”

My study of Colombia bears out my theory that the poor have diminished expectations about redistribution. Just as support for redistribution is less polarized in countries where spending is less progressive, support is less polarized in social policy domains in which formal and informal barriers exclude the poor. Respondents who do not receive benefits and do not expect to benefit are less likely to support social expenditures. Tellingly, citizens name concerns about whether the poor will receive benefits as the main reason not to expand social spending. [End Page 585]

Click for larger view
View full resolution

Figure 8.

Concerns about Increased Social Spending by Class

Source: Bogotá Survey

VI. Conclusions

Esping-Andersen famously argued that the amount of social spending is epiphenomenal to its content.⁹⁶ In some nations and policy domains, tax and spending policies do little to aid the poor. In other nations and policy domains, the welfare state is much more inclusive and progressive in terms of who pays and who benefits from state expenditures. These alternative scenarios, I argue, have very different implications for the structure of public opinion because they shape what each income group expects to receive from the welfare state. When social expenditures actually redistribute resources, the preferences of income groups diverge sharply. When social expenditures are truncated through exclusionary formal designs or through informal access barriers, the preferences of the poor and the rich look similar. Several types of empirical evidence support the argument: welfare beneficiaries are more supportive of redistribution across Latin America, income is more predictive of redistributive preferences in countries and policy areas with greater welfare state coverage and progressivity, and the poor do not expect to benefit on net from social expenditures.

These findings contribute to research on the political behavior of the [End Page 586] poor. Since Marx (at least), scholars have doubted the poor’s ability to understand their class interests without mobilizing forces like unions and programmatic political parties. The collapse of these actors in many developing countries raises questions about the poor’s ability to express and vote for their material interests. But such accounts overlook a much simpler explanation for the weaker relationship between class and redistributive preferences in developing countries. In Latin America, the poor often expect less from welfare states, and for good historical reason. Welfare states remain truncated due to their heavy reliance on social insurance, flat or regressive benefits, and informal access barriers.

Because of data limitations, I treated the relationship between public opinion and policy design as endogenous. Future work, especially as longitudinal public opinion surveys become available, may unpack this causal path to establish whether welfare state inclusion leads to attitudinal change. If the poor’s attitudes shift as social policies reach them, then substantial change is possible when political parties take the lead to expand social policy coverage. The rise of left parties and, more broadly, the extension of social programs to outsiders may pave the way to more coherent class politics.

This article’s emphasis on how public opinion is structured by the coverage and progressivity of benefits also provides a bridge to work on social policy preferences in Europe. There’s even more meaningful variation in the extent of inequality reduction when looking beyond advanced industrial economies. Figure 9 provides a preliminary view of this variation, using a combination of data from the International Social Survey Programme (issp) (which includes a small subset of Latin American and Asian countries), lapop (for waves that used the issp question formulation), and the Luxembourg Income Study (lis) (which measures fiscal progressivity). While Beramendi and Rehm show that income polarization is stronger in more progressive welfare states, Figure 9 suggests that the relationship is stronger and clearer when the set of cases is expanded beyond advanced industrial economies.⁹⁷ The implication is that behavioral models developed for advanced industrial welfare states apply to developing countries, such as Argentina and Uruguay, that have moved further to extend program coverage, but their underlying assumptions are not met in truncated welfare states in the developing world, such as Guatemala and Peru.

Although in many advanced industrial economies the primary variation in progressivity stems from formal program design, I underscore [End Page 587] the importance of informal access barriers in shaping experiences of the welfare state. Social work systems were central to the transformation of welfare states in advanced industrial economies, yet they remain underdeveloped and understudied in many developing countries. The absence of state social workers opens up space for local politicians to act as welfare brokers. Rather than directly controlling disbursements, politicians help poor voters learn about and navigate welfare programs that are, in theory, targeted or universal. Problems of uptake thus contribute to the truncated nature of welfare regimes in the developing world. Policy efforts that help the poor gain access to benefits, even without changes in formal program design, may be critical to raise expectations and cement support for redistribution.

Click for larger view
View full resolution

Figure 9.

Income Coefficients and Redistributive Impact in OECD and Select Latin American Countries

Source: issp 2006, lapop 2010, and lis Fiscal Redistribution Database 2011

Supplementary Material

Supplementary material for this article can be found at https://doi.org/10.1017/S0043887118000096. [End Page 588]