You are on page 1of 207

Poverty, Inequality and the Political Prospects of



Fabiana Velasques de Paula Machado

Submitted in Partial Fulfillment

of the

Requirements for the Degree

Doctor of Philosophy

Supervised by

Professor G. Bingham Powell

Department of Political Science

Arts, Sciences and Engineering
School of Arts and Sciences

University of Rochester

Rochester, New York


For Diego and Eduardo.


Curriculum Vitae

The author was born in São Paulo, on November 26, 1973. She attended

Universidade de Brası́lia from 1995 to 1999, and graduated with a Bache-

lor of Arts degree in International Relations. She earned an International

Certificate of Political Studies (CIEP) by the Institut d’Etudes Politiques

de Paris, France, in 2000. After a summer internship at the United Nations

Volunteer Program in Bonn, Germany, and working as an Evaluation Asso-

ciate at a non-profit organization in New York , NY for two years, she came

to the University of Rochester in the Fall of 2004 and began graduate studies

in Political Science. She pursued her research in the fields of Comparative

Politics, Mathematical Modeling and Econometrics. She was awarded the

William H. Riker Fellowship in the summer of 2005 and the W. Allen Wallis

Fellowship in Political Economy during the academic year 2006-2007. She

received a Master of Arts degree from the University of Rochester in 2007.

In the fall of 2007 she was awarded a full year pre-doctoral fellowship by the

Program on Democracy at Yale University under the direction of Professor

Susan Stokes. The following year she was invited to conduct research at the

Inter-American Development Bank in Washington, DC, where she will soon

become a Research Economist.



First and foremost I am grateful to my husband, Eduardo Leoni, who al-

ways believed in my potential as a scholar and from whom I received constant

encouragement and invaluable advice. Throughout the years he provided me

with a safe and peaceful harbor away from the stresses of graduate life. I

am indebted to my son, Diego, for his patience and responsibility, which al-

lowed me to dedicate long working hours to this project. I owe him as many

hours of quality time together, which I look forward to. I thank my parents,

Maria Cristina Velasques de Paula Machado and Sebastião Antônio Zuffo de

Paula Machado, for their efforts and sacrifices to guarantee I receive the best

possible education, opening up doors and teaching me its immense value. I

am grateful for the years spent under the guidance of Elizabeth McDonald

conducting research and evaluation on behalf of victims of crime. This ex-

perience demonstrated the importance of research and the opportunity to

use it in helping others.

This dissertation would not be possible without all I learned at the Uni-

versity of Rochester. I thank all of the great professors I had and my class-

mates, Daniel Gillion, Navine Murshid, Adam Ramey, Yoji Sekya and Mar-

tin Steinwand, for making the learning process enjoyable and effective. Even

though we now part our ways, these and many other friendships will remain

strong. I thank all the members of my dissertation committee. I am grateful

to Bing Powell for the guidance and the knowledge shared. His passion for

the study of Political Science will always be a source of inspiration. I had the

pleasure of having many great conversations and rich exchange of ideas with

Gretchen Helmke, a valued colleague in our shared interest in Latin Amer-

ica. Jim Johnson provided rich intellectual stimulation. His pragmatic and

upbeat attitude inspired a sense of confidence and trust in my own work

that was indispensable to the crafting of this thesis. To Mark Fey I owe

most of my knowledge of game theory and mathematical modeling, taught

with impressive clarity and effectiveness. They all provided me with invalu-

able comments, advice and encouragement throughout the various stages

of the project. I am greatly indebted to Paulo Barelli, on whose door I

knocked so many times with questions and doubts so diligently addressed.

I am also grateful to Randall Stone, Kevin Clarke, Curtis Signorino, John

Duggan, Hein Goemans and Tasos Kalandrakis for their teachings and guid-

ance. They taught me invaluable lessons whose scope goes well beyond the

academic realm. A special thanks goes to Pamm Ferguson, Lisa Zimmerman

and Kelly Corbet Wicks, for their kindliness and professionalism.

I would also like to thank the Program on Democracy at Yale Univer-

sity, in particular Susan Stokes, Thad Dunning and Marcelo Nazareno for

providing a stimulating intellectual environment and critical support in the

early stages of my dissertation research. I am indebted to Carlos Scar-

tascini for his support and friendship, and the Research Department at the

Inter-American Development Bank for providing an outstanding working

environment where important parts of the dissertation originated and came

to fruition, and where I was offered the best incentive a graduate student

can have to finish the dissertation: a great job.



This study addresses the question of why unequal new democracies tend

to display low levels of redistribution for long spells of time. Relative to

the more conventional accounts, I focus on two basic but crucial factors:

individuals’ preferences for redistribution and uncertainty about candidates

policy positions, when voters choose delegates to represent these preferences.

On citizens’ preferences I develop a theoretical model that explores the

heterogeneity of redistributive policies. I show that preferred levels of redis-

tribution, including that of poorer constituencies, might decrease as poverty

and inequality increases. This in turn reduces the incentives of politicians

to advocate for certain forms of redistribution. Empirical analysis of three

waves of public opinion surveys in Brazil provide evidence consistent with

the model’s results.

Once preferences are determined, the next important step in evaluating

the political prospects of redistributive policies is to assess representation.

Do the policies advocated by representatives reflect those preferred by their

constituencies? To answer this question I emphasize uncertainty and develop

a model of elections as a game of incomplete information, where candidates

display mix motivations. I show that uncertainty can account for the per-

sistence of low redistribution in unequal democracies, among other reasons,

because it might be optimal for poor voters to reelect incumbents that fa-

vor less redistribution than they do. I analyze electoral survey data about

each presidential election in Brazil since democratization and find evidence

consistent with the models’ assumptions and observable implications.



1 Introduction 1

1.1 Inequality and Redistribution in Latin America: a Brief Overview 1

1.2 The Argument . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.3 The Scope and Methodology . . . . . . . . . . . . . . . . . . 11

1.4 The Organization of Chapters . . . . . . . . . . . . . . . . . . 14

2 From High Expectations to Low Outcomes 19

2.1 Taking Stock of the Literature . . . . . . . . . . . . . . . . . 19

2.2 From Developed to Developing Countries . . . . . . . . . . . 26

2.3 A Full Information Model of Political Support for Redistri-

bution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3 Preferences: Citizens Views on Poverty, Inequality, and Re-

distribution 60

3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.2 Different Groups, Different Preferences? . . . . . . . . . . . . 63

3.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4 Representing Preferences: Candidate Selection Under Un-

certainty 81

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

4.2 Elections and Representation . . . . . . . . . . . . . . . . . . 84

4.3 Voting with Uncertainty: Policy Positions and the Choice of

Representatives . . . . . . . . . . . . . . . . . . . . . . . . . . 89

4.4 Some Implications to Other Theories . . . . . . . . . . . . . . 112

4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

5 Reaching a Decision: Evaluations of Candidates’ Policy Po-

sitions and Vote Intention 119

5.1 How Much do Respondents Know About Candidates Policy

Positions? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

5.2 Socioeconomic Characteristics and Vote Intention in Presi-

dential Elections . . . . . . . . . . . . . . . . . . . . . . . . . 136

5.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

6 Conclusion 158

Appendices 162

Bibliography 186

List of Tables

2.1 Public expenditure per student by grade . . . . . . . . . . . . 38

3.1 Support for social policies . . . . . . . . . . . . . . . . . . . . 65

3.2 Ratio mean income to income of the fifth quintile . . . . . . . 67

3.3 Logit results: taxes and health and education spending . . . . 70

3.4 Logit results: education as an important problem . . . . . . . 76

3.5 Logit results: health as an important problem . . . . . . . . . 78

4.1 Justification of absence in presidential elections by educa-

tional attainment . . . . . . . . . . . . . . . . . . . . . . . . . 85

5.1 Changes in indicators of poverty and inequality during pres-

idential terms . . . . . . . . . . . . . . . . . . . . . . . . . . . 149


List of Figures

1.1 Inequality around the world . . . . . . . . . . . . . . . . . . . 4

1.2 Gini coefficient of selected Latin American countries . . . . . 7

2.1 Percentage of government spending on selected programs ac-

cruing to the lowest and highest income quintiles. . . . . . . . 33

2.2 The indifference curves of younger voters under low inequality 49

2.3 An example of the Pareto set of the poor . . . . . . . . . . . 53

2.4 An example of the Pareto set of the rich under low inequality 54

3.1 Predicted probabilities of agreeing with same or more taxes

and services. . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.2 First differences in predicted probability comparing the young

poor with the young rich. . . . . . . . . . . . . . . . . . . . . 74

3.3 Predicted probabilities of mentioning health and education as

the most important problem facing the country. Bars corre-

spond to 95% confidence intervals. . . . . . . . . . . . . . . . 77

4.1 Equilibrium conditions: Proposition 4.1 (left) and 4.3 (right) 98


4.2 Equilibrium conditions: Proposition 4.5 and 4.7 . . . . . . . . 103

4.3 Parameter values for which the equilibrium in Proposition 4.9

holds given b and w. . . . . . . . . . . . . . . . . . . . . . . . 108

4.4 Parameter values for which the equilibrium in Proposition

4.10 holds given b and w. . . . . . . . . . . . . . . . . . . . . 111

5.1 Incidence of “don’t know” answers to candidate placement

question . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

5.2 Placement of candidates and parties on ideological scale (1997-

1998) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

5.3 Placement of candidates on pro-poor (left) to pro-rich (right)

scale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

5.4 Posterior distribution of estimated mean ideological placement 132

5.5 Logit regression results: votes for Lula vs. opponent . . . . . 141

5.6 First differences: probability of voting for Lula vs. opponent . 144

5.7 Trends in social expenditure in current US of 2000 (health,

education and housing) . . . . . . . . . . . . . . . . . . . . . 151

5.8 Trends in social expenditure in current US of 2000 (social

security and total) . . . . . . . . . . . . . . . . . . . . . . . . 152

5.9 Predicted probabilities of voting for Lula vs. opponent . . . . 153

1 Deviations from py ’s ideal point when inequality is low . . . . 169


Chapter 1


“Suddenly the children were wearing shoes,” says the teacher. A man

went to see Dirk and Claudia Haarmann. Beaming from ear to ear, he

asked: “Don’t you see?” They asked him what he meant. “Don’t you see? I

now have trousers and a t-shirt. I am now a person.”’

— Dialika Krahe on How a basic income program saved a Namib-

ian village, Der Spiegel, 2008

1.1 Inequality and Redistribution in Latin Amer-

ica: a Brief Overview

Poverty and inequality go hand-in-hand. At certain levels they are both

objectionable on fairness grounds, but also for their negative effect on other

desirable goals such as political and economic development (Alesina and

Section 1.1 2

Rodrik 1994; Persson and Tabellini 1994; Bourguignon and Verdier 2000)

and social cohesion (Glaeser, Scheinkman and Shleifer 2002; Gylfason and

Zoega 2003).

According to international rankings, whether compiled by the CIA or the

United Nations, among the fourteen most unequal countries in the world,

half are in Latin America and the other half in Sub-Saharan Africa (see

Figure 1.1). Nine of them1 , including the top three, have been considered

democratic2 for at least the past thirteen years.

In the African continent Botswana is considered a success story, both

in the political and the economic realm. Yet, it is one of the most unequal

countries in the region (Acemoglu, Johnson and Robinson 2003, chap. 4)

with a Gini coefficient estimated at 0.61. Latin American countries, in

turn, have made considerable progress towards democratization and some

have been particularly successful in implementing much needed economic

reform. However, Latin America is considered the most unequal region in

the world (Clements, Faircloth and Verhoeven 2007; Human Development

Report 2007). Poverty levels have remained quite high relative to the region’s

level of development, with an estimated 41% of the population living below

the poverty line. A number that has not shown much improvement over the

years and that, for many countries, was worse in the beginning of the decade

than in the eighties (Londoño and Székely 2000).

Namibia, Lesotho, Botswana, Bolivia, Colombia, Brazil, Paraguay, South Africa and
Receiving a polity score of 6 or higher. The polity score varies from -10 (very un-
democratic) to 10(very democratic).
Section 1.1 3

In line with such trends, Savedoff (1998) observes that life expectancy

and education are low in Latin America compared to other regions at the

same levels of income, and Bénabou remarks that “the incidence of public

education and health services, is much more egalitarian in East Asia than

in Latin America”(2000, 96).

One of the fundamental roles of government is to collect and reallocate

resources among its citizens. Reallocation is usually done in the form of

services, such as public provision of health and education. It can also be

monetary, such as old age pensions and cash transfers to the poor. The

extent to which governments around the world take on this role varies con-


According to data from the Organization for Economic Cooperation and

Development (OECD), governments in Sweden and France spend the equiv-

alent of 30% of their gross domestic product (GDP) in social programs.

They range from family allowances to the provision of job training, hous-

ing and retirement pensions. Meanwhile, in the United States and Mexico,

these expenditures were evaluated at 16% and 7% of the GDP respectively

in 2005. These figures, however, conceal tremendous variation in the kinds

of social programs that actually get funded in different countries.

In a 2007 report on government spending in Latin America, Clements,

Faircloth and Verhoeven find that over the past decade there has been an

increase in social spending in the region. On average, social spending ac-

counts for 13% of the GDP, or about one half of government expenditure in

Latin American countries. As the authors show, this is high compared to

what emerging countries in Asia, for example, spend on social policy, but low
Figure 1.1: Inequality around the world
Section 1.1

Gini coefficient

Data source: United Nations Human Development Report 2007-2008.

Section 1.1 5

compared to OECD countries. The authors claim that the growth in spend-

ing is partially due to an increase in revenues that have also contributed to

the improvement of the fiscal position of these countries.

In terms of type of programs funded, efficiency of allocation, and re-

sulting changes in social indicators, countries differ and results are mixed.

Bolivia, Colombia and Honduras, for example, are the countries where social

spending grew the most. In the first two countries the increase was mainly

due to social insurance and social assistance programs, while in Honduras

education accounted for a good share of the spending growth.

Regarding social indicators, the region performs poorly, specially given

its level of economic development. Expenditures on both health and educa-

tion tend to be poorly targeted, leading to low rates of human capital de-

velopment. Just to cite an example, secondary enrollment rates are low and

repetition rates are high, while a considerable share of educational spending

is targeted to tertiary education, and thus to the elite. In many countries

the report finds that patterns of investment in education are highly volatile,

thus contributing further to inefficiencies in provision.

When it comes to poverty and inequality, the authors of the report agree

with those of most other studies on Latin America arguing that “despite high

social spending, poverty rates remain high and are the region’s most glaring

developmental lag” (2007, 22). Part of the reason lies in the regressive

pattern of social spending, specially those in education and social insurance.

As shown by Lindert, Skoufias and Shapiro (2006), social insurance in Latin

America is funded to a great extent through general tax revenues, thus

turning it into a highly regressive transfer scheme. Recently, however, the

Section 1.1 6

increased implementation of well-targeted social assistance programs – such

as conditional cash transfers – is contributing to a significant reduction in

levels of poverty and inequality (see Figure 1.2). The amount of spending

allocated to such policies, however, remains very low (according to the study

at around 1% of the GDP).

While the goal of redistribution is typically to promote social equality,

achieving the opposite is, regrettably, not an uncommon outcome. Even in

some democratic countries where a majority would benefit from government

transfers of different sorts, we observe low levels of redistribution and bad

targeting. Countries in Latin America made very little or no progress in

curbing poverty and inequality up until the turn of the century (O’Donnell

1996; Londoño and Székely 2000). Only then, with the rise of leftist gov-

ernments, significant improvements in some countries started to be recorded

(ECLAC 2004; Bourguignon, Ferreira and Lustig 2005; Clements, Faircloth

and Verhoeven 2007).

Two issues emerge out of this brief overview. First, given that most of

these countries have been democratic for decades, we are left wondering why

redistribution has remained low when a majority could benefit from higher

levels. As we have seen, budget constraints do not seem to be the main

reason, as already large sums are being spent on social policies in some of

the countries. The second issue deserving investigation is why it took so

long for redistribution to start increasing, as illustrated by the recent and

slow trend of declining poverty and inequality in certain countries.

In this study I offer an account to both questions based on two funda-

mental issues usually overlooked in the broad literature on redistribution:

Section 1.2 7

Figure 1.2: Gini coefficient of selected Latin American countries

Argentina Brazil Chile Costa Rica





1985 1990 1995 2000 20051985 1990 1995 2000 20051985 1990 1995 2000 20051985 1990 1995 2000 2005

El Salvador Mexico Panama Paraguay





1985 1990 1995 2000 20051985 1990 1995 2000 20051985 1990 1995 2000 20051985 1990 1995 2000 2005

Source: SEDLAC

individuals’ preferences for different redistributive policies and uncertainty

in selecting delegates to represent these preferences in the policy-making

Section 1.2 8

1.2 The Argument

As illustrated by the case of Latin America, while low redistribution is an

issue in unequal new democracies, the possibility of change in its levels also

needs to be accounted for. The analysis developed in this study contributes

to our understanding of both. Relative to more conventional approaches I

call attention to two relatively basic factors that can lead to low levels and

delays in redistribution in unequal new democracies.

The first refers to voters’ policy priorities, which ultimately affect rep-

resentatives’ incentives to advocate for certain policies. While it is usually

assumed that the poor invariably favors high redistribution, there are inter-

esting new evidence pointing to the opposite. That is, the poor’s preferred

levels of provision of key policies that affect the extent of poverty and in-

equality can be low. In the particular case of public provision of education,

for example, borrowing constraints can lead the poor to prefer lower levels of

provision than they would were they able to borrow against future earnings.

The implications of low demand to the extent of redistribution in demo-

cratic systems is straightforward. If the majority prefers relatively low levels

of provision of certain policies, democracies achieving high representative-

ness are likely to display relatively low levels of provision. In other words,

even under the assumption that the political process is well functioning, we

would likely observe low provision of such policies. This means that if we

fail to incorporate into the analysis reasonable assumptions about the under-

lying distribution of preferences, we risk overstating and misunderstanding

Section 1.2 9

the effects of other factors on policy outcomes, specially the effects of the

often highlighted flaws in newly established democratic institutions.

What if representation is somehow compromised? Then outcomes de-

pend on the reasons why that happens. It is common in the literature to

argue that low representation can result from the elite manipulation of the

political process to avoid high levels of redistribution favored by the poor.

While this is highly plausible when preferences for redistribution by the poor

is on the high end, it would be unnecessary when demand is low. Thus we

would likely observe low provision on both cases and gain little insight into

the recent upward trends in redistribution.

Bribery, corruption, unequal political clout, and subversion of political

institutions are not the only ways in which representation can be compro-

mised. In this study I explore a more basic factor affecting the extent of

representation that, to my knowledge, has never been emphasized in the

context of redistribution. That is uncertainty about the actual policy ori-

entations of candidates during elections. If voters have limited information

about the ideological inclination of candidates, it is difficult for them to base

their choices primarily on policy preferences. Hence representation might fall

short of what is expected given the rules governing the political system.

This factor reflects a common feature of most countries in Latin Amer-

ica, namely that they are all relatively new democracies. In well established

democracies, parties tend to have a relatively stable constituency. In new

democracies, however, specially where a wave of new parties form with de-

mocratization, voters are not informed about the type of policies each candi-

date stands for. This leads to a process of learning, where citizens can only
Section 1.2 10

update their beliefs about representatives’ ideological positions once they

have the opportunity to observe the actual policy choices of incumbents.

On the political prospects of redistributive policies we would have two

possible scenarios given voters’ preferences. On the one hand, in instances

where redistributive policies are highly demanded, low representation can

lead to low provision. On the other, in those cases where demand is low,

lack of representation can lead either to higher or to even lower provision.

As will be shown, uncertainty can lead a majority of poor voters favor-

ing high redistribution to reelect incumbents that are not that redistribu-

tive, even if they are fully aware of the incumbents’ preferences. Conversely,

highly redistributive incumbents might also get reelected by a majority fa-

voring rather low provision. While this last case is less likely to arise, it

can play an important role in fostering certain redistributive policies, such

as public provision of education. Thus compared to a situation of full infor-

mation, the chances of observing high redistribution under uncertainty are

further reduced, but not completely eliminated, when demand is low.

These two basic factors provide a framework of analysis on which one

can build more encompassing accounts. As will be discussed, some studies

still rely on unlikely assumptions about levels of preferred redistribution and

amount of information available to voters during elections. While the focus

of these studies might not be on these particular factors, relying on unreal-

istic assumptions about them surely affects the results obtained. Moreover,

while some of the accounts reviewed in this study abstract from these basic

features, they are likely to have important implications to these theories.

Section 1.3 11

1.3 The Scope and Methodology

As with any broad theme, in order to make the analysis tractable and clear

the scope of the study needs to be restricted. Of all the steps making up

the political process leading to the implementation of policies I focus on

elections. Democracies are representative governments. This means citizens

need to choose individuals to represent their interests and to whom policy

decisions will be delegated. Thus, elections are a fundamental stage in the

democratic political process and the main instruments for citizens to affect

political decisions (Powell 2000).

The selection of representatives bears heavily on the likelihood that re-

distributive policies will emerge as an outcome of the political process. How

well this representation is achieved is fundamental to understanding what

follows in the next stages. Based on the voter-representative link, we can

begin to identify whose interests, if any, are being excluded right at the onset

of the process. If, for example, we find that the interests of the majority are

not being represented, it is likely that we would not need complex theories

of strategic alliances and subversion of institutions to explain why outcomes

differ from the will of most citizens.

We can think of the political process linking demand and supply of differ-

ent forms of redistribution to be divided into three stages: elections, policy

making and policy implementation. In this project I focus on the electoral

phase, providing a baseline for the analysis of subsequent stages of the pro-

Section 1.3 12

Another important feature of this study refers to the application of what

we might label a though experiment. Many authors have pointed out that

redistribution tends to be low in many democracies simply because they do

not work as they should. There are asymmetries of power, corruption and

other obstacles to the fully-fledged exercise of people’s democratic rights

and will. These are indeed relevant factors. They are, however, likely to be

affected by more fundamental issues that are at play whether institutions

are well-functioning or not. And the best way to isolate these more basic

effects is to assume the political system works.

New democracies created their systems based on existing ones. The

“transplantation” of these existing systems to different environments alone

can cause the expected outcomes to vary. Levels of redistribution, for in-

stance, should be different in different countries even if their institutions

are exactly alike. Both factors explored in this study can illustrate the

point. Two countries with the same political system but different levels of

poverty and inequality would respond to different underlying distribution of

preferences, possibly reflecting different levels of demand for redistribution.

Moreover, delegation requires information that is not readily available right

at the onset of new political systems. Thus in newer democracies, while

voters acquire this information, that is, learn about the ideological inclina-

tions of candidates and parties, uncertainty can hinder representation, thus

affecting policy outcomes.

Throughout the analysis I rely on two main research tools: mathematical

modeling and the analysis of individual level survey data.

Section 1.3 13

With respect to the empirical exercises, the proposed methodological

approach differs from what is common in the literature on inequality and

redistribution in mainly two ways. First, most studies tend to rely on cross-

country comparisons using data that is aggregated at the national level.

This approach can conceal important within-country disparities, since it

uses average estimates. It also ignores the fact that the same redistributive

policies do not necessarily mean the same thing across countries and time.

In Brazil, for example, unemployment insurance and pensions tend to be re-

gressive (Velez, Ferreira and de Barros 2003; OECD 2005), while in OECD

countries these policies are shown to redistribute the income in favor of the

poorer segments. More importantly, according to the mechanisms proposed

in this study, countries of different income and inequality levels should dis-

play different levels of demand and therefore of supply of redistribution.

Analyses conducted on pooled data, however, make it impossible to discern

among contending underlying dynamics, which is one of the main goals of

this study.

Second, and a likely consequence of the cross-country approach, studies

tend to rely on one global measure of redistribution. This implicitly assumes

redistribution is one homogeneous policy, favoring the poor segments of

society. However, as will be discussed in Chapter 2, redistribution can take

many forms, each benefiting a specific group in society that is not necessarily

defined in terms of income alone.

The within country approach is particularly suitable to such a task. It

avoids the problems involved in cross-country comparisons where measures

of political institutions, levels of redistribution and income are hardly, if at

Section 1.4 14

all, comparable. I focus on one country, Brazil, for the following reasons. It

regained the status of democracy in 1989, with the return of direct elections

for president, and has been upgraded to the status of full democracy by

research organizations such as the Freedom House in 2002, when the left

won the presidency for the first time. It is, however, one of the most unequal

countries in the world, with a Gini coefficient currently at 0.58 and with a

significant portion of the population living at high levels of poverty.3 .

1.4 The Organization of Chapters

The dissertation is organized in two parts. In the first part – comprising

chapters 2 and 3 – I investigate individuals’ preferences over redistributive

policies. In the second part – chapters 3 and 4 – I consider how these indi-

viduals choose delegates to represent their preferences in the policy-making

stage. In the first part, given the focus on preferences, I keep the decision

process simple. In the second, given the focus on elections, I make simplify-

ing assumptions about preferences – mostly following from the results in the

first part – and concentrate instead on a more rich account of individuals’

choice of representatives.

In chapter 2 I discuss the literature linking individuals’ preferences to

levels of redistribution. I begin with some of the most common arguments

According to the Census in 2000 60% of Brazilians aged 10 and up were living on
a monthly income of one minimum wage or less (minimum wage in 2000 = US 68.00).
About one third of the population was living with incomes below the poverty line and
16% were living on extreme poverty. These numbers have seen considerable improvements
since. In 2007, according to IPEA, about one fourth of Brazilians had incomes below the
poverty line and 8% of the population lives under extreme poverty (income below US 1.00
a day)
Section 1.4 15

emphasizing the role of inequality on preferences and thus on levels of ob-

served redistribution. I briefly review the long line of arguments that con-

clude inequality leads to high redistribution, and cover in more detail the

literature making the opposite claim. I argued that in order to better un-

derstand preferences for redistribution we need to consider the actual forms

it can take – e.g. transfers, services, insurance – in addition to the levels of

poverty and inequality in society. In light of this discussion I evaluate the

applicability of the arguments reviewed to the case of new democracies, and

propose an alternative account.

By means of a theoretical model I argue that demand, including that of

poorer constituencies, for certain redistributive policies decreases as poverty

and inequality increases. To illustrate these dynamics the model assumes

a two dimensional policy space where individuals decide the levels of tax-

ation and the allocation of proceedings between two different types of re-

distributive policies: pension transfers and public provision of education.

After solving for the ideal points of voters, I consider the incentives rep-

resentatives have to adopt these different policies based on a standard full

information model of elections. That is, assuming new democracies work as

well as those established long ago. The main result is that even democracies

achieving the praised quality of being highly representative might display

low levels of redistribution, if they are characterized by high levels of poverty

and inequality and lack of credit markets serving the poor.

In Chapter 3 I investigate some of the observable implications of the

theoretical model developed in the previous chapter. The analysis is mostly

based on individual level data. The chapter opens with a brief description
Section 1.4 16

of citizens’ views on poverty, inequality and redistribution. The remainder

of the chapter explores two particular issues following from the previous

theoretical discussion. One is that public provision of health and education,

while important policies to fight poverty and inequality, are in some cases

subject to low demand. The other refers to the effects of poverty and in-

equality on individuals’ preferred levels of redistribution. In line with the

theoretical model, results suggest that poorer individuals might indeed face

important trade-offs that lead their preferred levels of public provision of ed-

ucation to be lower than what is usually assumed. Furthermore, I find that

inequality can have a negative effect on the level of redistribution preferred

by wealthier individuals.

Chapter 4 introduces the second part of the dissertation. While on the

first part I explored the effects of preferences only – assuming perfect rep-

resentation – in this part I investigate factors that might compromise the

correspondence between voters’ interests and those of their representatives.

After a brief review of the literature on the subject, I build a theoretical

model that explores the effects of uncertainty about candidates ideological

inclinations on voters choice at the elections. I model a polity with three

types of candidates – left, moderate and right – and three corresponding

groups of voters – poor, middle and rich. Candidates are allowed to display

mixed motivations. That is, they can range from purely policy oriented to

office seekers. The main assumption is that voters do not know candidates’

types in the first period and update their beliefs about the incumbent after

observing her or him serve a first term in office.

Section 1.4 17

The model yields a number of important insights to the puzzle. I em-

phasize one particular equilibrium found, arguing it provides a reasonable

account of both low levels of redistribution in unequal new democracies, and

the possibility of change in those levels. This equilibrium states that as long

as voters perceive left type candidates to be rare relative to moderates and

right types, a majority favoring relatively high redistribution would still re-

elect a moderate incumbent – that can be made arbitrarily close to a right

type – even if they have complete knowledge of this incumbent’s ideological


I also explore the political prospects of redistribution beyond the levels

preferred by a majority of the population. I establish the results of the

equilibrium under the assumption that there exists candidates whose ideal

levels of redistribution lie above that of the poor majority. Similar result

are obtained. As long as the levels of redistribution preferred by a majority

lie closer to the ideal policy of a left type than to that of a right type, this

majority would still reelect these relatively more redistributive incumbents,

even if they are fully aware of the incumbents’ policy inclinations.

In Chapter 5 I investigate empirically the plausibility of both the main

assumption of the model and the conditions supporting the separating equi-

librium just discussed. I analyze electoral surveys about all five democratic

presidential elections in Brazil. I present evidence suggesting voters do in-

deed face uncertainty about the policy preferences of candidates when de-

ciding who to vote for. In addition, the analysis suggests voters are likely to

a) believe left types are rare among candidates, b) update their beliefs about

the policy orientation of incumbents after observing one in office, and c) use
Section 1.4 18

the information obtained about an incumbent when deciding who to vote

for in elections where that incumbent is running for a second term. Taken

together, all of these findings imply the separating equilibrium found in the

previous chapter provides a good characterization of elections in unequal

new democracies. In particular the fact that they might usually lead to low

levels of redistribution, and in some occasions to visible increases.

The dissertation concludes with a summary of the theory and empirical


Chapter 2

From High Expectations to

Low Outcomes

“Notice that the necessary aim of those without property is to obtain

some: all the means which you grant them are sure to be used for this pur-
pose. If, to the freedom to use their talents and industry, which you owe
them, you add political rights, which you do not owe them, these rights,
in the hands of the greatest number, will inevitably serve to encroach upon
— Benjamin Constant (1767-1830)1

2.1 Taking Stock of the Literature

The redistributive consequences of extending political rights to the masses

has concerned scholars and practitioners for centuries. The debate has

long been dominated by “fear of the marauding rabble of the dispossessed ”

(Shapiro 2002, 120). This is clear both in the writings of philosophers, no

matter of which school of thought, and the documented consternation of

statesmen. It is indeed so intuitive that it usually plays a central role in

Constant (1988, pg. 215)
Section 2.1 20

theories that go beyond explaining levels of redistribution itself. Many po-

litical theories of regime transition, for example, suggest that the prospect

of excessive redistributive pressures had profound impact on the way polit-

ical suffrage was extended to the masses in societies at varying degrees of

equality (Acemoglu and Robinson 2000, 2006). On the post transition eras,

economists have often applied that same logic to explain why inequality

seemed to be associated with low growth (Romer 1975; Alesina and Rodrik

1994; Persson and Tabellini 1994). The reasoning being that high inequality

leads to high redistribution, which in turn, depresses investment and hence


The seminal work of Meltzer and Richard (1981), based on the median

voter theorem, offers a clear exposition of the logic behind such high re-

distributive expectations in the presence of high inequality. The authors

show that a flat income tax schedule and equal division of the proceedings

among citizens would result in a net gain to all those living with less than

the average income in society. If this redistributive decision is based on

majority voting, it is implemented as long as more than half of the voters

stand to gain from it. Thus higher inequality, which means lower income

of the median voter relative to the mean, should be associated with higher

redistribution in democratic societies.

This logic has been the underpinning of many theories of redistribution

(Romer 1975; Roberts 1977; Perotti 1993; Hicks and Swank 1992). More

recently, however, it is less frequently found in studies of levels of redis-

tribution than embedded in theories explaining other phenomena – such as

transitions, growth and even the effect of international competition on social

Section 2.1 21

expenditures (Kaufman and Segura-Ubiergo 2001). Part of the reason lies in

the lack of empirical evidence supporting the claim. Notwithstanding major

measurement problems that beset the study of inequality, empirical studies

are often unable to find evidence that more inequality leads to more redis-

tribution or actually find the opposite relationship (Bénabou 1996, 2000;

Milanovic 2000; Moene and Wallerstein 2001; Kaufman and Segura-Ubiergo


In response to the newly accepted empirical regularity, an extremely pro-

lific literature developed that addresses the issue of why inequality does not

necessarily lead to high redistribution. This broad literature can be divided

into two groups. The first focuses on preferences, independent of the work-

ings of political institutions. The second pays particular attention to how

the functioning of political institutions, especially their shortcomings, lead

to disregard for the poor majority’ s wishes, and thus to low redistribution.

I deal with them in turn.

In the first part of this study I pay particular attention to preferences

and their more straightforward effect on levels of redistribution. In the sec-

ond part, beginning in Chapter 4, I shift focus to the process of selection of

candidates to represent these preferences. At this stage I discuss the rele-

vant literature belonging to the second group.

Milanovic (2000) employing a different measure of inequality, incorporating pensions
and net of all other transfers, actually finds that more inequality leads to more redistri-
bution in a sample of developed countries. However, he finds no significant relationship
between the income of the median voter and levels of redistribution
Section 2.1 22

I start this first part with a brief discussion of theories of individual

preference for redistribution, drawing attention to their strengths and limi-

tations in addressing the case of Brazil. This discussion is then followed by

a formalization of the mechanisms unveiled in order to better understand

their interplay and verify their consistency. The results obtained with this

exercise provides the theoretical framework guiding the empirical analyses

of individual level data conducted in the next chapter.

The logic described by Meltzer and Richard leads to an extreme outcome

commonly referred to as “the poor soaking the rich”. The first limitation

to this result regarding preferences is that citizens are probably aware that

high taxation can be inefficient. That is, that it can reduce the incentives for

investment and growth thus reducing the pool of resources available to be

divided. This expectation alone can depress demand for excessive levels of

taxation used to finance redistribution (Harms and Zink 2003). As argued

by Jencks “most thoughtful liberals, including Rawls, also recognize that re-

warding people for producing more goods and services will often improve the

absolute well-being of the least advantaged” (2002, 51). The extent to which

demand for redistribution is reduced under this logic is not clear. Most

studies still rely on the basic assumption that preferences for redistribution

tends to be high, especially among poorer individuals.

One mechanism linking this basic assumption to low redistribution is

that placing particular emphasis on political salience. Deciding on an al-

location of resources to different policies is usually subject to a constraint.

This means trade-offs are involved in the decision process possibly leading

individuals to accept some losses on the redistributive dimension in exchange

Section 2.1 23

for gains in other more important ones (Harms and Zink 2003). Thus the

lower the political salience of redistribution the less we would expect redis-

tributive outcomes to clearly reflect the preferences of citizens.

These dynamics were explored in a number of studies. Stasavage (2005),

for example, attributes differences in the form of provision and government

levels of expenditure on education to the importance citizens attach to this

service. Besley and Coate (2001) emphasize the effect of what they call “is-

sue unbundling”, arguing that “In a representative democracy, the bundling

of issues, together with the fact that citizens have only one vote, means that

policy outcomes on specific issues may diverge far from what the majority of

citizens want.”

Still other authors explore salience in the context of racial or ethnic cleav-

ages. They posit the existence of an alternate politically salient dimension

that is more relevant than and orthogonal to redistribution (Roemer 1998a;

Alesina, Baqir and Easterly 1999; Austen-Smith and Wallerstein 2003; Fer-

nandez and Levy 2005). That is, preferences for policies in this second di-

mension are not correlated with preferences for redistributive policies. This

leads the majority interested in high redistribution to split their votes on

other matters, thus failing to elect a representative that would advocate for

more redistribution.

While common to assume the poor prefers high redistribution, some

argue this is not necessarily the case when it comes to the median voter.

The fact that in some cases median voters tend to be around the center of

the distribution raises doubts about which side they actually feel closer to.

According to some of these accounts, even if the median voters’ income is

Section 2.1 24

below the mean, they might still prefer low redistribution and thus tolerate

more inequality. The relationship among the different income classes is

central to all these accounts. First, many authors argue that individuals’

prospect of upward mobility is an important determinant of tolerance for

inequality. As summarized by Harms and Zink “if today’s policy cannot be

easily reversed and if the politically decisive voter expects to earn an income

greater than the average in the future, there may be a preference for modest

redistribution although an individual’s current income is low” (2003, 665).

Kristov, Lindert and McClelland explored a similar line of reasoning

where the decisive group opts for higher redistribution if their members are

“more likely to trade places with those currently poorer than with the more

isolated rich elite”, and lower redistribution otherwise. This line of argument

is usually based on empirical findings for the US (Alesina and La Ferrara

2001) and has also been formalized by Roemer (1998b) and Piketty (1995).

Once the position of the pivotal voter is identified, Saint-Paul (2001)

argues that changes in the levels of inequality caused by an increase in the

poverty of the bottom class has minimal impact on this voter’s redistributive

preferences. If anything, the argument goes, the median voter might favor

even less redistribution. This can happen for different reasons. On the one

hand, some scholars explore the gains in social status that a gap between

poor and middle-class can afford the decisive voter. This gap can generate

tangible gains to the middle group – e.g. skill premiums (Harms and Zink

2003; Zink 2005) – or just a sense of social distancing and hierarchy, that

might be reinforced by the poor accepting their inferior ranking (Shapiro

Section 2.1 25

Moffitt, Ribar and Wilhelm (1998), on the other hand, argue that redis-

tribution might be lower when inequality is higher due to “qualified altru-

ism”. The idea is that redistribution garners political support based on the

benefit to income ratio. Thus if the income of the poor drops further, the

benefit needs to drop too to avoid, for example, disincentives to go on the

job market. If the benefit is too high relative to the income the poor expects

to earn, the incentive to work is reduced.

Finally, a couple of studies explore what I believe is the most fruitful

way of understanding support for redistributive policies. That is, taking

their heterogeneity into account. In the words of Osberg, Smeeding and

Schwabish, “Social insurance, targeted social assistance, and universal ben-

efits programs (like child allowances) may reflect different tastes, values, and

mechanisms for redistribution” (2003, 31). The vast majority of studies on

redistribution continue to base their analysis on social expenditures as a

proxy for governments’ redistributive efforts, thus ignoring the incredible

diversity of policies that lie beneath. Given that this diversity has impor-

tant implications to who supports and who opposes each type of policy,

this omission can be a serious one. Moene and Wallerstein (2001), for ex-

ample, find that preferences for social policy vary depending on whether

it was created for insurance or for pure redistributive purposes. Moreover,

Casamatta, Cremer and Pestieau (2000) pay particular attention to the ex-

tent of coverage – the overall size – and the redistributive impact – who gets

how much – of policies in order to determine preferences for them. Finally,

Donder and Hindriks (1998) emphasize the effect of targeting on what kinds

of constituencies form around different types of redistributive policies.

Section 2.2 26

As will be argued, which aspects are relevant to understand levels of

support for different kinds of redistributive policies depends on the economic

structure and the levels of poverty and inequality of a society

2.2 From Developed to Developing Countries

Most of the literature reviewed in the previous section focuses on devel-

oped countries. How likely are these arguments to apply to the developing

world? To answer this question I take Brazil as a benchmark. Based on the

overview provided in the introduction, Brazil is a typical Latin American

case. That is, all of the issues raised for Latin America in general pertaining

to the puzzle addressed in this study fit well with the Brazilian case. Most

importantly, Brazil is a highly unequal country that, like many of its Latin

American neighbors, spends relatively little on programs that redistribute

from the rich to the poor (Lindert, Skoufias and Shapiro 2006). Thus an in-

teresting case study to investigate the arguments summarized above. Since

they are about individual preferences without regard to the institutional

setting, the within country approach is ideal.

As just mentioned, most of the theories outlined were built and tested

based on the developed world. This alone raises a number of issues regarding

their applicability to developing countries. In particular, developing coun-

tries tend to be considerably more unequal and have much more widespread

poverty. This has major implications for the definition of income classes and

how they relate to each other, thus bearing heavily on the identity and pref-

erences of the pivotal voter. Moreover, due to these differences in economic

Section 2.2 27

strucure, policies that look similar on paper have very different redistributive

impact, thus affecting the levels of individuals’ support for them.

The argument advanced in this section is that, as a result of the interplay

of these factors, overall demand, including that of the poor, for types of

redistribution that are crucial to fostering equality, are likely to decrease as

inequality and poverty increase. As a consequence, elected representatives

may face little incentive to advocate for such policies. Thus, aside from

the conceivable flaws plaguing the workings of newly established democratic

institutions, a fully working democracy may still face obstacles in promoting

more equal societies.

The Pivotal Voter

Many of the arguments discussed in the previous section rely on some defi-

nition of different classes and their relationships in determining preferences.

But there does not seem to be a consensus on how these groups should be

defined. Studies rely on different factors to reach a categorization. Thus it

is important to be explicit about the definition used in this study in order

to asses which arguments seem to hold and which do not.

It is not generally straightforward in highly unequal countries with rela-

tively widespread poverty like Brazil – and most Latin American countries

– to know where to draw the line between the poor and the middle class. It

is important, however, to define the relative sizes of these groups given that

preferences will be linked to them and election results will depend on how

many votes each of these groups control.

Section 2.2 28

Much of the confusion revolves around the middle-class. The middle class

is usually loosely defined as the class between the elite and the workers class.

In many developed societies this definition coincides with the statistical

definition of average income and, in some cases lie close enough to that of

median income. In Brazil, and most likely in many other Latin American

countries, the socioeconomic – that is, including levels of education and class

– and income based definitions do not, however, coincide.

In more egalitarian societies, most citizens enjoy incomes around the

average of the whole population, and there is less controversy as to the

identity of the middle class. When the distribution of income is more skewed,

however, there are a lot more people earning incomes below the mean than

above it. Thus the middle-class, defined as those earning the mean income,

tend to be further to the right than the median. The more the distribution

is skewed, the more the middle-class will rather be part of an elite, and the

more encompassing the category of poor will be. In countries like Brazil,

given the socioeconomic definition of middle-class, which usually denotes

high skilled workers with college education, their income will in fact lie well

above the average income. Thus speaking about a middle-class in general

terms is all but straightforward.

The monthly average household income per capita in Brazil in 2007 was

estimated at R$390 (in Reais of 2002). This is roughly about one minimum

wage (or U S$195) per member of the household. Taking a household of four

members as a benchmark, this amount lies well below the income earned by
Section 2.2 29

high skilled and college educated citizens in the country3 . This average

household income corresponds to the average per capita income of the 7th

decile of the Brazilian income distribution in that same year. This means

that those living on about one minimum wage a month are among the richest

30% of the population. That is, about 70% of the population earns per

capita income below that average person. Now, the monthly household per

capita income that divided the population into two equally sized groups in

2007 was R$200 – or U S100 – per month. Thus it is natural to assume, in

the case of Brazil and probably of other unequal countries where poverty is

prevalent, that the poor form a majority.

Even if we take subjective evaluations of class membership, a similar

pattern arises. In a study about inequality in 2001 in Brazil, adult respon-

dents were asked to place themselves in one out of six social classes: poor,

workers, lower middle, middle, upper middle and upper class. Around 67%

of the respondents placed themselves either under poor or worker’s class

and a further 20% under lower middle-class. While no income information

is available about the Brazilian electorate, according to the data available

at the Supreme Electoral Tribunal, 71% of the electorate in 2002 had pri-

mary education or less. This number was 66% in 2006. This numbers are

meaningful given that voting is compulsory and enforced in the country.

Now that we have a better idea of who the poor and the median citizens

are in the country we can assess the arguments about their tastes. Per-
According to the POF – Pesquisa de Orçamentos Familiares – conducted by the IBGE
in 2002 and 2003, the average income of a household with at least one working member
having a college degree was around R$3800. That amount jumped to around R$6000
when two household members had tertiary education.
Section 2.2 30

ceptions of upward mobility, or feeling closer to the better-off rather than

the worse-off in the income distribution are likely a function of relative dis-

tances. That is, in very unequal countries the “upward move” necessary

to get you closer to middle and high class can be unrealistically high. In

such cases, Kristov, Lindert and McClelland (1992), among others, would

actually predict more progressive redistribution. In terms of more tangible

benefits, such as preserving the advantages afforded by skill premiums, it

is also likely that the pivotal voter in societies with an income distribution

close to the Brazilian one is less affected. Those benefiting from gaps in

education tend to belong to a very select minority.

What about the political salience of redistribution? As reviewed ear-

lier, some contend that even if the poor are a majority they might split

their votes on other more salient issues that would divide them along, for

example, racial or ethnic lines. Two factors limit the applicability of this

line of analysis to the case of some developing countries. First, it may be

difficult to find a politically salient dimension, such as race or religion, that

is not highly correlated with income. In Brazil, while the elite is made up

almost entirely of white people, blacks tend to be overrepresented among

the poor. In Guatemala, where indigenous Mayan and those of African

descent makeup about 40% of the population, according to the United Na-

tions Refugee Agency, almost all of the indigenous population (86%) live

below the poverty line (State of the World’s Minorities and Indigenous Peo-

ples - Guatemala 2009). The situation is similar in Bolivia, Honduras and

Nicaragua according to a report by Minority Rights Group International (In-

Section 2.2 31

digenous Peoples and Poverty: The Cases of Bolivia, Guatemala, Honduras

and Nicaragua 2003).

Second, even if such a dimension exists, given that poverty is an impor-

tant issue in many countries, it is unlikely that it will be more important

than or equally important as income in determining political preferences.

In fact, this logic may no longer apply even to the United States, the typ-

ical case under this approach. McCarty, Poole and Rosenthal argue that

“previously orthogonal conflicts have disappeared or been incorporated into

the conflicts over economic liberalism and conservatism. Most importantly,

issues linked to race are now largely expressed as part of the main ideological

division over redistribution” (2003, 1).

That being said, however, differences in taste, even among the poor,

might arise within issues of redistribution. As previously mentioned, re-

distributive policies are extremely heterogeneous in terms of targeting and


What Policies Redistribute to the Poor?

Redistributive policies come in all shapes and sizes. As suggested by some

of the studies reviewed earlier, different characteristics of policies might

have distinct effects on their political support. Not to mention the fact

that societies differ on their tastes and views regarding the aims of social

policy and the identification of “deserving recipients”. We saw examples of

studies that have focused on the insurance versus the redistributive aspect of

policies, and on size versus targeting. Particularly relevant to this project,

Section 2.2 32

however, is the fact that policies that look exactly the same on paper – social

protection systems in Latin America are largely based on the European

model – have very distinct redistributive impacts because of differences in

the underlying economic structure (e.g. levels of informality, poverty and

inequality) of countries.

Thus, aside from variations in taste, similar policies might draw differ-

ent levels of support simply because they affect the distribution of wealth

differently. This means that policies deemed to progressively redistribute

income in the developed world might do just the opposite in the developing

world. Hence, some of the aspects that have been singled out in studies

about western democracies, might not be the most relevant to discuss in the

context of Latin America.

The goal of this section is thus to present evidence regarding targeting

and the redistributive impact of the most common forms of social policy –

pensions, cash-transfers, education and health – to help identify the features

that are crucial in determining political support for them.

To begin, contrary to the conventional wisdom, not all redistributive

policies are progressive. That is, not all redistribute from the rich towards

the poor and some actually entail a considerable amount of redistribution

in the opposite direction (Camargo and Ferreira 2002; Velez, Ferreira and

de Barros 2003; Lindert, Skoufias and Shapiro 2006). This is an important

misconception that needs to be clarified as many studies, in particular em-

pirical ones, rely on measures of social spending by the government to gauge

how redistributive – implying that they benefit the poor – governments are.
Section 2.2 33

A study on social policy in Latin America conducted by the World Bank

in 2003 has found that the effects of specific redistributive policies on inequal-

ity vary considerably. The study’s results are shown in Figure 2.1, where

we can identify two striking patterns. The first is that classical welfare poli-

cies such as pensions, unemployment benefits and adult skills training are

actually regressive. This means that their benefits accrue to a much larger

extent to the better-off than to the worse-off.

Figure 2.1: Percentage of government spending on selected programs accru-

ing to the lowest and highest income quintiles.

Graph source: World Bank Report 2003

Section 2.2 34

Currently in Brazil more than half of the total economically active pop-

ulation is working in the informal sector – this is about the average for Latin

America. This automatically excludes this group from contribution based

benefits such as unemployment insurance and pension schemes. As revealed

in a study by the Instituto Brasileiro de Geografia e Estatı́stica (IBGE and

SEBRAE 2003), the informal sector tends to employ in higher proportion

those with lower educational attainment. If we take into account that half of

total social expenditures goes towards pensions, we have a situation where

the poor is excluded from most of social spending.

The second pattern refers to policies in the area of human capital de-

velopment. On the one hand, early age education and related services –

day care, kindergarten, primary school and school meals – are progressive.

However, public expenditures on secondary education, adult skills training

and tertiary education are all regressive. This picture reflects an important

problem in the redistributive system, where the policies that are most likely

to help lift people out of poverty are the privilege of a few. Notice, that

public services like education are non-excludable. This means that there

are barriers other than blatant exclusion preventing the worse-off from tak-

ing advantage of available secondary and tertiary education. In the case of

tertiary education the explanation is simple. Without good primary and

secondary schooling public universities are rarely an option to those born

in poorer households. Public universities are highly sought after making

acceptance difficult to the less educated (entrance is determined based on

scores on a standardized test). In the case of secondary education the prob-

lem is probably related to the expected returns being low relative to the
Section 2.2 35

investment required. Taking full advantage of secondary education requires

time that could otherwise be allocated to work and thus income earning.

The pervasiveness of the system is also felt in the area of health, where

more expensive procedures tend to be dispensed mostly to the rich (Arretche

2003). In both cases – health and education – private provision is available

to the minority who can afford. This can have negative consequences, as

will be explored below, to the level of support for public provision.

Most of the population, especially lower income groups, rely on services

provided by the government. In a survey conducted under the auspices of

the Brazilian Electoral Studies (ESEB) in 2002 65% of respondents claimed

their children attended public schools, 13% claimed private schools4 , and

9% claimed they attended both private and public schools. Regarding the

respondents themselves, 75% reported having attended public schools and

only 8% said they attended private institutions. These figures have not

changed much over the years. In 2007, according to the LAPOP surveys,

these percentages were: 72% of respondents with children attending public

school and 11% attending private schools, while 81% of respondents them-

selves attended public schools and 5% were getting private education.

The numbers relative to health are not much different. In that same

study by ESEB, 70% of respondents reported using public hospitals and

clinics, 24% claimed using private health services only, and 5% both. In fact,

over the years, between one third and one fourth of the population relied on

private health insurance, while the remaining relied mainly on public care.
13% don’t have school aged children or they do but the kids don’t go to school
Section 2.2 36

Thus public services attend a significant portion of the Brazilian population,

while a minority resort to private exit options.

This first study by the World Bank does not include, however, trans-

fers made in cash to the beneficiaries. Given that some of such programs

are relatively new (and became quickly popular), a more recent study was

primarily devoted to their analysis. There are two main venues of redis-

tribution in cash, henceforth referred to as “transfers”: conditional cash

transfers (CCT), and pension schemes. The first, started in the late eighties

as a municipal program in a small number of large cities. It was then im-

plemented nationwide in 2001, but with very narrow targeting. Only poor

families living in municipalities falling below the national average on cer-

tain indicators were eligible. In 2005 the program was greatly expanded

and all benefits brought under the umbrella of the newly established Bolsa

Famı́lia program,which also includes food and gas vouchers. The benefit

per family is very low, capped at around R$1705 , and it targets the young,

school-aged, population of very low income – eligibility is set at a monthly

per capita income of less than R$70 .

The second type of transfer includes three programs. First, there are

the regular contribution based pensions. Second there is the Rural Pension

(Previdência rural) created in 1992 reaching those who live from subsistence

farming and have worked as such for the same amount of years required

of regular employees in the formal sector – 30 to 35 years. The benefit

corresponds to one minimum wage. The third is the BPC (Beneficio por
Overall, while it benefits around 45 million people, the program’s budget corresponds
to no more than 0.3% of the GDP
Section 2.2 37

Prestação Continuada) implemented in 1996 which pays one minimum wage

to the elderly and handicapped whose monthly family per capita income falls

below one fourth of the minimum wage. Contrary to the rural pension, the

BPC is not part of the pension system and is much less encompassing than

the regular contribution based benefit which also covers sickness, death, and

so forth.

From an expenditures perspective, the proportion of total social expen-

ditures that is allocated to each of the policies discussed so far shows a

revealing picture. Brazil spends about 20% of its GDP on social policies.

Practically half of it goes towards regular pensions, around 4% towards

education and another 4% towards health. As shown in Table 2.1 the dis-

crepancy between regular school and college are considerable, although it

has been declining over the years.

This goes to show that even when expenditures are disaggregated into

different components such as pensions, health and education, as advocated

by Kaufman and Segura-Ubiergo (2001), the researcher needs to be careful,

since they can be financing different subtypes of services that benefit dif-

ferent segments of the population. As we can see, the financing of public

universities and their students, which is highly regressive, consumes most of

expenditures on education while benefiting a tiny minority. In health this

is more nuanced. Each country has a different scheme and comparisons are

difficult. In Brazil, as noted earlier, the most expensive procedures covered

by public health benefits the richer. Since in amounts relative to basic health

this is small, public expenditures on health in general show up as equally

benefiting the poor and the rich.

Section 2.2 38

Table 2.1: Public expenditure per student by grade

year primary secondary high school college college/basic education

2000 1,289 1,315 1,250 14,485 11.1
2001 1,274 1,433 1,422 14,317 10.5
2002 1,488 1,382 1,001 13,573 10.1
2003 1,441 1,369 902 11,892 9.0
2004 1,547 1,564 1,069 12,039 8.2
2005 1,731 1,648 1,082 12,243 7.9
2006 1,906 2,093 1,480 12,347 6.7
2007 2,166 2,317 1,572 12,322 6.1

Notes: Source: MEC/Indep. Values in Reais of 2007.

Overall, more is spent on contribution based benefits, even when it has

been shown that the country’s tax system is actually regressive. Most impor-

tantly, as argued by Lindert, Skoufias and Shapiro (2006), pensions, usually

seen as contribution based insurance schemes, actually work in Latin Amer-

ica pretty much like social transfers. The reason is that “almost all social

insurance regimes in LAC run significant deficits, with general tax revenues

financing a considerable share of benefits.” (Lindert, Skoufias and Shapiro

2006, 5). Hence distinctions based on insurance versus redistributive pur-

poses, for example, can be blurred.

Now that we know who benefits from these different types of expendi-

tures we can begin to understand their effects on general levels of poverty

and inequality. According to the 2003 World Bank study (Velez, Ferreira

and de Barros 2003) the poverty rate among the old was reduced after pen-

sion transfers, while that of the younger cohort was made worse. In the

newest report released in 2006 (Lindert, Skoufias and Shapiro 2006), that
Section 2.2 39

situation had changed somewhat, with programs like Bolsa Famı́lia being

commended by their targeting and significant effect on poverty reduction

among young recipients. While the benefit is enough to lift the very poor

out of misery, it is, however, low compared to the programs targeted to

the elderly. Recent studies found that the main contributor to declines in

poverty and inequality has been the BPC (Salomon 2006), and the reason is

that the benefit is relatively large, at one minimum wage. Interestingly, the

choice of setting the benefit at one minimum wage was made many years

before the significant raises that more than doubled its amount.

The effect of income transfers on pre-tax-and-transfers inequality, how-

ever, is rather limited. As cited in Osberg, Smeeding and Schwabish (2003)

in reference to public expenditures on human capital “as the primary route

to overcome inequalities of opportunity and circumstance” (2003, 2), R. H.

Tawney remarked that “the standard of living of the great mass of the na-

tion depends, not merely on the remuneration which they are paid for their

labour, but on the social income which they receive as citizens” (1964, 133).

There are roughly speaking two strands in the literature with respect to

the conceptualization of inequality. They differ on what is it that a policy

should try to equalize. The first group, where most studies on redistribu-

tion fall, considers income as the equalisandum, thus addressing monetary

transfers as the ones just discussed.

Recently, though, scholars and practitioners’ attention has shifted to

what has been termed “inequality of opportunity” (Roemer 1998a; Bour-

guignon, Ferreira and Menéndez 2005; Cogneau and Gignoux 2005), where

redistribution is not simply a tool for reallocating income, but a way of en-
Section 2.2 40

abling people to do better independent of the circumstances under which

they were born. The idea behind this approach is that a person’s income

is determined both by the effort she spends and the circumstances under

which she was born and raised. While a person can choose a level of effort

she cannot choose a place of birth, or the wealth and educational attainment

of her parents. Thus equalizing opportunities means implementing policies

that reduce the effect of circumstance in the determination of a person’s

income level.

Under this approach, forms of redistribution other than income transfers

become important. As noted by Roemer (1998a; 2005; 2003) and numer-

ous other scholars, education is one fundamental piece in achieving equality

of opportunity. Amartya Sen, for example, contends that education can

help increase people’s living standards both through the prospect of getting

better paid jobs, but also through better health and better informed partic-

ipation in the political process. Paes de Barros, Henriques and Mendonça

(2002) argue that education, by increasing workers productivity, contributes

to economic growth, increases in salaries, and decreases in poverty. Also,

given its “non-transferable nature”, it is easy to target, becoming an efficient

means of reducing inequality and promoting social mobility. In a study of

Brazil, they show that the main determinants of inequality are the hetero-

geneity of schooling levels in the population and the skill premiums paid to

more educated workers.

To summarize, while cash transfers do not necessarily reduce pre-tax-

and-transfers inequality - it simply levels post-tax-and-transfer incomes -

equality of opportunity is likely to do so. This is one of the main motivations

Section 2.2 41

to take human capital enhancement policies seriously when we are concerned

about inequality.

A couple of interesting observations emerge from this brief overview.

First, there seems to be some discrepancy between the benefits enjoyed by

the older population and those received by the young. In terms of transfers

this is understandable given the insurance character of pensions and the fact

that disincentives to look for work are only a concern in the latter case. In

terms of human capital, however, we saw that education, which is mostly

targeted to the young, consumes only a small portion of social expenditures

and seems even more regressive than health. Likewise, there also seems

to be a divide in terms of income. Most of the lower income people rely

on under-provided public services, while the wealthier enjoy relatively good

quality public colleges and better health coverage.

In the next section I formalize these ideas to explore their implications

to the preferences of different socioeconomic groups for distinct types of

redistributive policies and thus the incentives of political representatives to

support each of them. As argued by Osberg, Smeeding and Schwabish “Most

models are of a reduced form nature with little attention paid to desired levels

of redistribution (or national differences in the taste for redistribution) in

combination with the institutions and voting mechanisms (parties, lobbies,

etc.) legitimizing these tastes.” (2003, 36). I begin to fill this gap by

modeling preferences in the context of electoral competition.

To model preferences I employ a two dimensional policy space. This al-

lows a little more room to explore the heterogeneity in redistributive policies

just discussed. To capture the ideas presented above, I focus on two particu-
Section 2.3 42

lar aspects. First, I take the possible age conflict into account by considering

both pensions and education as forms of redistribution competing for funds.

Second, I explore the fact that some policies dispense benefits in cash, while

others do it in the form of services. While cash has a more direct impact

on an individuals’ utility, services are less straightforward. Education is an

interesting case in that respect. In models of redistribution it is usually

assumed that the poor has preferences for high provision. However, if we

rely on the economic literature on human capital development this is not

necessarily the case. Shortcomings in the functioning of markets, especially

credit markets, can constraint the preferred levels of provision of education

by the poor to relatively low levels. If this is the case, the observed lower

provision might have its roots on the malfunctioning of markets rather than

of the political system.

2.3 A Full Information Model of Political Support

for Redistribution

In this section I explore individuals’ preferences for two types of redis-

tribution – pensions and education – and their effects on the incentives

politicians have to advocate for them. I focus on the standard approach to

modeling electoral choice where full information is available to all players.

The approach taken in this first step differs, however, in one significant

way from the mainstream literature on redistribution. Instead of looking

for deficiencies in the way democracy works in developing countries I begin

exploring more fundamental issues related to preferences for different types

of redistribution. The question is whether deficiencies in the system tell

Section 2.3 43

the whole story, or if there is an additional and more basic factor at work.

This means the formal model was built with the following scenario in mind:

Suppose we have fully working democracies in the developing world (that is,

universal suffrage, free and fair elections, and representatives whose policy

positions coincide with those of their constituencies and who are able to make

credible commitments). Should we expect high levels of redistribution?

The model is a variant of a model developed by Levy on the politics of

public provision of education (Levy 2005). An attractive feature of the model

is that the source of conflict among the poor is age differentials. Given the

importance of education, age becomes a relevant factor since public provision

of education is a form of redistribution from the old to the young.

The version of the model presented here differs from Levy’s set up in var-

ious ways. First, the choices available to voters are education and pensions,

as opposed to education and monetary transfers that are equally distributed

among all citizens. The choices adopted are a reasonable approximation to

instituted types of redistribution. It must be noted that this study takes

these different forms of redistribution as a given and does not attempt to

explain how they originated. In general, such policies are not easily re-

versible or significantly changed in the short run. One of the important

insights is that these instituted forms of redistribution act as constraints on

agents’ preferences, which, in turn, affect the agents’ representatives policy


As a first cut I focus on two party competition under majority rule.

Despite the existence of multiple parties in Brazil, recent studies on party

ideology have pointed to the bimodality of preferences. This means parties

Section 2.3 44

can be easily divided into two groups: one sided with the government and

the other with the opposition. Thus the two-party simplifying assumption

is not too far a stretch from what is observed. Most importantly, extending

the analysis to multiparty competition does not significantly alter the main


The Model Setup

I consider two candidates, one affiliated with party L, representing the in-

terests of the poor, and the other with party R, representing the interests

of the rich. Candidates are policy motivated, choose a platform that maxi-

mizes the welfare of their constituency, and simultaneously announce their

policy platforms on a two dimensional policy space. The first dimension

corresponds to a level of taxation, t ∈ [0, 1], that is uniformly applied to

all citizens. The use of a uniform tax rate is not only the most common

approach for tractability reasons, but also the closest to the way taxation

is actually implemented. Despite the complexity of taxation systems, there

is much agreement that in most Latin American countries taxes are neither

progressive nor regressive, with some claiming it is actually regressive.

The second dimension corresponds to the level of public provision of

education g. I assume that the full amount collected through taxation is

redistributed towards education and/or pensions, subject to a budget con-


tw̄ = θqg + p
Section 2.3 45

where q is the price of public education, p is the amount to be redis-

tributed equally among the old in the form of pensions, θ is the proportion

of young people in the population, and w̄ denotes the mean income in society.

The budget constraint for public education is

g 6 tw̄/θq,

The portion of the amount collected that is not spent in education is assumed

to be allocated towards pensions.

Voters are divided into four groups according to their income level and

age. These groups are the old poor, the young poor, the old rich and the

young rich, denoted po , py , ro , and ry , respectively. Let wl denote the

income of the poor, wh the income of the rich, and w̄ the mean income,

where wl < w̄ < wh . Voters in the same group are assumed to have

identical preferences for policies. The size of each group is given by θ, the

proportion of young in the population, and π, the proportion of poor in the

population. In this game voters vote sincerely, that is, they cast their votes

for the candidate presenting the platform that gives them higher utility. In

case a voter is indifferent between two platforms she votes with probability

1/2 for each candidate.

In order to solve the model I proceed as follows. First, I present vot-

ers utility functions and solve for their ideal points and indifference curves.

Next, I determine the ideal points of the candidates from each party and

describe the winning platforms for a given set of values of the model param-

eters. Throughout the analysis I assume the poor form a majority.

Section 2.3 46

Voters’ Preferences

The older voters

Older voters are assumed to care about pensions and do not benefit from

public education. The assumption is motivated by the view that parents are

more directly affected by their children’s education than grandparents do.

The utility of the older voters is a function of the level of taxation and the

level of pensions they receive, thus

u((1 − t)wi + (tw̄ − θqg)/(1 − θ))

where um > 0 and umm < 0 6 . That is, the utility for money (pensions

are monetary transfers) is very high at low levels of income and declining,

but never negative, as income increases. The intuition for this assumption is

that a $1 raise in a poor person’s income yields more utility than the same

raise in a rich person’s income.

Lemma 2.1: In the (t,g) policy space, the ideal policy of ro is (t = 0, g =

0), and that of po is (t = 1, g = 0) . The indifference curves of ro ( po ) are

w̄−wh (1−θ)
linear, with slope θq ( w̄−wθq
l (1−θ)

According to Lemma 2.1, the rich old voter is indifferent between a relatively

high level of taxation, with all proceedings going towards pension, and a
The subscript m is used to denote the derivative with respect to money. It is done to
avoid confusion when we get to the utility of the young, which will be additively separable
over money and education.
Section 2.3 47

lower level where all goes towards public education. The old poor, on the

other hand, is indifferent between this same higher level of taxation allocated

in full towards pensions and even higher taxes with a fraction going for public

education. This is because the old poor is made better off with transfers,

since it contributes less than it receives.

The younger voters

Younger voters in the model are assumed to care both about their income

and education. While the first is straightforward, the intuition behind the

second is that younger voters are the primary beneficiaries of education.

The more education they get, the higher the income they expect to receive

in the future, which also raises their prospects of higher savings for old

age. Parents of school-aged children, on the other hand, can benefit from

education in two ways. First, if in the absence of public schooling parents

would be paying for private education for their children, public provision can

be seen as a reduction in their expenses today. Second, by educating their

children, parents expect them to receive a higher income in the future. This

makes the children better able to support their parents once they retire.

Younger voters strategies involve two decisions. First they need to choose

a policy position, as older voters do. Second they need to choose a level of

consumption of private education, denoted s, to supplement the level of ed-

ucation that is publicly provided (if the amount of education that is publicly

provided is at the optimal level for a voter, then she will not buy private

education). The utility of the younger voter is assumed to be additively

Section 2.3 48

separable over the two goods they care about. The utility for education,

which is a function of the level of public and private education consumed,

is assumed to be strictly concave7 . The utility for money is the same as for

the older voters. However, in the younger voter’s case, since they do not

receive pensions, the utility for money is just a function of the income they

keep after paying taxes minus the money they spend on private education,

if any. Formally, it can be written as

u((g + s), (wi (1 − t) − αqs)),

where α > 1 captures the idea that private education is costlier than educa-

tion that is publicly provided. In particular I make two assumptions about

the consumption of education.

A1. Private education is too expensive for the poor to acquire, thus wl < αq

for all s > 0.

A2. Let g denote the level of public education, then for all g 0 ∈ [0, ḡ),

ue (g 0 ) > um (t0 ) for all younger voters, where t0 = g 0 qθ/w̄.

Given that all py receives in return for their taxes is education, for them

to demand any public education it must be the case that the loss in utility

caused by a lower post-tax income is outweighed by the corresponding gain

in utility from public education. All assumption 2 does is make sure that

the income of the poor, while not enough to acquire private education, is
Some economists argue that the utility for education is in fact “S” shaped, since the
returns from secondary and college education are disproportionally higher than those from
primary and graduate levels. But the gist of results would just be reinforced with this
Section 2.3 49

still high enough for them to value some of the cheaper education provided

by the state.

Lemma 2.2: In the (t,g) policy space, the ideal policy of ry is (t = 0, g =

0) if θ
α > w̄
wh , and (t = t∗r , g = tw̄/θq) otherwise, and the ideal policy of py is

(t = t∗p , g = tw̄/θq), where t∗p < t∗r . The indifference curves of ry are weakly
wh wh um
concave with initial slope αq and ue thereafter. The indifference curves
wl um
of py are concave with slope ue .

Figure 2.2: The indifference curves of younger voters under low inequality

education (g)


(t∗r , t∗r w̄/θq)

(t∗p , t∗p w̄/θq)

taxes (t)
(0,0) (1,0)

Lemma 2.2 yields a number of interesting results, some of which are

depicted in Figure 2.2. The young rich have a much higher demand for ed-

ucation in general than the poor, due to the income effect. That is, the rich

are situated at a level of wealth where the utility gained by attending school

outweighs the loss in income resulting from the cost of receiving that educa-
Section 2.3 50

tion. The poor, on the other hand, can only afford some education, if any,

in the absence of credit markets. As argued by Perotti “The largest compo-

nent of the cost of education, particularly secondary education, is foregone

income. In this case, if the marginal utility of consumption at very low lev-

els of consumption is very high, poor individuals who cannot borrow will not

invest in education. Since their children will start out in the same position,

that dynasty will be caught in a poverty trap, with no investment in human

capital” (1992, 152). This mechanisms is also raised in Gregorio (1996);

Saint-Paul (2001); Christou (2001) and Grossmann (2008). The important

implication of this process is that the poor might prefer to pay lower taxes

and have lower provision of public education than it would otherwise be the

case – that is, in the absence of borrowing constraints.

Another important insight in Lemma 2.2 is that overall demand for pub-

lic education declines as inequality increases. While the poor always de-

mands a lower level, inequality reduces the demand by the rich. This is

because the monetary burden of providing widespread public education be-

comes excessive for the rich to bear. They are better off consuming private


Indeed, Osberg, Smeeding and Schwabish, for example, find that “ high

income families, who can afford private substitutes for public services, have

less incentive to vote for the funding of public services. Noncash benefits

clearly play a large and equalizing role for most families but greater inequality

means that private health insurance or private schooling may be an attrac-

tive alternative to public sector provision for high-income families” (2003,

18). The authors proceed arguing that “education may be a particularly

Section 2.3 51

important case where increased spending leads to more and better education

and thereby promotes economic growth directly, as well as indirectly through

increased social equality and cohesion” (2003, 32)

Therefore, contrary to the demand for pensions by the old, which is ever

increasing given its monetary nature, public education, and other services of

similar nature, are subject to more complex dynamics. Most models of re-

distribution limit their scope to monetary transfers, even when discussing in

kind forms of redistribution. It is common to assume that government trans-

fers the amount necessary to agents and that agents invest the full amount

towards the designated service (for a brief overview see Bourguignon and

Verdier (2000)). This modeling approach obscures the fact that below a

certain income the poor has little choice but to allocate their proceedings

towards subsistence. Investment in human capital is not necessarily a pri-

ority. In a study conducted by the IBGE in 1996 on living standards, out

of five expenditure categories, education was the one where the least was

spent on by poorer households.

Parties and Elections

As a starting point, I consider two exogenous parties L and R. I assume that

party L represents the interests of the poor and that party R represents the

interests of the rich voters. I make the following initial assumptions about

the ideal points of parties.

A3. Parties can only propose platforms that are located in their constituen-

cies contract curve (or Pareto set).

Section 2.3 52

A4. Parties present platforms that weight the ideal points of its constituen-

cies according to their size.

Assumption 3 refers to the credibility of a party in truly representing

the constituency it claims to represent. The chosen platform has to be a

policy that is preferred by all its members to policies outside this set. By the

definition of a Pareto set, platforms outside the contract curve can always

be “improved upon”, in the sense of making all constituencies better off, or

at least one group better off while the other remains indifferent.

Pareto optimality, however, does not take issues of redistribution into

account. In particular, there are policies in the contract curve of two groups

that may favor one group disproportionately over the other. An example of

such a point can be seen in Figure 2.3. The Pareto set of L extends from the

ideal policy of po to the ideal policy of py . Clearly, though, if L announces

(t = 1, g = 0) it is redistributing all taxed income to the old, leaving the

young with no income at all. In such a case it would be misguided to argue

that L represents the interests of the younger poor voters. Assumption 4

comes into play to avoid such extreme cases.

Lemma 2.3: With two policy oriented exogenous parties, L representing

the poor, and R representing the rich according to A3 and A4:

(i) the R party’s ideal policy x∗r is (t = 0, g = 0) when θ

α > w̄
wh and

(t = θt∗r , g = tw̄/θq) otherwise.

(ii) the L party’s ideal policy x∗p is (t = (1 − θ) + θt∗p , g = t∗p w̄/q)

Section 2.3 53

An example of the Pareto set of the Left and Right parties is depicted

in Figures 2.3 and 2.4, respectively. Notice that the lower the demand by

the constituencies of each party, the lower the level of public provision of

education each candidate will announce. Following the results in Lemma

2.2, the higher the inequality in society the lower the demand for education

will be. If we assume platforms are binding, the higher the inequality in

income, the lower the chances of achieving more equality through education.

Figure 2.3: An example of the Pareto set of the poor

education (g)



(t∗p , t∗p w̄/θq)

taxes (t)

Proposition 2.1: If π > 1/2, θ > 1/2, no group forms a strict majority

and A1 though A4 hold, then

θ w̄
(i) if α > wh and θ ∈ Ω, the L party wins the election and implements a

level of education below what is demanded by its younger constituency,

but above the level proposed by R.

Section 2.3 54

Figure 2.4: An example of the Pareto set of the rich under low inequality

education (g)

(t∗r , t∗r w̄/θq)

(t∗p , t∗p w̄/θq)


taxes (t)

θ w̄
(ii) if α > wh and θ 3 Ω, the R party wins and no redistribution occurs.

(iii) if θ
α < w̄
wh and θ < θ∗ , the R party wins and implements a level of

education that is above that proposed by the L party.

(iv) if θ
α < w̄
wh and θ > θ∗ , the L party wins the election and implements

a level of education below what is demanded by both young rich and


where Ω and θ∗ are defined in the appendix as the set of values of θ, such

that the poor strictly prefers L’s policy to R’s policy, and the value of θ such

that py is indifferent between the policy advertised by R and that advertised

by L, respectively.
Section 2.3 55

According to these results, the highest level of public provision of edu-

cation under low inequality lies between the levels demanded by the young

rich the those demanded by the young poor. In high inequality, on the other

hand, this level drops to below what is demanded by the young poor, because

the young rich now prefer to get private as opposed to public education. No-

tice that the income effect still plays an important role. Lower inequality in

a poor country would yield relatively lower levels of demand, possibly below

the levels demanded under higher inequality in richer societies.

Moreover, if the L party weights the interests of its older constituency

heavily enough, the young poor would find it profitable to elect the right

wing party instead. This instance of the poor splitting their votes on issues

of redistribution itself is to some extent a consequence of them deriving

utility from completely different policies.

Proposition 2.1 relies on somewhat restrictive assumptions about the

flexibility of parties to act strategically. Allowing parties to deviate from

their ideal points in order to gain votes raises a number of important issues,

in particular with respect to the meaning of representation. In the following

proposition we relax assumptions A3 and A4 and allow parties to deviate

from their ideal points, as long as their constituencies either remain indif-

ferent or are made better off by the move. That is, if their expected utility

from the move is at least as great as the expected utility from the party

choosing its ideal point, given the other party’s policy. Formally:

A5. Let j ∈ {L, R} and j(i) denote the ith constituency of j, where i ∈

{old, young}, then parties may propose a policy x0j 6= x∗j if:
Section 2.3 56

P r(j wins|x0j , x−j )uj(i) (x0j ) + (1 − P r(j wins|x0j , x−j ))uj(i) (x−j )) ≥

P r(j wins|x∗j , x−j )uj(i) (x∗j ) + (1 − P r(j wins|x0j , x−j ))uj(i) (x−j )) for all

i and j.

Proposition 2.2: If π > 1/2, θ > 1/2, A1 and A2 hold, and no group

forms a strict majority, then

θ w̄
(i) if α < wh and A5 holds, both parties offer the ideal point of the young


θ w̄
(ii) if α > wh , no platform above the Pareto set of the poor will be offered.

One of the important insights here is that the level of public provision

of education advertised is largely determined by the demand expressed by

the young poor constituency. Especially in a situation of high inequality,

there is no reason to expect a level of provision above that demanded by the

young poor. Allowing multiple parties to compete under such circumstances

does not change this particular result.

In sum, the observable implications of the model can be summarized as


(i) The income effect: If poor individuals face borrowing constraints, the

lower their income the higher the utility from current consumption,

such as food and shelter. The poor will only start investing in educa-

tion once they reach an income level where the marginal return from

education is above that of current consumption. Thus at low income

levels, the ideal level of public provision of education will tend to be

Section 2.4 57

(ii) The inequality effect: The higher the inequality the lower the level of

public provision of education demanded by the richer constituencies.

The reason is that it becomes cheaper for this minority to invest in

private as opposed to widespread public education.

(iii) If the preferences of elected officials closely mirrors the preferences of

constituencies, the bigger the proportion of poor voters and the higher

the inequality, the lower the level of public provision of education one

should expect.

Under these circumstances, preferred levels of public provision of edu-

cation will tend to be low in the presence of poverty and inequality. For

elected representatives this means lower incentives to advocate for them.

This is particularly true if the most praised quality in a democracy, that is

representativeness, is present.

2.4 Conclusion

In this chapter I addressed the first of two factors emphasized in this study

that contributes to low levels of redistribution in unequal new democracies:

individuals’ preferences for redistributive policies. I first outlined some of

the strengths and limitations of the existing accounts of such preferences,

using the Brazilian case as a benchmark. Two important points were made.

The first refers to the identity of the pivotal voter, a central player in many

theories explaining variation in levels of redistribution across societies. The

Section 2.4 58

poor – in Brazil, and very likely in most other Latin American countries

– form a majority. That is, in countries where poverty and inequality are

pronounced, the median voter tends to be relatively poor and therefore

considerably distant from the small elite making up the top quintiles of the

distribution. In light of that, any account of low redistribution in unequal

democracies relying on expectations of social mobility and benefits accruing

to the pivotal voter as a result of sustained inequality looses appeal.

The second point referred to the importance of taking the heterogeneity

of redistributive policies into account when studying the determinants of

preferences for them. On the one hand, based on who benefits from these

different types of transfers and services, different cleavages might arise. I

explored two possible ones, the traditional income based cleavage and age.

On the other hand, different types of redistributive policies may be sub-

ject to different preference setting dynamics. While the effect of monetary

transfers on the utility of individuals is straightforward, that of in kind type

of transfers is not. I relied on the case of public provision of education – a

service found to have a considerable impact on lowering poverty and pre-tax-

and-transfer income inequality – to illustrate the point. First, in line with

many recent empirical studies, I find that demand for education by the poor

might be low. Second, in kind forms of redistribution are also susceptible to

low demand by the wealthier given the availability of private exit options.

This effect is particularly pronounced when levels of inequality are high.

As inequality increases so does the share of the rich in financing widespread

public provision of services. This makes the exit options increasingly attrac-

tive. An important point that emerges as a consequence is that while the

Section 2.4 59

preferred levels of provision by the poor might be constrained to lower levels,

when inequality is high these levels might still lie above those preferred by

the wealthy.

These dynamics were explored in the context of a theoretical model

taking the electoral process into account. Under the conventional full infor-

mation approach to modelling elections, I find that low demand for certain

policies can reduce the incentives of politicians to advocate for them. That

is, even under the assumption that the political system works well, achieving

the praised quality of high representativeness might still result in low pro-

vision of certain important policies that help reduce poverty and inequality.

Chapter 3

Preferences: Citizens Views

on Poverty, Inequality, and

In this chapter I investigate preferences for redistributive policies em-

pirically. I begin with an overview of citizens’ perceptions of poverty and

inequality based on a study conducted by the IUPERJ – Instituto Univer-

sitário de Pesquisas do Rio de Janeiro –in 2001. I then provide a more

detailed analysis of preferences for selected redistributive issues.

The analysis follows directly from the framework provided by the the-

oretical model in the previous section. First the model explored cleavages

among voters both along age and socioeconomic lines. These were captured

by comparing pensions and public provision of education. I evaluate the ex-

tent to which these distinctions are present in the data in light of available

questions related to redistribution.

Second, the model yielded two important observable implications with

respect to the effects of poverty and inequality on preferences for redistri-

Section 3.1 61

bution. The first was an income effect, where poor voters are expected to

prefer lower levels of provision of education given borrowing constraints.

The second was the inequality effect, where richer voters are expected to

display lower levels of support for redistribution as inequality increases. In

the particular case of service provision, increases in inequality can lead the

rich to bear a larger share of the costs of widespread provision. This, in

turn, makes it profitable for the wealthier to invest in private exit options

instead of supporting public provision.

Given that data on each type of redistribution considered in the theoret-

ical model are not available, I evaluate the model assumptions and results

based on two issues captured by the existing surveys. First I examine pref-

erences for taxation and social expenditures on health and education taken

together. Then I look at individuals’ declared priority assigned to health

and education as problems that need to be addressed. The chapter con-

cludes with a summary of the main findings, in particular those that will

inform the analysis in the second part of the dissertation.

3.1 Overview

Inequality is acknowledge by a vast majority in Brazil. According to

the study conducted by the IUPERJ in 2009, 83.6% of respondents strongly

agree that inequality is high. Tolerance for inequality, however, is not neces-

sarily low. It is ranked as the biggest problem by a mere 3% of respondents,

as a second most important problem by 4%, and as the third biggest prob-

lem by 7%. Moreover, 40% of respondents agree that big wage differences
Section 3.1 62

are necessary for the country to prosper. In terms of reducing it, most

respondents believe the government is the main actor capable of doing it

(58%) and actually the one responsible for that task (88%). The means for

achieving more equality, however, do not necessarily involve redistribution.

A majority (57%) believes that only economic growth can foster equality,

even in the absence of any redistributive intervention by the government.

Thus inequality is acknowledged but is not seen as a priority. It is also

a problem that respondents believe the government can and should address,

but not necessarily through direct redistributive intervention. Policies that

affect redistribution are not limited to the provision of services and trans-

fers. Fiscal policy is a well studied redistributive tool. In that regard most

respondents believe the rich should pay much more (49%) or more taxes

(33%) than other citizens, independent of how the tax money is spent.

When it comes to financing education, health and housing for those in

need the country seems divided. About half of respondents (49%) percent

believes the government should raise taxes to increase these benefits, while

40% disagree with that. This all the while believing that improving the qual-

ity of public services is the best way to combat inequality (38%). The idea

of poverty as a rural phenomenon seems to be still prevalent, even though

less than 20% of the population is rural. Almost one third of respondents

(27%) believe that land reform is the best policy to curb inequality. Interest-

ingly enough, in second and third places come policies that would increase

worker’s share in their firms’ profits.

Given this general panorama the natural question that arises is whether

different socioeconomic groups display significantly different preferences as

Section 3.2 63

suggested by the model in the previous chapter. Conflicts between rich

and poor are perceived to be strong in the country – 62% of respondents

believe that to be the case – as is the related conflict between employers

and workers (64%). More people see these conflicts as important than those

between whites and blacks (50%), low and middle classes (49%), and old

and young individuals (33%). The next section investigates the extent to

which these perceptions reflect actual underlying differences in tastes for


3.2 Different Groups, Different Preferences?

The model in the previous chapter revealed a number of mechanisms link-

ing the socioeconomic characteristics of voters to the levels of poverty and

inequality in determining preferences for redistributive policies. I use these

findings to guide the analysis that follows. Before proceeding to the results,

however, a number of issues related to the data needs to be discussed.

The model was designed to capture group conflicts based on the trade-

offs citizens make between paying taxes and receiving benefits in return. A

number of items in the Latin American Public Opinion Project (LAPOP)

in 2006 and 2008 and the IUPERJ study in 2001 taps into this issue in

slightly different ways. In 2001 respondents to the IUPERJ study were

asked whether they would agree that taxes should be raised in order to

guarantee better education, health and housing to the needy. The LAPOP

project asked the question from the opposite perspective, referring to health

and education services in a non-targeted manner. Respondents participating

Section 3.2 64

in both waves were asked whether they agree that the government should

reduce the provision of health and education so that taxes could be lowered.

The two types of question differ in ways that might be important, and we

need to keep this caveat in mind when considering the results.

The choice of policies mentioned and the bundling of them into one ques-

tion also impose some limitations. Since the marginal returns of health and

education differ in significant ways – with health bringing higher immediate

benefits to individuals – it is hard to gauge the income effect resulting from

the model. Moreover, age conflicts might get blurred by the fact that health

benefits both. To make up for this limitation I also analyze respondents’ pol-

icy preferences based on the issue they believe should be prioritized by the

government. I isolate in particular three relevant areas: education, health

and poverty and inequality. I consider each of these individually to check

whether any of the socioeconomic groups defined in the model is more or

less likely than the others to name the issue. Since these do not capture

preferences for redistribution itself, the analysis is kept simple. It only in-

cludes the basic socioeconomic characteristics of individuals controlling for

the levels of poverty in their respective states. The dynamics relating in-

equality to the redistributive preferences of the rich should not apply in this


Table 3.1 reports the answers to all these items in each of the surveys an-

alyzed. On the trade-off between taxes and public services provision about

half of the respondents in 2001 agreed that taxes should be raise to guar-

antee health, education and housing for those in need. In 2006 and 2008 a

significant majority disagreed with reductions in health and education ser-

Section 3.2 65

Table 3.1: Support for social policies

2001 2006 2008

Prefers More (or Same) Levels
of Taxes and Services 55% 83% 91%
Most Important Problem:
Education 4.5% 2.3% 4.2%
Health 18.7% 9.3% 18%
Poverty and Inequality 14% 8.5% 10%
Three Most Important Problems:
Education 31%
Health 57%
Poverty and Inequality 49%
N 2000 1196 1481

Notes: Source: Scalon (2001) and The AmericasBarometer by the Latin American Public Opinion
Project (LAPOP),

vices so that taxes would be lower. The differences observed can come from

three factors or a combination of them. First, there might be a trend to-

wards higher tastes for redistribution. Second targeting might be a crucial

factor accounting for why less respondents agreed with higher taxes to fi-

nance services in 2001 – where the question referred specifically to the needy

as opposed to all. Finally, the observed differences between 2001 and more

recent years can be attributed to the number of individuals who agree that

current levels of tax and service provision should remain unchanged. Those

who believe that to be the case would disagree with more taxes in 2001,

but agree that taxes and services should not be lowered in 2006 and 2008.

While these are interesting issues, they need not to detain us here. In either

case, if any difference in tastes between different groups exist, they should

be noticeable.
Section 3.2 66

For analytical purposes it is worth noting that the very high proportion

of respondents falling in the same answer category in 2006 and 2008 can

make it hard to detect differences in tastes. The reason is that there is too

little variation to be explained. This is probably why surveys conducted else-

where, in particular in Europe and the US, usually cover the whole spectrum

of answers. That is, respondents can choose from a wider menu of options

including: increasing tax and benefits, keeping them unchanged, reducing

both, or simply reallocating funds among different forms of redistribution.

Unfortunately this rich set of items was never included in surveys conducted

in Latin American countries.

With regards to explanatory variables, I begin with the central measure

for this study: inequality. As a growing number of studies point out each

measure of inequality emphasizes a different aspect of the income distri-

bution. The widely utilized Gini coefficient, for example, places particular

emphasis on the middle portion of the distribution, being very sensitive to

changes in that range but unaffected by changes in the tails of the distribu-

tion. In a similar fashion, each income ratio that can be computed captures

the action in the particular quantiles of the distribution used. Most impor-

tantly, all these possible indices are not necessarily highly correlated. Thus

the choice of measure of inequality should always be justified based on clear

theoretical grounds.

In the model in the previous chapter inequality is found to affect in par-

ticular the preferences of the wealthier voters. The idea being that the higher

the inequality, all else equal, the higher the tax burden falling on the rich to

finance widespread provision of services. This makes it profitable for them

Section 3.2 67

to opt out of public provision and finance only their own private consump-

tion, thus reducing their political support for public provision. Contrary to

the effect of poverty on the preferences of the poor, the effect of inequality

on the wealthy is not particular to the type of redistributive policy studied

(e.g. education). Any form of redistribution that excludes the elite or to

which an exit option is available would be compatible with the proposed

mechanism. This means that the grouping of health, education and housing

for the needy into one question, should not affect the results.

The relevant inequality measure in this case, as described in the model,

is the ratio of mean income to the elite’s income. This is exactly the measure

employed in the analysis. More specifically I use the income of the richest

20% as a proxy for the income of the wealthiest. Notice that the higher the

ratio the lower the inequality. This measure is computed for each state in

the country, which corresponds to the lowest meaningful level of disaggrega-

tion. Many of the services studied have been decentralized with states and

municipalities working together to finance and manage them.

Table 3.2: Ratio mean income to income of the fifth quintile

2001 2006 2007

min 0.301 0.299 0.304
mean 0.326 0.345 0.348
max 0.365 0.385 0.382
N 26 20 21

Notes: Source: IPEA Data.

As is clear from Table 3.2, there is not much variation across states on

levels of inequality. The average per capita income in each state tends to
Section 3.2 68

represent about one third of the average per capita income earned by the

richest quintile. The lack of variation in inequality across states makes it

harder to detect statistically significant effects of inequality on preferences

for redistribution.

Per capita income categories are based on reported household income

levels and the number of people in the household. The IUPERJ study

provides a continuous variable of household income. The LAPOP survey,

however, breaks it down into categories. To recover a continuous measure I

assign to each individual the mean of the log of the income range. Estimating

the actual wealth of respondents is no easy task. Proof of that is the rich

and complex set of questions regarding income included in household surveys

nowadays. A simple question like the one asked in public opinion surveys

is not likely to capture all the complexities involved in that assessment, but

rather a self reported snapshot.

Since wealth and educational attainment are closely related I also include

in the estimations information on schooling. The educational attainment of

individuals is measured based on categories for each schooling level. The

first group being illiterate with no schooling, the second corresponding to

primary school, the third to secondary school, the fourth to high school and

finally the fifth including those who attended college.

To capture the extent of age conflict, the specifications include age mea-

sured in years. As is common in the literature I also include age squared as

a control. Finally, to capture the extent of poverty in each state I employ

the log of the average income of the poor. Since poverty and inequality
Section 3.2 69

are closely related, I control for poverty levels not to attribute to inequality

whatever effects it might have.

The models are specified as follows:

P r(yi = 1) = logit−1 (α + β1 IN COM E + β2 EDU CAT ION + β3 AGE

+ β4 RAT IO + β5 AGE ∗ AGE

+ β6 RAT IO ∗ IN COM E


where i denotes the individual respondent, and y = 1 means the respondent

agreed with more taxes to finance more services to the poor in 2001, and

disagreed with reducing health and education services to lower taxes in 2006

and 2008.


Given the nonlinearity of the specification, further complicated by the in-

troduction of interaction effects, the coefficients of the estimation (shown

on Table 3.3) tell us very little about the actual estimated effects and their

statistical significance. The marginal effect of each variable depends on the

values taken by the other variables in the model. We work instead with

predicted probabilities, which are directly linked to the hypotheses being

tested and easily interpreted.

The model in the previous chapter was designed to explore two kinds of

conflicts: one between different age groups and the other between different
Section 3.2 70

Table 3.3: Logit results: taxes and health and education spending
IUPERJ 2001 LAPOP 2006 LAPOP 2008
Intercept 17.04 12.45 −2.86
(3.97) (7.22) (11.43)
Income −0.26 −1.00 −0.39
(1.12) (1.42) (2.47)
Age −0.03 0.05 −0.00
(0.02) (0.02) (0.05)
Age squared 0.00 −0.00 −0.00
(0.00) (0.00) (0.00)
Education −2.08 −2.25 3.03
(1.50) (1.85) (3.08)
Ratio Mean/Rich −15.73 −27.23 39.80
(12.25) (19.47) (29.23)
Ratio * Income 0.24 3.22 1.43
(3.43) (4.18) (7.19)
Ratio * Education 5.84 6.89 −8.65
(4.59) (5.45) (8.94)
Log Poor Income −2.45 ∗ −0.74 −2.01
(0.47) (0.71) (1.85)
N 1629 1116 771
AIC 2147.06 1005.09 436.50
BIC 2341.31 1185.72 603.81
log L −1037.53 −466.54 −182.25
Standard errors in parentheses

income strata, with particular emphasis on the effect of personal income

on preferences for education. Thus I discuss the results based on the four

groups that emerge following these classifications. For consistency I employ

the same labels used in the model: young poor (py), old poor (po), young

rich (ry) and old rich (ro). To compute the predicted probabilities I first

create one profile for each group of respondents. The values are set as

Section 3.2 71

Young Poor Age is set to 20 years old, the educational attainment to

primary schooling and the income to the first quintile group.

Old Poor Age is set to 60 years old, the educational attainment to primary

schooling and the income to the first quintile group.

Young Rich Age is set to 20 years old, the educational attainment to col-

lege and the income to the fifth quintile group.

Old Rich Age is set to 60 years old, the educational attainment to college

and the income to the fifth quintile group.

Poverty levels are kept at the mean sample value and inequality is allowed

to vary according to the effect being investigated.

I begin with a simple plot of predicted probabilities for each group letting

levels of inequality vary. Figure 3.1 displays the results for each survey-year.

The first thing to note is that the age conflict is not very pronounced. As

previously mentioned, one of the reasons might the bundling of different ser-

vices together in one question. In the model I was able to isolate two services

with clear distinct constituencies in terms of age. This is unfortunately not

the case here.

Age differences turn out to be statistically significant over the whole

range of inequality in the 2001 sample. In that case, older respondents

tended to be less likely to agree to higher taxes and higher targeted spend-

ing on the destitute. It is difficult, however, to know whether this conflict

has reduced over time or whether it was prompted by the wording of the

item. As already mentioned, in 2001 the question was posed differently. It

Section 3.2 72

Predicted Probabilities: Same or More Services 1.0 1.0

Predicted Probabilities: Same or More Services

● ●

● ●
● ● ●

● ● ●
0.8 0.8 ●

! ●
● ●

● ●
0.6 ● ● ● ●
● ● ● ●
● ●

0.4 ● 0.4

0.2 0.2

0.30 0.31 0.32 0.33 0.34 0.35 0.36 0.30 0.31 0.32 0.33 0.34 0.35 0.36
Equality (Mean to High Income Ratio) Equality (Mean to High Income Ratio)

(a) IUPERJ 2001 (b) LAPOP 2006

Predicted Probabilities: Same or More Services

● ● ● ● ● ●

● ● ●
● ●
● ●
● ●

0.8 ● Income

● ! poor
! rich

! old

0.30 0.31 0.32 0.33 0.34 0.35 0.36
Equality (Mean to High Income Ratio)

(c) LAPOP 2008 (d) Legend

Figure 3.1: Predicted probabilities of agreeing with same or more taxes and

asked respondents whether they agree to pay more taxes to guarantee bet-

ter housing, health and education to the needy. The LAPOP items make

no reference to particular beneficiaries and asks whether respondents would

0.0 0.5
less education
and health
in order to 2.5
reduce the tax
Section 3.2 73

Particularly relevant to this analysis is the result regarding the effect

of inequality on the preferences of wealthier respondents. In every survey-

year we observe a trend among richer respondents to be more likely to agree

with higher services provision, including targeted ones, as equality increases.

This effect is statistically significant in 2006, but fails to attain conventional

levels of significance in the other two years. One of the reasons might be

the low levels of variation in the measure of inequality, as noted earlier.

Regarding the income effect on the preferences of the poor we obtain

the following results. First, notice that the levels of agreement by the poor

are estimated to be similarly high in all cases. Richer respondents seem

to be more sensitive to the wording of the question. While they tend to

be less likely to agree with more taxes and more targeted provision than

the poor, the reverse is true when no targeting is mentioned. In the latter

case they tend to be more likely than the poor to disagree with provision

being lowered so that taxes are reduced. These differences are statistically

significant (see Figure 3.21 ) both in 2001 – where the poor are more likely

to agree with improved targeted provision of services – and 2006 – where at

higher levels of equality, the rich is more likely to agree with same or higher

levels of provision of services.

Hence, all policies considered, results indicate that the poor are likely

to display stronger support for improved provision when they are targeted,

but that their support for non targeted provision lies below that of wealthier

respondents. As shown in the previous chapter, total public expenditures on

both health and education are not necessarily progressive. In the particular
The same pattern holds if we consider older respondents.
Section 3.2 74

First Differences Among Young: Poor − Rich 0.8 0.8

First Differences Among Young: Poor − Rich

0.6 0.6

0.4 ●
● 0.4

0.2 ● 0.2

● ●

0.0 0.0 ●

−0.2 −0.2 ●

● ●

−0.4 −0.4

−0.6 −0.6

0.30 0.31 0.32 0.33 0.34 0.35 0.36 0.30 0.32 0.34 0.36 0.38
Equality (Mean to High Income Ratio) Equality (Mean to High Income Ratio)

(a) IUPERJ 2001 (b) LAPOP 2006

First Differences Among Young: Poor − Rich




0.0 ● ● ● ● ●
● ●
● ●
● ●




0.30 0.31 0.32 0.33 0.34 0.35 0.36

Equality (Mean to High Income Ratio)

(c) LAPOP 2008

Figure 3.2: First differences in predicted probability comparing the young

poor with the young rich.

case of education, statistics show that more is spent on tertiary levels – which

favors the rich disproportionately – than on primary and secondary educa-

tion. The results just described might simply reflect individuals’ knowledge

of this fact.
Section 3.2 75

The actual effect of income uncovered in the previous chapter, however,

applies in particular to services whose initial marginal returns are low rela-

tive to that of income. Long term investments such as education are good

examples. Health and housing, however, can have relatively large marginal

returns in the short run. This means that effect cannot be fully recovered

based on the dependent variable employed so far.

In order to gauge that effect, I consider a different question common to

all these surveys. It asks respondents to name the most important problem

facing the country. I recode the answer to this question into a series of

dummy variables where 1 denotes the respondent named a particular issue

and 0 that she named some other problem. I then specify logistic regressions

on just two of the items available: education and health. The specification

is kept simple. The primary aim is to check whether different socioeconomic

groups hold different priorities concerning these services.

I base the discussion on predicted probabilities. For reference, the es-

timated coefficients and standard errors are reported in Table 3.4 – where

education is the dependent variable – and Table 3.5 – where health is the

dependent variable. Note again that since these are nonlinear models, the

significance of the estimated effects is a function of the values taken by all

the variables. Hence, predicted probabilities with confidence intervals help

shed light on the statistical significance of the differences reported.

It is worth noting that in 2001 respondents were given the option to

choose up to three issues and were then asked to provide a ranking of the

most pressing problems. The results displayed in both Tables 3.4 and 3.5

and on Figure 3.3 are based on whether the respondent cited each issue at
Section 3.2 76

Table 3.4: Logit results: education as an important problem

IUPERJ 2001 LAPOP 2006 LAPOP 2008
Intercept 1.53 −7.34 −2.45
(1.74) (7.49) (7.95)
Income 0.02 0.23 0.15
(0.04) (0.17) (0.14)
Age 0.02 0.07 0.05
(0.02) (0.08) (0.08)
Age squared −0.00 −0.00 −0.00
(0.00) (0.00) (0.00)
Education 0.15∗ 0.68∗ 0.62∗
(0.06) (0.26) (0.20)
Log Poor Income −0.73 −0.27 −1.03
(0.42) (1.74) (1.85)
N 1817 1116 828
AIC 2238.79 240.61 286.51
BIC 2370.91 361.03 399.77
log L −1095.40 −96.31 −119.26
Standard errors in parentheses

indicates significance at p < 0.05

all, either as a priority, as a second ranked problem or as a third ranked one.

Running the analysis using only the first ranked problem in 2001 yields very

similar results to those obtained with the LAPOP data.

Looking at Figure 3.3, we observe that in every survey-year the poor

were significantly less likely to name education as an important problem than

wealthier individuals. Moreover, in all three years they are significantly more

likely to name health rather than education. These findings are consistent

with the propositions of the theoretical model. The investment characteristic

of education combined with the lack of access to credit markets can severely

constraint the poor’s preferred levels of public provision of education.

Section 3.2 77

Predicted Probabilities: Education Important Problem

Predicted Probabilities: Health Important Problem

0.7 0.7

0.6 0.6 ●

0.5 0.5

0.4 ●
● 0.4

0.3 ● 0.3

0.2 0.2

0.1 0.1

0.0 0.0

po py ro ry po py ro ry
Groups Groups

(a) Education IUPERJ 2001 (b) Health IUPERJ 2001

Predicted Probabilities: Education Important Problem

Predicted Probabilities: Health Important Problem

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1 ●

● ●
● ●
● ●
0.0 0.0

po py ro ry po py ro ry
Groups Groups

(c) Education LAPOP 2006 (d) Health LAPOP 2006

Predicted Probabilities: Education Important Problem

Predicted Probabilities: Health Important Problem

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2 ● ●

● ●

0.1 ● 0.1

● ●
0.0 0.0

po py ro ry po py ro ry
Groups Groups

(e) Education LAPOP 2008 (f) Health LAPOP 2008

Figure 3.3: Predicted probabilities of mentioning health and education as

the most important problem facing the country. Bars correspond to 95%
confidence intervals.
Section 3.3 78

Table 3.5: Logit results: health as an important problem

IUPERJ 2001 LAPOP 2006 LAPOP 2008
Intercept −4.17∗ −6.60 1.50
(1.63) (3.87) (4.26)
Income −0.06 −0.02 0.10
(0.04) (0.08) (0.07)
Age −0.00 0.17∗ 0.06
(0.02) (0.04) (0.04)
Age squared 0.00 −0.00∗ −0.00
(0.00) (0.00) (0.00)
Education −0.15 ∗ −0.16 −0.21∗
(0.06) (0.12) (0.10)
Log Poor Income 1.19∗ 0.33 −0.92
(0.40) (0.89) (1.00)
N 1817 1116 828
Standard errors in parentheses

indicates significance at p < 0.05

It is important to notice that one of the best predictors of prioritizing

education is educational attainment itself. This suggests that as public

provision of education improves, preferences for education might change

too. Most importantly, they might change in a direction that would put

more pressure on representatives to advocate for further improvements.

The results regarding health are mixed. Taking all three options given

to respondents into account in 2001, the poor, both old and young, were

significantly more likely to cite it as an important problem. If we focus on

respondents’ first options only, the pattern in 2001 is very similar to that

of 2008, where no significant distinction exists among the different groups.

The trend is for older voters to be more likely to choose health. Indeed in

2006 they are significantly more likely to do so.

Section 3.3 79

3.3 Conclusion

In the empirical analysis conducted in this chapter I looked for evidence

of four issues raised in my theoretical model. First, I explored the extent

of variation in preferences for redistribution among individuals from differ-

ent socioeconomic backgrounds. I found statistically significant differences

whether the dependent variable was about taxes and the provision of health

and education taken together, or about health and education as govern-

ment’s policy priorities.

The second issue studied refers to age cleavages in tastes for redistribu-

tion. The evidence to that effect was weaker. It was statistically significant

only in 2006 regarding health as a government priority. Older respondents

were more likely to rank it as a top priority compared to younger respon-

dents. In other cases, differences based on age might have been less clear

for several reasons. To begin with, the theoretical model relied on an ex-

treme case, where players were particularly self-interested. In the presence

of altruism, differences in preferences among distinct age groups are likely

to decrease. Another possible reason for the lack of significant differences

might be related to the composition of poorer households. They tend to

be larger, not only due to a higher number of children, but also to the fact

that both young and old share the same home. Whenever this is the case,

the monetary benefits accruing to the old in the form of pensions, for ex-

ample, might also benefit the younger members of the household. In order

to better evaluate age cleavages in tastes for redistribution we would need

Section 3.3 80

survey questions that make clear distinctions between the different types of


The third issue relates to the income effect on levels of preferred provision

of services. In line with the model results, poorer respondents were signif-

icantly less likely to rank education as a government priority than richer

participants. The results regarding preferences for taxation versus the qual-

ity of provision were interesting. Poorer respondents were more likely than

richer respondents to agree that taxes should be raised in order to provide

better health, education and housing to the needy. The difference was es-

pecially pronounced where inequality is higher, as suggested by the model.

When the question was asked in general terms – simply whether health

and education services should be reduced in order to lower taxes – poorer

participants were more likely to agree that health and educational services

should be reduced so that taxes could be lowered. Given that in many cases

both these services benefit the better-off to a greater extent, the pattern

uncovered might just reflect individuals’ knowledge of this fact.

Finally, the fourth issue concerned the effect of inequality on the pre-

ferred levels of service provision by the wealthy. While results obtained were

in line with the theoretical expectations, they failed to attain statistical sig-

nificance in two out of three survey-years. One reason for that might be

the very little variation in levels of inequality across the different states in


Chapter 4

Representing Preferences:
Candidate Selection Under

4.1 Introduction

In the first part of the dissertation I paid particular attention to preferences

and how poverty and inequality might negatively affect the levels of political

support enjoyed by different redistributive policies. The next step, and the

focus of this part of the dissertation, is to investigate how voters choose

candidates to represent these preferences.

In the first model I kept this decision process simple, by adopting the

conventional electoral framework of perfect information. In the words of

Kitschelt and Wilkinson (2007) I relied on the “responsible party govern-

ment” model. Under this assumption, I found that politicians have little

incentive to advocate for redistributive policies characterized by low levels

of support. This means that even democracies achieving high levels of rep-
Section 4.1 82

resentativeness would likely display low levels of provision of such policies.

While useful to isolate the effects of preferences on policy outcomes, there

are many reasons to believe that the “responsible party government” model

does not offer the best characterization of elections in new democracies.

One possible reason for why redistribution might be low in unequal coun-

tries – even lower that what is preferred by a majority – is the mismatch

between the policies representatives stand for and the policies preferred by

their constituencies. Representation failures can result from many factors.

In this chapter I follow a similar approach from to the one adopted in the

first part of this dissertation. While it is possible that institutional failures

in the workings of newly established democracies may hinder representation,

I concentrate on a more basic factor: uncertainty on the part of voters about

the actual policy position of candidates. Given the focus of this study on

relatively new democracies, it is reasonable to expect uncertainty to play

an important role, making it a natural starting point in studying the voter-

candidate link.

The main claim in this chapter is that uncertainty can account for both

lower than expected levels of redistribution, taking individuals’ preferences

as given, and delays in the implementation of policies favored by a majority.

The lack of information about candidates’ ideal policies leads to a learning

process where voters can update their beliefs about the stance of a candidate

once she serves as an incumbent. During this learning process, however, de-

lays can happen due to the probabilistic nature of the electoral choice and

the fact that a majority might reelect incumbents that do not closely mirror

their preferences – even under full information about these preferences. In

Section 4.1 83

particular, poor voters might find it optimal to reelect incumbents imple-

menting less redistribution than they favor.

The chapter is organized as follows. I begin with a brief review of some

of the main arguments in the literature about why elections might yield

low representation. I then introduce the central theme of the chapter and

develop a theoretical model to explore the effects of uncertainty on the

likelihood of electing representatives that advocate for redistribution.

In order to provide a richer account of this voter-candidate link I make

simplifying assumptions about individuals’ preferences relative to the theo-

retical model developed in Chapter 2. First, given the lack of strong evidence

in support of age cleavages among individuals, I abstract from these consid-

erations in the analyzes that follow. Second, one of the main points in the

previous chapter was to show that levels of preferred provision of certain re-

distributive policies can be low. I explore this result from two perspectives.

First, under high inequality, the model in Chapter 2 suggested that while

demand by the poor might be low for services like education, it would still

lie above the levels preferred by the rich. Thus on a scale going from high

redistribution favoring the poor to low redistribution, the poor would still

be located to the left of wealthier voters. This is also clearly the case when

it comes to transfers. Thus I assume the poor are located to the left of the

ideological spectrum.

Second, given the assumption of full information in the previous model

we found that the political prospects of policies enjoying low levels of support

are meager. The maximum level of provision that might arise corresponds to

the levels preferred by the majority. Thus one of the questions that arise is
Section 4.2 84

the following: are these societies doomed to an equilibrium of low provision?

To answer this question I explore the political prospects of redistribution

beyond what is preferred by a majority. As will be shown, the fact that

uncertainty can lead to mismatches in representation means that incumbents

choosing higher levels of provision might still get reelected. Hence it is

possible to observe instances of considerable increases in the provision of

certain redistributive policies, as long as there are candidates who endorse


After presenting the results of the model I discuss some important impli-

cations it has to the theories reviewed in the first section. The chapter then

concludes with a summary of findings and some observable implications to

be explored in the next chapter.

4.2 Elections and Representation

Three issues that might compromise the match between voters’ preferences

and that of their representatives received a great deal of attention from

scholars. First, many researchers sought an explanation to low levels of re-

distribution on low turnout by the poor (Hicks and Swank 1992; Franzese Jr.

2001; Shapiro 2002; Gylfason and Zoega 2003; Harms and Zink 2003; Jencks

2002; Robinson 2009). That is, although the poor might be a majority in

the population, they are not a majority among those who actually turn out

to vote. As explained by Jencks:

Section 4.2 85

If everyone votes, the electorate is by definition representative

of the population and politicians need to keep all income groups
happy. When people stop voting, turnout almost always falls the
most among the poorest and least educated. As the income gap
between those who vote and the population as a whole widens,
politicians have less incentive to push legislation that benefits the
lower half of the income distribution. (2002, 63)

Brazil is an interesting case in that respect given that it is highly un-

equal, redistribute relatively little to the poor and has enforced compulsory

voting. If low turnout was one of the major reasons why the poor do not get

redistribution, it looks like Brazil would be much more redistributive than it

is today. According to data available from the Brazilian Electoral Tribunal

the poor – proxied by low levels of educational attainment – make up about

70% of the electorate (see Table 4.1). Given that turnout is usually around

83%, even if we assume all those not turning out to vote are poor they would

still be a majority among those who vote.

Table 4.1: Justification of absence in presidential elections by educational


Percentage of Absent Percentage of

Voters by Category Electorate
2002 2006 2002 2006

Primary or less 4.5% 6.1% 71% 66%

Secondary 5.8% 7% 24% 28%
College 8.2% 7% 5% 6%

Notes: Absent voters are those who did not vote but justified their absence to the electoral

As shown in Table 4.1 not all absent voters are poor. While data on

turnout by education levels is not available, we know the educational at-

Section 4.2 86

tainment of those who justified their absence on the election day1 . They

correspond to about half of the voters who did not vote. Thus the figures

suggest that turnout is not necessarily lower among the less educated.

The second issue refers to illicit practices such as vote buying and pa-

tronage, where the weight of policy relative to other considerations during

elections is assumed low. A prominent line of argument is that in many

countries clientelist links develop between parties and voters. In their quest

to hold office, clientelist parties distribute goods and patronage instead of

conquering voters by implementing the policies they favor. Voters, in turn,

especially the poor, engage in such exchanges by “selling” their votes for the

handouts and favors. That is, voters are assumed to value the goodies more

than policy, either because of ignorance (Kitschelt and Wilkinson 2007),

time horizons, or poverty itself – which makes them value goods relatively

high (see Stokes (2007)).

Although not developed to explain the relationship between inequality

and redistribution, theories of clientelism tend to highlight the link between

poverty and the pervasiveness of such practices. Robinson (2009), for exam-

ple, argues that parties engaging in clientelism have no incentive to improve

the lot of the poor through redistributive policies since this would increase

the price of their votes. In a similar vein Medina and Stokes (2002) contend

that “(r)edistribution hurts clientelistic incumbents” because it “undermines

the incumbent patron’s advantage over any challenger” (2002, 17). Robin-

son and Verdier (2003) argue further that “clientelism becomes a relatively
Compulsory voting is enforced through fines in Brazil. The only way for people not
to vote and not to pay the fine is to justify the absence on the day of the elections. This
is allowed, as long as the individual is 100 km away from her designated pooling station.
Section 4.2 87

attractive political strategy in situations with high inequality” (2003). The

consequences are thus clear. Low redistribution would be more likely exactly

where poverty and inequality are more endemic.

The third issue refers to the effect of resources in biasing representa-

tion towards the interests of the better-off – through bribes and other illicit

means or through campaign contributions. By means of such tactics only

certain candidates might be competitive and have a chance of winning elec-

tions. If those who enjoy this advantage are candidates representing the

interests of the wealthy, while a majority in the electorate is low-income, we

would again observe poor representation. Studies have attempted to gauge

the extent of that effect empirically by weighting the political influence of

citizens according to their incomes. The results, however, seem to account

only for a small part of inequality (Bénabou 2000; Ferreira 2001).

There is, however, a more fundamental issue that is rarely explored

in this context and that has important implications to all these accounts,

namely lack of information. Not knowing the ideological inclinations of

candidates can often lead to inconsistencies in representation. Clear ideo-

logical positions are not features inherent to any political system. Parties in

developed democracies established their reputations for being left or right

over years serving terms in office, with voters observing their behavior and


One shouldn’t expect citizens of new democracies, where new parties

and new faces constantly emerge in the political scene, to display the same

levels of knowledge about the policy preferences of candidates as citizens of

mature democracies. When information level is low, it might be reasonable

Section 4.3 88

for voters to base their choice on personal qualities and attributes or even

gifts. As voters learn from the choices of incumbents, policy can become

increasingly important, strengthening the programmatic links between vot-

ers and their representatives. How this process unfolds, however, depends

on the motivations and electoral strategies adopted by candidates and their


In order to explore these dynamics formally I model elections as an

incomplete information game. I assume voters have no knowledge about

candidates’ policy positions in the initial period and update their beliefs

about it once candidates serve as incumbents. An important issue explored

is that of politicians’ motivations. Candidates’ policy choices depend on the

extent to which they value policies relative to simply holding office. Thus

candidates’ strategies will be a function of the possible trade-offs between

seeking votes and adhering to a particular ideological stance, as suggested,

for example, by the clientelism literature.

In contrast to the model developed in the first part, in this section I

simplify assumptions about preferences to focus on a richer account of can-

didate selection. The model yields a number of interesting insights that

have important implications to the more conventional accounts summarized

Section 4.3 89

4.3 Voting with Uncertainty: Policy Positions and

the Choice of Representatives

The Model

I model a polity with three groups of voters, i = {p, m, r}, where p stands

for poor, m for middle and r for rich, and three types of candidates, t =

{L, C, R}, where L stands for left, C for center, and R for right. I consider

a unidimensional policy space where policies, denoted by x, can take the

following values: x ∈ {0, 21 , 1}. Each member of a group of voters share the

same ideal policy denoted by x̃i . Each type of candidate is associated with

an ideal policy that I identify by x̃t . These are given by:

• x̃L = x̃p = 0

• x̃C = x̃m = 2

• x̃R = x̃r = 1

I model a two-period game of incomplete information where voters know

their own types, but do not know the candidates’ types. All players share a

common prior distribution of these types. This is given by pL > 0, pC > 0

and pR > 0, denoting the probabilities of a candidate being at the left, the
center or the right of the ideological space, and where pt = 1. Candidates

are assumed to have full information. They know their own type and those

of the voters. The distribution of voters in the population is known to all

players and is given by pp , pm and pr , where pi = 1.
Section 4.3 90

I assume the group of poor voters form a majority in the population,

that is pp > 12 . This is a reasonable assumption for many new democracies.

Furthermore, I assume that they all vote in each election.

I model the utility of candidates as a function of their preferred policies

(x̃t ) and the extent to which they are office oriented (Calvert 1985; Duggan

and Fey 2005). This is captured by the term w ≥ 0. When w = 0 candidates

are said to be purely policy oriented. As w increases the value of office grows

relative to that of policy and candidates become increasingly office oriented.

Candidates’ utilities are given by:

Ute (x) = −(x − x̃t )2 + w,

Utne (x) = −(x − x̃t )2 .

where e stands for “elected” and ne stands for “non-elected”. I assume that

candidates and voters have Euclidean preferences for policy. Voters’ utilities

are simply a function of policy, and can be written as:

Ui (x) = −(x − x̃i )2 .

Candidates can be reelected once, that is, they do not serve more than

two terms. The game sequence can be described as follows.

(1) Nature draws the first incumbent2 ;

Since in the first period voters have no information about candidates’ types, I assume
they “flip a coin” in deciding who to vote for.
Section 4.3 91

(2) Incumbent implements policy x1 , where the superscript 1 denotes the

first period;

(3) Voters observe x1 and, whenever possible, update their beliefs about

the incumbent’s type via Bayes’ Rule.

(4) Nature draws a challenger;

(5) Voters cast their votes;

(6) Winner becomes the incumbent in the second period and implements

x2 . Game ends.

Candidates strategies consist of a pair of policies denoted st = (x1t , x2t )

chosen to maximize their expected utilities as incumbents in the first and sec-

ond periods, given the strategies of the other players. Let S = (sL , sC , sR ).

Candidates’ expected utility can be written as follows:

EUt (st , s−t ) = Ute (x1t ) + Γ(γi )[Ute (x2t )] + (1 − Γ(γi )) pt Utne (x2t )

where Γ is a function that maps voters’ strategies denoted by γi (x1 ) (de-

fined below) to a probability of the incumbent being reelected. Let σ =

(γp , γm , γr ), then Γ : σ → [0, 1].

Voters’ strategies consist of deciding whether to vote for the incumbent,

γi (x1 ) = 1, or the challenger, γi (x1 ) = 0. If indifferent, I assume a voter

votes for the incumbent with probability 21 .

In deciding who to vote for, voters compare their expected utility from

reelecting the incumbent (I) with that of electing the challenger (Ch) instead.
Section 4.3 92

These expected utilities are given by:

µ(t|x1 ,S) Ui (x2t ) , and
EUi (I) =
pt Ui (x2t ) .

EUi (Ch) =

where µ(t|x1 ,S) is the updated probability that the incumbent is of type

t given the policy she implemented in her first term. This probability is

updated whenever possible following Bayes’ Rule.

Given the features of the model, I employ the Perfect Bayesian Equi-

librium concept. A complete characterization of such equilibria includes a

full description of beliefs, that is, not only those updated beliefs following

equilibrium strategies but also those following off-the-equilibrium-path ac-

tions. Given the latter can take numerous forms generating a multiplicity

of equilibria, I apply the divinity refinement.

γi (x1 ) is a step function, defined as follows:

γi (x1 ) = 1 if EUi (I) > EUi (Ch)

γi (x1 ) = 0 if EUi (I) < EUi (Ch)

γi (x1 ) = if EUi (I) = EUi (Ch)

Γ(γi ) is simply a function that weights each of the groups’ probabilities

by the groups’ sizes to determine whether the incumbent gets reelected or

Section 4.3 93

Clearly, voters should consider how the other players vote in making

their final decisions. Given the assumption that the poor form a majority

of the electorate and always turn out to vote, unless they are indifferent

between the two contenders, their vote is decisive and strategic voting is not



The model yields three different types of pure strategy equilibria. In the

first type, candidates value office to a relatively high extent. They find it

thus profitable to choose the policy that pleases the majority of poor voters

in the first period to secure a chance at reelection. As a result voters are

indifferent between the incumbent and the challenger.

In the second equilibrium, the value of office is low enough to make the

right-wing candidate choose her ideal policy in the first period even if that

means no reelection. It is, however, still profitable for the moderate type to

mimic the left candidate and implement the ideal policy of the poor. Voters,

in turn, reelect only incumbents who implement their preferred policy in

their first term.

Finally, for relatively low values of office we get an equilibrium where

each candidate implements their ideal policy, revealing their exact type.

Both the left and the moderate types get reelected. That is, even with full

information about the incumbent’s type, poor voters reelect a non-left rep-

Section 4.3 94

Proposition 4.1: If w > 2 − pL − p4C , and pR < 14 (1 − pC ) there exists a

unique Perfect Bayesian Equilibrium where:
• candidates play:
S ∗ = (x1L = 0, x2L = 0), (x1C = 0, x2C = ), (x1R = 0, x2R = 1) ;

• voters’ beliefs are:

µL|0,S ∗ = pL , µC|0,S ∗ = pC , µR|0,S ∗ = pR , µR|{ 1 ,1},S ∗ = 1;


• voters choose: γi (0) = 2 for all i, and γp ( 21 ) = γp (1) = 0.

The first equilibrium is described in Proposition 4.1. Because every in-

cumbent decides to implement the same policy, voters cannot learn about

their ideological inclinations, making them indifferent between the incum-

bent and the challenger during elections. This equilibrium is supported by

two conditions. First, candidates are sufficiently office oriented, as opposed

to caring mainly about policy. That makes it profitable for both the right

and moderate type to concede on policy – by implementing a leftist policy

in the first period – for a chance to remain in office for a second term.

The second condition is the shared belief that a right type is relatively

rare. This means the expected utility from an unknown challenger is higher

than that expected from the moderate type. Thus poor voters do not reelect

incumbents implementing x1 = 21 , guaranteing C has no incentive to deviate.

Another possible equilibrium (stated in Proposition 4.2) is one where

only the left and the moderate candidates choose the preferred policy of the

poor in the first period and get reelected. The Right candidate plays her

ideal policy in both periods, but does not remain in power for a second term.
Section 4.3 95

Proposition 4.2: If 14 ( pCp+pC

− pC ) < pR < 14 (1 − pC ) and p4C < w <
1 − pL − p4C , there exists a unique separating Perfect Bayesian Equilibrium

• candidates play:
∗ 1 2 1 2 1 1 2
S = (xL = 0, xL = 0), (xC = 0, xC = ), (xR = 1, xR = 1) ;

• voters’ beliefs are’:

pL pC
µL|0,S ∗ = , µC|0,S ∗ = ,µ 1 ∗ = 1;
pC + pL pC + pL R|{ 2 ,1},S

• voters choose: γp (0) = 1 and γp ( 12 ) = γp (1) = 0.

Two conditions need to be satisfied to support this equilibrium. First,

the probability of a right type needs to be relatively low, but not too low.

The upper bound on pR guarantees that poor voters would not reelect an

incumbent who implements x1 = 21 , which would encourage C to deviate.

The lower bound guarantees that poor voters reelect the incumbent after

observing the leftist policy, since they know there is a chance the incumbent

is a moderate and will thus implement x2 = 21 .

The second condition specifies a range on the value candidates place on

holding office. The intuition for the upper bound is that the Right candidate

needs to care enough about policy relative to office in order to prefer her

ideal point as opposed to deviating to get votes. The lower bound plays

the opposite role with respect to C. Since C is not implementing her ideal

policy, she needs to care enough about office not to have an incentive to

deviate to her preferred policy outcome.

Section 4.3 96

Finally, in a third possible equilibrium (stated in Proposition 4.3) each

type implements her own ideal policy in both periods. Thus after observing

an incumbent in office, voters know her ideological preferences for certain.

Even then, both left and moderates get reelected.

Proposition 4.3: If w < 14 (1 − pC ) − pL , there exists a unique separating

Perfect Bayesian Equilibrium where:
• candidates play:
∗ 1 2 1 1 2 1 1 2
S = (xL = 0, xL = 0), (xC = , xC = ), (xR = 1, xR = 1) ;
2 2

• voters’ beliefs are:

µL|0,S ∗ = 1, µC| 1 ,S ∗ = 1, µR|1,S ∗ = 1;


• voters choose γp (0) = γp ( 12 ) = 1, and γp (1) = 0.

The condition supporting this equilibrium is that candidates be suffi-

ciently policy oriented. This guarantees that the right type has no incentive

to deviate to a moderate policy in the first period to get reelected. A

corollary of this condition is that the prior probability of a left type be rel-

atively small. This leads voters to expect policies closer to the right from

an unknown challenger, thus leading the poor to reelect a moderate. These

expectations also play a role in the utility of the right type. Not getting re-

elected is not too bad if the chances that another right candidate will come

to power and implement her preferred policy are high.

Figure 4.1 displays the values of the parameters supporting the equilibria

described in Propositions 1 and 3 . Given the assumption of risk aversion,

Section 4.3 97

no pure strategy equilibrium is supported in a range of parameter values.

Note that the equilibrium described in proposition 2 – not displayed in the

Figure – holds for the same probability distribution of types in equilibrium

1, except that the values of office lie between those of proposition 1 and 3.

In general, Proposition 4.3 is the one that most closely reflect the reality

in many unequal democracies. Politicians who are genuinely interested in

the welfare of the worse-off and able to implement redistributive policies are

relatively rare. As a consequence, poor voters are likely to settle for policies

further away from their ideal with the overall result being possibly long

spells of low redistribution. Interestingly, however, even in the presence of

uncertainty, the lack of redistribution in unequal societies is not necessarily

a consequence of parties being office oriented and caring thus little about

policy (this would actually lead to more rather than less redistribution).


How far to the right?

Given that it can be rational for the poor to reelect non-leftist incumbents

even under full information about their policy preferences, a natural question

that arises is how far to the right an incumbent can be and still get reelected.

The answer is stated in Proposition 4.4.

Let x̃C = a, where a ∈ (0, 1). We can restate Proposition 4.3 as follows:

Proposition 4.4: If w < (1 − a)2 (1 − pC ) − pL , there exists a separating

Perfect Bayesian Equilibrium where:
Section 4.3 98

Figure 4.1: Equilibrium conditions: Proposition 4.1 (left) and 4.3 (right)


Value of Office
Probability of Left Type

0.6 (1.5,1.62]

0.0 0.2 0.4 0.6 0.8 1.0
Probability of Right Type

Notes: The values of office displayed correspond to the lower bound in the case of
Proposition 4.1 – condition supporting equilibrium is w > 2 − pL − p4C – and the
upper bound in the case of Proposition 4.3 – condition supporting equilibrium is
w < 41 (1 − pC ) − pL . For illustration purposes I categorized the values of office.
Section 4.3 99

• candidates play:

S ∗ = (x1L = 0, x2L = 0), (x1C = a, x2C = a), (x1R = 1, x2R = 1) ;

• voters’ beliefs are:

µL|0,S ∗ = µC|a,S ∗ = µR|1,S ∗ = 1;

• voters choose γp (0) = γp (a) = 1, and γp (1) = 0.

The conditions supporting an equilibirum where poor voters reelect a

“moderate” incumbent becomes increasingly restrictive as her ideal policy

becomes closer to the rightmost policy. As a → 1, pL → 0 and w → 0. That

is, the more players perceive a redistributive type as a rare type, and the

more candidates care about policy relative to office, the more poor voters

would be willing to reelect an incumbent who does not redistribute much.

Again, this holds with full knowledge of the incumbent’s policy preferences.

Risk neutrality

So far I have assumed concave utilities over policies. That is, players are

risk averse. It is indeed common to attribute risk aversion to individuals,

in particular poor ones3 . Thus it is interesting to understand what role this

assumption plays in the equilibria discussed so far. In order to do it I char-

acterize similar equilibria under the assumption of risk neutrality drawing

Interesting experiments were run in Canada and Peru to capture the degree to which
the poor are risk averse and have short term horizons. For results see Engle-Warnick,
Escobal and Laszlo (2006) and Eckel, Johnson and Montmarquette (2004)
Section 4.3 100

attention to differences between them. Under this alternative assumption,

candidates’ utilities are given by4 :

Ute (x) = −|x − x̃t | + w,

Utne (x) = −|x − x̃t |;

and voters’ utilities by:

Ui (x) = −|x − x̃i |.

I begin by examining the semi-pooling equilibrium, where all candidates

choose the leftmost policy in their first term and their ideal policies in the

second term. The conditions assuming the poor are risk neutral are stated

in Proposition 4.5. We notice that the equilibrium holds with less restrictive

conditions than when risk aversion is assumed. In particular, we see that

the value of office can be smaller than that stated in Proposition 4.1, while

the shared probability of a right type can be higher.

Proposition 4.5: If w > 2 − pL − 2 , and pR < 12 (1 − pC ) there exists
a Perfect Bayesian Equilibrium where:

• candidates play:
∗ 1 2 1 2 1 1 2
S = (xL = 0, xL = 0), (xC = 0, xC = ), (xR = 0, xR = 1) ;
Notice that while both the poor and rich are risk neutral in this case, moderates are
not. Given their central position their utility function is tent-shaped
Section 4.3 101

• voters’ beliefs are:

µL|0,S ∗ = pL , µC|0,S ∗ = pC , µR|0,S ∗ = pR , µC| 1 ,S ∗ = µR|1,S ∗ = 1;


• voters choose: γi (0) = 2 for all i, and γp ( 21 ) = γp (1) = 0.

Consider now the equilibrium where the moderate pools with L in the

first period – choosing x1 = 0 – and R plays her ideal policy in both peri-

ods. As stated in Proposition 4.6, under risk neutrality this equilibrium still

holds. Compared to the conditions in Proposition 4.2, however, the range

of values of office supporting the equilibrium is now smaller and the values

of pR are higher.

Proposition 4.6:If 21 ( pCp+p

− pC ) < pR < 12 (1 − pC ) and p2C < w <
1 − pL − 2 , there exists a separating Perfect Bayesian Equilibrium where:

• candidates play:
∗ 1 2 1 2 1 1 2
S = (xL = 0, xL = 0), (xC = 0, xC = ), (xR = 1, xR = 1) ;

• voters’ beliefs are :

pL pC
µL|0,S ∗ = , µC|0,S ∗ = , µ 1 ∗ = µR|1,S ∗ = 1;
pC + pL pC + pL C| 2 ,S

• voters choose: γp (0) = 1 and γp ( 21 ) = γp (1) = 0.

In the most interesting case, the separating equilibrium where both left-

ist and moderate candidates are reelected, risk neutrality also leads to less

restrictive conditions on the value of office and, as a consequence on the

shared prior probability of leftist candidates. As stated in Proposition 4.7

Section 4.3 102

the equilibrium holds for higher values of pL and of w when compared to

Proposition 4.3.

Proposition 4.7: If w < 12 (1−pC )−pL , there exists a separating Perfect

Bayesian Equilibrium where:

• candidates play:

∗ 1 2 1
S = (x1L = 0, x2L = 0), (x1C 1 2
= , xC = ), (xR = 1, xR = 1) ;
2 2

• voters’ beliefs are:

µL|0,S ∗ = µC| 1 ,S ∗ = µR|1,S ∗ = 1;


• voters choose γp (0) = γp ( 12 ) = 1, and γp (1) = 0.

Thus two of the equilibria found under risk neutrality – one where all

candidates choose the same policy and another where they all pool on the

policy preferred by the poor – hold for a higher range of parameter values

than the corresponding equilibria found under risk aversion. This difference

is clear when we compare Figures 4.2 and 4.1. This suggests that, given the

assumptions made, risk aversion leads poorer voters to reelect incumbents

that are relatively more redistributive than they would were they risk neu-

Section 4.3 103

Figure 4.2: Equilibrium conditions: Proposition 4.5 and 4.7



Value of Office
Probability of Left Type

0.6 (1.25,1.38]

0.0 0.2 0.4 0.6 0.8 1.0
Probability of Right Type

For illustration purposes I categorized the values of office. They are actually con-
tinuous, but given the jump from 0.5 to 1 that occurs from one equilibrium to the
next, a continuous shading would not be very informative
Section 4.3 104

Proposition 4.8: If w < |1 − a|(1 − pC ) − pL , there exists a separating

Perfect Bayesian Equilibrium where:
• candidates play:

S ∗ = (x1L = 0, x2L = 0), (x1C = a, x2C = a), (x1R = 1, x2R = 1) ;

• voters’ beliefs are:

µL|0,S ∗ = µC| 1 ,S ∗ = µR|1,S ∗ = 1;


• voters choose γp (0) = γp (a) = 1, and γp (1) = 0.

As stated in Proposition 4.8, the lower the value of office and, as a

corollary, the lower the shared probability of a left candidate, the less re-

distributive an incumbent can be and still get reelected. Compared to the

results under the assumption of risk aversion stated in Proposition 4.4, under

risk neutrality candidates might care more about office and left candidates

can be perceived as relatively more common and the equilibrium still holds.

Thus under this set up, and as noted earlier, risk aversion can work to the

advantage of poor voters.

Low Demand and Uncertainty: Investigating the Prospects of


In the first part of the dissertation we saw that the preferred policy of

the poor does not always correspond to maximum redistribution. In the

particular case of education under borrowing constraints the poor’s ideal

levels of public provision of education might be below the optimal feasible

level. That is, in order to reach this ideal level less taxation and less public
Section 4.3 105

investment in education are required than it would be were the poor not

constrained in their capacity to borrow against future earnings.

The other result obtained in the first part was that even if there exists

politicians whose ideal policy correspond to more redistribution than de-

manded by the poor, they would not get elected. This result followed from

the full information and responsible party model employed earlier. What

about under uncertainty? Would the same null prospects of redistribution

beyond that preferred by the poor obtain? Not necessarily.

As mentioned in the introduction an important part of the puzzle is

understanding the delay in the adoption of redistributive policies in unequal

countries. While levels have remained low in many countries for many years,

in others they have seen recent increases (Robinson 2009).

In this section I explore the political prospects of redistribution beyond

what might be preferred by the poor majority. Even when no particular

group of voters favor certain policies, personal convictions and tastes might

lead individual candidates to have a high stake on them. Policies that are

most likely affected by low demand by the poor tend to relate to human

capital development, which is found to have major implications for growth,

the functioning of democracy and other important aspects of individuals’

lives. These are thus policies held by many in particularly high esteem.

International development organizations certainly do. With the increasing

amount of funding they are making available to improve the delivery of such

services, the incentives for incumbents to adopt them are also growing.

An interesting question that arises then refers to the political prospects

of such incumbents. How likely are they to win elections and remain in power
Section 4.3 106

for a second term if they choose, for example, to implement higher levels of

public provision of education than demanded by the poor? To explore this

issue I adapt the model developed in the beginning of the chapter. Relative

to that baseline set-up I simply move the ideal policy of the poor towards the

right and assume x̃p = x̃m = x̃C = b. I focus on the case where b ∈ (0, 21 )5 .

Under these assumptions, there exists two equilibria where a leftist type

gets reelected after implementing her ideal policy. They are analogous to the

equilibria described in Propositions 1 and 3. Compared to these equilibria,

however, under “low demand” by the majority the set of parameter values

supporting the reelection of the leftmost candidate after she implements the

leftmost policy in her first term is smaller. Notice that a third equilibrium

found earlier, where both L and C choose x1 = 0 and get reelected while

R chooses x1 = 1 and does not get reelected, does not exist under the new


In the first equilibrium (described in Proposition 4.9) all types choose

x1 = 0 and all deviations are expected to come from the R type. That is,

if for some reason – e.g. politicians get a flow of funds from development

organizations to deliver a state of the art public educational system – incum-

bents implement higher redistribution than preferred by the poor majority,

they would still have a chance at reelection.

The second equilibrium is a separating equilibrium where each type

chooses their own ideal policy in both periods. Both L and C get reelected
5 1 1
Note that by symmetry, when b > 2
the results that hold for L when b < 2
hold for R and vice-versa.
Section 4.3 107

in this case. Thus an outcome of high redistribution is not impossible. I

discuss the conditions supporting these equilibria in turn.

3 2 3 4
Proposition 4.9: If 2 − (1 − b)2 pC < w < (2b −b )pC +(1+2b
2b2 −1
−2b)pR −b
, and
b ∈ (0, 12 ) there exists a Perfect Bayesian Equilibrium where:

• candidates play:

S ∗ = (x1L = 0, x2L = 0), (x1C = 0, x2C = b), (x1R = 0, x2R = 1) ;

• voters’ beliefs are:

µL|0,S ∗ = pL , µC|0,S ∗ = pC , µR|0,S ∗ = pR , µR|{b,1},S ∗ = 1;

• poor voters choose γp (0) = 2 and γp (b) = γp (1) = 0.

The equilibrium described in Proposition 4.9 only holds for relatively

high values of office (w > 1). The lower bound on w is what guarantees a

rightist type does not have an incentive to deviate to its ideal policy. The

right type needs to care enough about office not to do it, since any policy

other than the leftmost does not get her reelected. The upper bound refers to

the condition leading voters to believe that an incumbent who implements

policies other than the leftmost is type R. Notice that if voters were to

believe the type more likely to deviate to b was C, than they would reelect

and incumbent implementing b and C would deviate to b. And this would

not be an equilibrium.

For illustration purposes I plot in Figure 4.3 the values of the parameters

supporting this equilibrium given the ideal policy of the poor and the value

of office. For comparison purposes I set two values of ideal policy of the
Section 4.3 108

poor ( 14 and 13 ) and three different values of office (1.1, 1.25 and 1.5). These

values of office are relatively high. Thus candidates need to be significantly

office oriented for such an equilibrium to occur.

1.0 ●
● 1.0 ●


● ●




● ●




●● ●


0.8 ●

● 0.8 ●

● ●
Probability of Left Type

Probability of Left Type

Probability Probability
of Type C of Type C
0.6 ● 0 0.6 ● 0
● 0.2 ● 0.2
0.4 0.4
● 0.4 ● 0.4

● 0.6
● 0.6
● 0.8 ● 0.8

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Probability of Right Type Probability of Right Type

(a) b = 1/4, w = 1.1 (b) b = 1/3, w = 1.1

1.0 ●
● 1.0 ●


● ●




● ●








● ●





0.8 ●






● ●

0.8 ●












● ●







Probability of Left Type

Probability of Left Type







● Probability ●







● Probability






● ●











●● of Type C ●




●● of Type C

0.6 ●




● ● 0 ●

0.6 ●


● ● 0



●● ● ●
● 0.2
● 0.4 ● 0.4

● 0.6
● 0.6
● 0.8 ● 0.8

0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Probability of Right Type Probability of Right Type

(c) b = 1/4, w = 1.25 (d) b = 1/3, w = 1.25

1.0 ●
● 1.0 ●


● ●




● ●








● ●





0.8 ●






● 0.8 ●












● ●







Probability of Left Type

Probability of Left Type









Probability ●



















●● of Type C ●










●● of Type C











●● ●











0.6 ●













● ● 0.6 ●













● ●















0 ●






























● ● ●















● ●












● 0.2 ●












● 0.2

0.4 ●


0.4 ●












●● 0.4 ●









●● 0.4










● ●







0.2 ●









● ● 0.6 ●

0.2 ●





●● ● 0.6








● ●










● ● 0.8 ●

● ● 0.8






0.0 ● ●


●● 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Probability of Right Type Probability of Right Type

(e) b = 1/4, w = 1.5 (f) b = 1/3, w = 1.5

Figure 4.3: Parameter values for which the equilibrium in Proposition 4.9
holds given b and w.
Section 4.3 109

As shown in Figure 4.3, as the value of office declines, the shared prob-

ability of a left type needs to be large for the equilibrium to hold. As

candidates value office to a greater extent, the probability of a left type can

be smaller and the equilibrium holds for a larger set of probabilities of types.

The other point to notice is that as the ideal policy of the poor moves further

away from the leftmost policy, the equilibrium only holds for a smaller set of

probability distributions over types. In other words, the less redistribution

is favored by the poor the lower the incentives for candidates to implement


In the next equilibrium all candidates choose to implement their own

ideal policy in the first period. While both L and C get reelected, R does

not. The conditions supporting this equilibrium are given in Proposition

4.10. The upper bound on the value of pL guarantees that R has no incentive

to deviate to x1 = b in order to get reelected. Note that this upper bound can

be rewritten in terms of w. Thus another way to express it is that candidates

cannot value office too much, otherwise they would have an incentive to cater

to the interests of the majority instead of implementing their own preferred

policy. Thus contrary to the previous equilibrium, in this one the value of

office needs to be sufficiently small.

The lower bound on pL guarantees that poor voters reelect incumbents

choosing x1 = 0. Even though their ideal policy is not located at x = 0, this

condition guarantees that their expected utility from reelecting L and C is

still higher than that of reelecting an unknown challenger.

Section 4.3 110

Proposition 4.10: If 1− (1−b)
pR < pL < (1−b)2 (1−pC )−w and b ∈ (0, 21 ),
there exists a separating Perfect Bayesian Equilibrium where:

• candidates play:

S ∗ = (x1L = 0, x2L = 0), (x1C = b, x2C = b), (x1R = 1, x2R = 1) ;

• voters’ beliefs are:

µL|0,S ∗ = 1, µC|b,S ∗ = 1, µR|1,S ∗ = 1;

• poor voters choose γp (0) = γP (b) = 1 and γp (1) = 0.

Again, for illustration purposes I plot in Figure 4.4 the values of the pa-

rameters supporting the equilibrium described in Proposition 4.10 for given

values of b, the ideal policy of the poor, and w, the value of office. As is

clear, the further away the ideal policy of the poor is from the leftmost pol-

icy the smaller the set of parameters supporting the equilibrium. Given the

incentives R have to deviate in search for votes, the value of office needs to

be relatively small.

To summarize, we notice that the results with respect to the value of

office remain the same as before. Candidates who do not share the same

preferences held by a majority have an incentive to cater to the interests

of that majority for votes when they value office sufficiently. Otherwise the

costs of implementing policies they dislike are too high, leading them to

implement their own preferred policies.

Regarding the effects of low demand, notice in Figures 4.3 and 4.4 that

in both equilibria, as the ideal policy of the poor is further away from the

leftmost point the set of parameter values becomes more restricted. That is,
Section 4.3 111

1.0 1.0

0.8 0.8
Probability of Left Type

Probability of Left Type

Probability Probability
of Type C of Type C
0.6 ● 0.6 ●



0 0





● ● ●








0.4 ●










● 0.4
















●● ● 0.4 ●








● ● 0.4















● ●



























● ● ●●













●● ●



























































● ●
0.2 ●



















●● ●
























0.8 ●














































● ●






















0.0 ●



























● 0.0 ●























0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Probability of Right Type Probability of Right Type

(a) b = 1/4, w = 0 (b) b = 1/3, w = 0

1.0 1.0

0.8 0.8
Probability of Left Type

Probability of Left Type

Probability Probability
of Type C of Type C
0.6 ● 0 0.6 ● 0
● 0.2 ● 0.2
0.4 0.4



● ● 0.4 ● 0.4













● ● 0.6


●● ● 0.6










●● 0.2


















● ● 0.8 ●








● ● 0.8
















● ●





























● ●●















0.0 ●



















●● 0.0 ●
















0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Probability of Right Type Probability of Right Type

(c) b = 1/4, w = 0.15 (d) b = 1/3, w = 0.15

1.0 1.0

0.8 0.8
Probability of Left Type

Probability of Left Type

Probability Probability
of Type C of Type C
0.6 ● 0 0.6 ● 0
● 0.2 ● 0.2
0.4 0.4
● 0.4 ● 0.4

● ● 0.6 ● 0.6


● 0.2





● ● 0.8 ● 0.8







● ●












● ●●





0.0 ●

● ●




●● ●

●● 0.0 ●






0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Probability of Right Type Probability of Right Type

(e) b = 1/4, w = 0.35 (f) b = 1/3, w = 0.35

Figure 4.4: Parameter values for which the equilibrium in Proposition 4.10
holds given b and w.

the less likely we are to observe an equilibrium where highly redistributive

incumbents get reelected.

Section 4.4 112

Compared to the stark results obtained from a full information model,

in the presence of uncertainty, while low demand can make it less likely that

certain redistributive policies emerge, they are not precluded from the set

of possible outcomes.

4.4 Some Implications to Other Theories

Three points emerge from the formal exercises. First, the extent to which

there exists a trade-off between seeking office and seeking particular policies

depends on the underlying distribution of preferences in the population. In

cases like Brazil, where poverty is widespread, candidates who care about the

well-being of the worse-off do not face that trade-off. Candidates favoring

less redistribution, however, do. For these candidates to have an incentive to

cater to the interests of the poor they have to value office to a great extent.

The second point refers to the importance of voters prior beliefs about

the distribution of types among candidates. The result that most closely

approximates what we actually observe in many countries – Proposition 4.3,

where right policies are more likely and non-left candidates get reelected –

is supported by the belief that candidates who care about the welfare of the

poor are rare. That is, the more they expect an unknown challenger to hold

preferences distant from their own, the more likely they are to reelect center

or center-right incumbents. This, in turn, can contribute to the persistence

of low redistribution.
Section 4.4 113

Finally, the third point refers to the electoral behavior of poor voters.

The fact that they reelect non-redistributive incumbents is not necessarily a

sign that they don’t care about policy or that they are being irrational or ig-

norant in assessing candidates’ policy positions. Even under the assumption

that they are as capable as any other voter to evaluate policy and update

beliefs about politicians preferences, such an outcome can occur.

Various implications to both clientelism and the asymmetries in political

weight theories follow from these points. Consider, for example, the result

that the trade-off between office and policy orientation only play out among

parties whose ideal policy does not coincide with that of the poor majority.

This result can be at odds with clientelist accounts. It is usually argued

that parties that engage in clientelism are not programmatic – that is, they

are office rather than policy oriented. According to the results of the model,

however, the opposite would be true. Right candidates who care a lot about

policy would be exactly the ones with a stronger incentive to seek the votes

of the poor majority through means other than contradicting their own

ideological views.

Moreover, if a party chooses to target private goods as opposed to imple-

ment general redistribution, they would be expressing a distaste for redis-

tributive policies, or, in other words, a preference for more rightist policies.

Unless, of course, it is cheaper or easier to rely on clientelism.

Information on the actual costs of targeting small benefits rather than

implementing a general eligibility redistributive program is not available.

What the literature on clientelism do seem to agree on is that building clien-

telistic machines is an incredibly costly enterprise (see for example Schaffer

Section 4.4 114

(2007)), which includes not only the cost of individual disbursements, but

also the costs of delivering and monitoring results. If we compare that to

the electoral returns from conditional-cash-transfer programs, for example,

there are reasons to expect the latter to be relatively inexpensive (in Brazil

it consumes about 0.3% of the GDP).

As pointed by Robinson in his policy recommendations to the United

Nations, one fundamental factor in promoting redistribution is encouraging

the creation of parties that share the interests of the poor. The author

claims that “If this [conditional cash transfer programs] played an important

role in getting President Lula reelected, Latin American politicians will have

understood this much faster than the UNDP” (2009, 50). That is, for these

politicians not to follow suite, some this distaste for such policies must be

at play. If this logic is correct, clientelistic parties would not only be policy

oriented, but also right rather than left leaning.

Now consider the second point raised above, that is, the importance of

individual beliefs about the distribution of types in society. As just men-

tioned, it is commonly argued that clientelistic politicians prefer to keep

redistribution low – so they can both please the rich, who are probably their

main constituency, and keep the poor poor – and disburse only enough pri-

vate benefits to secure an electoral victory. Many studies of vote buying

show that these individual disbursements can be relatively cheap. From the

poor voters’ perspective then, the utility derived from such small individual

disbursements is only enough to win their votes if their expected utility from

electing the challenger is particularly low. That is, the policy expectations
Section 4.4 115

of poor recipients should play an important role on the pervasiveness of


Finally, the consummation of these exchanges into clientelist practices

does not need to follow from assumptions of ignorance or disregard for policy

on the part of poor voters. Elections pitching “unknown challengers”, that

is, candidates on whom voters do not have enough information allowing

them to make distinctions based on ideology, can be a fertile ground for

clientelist practices to develop.

Now consider theories of unequal political clout following from the un-

equal distribution of resources in society. As mentioned earlier, asymmetries

of power at the electoral stage account for some of the deficiencies in repre-

sentation but not all of it. Since the distribution of resources tends to persist

over time and is likely to respond to how much redistribution occurs, it fails

to account for the complete puzzle, that is, for the introduction, gradual or

sudden, of redistributive policies.

The model, however, suggests a strong role of citizens’ perceptions of

such asymmetries on policy outcomes that is consistent with what we ob-

serve. The equilibrium that allows for consecutive periods of low redistri-

bution is supported by the belief that left types are rare. This belief can

result from various factors. Most importantly, it can follow directly from

theories of inequalities in political weight. A low pL might reflect, for ex-

ample, the high entry costs of elections. A high monetary cost of running

for office would negatively affect the entry of candidates representing poorer

constituencies while favor those representing the rich.

Section 4.4 116

The more voters believe rich constituencies have an advantage in terms of

number of candidates at elections, the lower the expected utility of the poor

from an unknown challenger, and the higher the probability of reelecting a

non-redistributive incumbent. Thus, while not attributing all distortion in

representation to unequal political clout, the model provides a framework

for incorporating and understanding its effects.

These dynamics remain practically unchanged under the assumption

that the preferred levels of redistribution by the poor are lower than the

maximum feasible level. Under this assumption, uncertainty about candi-

dates’ policy positions means redistribution beyond what is preferred by the

majority can still be the outcome of the political process. This result con-

trasts with those obtained under the assumption of full information, which,

while common in the literature, does not appear reasonable in relatively new

democracies as will be shown in the next chapter.

The main point is that uncertainty can bring about outcomes that differ

from those preferred by a majority. This can cut both ways. If it is indeed

the case that right types are more common than left ones, it might take

time for redistributive policies to arise, while more common for right ones

to take place. Low demand contributes to that delay even further.

The possibility, however small, of having more redistribution than de-

manded can be important in light of the empirical findings. They suggested

that educational attainment is an important determinant of rating the pro-

vision of education highly. Thus the chance of providing more education

than preferred under certain constraints can lead to increased demand for

it in the future.
Section 4.5 117

4.5 Conclusion

While the more conventional accounts discussed in this chapter raise relevant

shortcomings in the working of newly established democracies, they are each

an element of a broader context. In this chapter I suggested a more basic

factor – lack of information about candidates ideological positions – that

explains low redistribution in unequal democracies and also accounts for the

possibility of changes in the levels of redistribution over time.

There are clear reasons why low levels of information about parties’

and candidates’ ideological inclinations might be an issue, especially in

new democracies. This makes it a good starting point for analyzing the

voter-candidate linkage, which is central to any investigation of the politi-

cal prospects of policies. Moreover, the learning process that results from

uncertainty can provide a useful framework for understanding the interplay

of factors commonly addressed in the literature, as well as the opportunities

for some of them to arise.

Three important insights arise from the model. First, it calls our atten-

tion to the role of candidates’ motivations. Low redistribution in unequal

democracies can be a sign that candidates are relatively policy, as opposed

to office, oriented. Policy orientation leads candidates to choose their own

ideal policies once in office. That is both moderates and right types would

not try to “mimic” a left type in order to garner votes. If candidates were

purely office seekers, they would have an incentive to implement the policy

favored by a majority to win the election. If that were the case, however,

we would likely observe more rather than less redistribution.

Section 4.5 118

The second point refers to the importance of voters perceptions of the

ideological locations of candidates. The particular equilibrium addressing

the central question of this study suggests that voters believe candidates

sharing the preferences of the poor majority are relatively rare. This would

lead poor voters to reelect center or center-right incumbents, even if they

are aware of their preferences. This is because they expect a random chal-

lenger to be even more to the right. Hence the feeling of distance from and

disillusionment with most politicians – common in many new democracies –

can serve to perpetuate outcomes that are disadvantageous for the majority.

Finally, the model provides an account of voters’ behavior that does

not rely on assumptions of irrationality, ignorance, and disregard for policy.

Even under the assumption that they are as capable as any other voter to

evaluate policy and update beliefs about incumbents’ policy positions, it can

be optimal for them to reelect not so redistributive candidates.

In the last section of the chapter, I explored the political prospects of

redistribution beyond the levels preferred by a poor majority. Assuming

there are candidates favoring relatively higher levels of redistribution I find

that uncertainty can lead to their reelection. This is an important result

concerning redistributive policies subject to low demand. While full infor-

mation might preclude them from the set of feasible policies, uncertainty

may provide the opportunity for them to arise. In cases like public provi-

sion of education – where educational attainment was itself a good predictor

of higher support – that opportunity might be crucial in setting a virtuous

cycle of increased pressure for higher quality of provision.


Chapter 5

Reaching a Decision:
Evaluations of Candidates’
Policy Positions and Vote

In the previous chapter I analyzed the consequences of citizens’ lack of

information about the ideological positions of candidates to the prospects

of redistributive policies. In this chapter I analyze individual level data

to explore some of the observable implications of the model. I focus in

particular on three factors.

First, the model relies on the assumption that voters face uncertainty

about the actual policy position of candidates when making a decision at

elections. While this is an intuitive assumption, I gather some preliminary

evidence suggesting this is indeed the case.

Second, based on the model results, I look for evidence that the separat-

ing equilibrium characterized in the previous chapter holds. This was the

equilibrium that accounted for delays in the implementation of redistribu-

Section 5.0 120

tive policies, among other reasons because the poor would reelect not-so-

redistributive incumbents. The condition supporting this equilibrium was

that candidates be relatively policy oriented, as opposed to only caring about

winning the elections. The corollary condition was that poor voters perceive

candidates that share their preferences to be rare. While no data is avail-

able to test the main condition, I am able to explore the plausibility of its


Finally, if the separating equilibrium holds, incumbents would seek to

implement their own ideal policy. This means after observing an incumbent

in office, voters should be able to update their beliefs about this represen-

tative’s actual policy inclinations. This yields two observable implications.

First, uncertainty about the ideological location of an incumbent running

for reelection would be lower relative to “unknown challengers”.

Second, in elections where none of the contenders are former incum-

bents, due to uncertainty, voters’ would be unable to distinguish between

them on ideological grounds. As a consequence, the choice of representa-

tives would be more random with no significant socioeconomic cleavages in

voting patterns. Conversely, if one of the candidates is an incumbent, vot-

ers should have more information on his or her ideological inclinations and

use this information to make their decisions. That is, if different socioe-

conomic groups have different tastes for redistribution, once they are able

to distinguish candidates ideologically they should choose the one they be-

lieve is closer to their preferences. As a result, significant differences should

arise in voting patterns across groups. In sum, if the separating equilibrium

provides an accurate description of the situation in unequal new democra-

Section 5.1 121

cies we should observe: a) no significant differences among groups of voters

backing different candidates when they resemble “unknown challengers”, b)

significant differences when more information is available.

The chapter is divided into two sections. The first section deals with the

issue of uncertainty in individuals’ evaluations of candidates policy positions

and the extent to which individuals perceive pro-poor candidates as rare. As

we will see these two issues are related and should be discussed together. On

uncertainty, the section covers both the basic assumption that uncertainty

is an issue, and also the observable implication arising from the separat-

ing equilibrium that levels of uncertainty should differ between evaluations

about an incumbent and those regarding “unknown challengers”.

The second section deals with the observable implications following from

the separating equilibrium with respect to voting patterns. That is, whether

we observe significant differences between the voting choices of different

socioeconomic groups.

The analysis conducted in this chapter is based on individual level data

from a number of different sources. For the sake of clarity I discuss the

different sources and the variables employed in each exercise in the corre-

sponding section. It is worth noting that data collection in new democracies

is relatively rudimentary. This affects the degree to which we are able to

perform rigorous empirical analysis. I provide some suggestions about data

collection efforts that might contribute to a better understanding of the

issues investigated here.

Section 5.1 122

5.1 How Much do Respondents Know About Can-

didates Policy Positions?

The basic assumption in the model developed in the previous chapter is that

individuals are uncertain about the actual ideological orientation of candi-

dates during elections. In some of the equilibria found, because candidates

favoring different policies might implement similar policies once in office,

not a lot of learning takes place once voters observe a candidate serve as an

incumbent. In the equilibrium that most closely reflect the patterns we are

interested in explaining, however, incumbents choose their own ideal policies

and updating takes place. That is, after observing an incumbent in office

voters gain more information about her or his ideological stance.

Clearly the amount of information gained is a function of the systems’

“clarity of responsibility” (Powell and Whitten 1993; Powell 2000). Thus

cross-country comparisons can be complicated and a within country ap-

proach particularly suitable. In the analysis that follows I focus on indi-

vidual placements of candidates to the presidency on an ideological scale

ranging from pro-poor inclinations to pro-rich ones. The presidency is the

most salient post in Brazil, and the one where incumbents’ views should be

more easily conveyed to the public. The scale defined in terms of interests

of the poor versus those of the rich is particularly close to the subject of this

study and less susceptible to misinterpretations by survey respondents.

In terms of data availability a number of issues deserve attention. The

main limitation I face refers to lack of consistency across years. The data

comes from different sources and questions on the evaluations of candidates

Section 5.1 123

policy positions were asked in only three electoral years by four different

public opinion projects. In two of these the question was asked in terms

of “left” and “right” placement and the other two in terms of a candidate

being either “pro-poor” or “pro-rich”. In new democracies, and particularly

in Brazil, left-right scale ratings have been shown to be invalid measures of

policy evaluations. Respondents have very different understandings of what

these terms really mean. Thus I rely on the two surveys asking respondents

the extent to which each candidate from a list1 was considered pro-poor.

As a result of this lack of consistency, ratings are not available for all

the years and wherever they are available the question differs from one year

to the other. This means I have limited ability to compare the ratings of a

candidate as an “unknown challenger” with her or his rating as an incumbent

running for reelection. Moreover, comparisons of ratings across time are also

made more difficult. Despite these complications, the exploratory analysis

reveals remarkably high consistency with the model assumptions and results.

The data utilized in this section comes from three sources. Data for

the first presidential election after democratization comes from a survey

conducted in October of 1989 by Datafolha, a private polling firm. Both

urban and rural dwellers were interviewed in a sample of 4893 respondents

drawn to be representative of the Brazilian population aged 16 or older

(16 years old is the age where Brazilian citizens acquire the right to vote).

The 1998 survey was conducted in September of that year by Vox Populi,

another private polling firm. The data is available online at the Consórcio
The questions usually refer to the top four or three viable candidates, since the full
list can contain up to twenty candidates.
Section 5.1 124

de Informações Sociais. The sample (3266 respondents) was drawn to be

representative of the voting age Brazilian population. Finally data for 2006

– the latest presidential election in Brazil – comes from the LAPOP survey

conducted in 2006 with a representative sample of citizens 18 years old or

older (1214 respondents).

In 1989, when asked about the policies the candidate they favored stood

for that they thought was important, one third of respondents answered

they didn’t know. The survey was conducted less than one month before

the elections. This figure changed somewhat depending on the respondents’

level of education. Among those with primary or less – around 70% of

the sample – 39% answered “don’t know”. Among those with college only

16% did so, and among those with secondary education about one fourth

admitted not knowing. Since respondents were only asked to rate their own

chosen candidate, no comparisons can be made across candidates based on

this data.

In 1998 – the third presidential election, where Fernando Henrique Car-

doso (FHC) was the incumbent running for re-election – when asked whether

each of three contenders – FHC, Lula and Ciro – had will to improve the lot

of the poor answers varied. When the question referred to the incumbent,

FHC, only 13% answered “don’t know” and about half said “yes”. With

regards to Ciro Gomes, answers were very close to a random draw: one-

third answered “don’t know”, another third said “yes” and the remaining

third “no”. Evaluations of Lula lied in between with 46% of respondents

answering “yes” and 17% “don’t know”.

Section 5.1 125

On this particular topic – how pro-poor candidates were – the differences

in educational attainment were not very pronounced in terms of incidence

of “don’t knows” by candidate. Considering only those respondents with

primary education or less, the rates of “don’t know” were 36% for Ciro,

22% for Lula and 16% for FHC.

In 2006 – the fifth and latest presidential election, where Lula was the

incumbent running for re-election – the LAPOP survey asked respondents to

place the four main contenders on a scale going from pro-rich (1) to pro-poor

(10) – this scale was reversed for the purposes of this analysis. Inspection

of the individual placements show an interesting pattern of “don’t know”

answers. In evaluations of the incumbent, Lula, 98% of respondents ventured

an answer, even among the least educated. The challengers had much higher

rates of “don’t know” answers that went from around one third to one fourth

of respondents.

Figure 5.1 displays the incidence of “don’t know” in both years where

respondents were asked to rank the main contenders. While I cannot test

for the significance of these differences we observe a pattern compatible with

the model developed in the previous chapter. The rates of uncertainty are

higher for challengers who never served as incumbents before and lower for

the incumbent running for reelection. In particular if we compare the rates

of “don’t know” for Lula, who was a challenger in 1998 and the incumbent

in 2006, the differences suggest an increase in the number of respondents

who believed they were able to provide an estimate of his policy stance.

Looking solely at self-reported “don’t knows”, however, is a very crude

way of capturing uncertainty. Those providing an answer might still be

Section 5.1 126

Figure 5.1: Incidence of “don’t know” answers to candidate placement ques-

35 ●

Incidence of Don't Knows (Percentages)


25 incumbent
● no
20 yes
● ●
15 candidate
● ● Lula
10 ● others

1998 2006
Election Year

guessing the candidates’ positions. While there exists methods to cap-

ture the degree of uncertainty in respondents’ answers, both in terms of

“don’t know” and actual ratings (see for example Bartels (1986) and Al-

varez (1997)), I lack the data necessary to identify the proposed models.

In particular I lack data on individual placement of candidates on different

policy issues and individuals self-placement on these same ideological scales.

An additional way of evaluating uncertainty is by comparing the indi-

vidual ratings to more informed estimates of candidates’ or their parties’

placement. Given the data limitations, such comparisons are, again, very

exploratory, but insightful nonetheless. They can, in fact, shed light on both

uncertainty and the extent to which respondents believe right and moderate

types are more common than left ones.

Section 5.1 127

First, where the individual assessments diverge from other accepted es-

timations of the positions of candidates, there is, at least to some extent,

sign of uncertainty2 . That is, respondents might just be taking their best

guess at placing candidates on the scale. Second, if this is the case, then

the natural question is how likely are these guesses the result of a particular

belief about the distribution of candidates as opposed to a simple random

estimate? In other words, are the answers, or “guesses” we observe likely to

come from a prior belief that candidates sharing the preferences of the poor

are rare relative to moderate and rightist types?

The discussion that follows tries to bring insights to all these questions,

including the expectation of more uncertainty regarding candidates who

never served as incumbent compared to those who have. As pointed out

earlier, if the separating equilibrium found in the previous chapter provides

a good characterization of the situation in unequal new democracies, indi-

viduals should be able to update their beliefs about the actual ideological

inclination of the incumbent, because incumbents always try to implement

their own ideal policies. Thus if we find evidence that there is less uncer-

tainty regarding the incumbent, we find evidence in favor of the separating

equilibrium and against the other two pure strategy equilibria found.
Ideally this should be evaluated based on individuals self-placement or some other
reference point so that ratings from different individuals are in fact comparable. Roughly
speaking, however, the higher the disagreement between more informed estimates of can-
didates placement and individuals’ placement the more likely it is that uncertainty played
a role. The other reason can be that individuals make mistakes, that they don’t know
how to infer candidates’ policy positions based on what they observe. If mistakes by the
individual were the main reason, however, we would probably not find a higher congruence
of assessments with respect to incumbents. Individuals would likely be wrong about them
too, but that is not what the data suggests.
Section 5.1 128

Beginning with 1998, where FHC was the incumbent running for reelec-

tion, about 50% of respondents answered that FHC was committed to the

cause of the poor. This number was slightly lower for Lula (46%) and even

less for Ciro (34%). Those with primary education or less were slightly less

likely to see Lula as pro-poor (42%) than the average and less likely to see

him as pro-poor compared to FHC (53% thought FHC had the resolve to

improve the situation of the poor).

Lula, however, is considerably more to the left than FHC. Estimates of

the ideological positions of parties based on legislators’ assessment place the

PSDB (FHC’s party) around the center and both the PT (Lula’s party)

and the PPS (Ciro’s party) towards the left of the ideological spectrum in

1997(Power and Zucco Jr 2009). These estimates are plotted in Figure 5.2

together with the proportion of respondents in 1998 who rated each of the

candidates as being pro-rich3 .

In terms of the proportion of individuals rating FHC as pro-poor, we

can place him somewhere around the center close to his party’s location

according to legislators’ assessment. The evaluations of both Lula and Ciro,

however, seem considerably off-mark relative to the legislators’ party rat-

ings. The fact that both Ciro and Lula were ranked by respondents as more

rightist than they actually were, suggests the conditions supporting the sep-

arating equilibrium might in fact hold. That is, when asked to assess the

policy position of an “unknown challenger”, voters believing left types are

rare would tend to place that candidate towards the right.

These proportions were calculated coding the “don’t knows” as missing. If instead we
use the whole sample, the discrepancies between Lula’s and Ciro’s rating relative to their
parties’ placement is even starker.
Section 5.1 129

Figure 5.2: Placement of candidates and parties on ideological scale (1997-


Party Positions: Legislators Assessment




Ciro (PPS)


Lula (PT)

0.2 0.3 0.4 0.5 0.6 0.7 0.8

Candidates Positions: Individuals Assessment

Source: Data on legislators’ assessment of party positions were taken from Power and Zucco Jr
(2009). Vertical bars represent the 95% confidence intervals around the estimates. Candidates
positions correspond to the proportion of respondents to the Vox Populi survey in 1998 saying
that the candidate was pro-rich.

As we will see, the view of FHC as more leftist than Lula is also in line

with the results we obtain in the next section on vote intentions. In 1998

the poor were more likely to vote for FHC. The reason might be exactly the

one proposed in the separating equilibrium. Although poor voters saw him

as a moderate, they reelected him because they expected the challengers to

be further to the right.

Thus based on the 1998 data, figures suggest that there was more uncer-

tainty about the placement of the challengers than that of the incumbent.
Section 5.1 130

Moreover, the challengers tended to receive ratings placing them further to

the right of the policy spectrum. If respondents believed that left types were

more common, we would expect the opposite to hold. That is, challengers

being more likely to be placed towards the left of the scale. In addition,

the fact that FHC was seen as a moderate, lends plausibility to the results

obtained from the separating equilibrium. In particular the part saying that

moderates would be reelected even if poor voters were fully aware of her or

his position.

Moving now to 2006, based on the LAPOP data, respondents were more

likely to rate the challengers – Gerald Alckmin, Heloı́sa Helena, and Cristo-

vam Buarque – as either moderate or pro-rich. The answers given about

Lula, however, were considerably different. Figure 5.3 displays the densities

of responses corresponding to each candidate. As is visible in the figure, in

Lula’s case, respondents were more likely to rank him as “pro-poor”, to an

even greater extent than did respondents in 1998, when he was a challenger.

In fact, for the ratings of Lula in 2006 to be similar to those in 1998 we

would have to assume that anyone placing him around the third position or

higher on the ideological scale from 1 (pro-poor) to 10 (pro-rich), believed

he was not committed to the cause of the poor. Considering only those

respondents with primary education or less, we would need to assume anyone

placing Lula somewhat after the second position or higher believed he had

no will to improve the lot of the poor. Another way to see that is by simply

splitting the 2006 scale in two. Then 64% of respondents rated Lula as pro-

poor in 2006 compared to 46% in 1998. This figure goes to 69% among the

least educated in 2006 compared to 42% in 1998.

Section 5.1 131

Figure 5.3: Placement of candidates on pro-poor (left) to pro-rich (right)






0.2 0.4 0.6 0.8

Ideological Scale

Source: The AmericasBarometer by the Latin American Public Opinion Project (LAPOP),

In order to test for the observed differences in candidates’ ranking I run

a simple Bayesian analysis. First I rescale the answers to the 0 to 1 interval.

Assuming the ratings have a beta distribution I estimate the scale and shape

parameters in each case – one set of parameters per candidate. We can then

compare differences in the estimated distributions likely to have yielded the

answers we obtained from the survey for each candidate. The model is given

in the appendix. I follow the parametrization in Branscum, Johnson and

Thurmond (2007).
Section 5.1 132

Figure 5.4: Posterior distribution of estimated mean ideological placement





0.3 0.4 0.5 0.6 0.7

Estimated Mean Ideological Placement

Figure 5.4 displays the posterior density of the estimated mean of the

beta distribution likely to have yielded the responses in the LAPOP survey

by candidate. As we can see, the estimated mean in the case of Lula is

significantly to the left of his main challengers. The challengers tend to be

clustered towards the right side of the spectrum. An interesting question

that arises is how closely these placements based on individual assessments

are to other estimates of candidates position? Unfortunately, estimations of

the ideological placement of presidential candidates are not available. But

we can use those of their parties as done earlier.

Heloı́sa Helena belongs to the PSOL, an extreme left faction of the PT

that decided to split in response to the party’s moderation in the past couple
Section 5.1 133

of years. Thus the fact that Heloı́sa Helena is considerably to the right of

Lula is quite interesting. It is highly suggestive of the tendency of voters to

believe challengers are not likely to be pro-poor.

Cristovam Buarque ran as the candidate of the PDT, another left party.

His campaign in 2006 was heavily based on increasing the quality of public

provision of education. He was the head of the think thank that designed

the Bolsa Escola program, a conditional-cash-transfer program requiring re-

cipients to attend school on a regular basis. He was the first to implement

the program in the Federal District during his term as a governor (1995 -

1998) for the PT. During Lula’s first term in the presidency, Bolsa Escola

was incorporated into Bolsa Famı́lia, a more encompassing transfer program

that now benefits millions of poor families in the country. In sum, a member

of a left party and co-creator of the most popular conditional-cash-transfer

program in the country was rated as the most rightist candidate by respon-

dents. This suggests individuals ideological assessments of Cristovam are

very likely the result of a prior belief that “unknown challengers” are more

likely to be rightists.

In terms of his performance at the polls, Cristovam received less than

3% of the votes mostly from wealthier municipalities. In light of the findings

obtained in Chapters 2 and 3, this can be a sign that education might indeed

not be a priority among poorer voters.

Geraldo Alckmin, in turn, ran under the PSDB, the party of the for-

mer president FHC. Respondents placed him somewhat towards the right,

which seems compatible with the position of his party. Unfortunately it is

impossible to assess how much of this consistency is due to lucky guesses as

Section 5.1 134

opposed to informed responses. Alckmin received about 42% of the votes

in the first round of elections in 2006, mostly from the richer regions in the

country. He lost the run-off to Lula by a margin of 20% of the votes (Lula

won 61% and Alckmin 39%).

Finally Lula was the candidate for the Workers’ Party (PT) a relatively

cohesive left party. In order to discuss his placement it is worth considering

both the time trends and his position relative to his main contenders (both

from the PSDB) in 1998 and 2006. Based on legislators’ assessment4 of

party positions over the period 1990 to 2005, Power and Zucco Jr find that

during the period 1990 to 1993 the PT and the PSDB tended to converge to

positions at the left of the median legislator. From 1993 to 2001, however

they begin to diverge with both the PT moving to the left and the PSDB

moving to the right past the median legislator.

Going back to the individual assessments of candidates, the year where

respondents rated FHC as more “pro-poor” than Lula – 1998 – was the

year were their parties diverged the most on ideological leanings. Lula’s

ratings were then considerably more to the right than that of his party.

This suggests a right bias in his evaluations at a year where respondents

had not yet had the opportunity to observe him in action.

While the scale presented to legislators was the same that was presented to respondents
in the LAPOP survey (1 - 10), the wordings were different. Legislators were asked about
left (1) and right (10) ideological placement, while respondents were asked about pro-poor
(10) and pro-rich (1). This last scale was reversed for the analysis. These differences make
precise comparisons difficult. Moreover, the methodology employed by Power and Zucco Jr
differs from the one employed here, since they have data on legislators self-placement on
the same scale available for computations. For details on the methodology the reader is
referred to the authors’ paper.
Section 5.1 135

When we move to 2006, where some ideological convergence between the

PT and the PSDB takes place, Lula receives mostly pro-poor ratings, even

more than he had in 1998. Thus taking these party positions into account,

the shift in perception regarding Lula was considerable. Were his rating in

1998 to reflect his party position then, they should have been even more to

the left than those in 2006.

To summarize, the data suggests that voters face uncertainty about the

policy positions of candidates during elections. This uncertainty is visi-

ble both in terms of incidence of respondents admitting they do not know

about the ideological inclinations of candidates, and also among those who

provided answers. Compared to ideological placements based on legisla-

tors assessments, respondents’ ratings in the two surveys analyzed not only

differed to a greater extent when it came to rating challengers – those who

never served as incumbents before – as opposed to incumbents, but also sug-

gested that answers could have resulted from a prior belief that left leaning

candidates tend to be rare.

After observing FHC’s first term respondents rated him as a moderate,

while Lula was placed slightly to FHC’s right. The poor and less educated

were more likely to rate FHC as pro-poor than Lula relative to the average

respondent – who also rated FHC as more pro-poor but by a smaller margin.

As shown in Figure 5.2, in terms of proportions in 1998 we have FHC close

to the center and Lula towards the center-right. It is only in 2006, after his

first term in office, that the voters will rate Lula as considerably pro-poor,

more so than two of his contenders, Heloı́sa Helena and Cristovam Buarque,

who were also, if not more, left leaning than Lula. Challengers in both years
Section 5.2 136

tended to be placed on moderate to right positions. If any non-separating

equilibria found were supported by the data we would likely observe the

opposite pattern. That is, candidates who never served as incumbents re-

ceiving left to moderate rankings.

Regarding the ratings of incumbents – FHC in 1998 and Lula in 2006 –

individuals’ assessments seemed more consistent with legislators rankings of

their parties, thus suggesting that some learning about the policy positions of

incumbents does indeed take place. In the next section I explore differences

in voting patterns between elections with an incumbent and those without


5.2 Socioeconomic Characteristics and Vote Inten-

tion in Presidential Elections

In the analysis that follows I focus on two particular implications of the

model. First, assuming redistribution is a salient issue and that preferences

for it differ among voters from different socioeconomic groups, I look for

evidence that voting choices in elections where no incumbent is running for

reelection are less predictable – based on voters socioeconomic characteristics

– than the ones with incumbents. Second, in elections with incumbents I look

for evidence that respondents make their voting choices based on information

about the incumbent’s policy position acquired during the incumbent’s first

Section 5.2 137

To answer these questions I build a baseline econometric model specified

to assess how socioeconomic characteristics affect voting decisions. The

focus on socioeconomic characteristics is a function of the object of study.

Redistribution is one issue where a person’s income, age and educational

level should be particularly relevant in determining preferences.

To address the first question I compare the predicted probabilities of vote

intention of three different profiles of voters (poor, median and rich) based

on the chosen econometric specification. The goal is first to estimate the

extent to which respondents belonging to different groups intended to vote

for different candidates. Second, we want to know whether results differ

depending on who the candidates were: former incumbents or “unknown


Given that the surveys analyzed do not contain questions on vote inten-

tion and respondents’ preferred levels of redistribution, the second question

needs to be evaluated in light of the policy decisions of incumbent candi-

dates during their first term. The kinds of policies an incumbent implements

in her or his first term can affect voting decisions at two levels. One more

visible level refers to the actual observable outcomes of these policies. They

can be gauged through pragmatic questions such as: Has poverty declined,

increased or remained the same as a result?

The other, less directly observable level, thus requiring a higher degree

of sophistication and information on the part of voters, refer to the actual

contents of the policies or the chosen approach to fighting poverty and in-

equality. These refer to questions such as: What kind of social policies

were supported? Which ones were reversed from the previous period and
Section 5.2 138

which new ones were created? For practical reasons, and not to impose too

much on the level of sophistication of voters I focus on the more pragmatic


The data gathered for this section come from various sources and mainly

from private polling firms. The 1989 data comes from IBOPE (Brazilian In-

stitute of Public Opinion and Statistics) and was conducted between the first

and the second round of the presidential elections. The data for 1994 comes

from the CBPA (Brazilian Company of Research and Analysis) and was

conducted in August 1994 (about a moth before the actual election). These

two were obtained through the Roper Center. For 1998 I have data from Vox

Populi, another private pooling firm. The survey was conducted in October

1998, a couple of days before the elections. For the remaining two elections

data was collected by CESOP at Unicamp and FGV/Opinião in 2002 and

CESOP and Ipsos in 2006. Both are part of the ESEB (Brazilian Electoral

Study) and were collected a couple of months before the elections. The last

three datasets were obtained through the Consortium of Social Information

(CIS)5 . They were all designed to be representative of the Brazilian popu-

lation. Except in the case of 1994, they all include multiple cities and cover

all states. In 1994 only residents of five major capitals were interviewed.

Given the different times relative to the actual elections that each of these

studies took place we might worry about how closely the voting intentions

reflect the actual results.BUILD GRAPH

The analysis presented focus on voters choice between two candidates.

The actual number of candidates in each election varied from 6 up to 20

Available at
Section 5.2 139

contestants. Usually though, no more than 3 were considered “viable”, that

is, receiving at least 10% of the votes. In the 1989 election I use the question

on vote intention for the second round, where the choices were Collor and

Lula. In the 1994 election I selected only those respondents who intended

to voter either for Fernando Henrique Cardoso (FHC) or Lula. The third

most voted candidate was Eneas receiving only 7% of the votes. In 1998

I run the analysis on a question prompting respondents to chose between

Lula and FHC. In this election, where FHC won in the first round, the third

most voted candidate was Ciro Gomes, receiving about 11% of the votes.

In 2002 and 2006 I focus on the question prompting respondents to choose

between the two contestants who made it to the second round, they were

Lula and José Serra and Lula and Alckmin, respectively.

This analysis is thus based on the IIA (independence of irrelevant alter-

natives) property of multinomial logit models. This means that throughout

the analysis results should be interpreted in relative rather than absolute

terms6 .

To make the analysis comparable across the five election years, I recoded

the independent variables so that every year was measured using the same

scale. This, however, entailed identifying the “minimum common denom-

inator” for each variable. As a consequence I had to settle for the most

condensed categorization of answers. Below I present a brief description

of how each of these variables was measured along with some associated

The sample size for each election year and the actual number of observations used in
the analysis are as follows: 1989 - 3650,3109; 1994 - 1400,802; 1998 - 3168, 2802; 2002 -
2514,1931; and 2006 - 1000,785
Section 5.2 140

Income All surveys ask respondents the estimated income of the house-

hold, that is, of all incomes of all household members combined. Un-

fortunately only two of the surveys ask how many people live in the

household. Thus I am not able to work with per capita values. Three

of the surveys categorize income in terms of corresponding number of

minimum wages, while the remaining two provide the actual amount in

Reais as a continuous variable. The preliminary approach adopted was

to transform the minimum wage categories into current Reais amounts

(at 2008 values) and replace the value of each category with the mean

log value of that category range.

Education This indicator is composed of four categories of educational

attainment: 0 - from illiterate/no education to 4th grade; 1 - from

5th to 8th grade; 2 - high school (9th to 11th grade); and 3 - some or

complete college and beyond.

Age The age categories are coded as follows: 1 - 16 to 17 years old; 2 - 18

to 25 years old; 3 - 26 to 30 years old; 4 - 31 to 40 years old; 5 - 41 to

50 years; 6 - more than 51 years old.

Gender Females are coded as 1 and males as zero.

I take advantage of the fact that Lula was a candidate in all five elections

and set the intention to vote for him to 1 and the intention to vote for his

opponent to zero. Figure 5.5 shows the estimated coefficients along with

the 95% confidence intervals. There are two specifications being compared
Section 5.2 141

Figure 5.5: Logit regression results: votes for Lula vs. opponent
●● ●●
Intercept ●● ●● 0
Income ●● ●● ●● 0
●● −0.5

●● ●● ●
Education ●

● 0

Age ●● 0
●● ●●

Female ●● ●●
●● −0.5
Education x ●

● ● 0
Income ●










● Interaction
● No Interaction

Lines correspond to 95% confidence intervals. This graph was built based on code developed by
Kastellec and Leoni (2007).
Section 5.2 142

in this graph. One containing an interaction term of income and education

and one without it.

While these coefficients give us some idea of what mattered in the de-

cision of who to vote for, it is difficult to assess the model’s results based

solely on them. Following the approach adopted in the previous chapters I

compare predicted probabilities of meaningful profiles of voters to discuss

the results. While the theoretical model in the previous chapter was based

on three types of voters going from a majority of poor to a minority of

middle and upper classes, empirically it is difficult to distinguish between

the last two. The number of respondents to the surveys would need to be

considerably higher so that we could have a representative sample of the

two smaller groups. Thus when generating the profiles for comparison I ag-

gregate them under the label “rich”. Given that poorer and less educated

respondents form a majority I break down this group into two, simply to

explore possible differences. I divide this group into the very poor and the

median, which as argued in the previous chapter is rather low than middle


The very poor voter profile was created to represent the about one-third

to one-fourth of the population in Brazil considered to be living below the

poverty line. Their average income during the course of these years were

always around R$60 per capita in Reais of 20087 . Thus I set the income

of the poor profile to R$240, corresponding to a poor household with 4

members (with all the caveats mentioned above). In terms of in sample

All income values in this exercise were decided based on per capita income estimates
for each category based on household surveys. The data is available at IPEADATA.
Section 5.2 143

household income this figure corresponds roughly to the bottom 20%. The

education level of the poor profile was set to 0, meaning 4th grade or less,

the gender was set to male and the age to the mean, represented by the

group of 31 to 40 year olds8 .

The median voter, while better off than the very poor, is still poor with

a household income set at R$800 per month (which corresponds to about

U S350). The level of education was set to 1, corresponding to the category

4th to 8th grade, and the age and gender as specified for the poor. Finally I

set the rich voter profile to a household monthly income at around R$6600

(corresponding to the top quintile in the sample), college education and

mean age.

Do Voters from Different Socioeconomic Groups Vote Alike?

Going back to the first question, are elections without an incumbent less

predictable than the ones where an incumbent was running for reelection?

That is, can we forecast respondents voting decisions based on their so-

cioeconomic characteristics once they had the opportunity to observe the

policies implemented by an incumbent?

To answer this question I calculate predicted probabilities of intention

to vote for Lula for each profile generated. If specific constituencies based

on income and education formed around an incumbent we should observe

I also compared predicted probabilities setting the age to the other categories and
did not observe a significant difference in results on first differences. I did though observe
differences in the predicted probability of the poor voting for Lula instead of his opponent.
Younger poor voters (18 to 25) were more likely to vote for Lula in 1989, while older
voters (30 to 40, and 40 to 50) were significantly more likely to vote for Collor. The other
significant difference occurred in 1994, where both median and richer older voters (30 to
50) were significantly more likely to vote for FHC.
Section 5.2 144

significant differences in vote intention between the profiles. Conversely, if

citizens do not have information on the policy position of candidates, we

expect to observe a relatively random pattern of voting and an inability to

differentiate between the profiles.

Figure 5.6: First differences: probability of voting for Lula vs. opponent

● Poor − Rich

Poor − Median
Rich − Median

First Differences: Probability Vote Lula





Collor FHC FHC Lula Lula

1989 1994 1998 2002 2006

In Figure 5.6 I plot the estimated first differences along with the cor-

responding 95% confidence intervals. Starting with the very first election

we actually observe significant differences between voters with different so-

Section 5.2 145

cioeconomic backgrounds. With no previous experience enabling voters to

determine the ideological inclinations of candidates we see a clear pattern of

the poor being significantly more likely to vote for Collor (a candidate from

the poor Northeast) than the rich and the median voter. We also observe

the rich being significantly more likely to vote for Lula (a blue collar worker

from the state of São Paulo) than both the poor and the median.

Different reasons might be behind this result. This was the very first

presidential election after democratization, and possibly the election were

the least amount of information about candidates’ policy positions was avail-

able to voters. This could have led to two possibilities. First, under such

circumstances it would be natural for voters to rely on other features of can-

didates in making their decisions. Indeed Collor ran on an anti-corruption

platform that reflected the popularity he acquired during his term as a gov-

ernor of Alagoas – a poor state in the Northeast of the country – of being

tough on civil servants receiving public salaries and benefits without actu-

ally working. His great clout over the media, specially in the poorer regions

of the country, contributed to building a strong reputation, specially among

the poor.

Another view is that poor voters being more risk averse would be more

likely to favor a candidate with ties to previews governments and on which

they had more information. No one really knew what kinds of policies to

expect from Lula at that time, in particular whether his policies would

benefit the poor or not.

It is interesting to note, however, that Collor’s tenure as a president was

short lived, mainly because of his policy choices and, ironically, corruption.
Section 5.2 146

He only ruled for about half of his term, as corruption scandals and disas-

trous economic policy led to a motion for his impeachment and subsequent

resignation in 1992.

In the following “non-incumbent” election (1994) the distinction between

voters earning different incomes and of distinct educational levels was much

less stark. On the one hand, we observe a statistically significant difference

between the very poor and the rich and median respondents. According to

the estimation, the very poor were more likely to intend to cast their votes for

Lula than the other two groups. On the other hand, there was no statistically

significant difference between the median and the rich respondents. Given

that the median voter in Brazil is poor, this result conforms with the model


Like Collor, while FHC was not an incumbent to most voters, he had

built a reputation of being highly skilled at dealing with the most impor-

tant issue of the day: inflation. FHC served as a finance minister during

the previous government and was the mastermind behind the “Plan Real”.

This was the economic stabilization plan that put an end to years of hyper-

inflation, bringing the four-digit annual inflation rate of 1993 (2490%) down

to one digit in subsequent years. The plan was not implemented during

FHC’s tenure as finance-minister as he was running for office at the time

(the plan was implemented in June of 1994). Clearly the triumph of the

plan was used during his campaign to signal his competence (apparently

with a high degree of success as FHC started the race behind Lula gaining

increasing ground after the implementation of the Plan Real). Whether or

not this conveyed information on FHC’s ideological inclinations is however

Section 5.2 147

debatable. As with the case of corruption, we would expect the success of

the Real plan to favor every citizen in the country. Indeed individuals of

every socioeconomic strata were more likely to vote for him than for Lula.

In the next elections, when FHC ran for reelection, we observe a higher

degree of differentiation between the three groups of respondents, consistent

with the model expectations. Interestingly we observe a complete reversal

from the 1994 tendencies. In this election the poor were significantly more

likely to intend to vote for FHC than both the rich and the median. Also

the rich were significantly more likely than the median to intend to vote

for Lula instead of FHC. As mentioned earlier, according to the Vox Populi

poll one month before the election FHC was perceived to be more pro-poor

than Lula. His moderate position according to the survey results also seem

consistent with the overall impact of his policies on poverty and inequality

– explored in more detail below – specially if compared to his predecessor


In 2002, when no incumbent was running for office, we move to a state

of imperceptible differences, as suggested by the theoretical model. Lula

won this election after a second round where he received 61% of the votes

and his opponent, José Serra, received 38%. Finally at the last contest

(2006), where Lula ran for reelection, we observe marked differences in the

predicted intentions to vote for Lula as opposed to Alckmin. Again matching

the expectations derived from the theoretical model. The clear pattern is

the poor being most likely to intend to vote for Lula, followed by the median.

The rich are at the opposite end, being significantly more likely to choose

the challenger over the incumbent. As the results in the previous section
Section 5.2 148

suggested, Lula was perceived as significantly more pro-poor than Alckmin

and all the other main contenders.

Linking Respondents Voting Intentions to Incumbents’ Per-


One thing is assessing the degree to which we are able to distinguish between

voting intentions of different socioeconomic groups. Another is evaluating

whether the direction of the observed differences follow the expected pat-

terns. Ideally this issue should be investigated based on data containing both

individuals’ vote intentions and policy preferences in each election. Given

data limitations, this exercise is conducted based on background knowledge

on the policy choices of incumbents during their first terms.

The task of placing candidates on a unidimensional ideological scale is

simplified given the focus on redistribution. As mentioned earlier, I abstract

from the more sophisticated issue of qualifying specific policies in terms

of their approach to fighting, or not, poverty and inequality. I adopt the

objective route of basing the discussion on changes in indicators measuring

levels of poverty and inequality over the course of a candidate’s term.

In other words, whatever the subjective merits of different ways of help-

ing the poor (with or without conditionality, for example), I focus on their

ultimate outcome: whether they contributed to the increase or decline in

levels of poverty and inequality.

The crucial implicit assumption in this exercise is that redistribution,

as I defined it, is a relevant policy issue that we expect voters to respond

to first and foremost. This might be a strong assumption. However I take

Section 5.2 149

it as highly plausible given the fact that Brazil fares particularly bad in

both poverty and inequality, especially when compared to other countries

at similar levels of development.

While I am particularly interested in the two elections with an incumbent

candidate, in order to place candidates in perspective I also report on the

performance of the first democratically elected president. During Collor’s

short tenure, poverty and inequality remained high. Around 40% of the

population was estimated to be living below the poverty line, the income of

the poor declined somewhat and inequality levels reached a peak value of

the Gini coefficient of .64.

Table 5.1 reports some changes in selected indicators of poverty, using

data available at SEDLAC (Socioeconomic Database for Latin America and

the Caribbean). These data are based on the annual household surveys con-

ducted by the IBGE (Brazilian Institute of Geography and Statistics). Many

different indicators are available. Given that they are highly correlated, I

report a selection of them.

Table 5.1: Changes in indicators of poverty and inequality during presiden-

tial terms

Measure 1989-1993 1994-1998 2002-2005

Poverty as basic needs9 - 1.5% - 1.5% - 7%
Gini .64 .6 .59 to .56
Share below U S$2/day 6.5% - 2% - 14%
Child labor 3% - 28% - 26%
Child labor bottom quintile 33% - 16% - 6%

Source: SEDLAC (Socioeconomic Database for Latin America and the Caribbean)

Computed using an indicator of basic needs. This indicator includes the following
Section 5.2 150

What we observe is that FHC’s first term was marked by no drastic

change in the condition of the poor. In relative terms, however, compared

to what happened during the government of Collor, the situation of the

poor has started to reverse course. It was not until the government of Lula,

however, that considerable improvements were obtained, both in absolute

and relative terms.

These figures exclude the biggest poverty alleviation program the “Bolsa

Famı́lia”, which conditions cash transfers on school attendance and health

check-ups. FHC was the first president to adopt the program nationally,

focusing on municipalities with selected development indicators below the

country average. Thus only families living in very poor municipalities were

eligible. Lula extended the program to cover the whole national territory,

with eligibility based solely on household characteristics. The growth in the

number of families enrolled in the program was considerable during the gov-

ernment of Lula. The number of beneficiaries from governmental transfers

almost tripled, going from 4,824,542 individuals by the end of FHC’s last

term to 12,013,372 by the end of Lula’s first term (IPEA 2005). As noted by

Clements, Faircloth and Verhoeven, in Brazil, “the expansion of the Bolsa

(i) more than 4 persons per room
(ii) the household lives in ÒpoorÓ places (e.g. street, shanty towns)
(iii) the dwelling is made of low-quality materials (see section 7)
(iv) the dwelling does not have access to water
(v) the dwelling does not have an hygienic restroom (see section 7)
(vi) there are children aged 7 to 11 not attending school
(vii) the household head does not have a primary school degree
(viii) the household head does not have a high-school degree, and there are more than 4
household members for each income earner.
Section 5.2 151

Famı́lia during 2003 and 2005 contributed to a narrowing of income gaps

between the rich and poor and a reduction in poverty rates” (2007, 27).

Consistent with the figures just presented, social spending has seen a

steady increase over the past years as shown in Figures 5.7 and 5.8 based

on data from ECLAC.

Social Expenditures Per Capita


Social Expendinture Per Capita


FHC FHC Lula Lula

1994 1998 2002 2006

Figure 5.7: Trends in social expenditure in current US of 2000 (health,

education and housing)

Based on these figures we see a clear pattern of increased amelioration

in the situation of the poor. While they were timid during FHC’s first

term, they were markedly better than under his predecessor. With Lula,
Section 5.2 152

Social Expenditures Per Capita

1000 Total
Social Security
Social Expendinture Per Capita


FHC FHC Lula Lula

1994 1998 2002 2006

Figure 5.8: Trends in social expenditure in current US of 2000 (social secu-

rity and total)

the improvements are remarkable, especially if we take into account the

fact that less scope for improvement would lead to lower advancements in

percentage terms.

According to the separating equilibrium found in the previous chapter, if

a candidate displays relatively centrist rather than rightist traits she would

be receiving the votes of the poor, but not those of the rich. While we do

not observe this stark result we do observe patterns consistent with it (see

Figure 5.9). Most respondents reported the intention to vote for FHC when
Section 5.2 153

Figure 5.9: Predicted probabilities of voting for Lula vs. opponent


● poor
Predicted Probabilities: Vote for Lula





Collor FHC FHC Lula Lula

1989 1994 1998 2002 2006

he ran for reelection in 1998, which is reflected in predicted probabilities for

Lula that are below 50% for all groups. Thus the model prediction that the

rich would vote for the challenger is not observed.

One of the reasons – that deserves further exploration – can be that

the better educated are better able to evaluate the policy position of can-

didates. As pointed out by Power and Zucco Jr, during the government of

FHC the PSDB shifted considerably to the right. For the average voter,
Section 5.3 154

however, Lula was located even more to the right, which, as pointed out

earlier, could have resulted from a shared prior belief that “unknown chal-

lengers” are rarely leftist. Better educated voters, with access to more and

higher quality information might have been able to assess the candidates’

ideological positions with more accuracy.

In line with the results obtained in the previous section, we do observe

that the poor were more likely to intend to vote for FHC if compared both

to the rich and the median. The median, in turn, was also more likely to

vote for FHC if compared to the rich.

Now turning to the 2006 elections the observed predictions follow much

more closely the model results. The poor and the median were considerably

more likely to have voted for Lula, the incumbent, after observing his re-

distributive policies and experiencing considerable improvements. The rich,

however, were split between Lula and his opponent Geraldo Alckmin.

5.3 Conclusion

In this chapter I looked for evidence supporting the main assumption of the

theoretical model developed in the previous chapter and evidence consistent

with the claim that the separating equilibrium provides a good description

of elections in new democracies. That is, I sought empirical evidence that

a) voters face uncertainty during elections; b) voters believe that most can-

didates are either moderate or rightist; c) voters update their beliefs about

a candidate policy position after observing her or him serve a term in of-

fice; d) significant differences in vote intention among voters from different

Section 5.3 155

socioeconomic groups are more likely to arise when incumbents are running

for reelection – this is because voters have better information about the

ideological inclinations of incumbents and can thus base their decisions on

policy issues such as redistribution.

Throughout the chapter I focused on presidential elections in Brazil,

analyzing whenever possible, data corresponding to all elections since de-

mocratization in 1989. I relied mostly on individual level data collected

from private polling firms during the electoral years.

The empirical exercises suffered from a number of limitations imposed by

data availability. Data collection in new democracies is not as well structured

as it is in more developed countries. In that regard, the theory developed

in the previous chapter provides useful insights into the kind of data collec-

tion efforts that could contribute to a better understanding of how voters

choose candidates in elections. First, in order to gauge the degree of un-

certainty faced by voters during elections we need data on both individuals

evaluations of candidates policy positions on several issues and individuals’

self-placement on these same scales.

Second, in order to assess the extent to which these evaluations matter for

their voting choices, we need the same survey to contain both the assessment

questions and vote intention questions. Given that available data comes

mainly from private polling surveys conducted with the main purpose of

gauging vote intentions, these two conditions were not met. Despite these

limitations, the exploratory analysis performed in this chapter suggests a

number of relevant insights.

Section 5.3 156

Individuals responses to candidates’ ideological placements on a pro-poor

to pro-rich scale both in 1998 and 2006 suggested they do indeed face uncer-

tainty. Moreover, in line with the theoretical model, this uncertainty seemed

higher with respect to “unknown challengers” than incumbents. There were

thus clear signs that voters update their beliefs about the policy positions of

incumbents. The individual assessments of incumbents tended to be closer

to the evaluations of legislators about the candidates’ parties than individual

assessments of challengers.

Furthermore, comparisons between legislators’ evaluations of challengers’

parties positions and the individual assessments of challengers revealed a

bias towards the right in individuals’ evaluations. That is, respondents’

placements of challengers on a policy scale seem to have resulted from the

belief that left types are rare. This is exactly the condition supporting the

separating equilibrium found in the previous chapter. If any of the other

equilibria found held, we would observe the opposite tendency.

In terms of voting patterns results are mostly consistent with those fol-

lowing from the separating equilibrium. While in the first two elections after

democratization results differed somewhat from the model expectations, the

following three elections conformed remarkably well. One plausible reason

for the discrepancies in the first two elections is the fact that they were the

ones with most uncertainty about candidates’ policy positions. Two case

scenarios could have emerged. In the first we could argue that given the

ties of the first two candidates to their respective predecessors’ government,

risk aversion led poor voters to favor those candidates to a greater extent

than richer voters. The second is that under considerable uncertainty, voters
Section 5.3 157

had to choose candidates based on attributes, other than policy preferences,

that generated the unexpected cleavages. Better data and further analysis

would be necessary, however, to reach a conclusion. Results regarding the

last three presidential elections (1998, 2002 and 2006), however, suggest the

separating equilibrium might indeed provide a good account of elections and

hence of low levels or redistribution in unequal new democracies.


Chapter 6


In this study I emphasized two basic but crucial factors in accounting

for low levels and delays in the adoption of redistribution in unequal new

democracies: citizens’ preferences for different redistributive policies and

uncertainty about the ideological inclinations of candidates competing to

represent these preferences.

Regarding preferences I called attention to the diversity of redistributive

policies and how levels of support for each type can be affected by levels of

poverty and inequality. In particular, I argued that poverty and inequality

can lead to low demand for certain forms of redistribution – like public

provision of education – found to be a significant instrument in promoting

more equality.

Low income coupled with the lack of credit markets serving poorer cit-

izens can constraint their preferred levels of public provision of education

to values below those that would obtain, were they able to borrow against

future earnings. Inequality, in turn, can make it profitable for wealthier vot-

ers to opt for private provision of services instead of supporting improved

Section 6.0 159

quality of public delivery. As the gap between rich and poor widens, the

financial burden of providing widespread public services can become too

heavy for the rich, relative to the costs of private exit options.

I provided evidence consistent with these dynamics using individual level

data on support for taxation and public provision of health and education,

and on the priority assigned to both services as important issues the gov-

ernment should address. Poorer respondents tended to be significantly less

likely to prioritize education, for example. Moreover, despite the little vari-

ation in levels of inequality across states, I find evidence that under higher

inequality richer voters display lower levels of support for public provision

of such services.

In the second part of the dissertation I explored the effect of uncertainty

about candidates’ policy positions on representation. Based on a model

of elections as a game of incomplete information where candidates can be

either policy or office oriented – or a combination of both – I argued that

uncertainty can account for both lower levels of redistribution than expected

given individuals’ preferences, and delays in the implementation of policies

favored by a majority.

The lack of information about candidates’ ideal policies leads to a learn-

ing process where voters can update their beliefs about the ideological incli-

nation of a candidate once she serves as an incumbent. During this learning

process, however, delays can happen due to the probabilistic nature of the

electoral choice and the fact that a majority might reelect incumbents that

do not closely mirror their preferences – even under full information about

these preferences. In particular, poor voters might find it optimal to reelect

Section 6.0 160

incumbents implementing less redistribution than they favor. This happens

as long as voters share the belief that candidates truly committed to the

interests of the poor are relatively rare.

Analysis of survey data about each presidential election in Brazil since

democratization suggests that voters do indeed face uncertainty about can-

didates ideological placements. Furthermore, there is indication that when

evaluating candidates during elections, voters responses follows from the be-

lief that pro-poor candidates are rare relative to moderates and right types.

Regarding learning, there were clear signs that candidates were somewhat

more knowledgeable about incumbents’, when one was running for reelec-

tion, than about challengers’ policy positions. Moreover, voters seem to use

this additional information obtained from retrospective evaluations, when

deciding whether to reelect the incumbent or to elect a challenger instead.


Appendix 162

Proofs of Propositions in Chapter 2

Proof of Lemma 2.1: The older voters are indifferent between policies

that provides them the same income. Therefore, setting the utility to a

constant and solving for g, yields the older voters indifference curves. This

is equivalently done by taking the derivatives of u with respect to g and t

as follows. For wi ∈ {wl , wh }

θq w̄ − wi (1 − θ)
− um dt + um dg = 0
(1 − θ) (1 − θ)

dt w̄ − wi (1 − θ)
dg θq

The ideal point of each voter is the pair (t, g) that maximizes his util-

ity function. It is easy to see that for ro the utility function is maxi-

mized at (t = 0, g = 0). Since u is increasing in its arguments, ro is

better off keeping his entire income wh , then paying taxes and getting

(1 − t)wh (tw̄ − θqg)/(1 − θ) < wh . For po , on the other hand, since

wl < w̄ < wh , he is better off at (t = 1, g = 0) which yields him the

highest possible payoff u(w̄/(1 − θ)).

Proof of Lemma 2.2: The indifference curves of the younger voters is

given by

ue dt − wi um dg = 0

dt wi um
dg ue
Appendix 163

By concavity of u(e), ue > 0 up to e∗ , where ue (e∗ ) = 0, and ue < 0 there-

after. Thus the indifference curves of the younger voters are both initially

positive and increasing in g.

Young poor

Lets look at the case of the young poor first. Their private decision is to

maximize s given a policy (t, g), subject to s > 0. That is

max u((g + s), (wi (1 − t) − αqs))


that is

L = u((g + s), (wi (1 − t) − αqs)) + λs,

→ ue − αqum + λ = 0


λs = 0


That is s > 0 if ue > αqum and s = 0 otherwise. By A1, it is always the

case that s = 0 for the young poor. Thus, ue < αqum . The initial slope of

the young poor indifference curves is greater than

wl um
Appendix 164

Notice that the higher the price of private education the less steep the initial

slope. From this point on ue is decreasing and the slope gets steeper until it

becomes negative at g = e∗ . The ideal point of the young poor is going to be

the point where their indifference curve is tangent to the budget constraint

line. We denote this point by (t = t∗p , g = tw̄/θq).

Young rich

For the young rich, the derivation is similar with the exception that

these voters can afford private education. The optimal s for the young rich

is going to be positive whenever the level of publicly provided education is

below the optimum level g given t, denoted gr (t). Until that level is reached,

s > 0, ue = αqum , and the slope of the indifference curve is