Democracy Under High Inequality: Capacity, Spending, and Participation

Democracy under High Inequality: Capacity, Spending,
and Participation
Francesc Amat, University of Barcelona
Pablo Beramendi, Duke University
Contrary to the view that inequality reduces turnout, political participation among low-income voters is higher in
democracies with high levels of inequality and intermediate levels of state capacity. We address this puzzle by analyzing
the link between political mobilization and spending decisions at different levels of inequality and state capacity. Under
high inequality and low levels of capacity, parties find it optimal to mobilize low-income voters via targeted goods. Yet,
as capacity increases, targeted mobilization becomes a less effective strategy for voters’ mobilization. To evaluate the
implications of this argument we exploit a quasi-natural experiment, the anticorruption audits by the Brazilian gov-
ernment. We show that an exogenous increase in monitoring effort by the state breaks the equilibrium around high
participation of the poor and leads to a reduction in turnout, a reduction in the provision of targeted goods, and a
decline in the likelihood of reelection by incumbents.
I nequality is known to bias political influence, under-

mining the idea of citizens as political equals (Dahl 1991;
Przeworski 2010). The notion that inequality reduces, in
relative terms, the political participation by low-income
voters is a widely shared understanding in comparative pol-
and high levels of formal political equality? To address this
puzzle, we reason from the premise that turnout levels reflect
primarily political parties’ efforts to mobilize voters, especially
those situated in the lower half of the income distribution. The
explanation of turnout requires an account of not only voters’
itics.1 And yet, in younger, less developed, and very unequal incentives to engage in elections but also parties’ choices
democracies poor voters often seem as willing (if not more) to about (1) whom to target and (2) how to target.
engage in politics as their counterparts in rich democracies We argue that mobilizing low-income voters through
(Kasara and Suryanarayan 2014; Krishna 2008; Stokes et al. targeted spending is an optimal strategy in contexts of high
2013). In fact, the relationship between inequality and elec- inequality and low capacity, that is, in contexts in which weak
toral turnout in the developing world reverses the patterns administrative capacity allows incumbents to manipulate
observed in wealthier democracies: higher levels of inequality budgets for political benefit. In such an environment, parties
are associated with high, rather than low, electoral partici- see an electoral advantage in seeking the support of a large
pation in places like Mexico, Brazil, or Peru, even after one pool of low-income citizens. Accordingly, they prioritize
accounts for compulsory voting laws. policies that effectively work as targeted efforts toward low-
How does democracy work under very high economic income voters based on a conditional exchange for political
inequality to feature at once high levels of economic inequality support. It is in these settings where we expect to observe a
Francesc Amat (francesc.amat@ub.edu) is research fellow at the Institutions and Political Economy Research Group at the University of Barcelona. Pablo
Beramendi (pablo.beramendi@duke.edu) is professor of political science at Duke University.
We gratefully acknowledge financial support from the Korean National Science Foundation (grant SES-1227237). This project has also received funding
from the European Research Council through the Partydemocracy project (grant 694318). Francesc Amat has received financial support through the
Postdoctoral Junior Leader Fellowship Program from La Caixa Banking Foundation. Data and supporting materials necessary to reproduce the numerical
results in the article are available in the JOP Dataverse (https://dataverse.harvard.edu/dataverse/jop). An online appendix with supplementary material is
available at https://doi.org/10.1086/707397.
1. The effect occurs through a variety of channels. Inequality limits the resources poorer individuals need to engage in politics (Solt 2008; Verba,
Schlozman, and Brady 1995), alters the structure of informational networks (Bond et al. 2012), privileges wealthier voters via campaign contributions
(Campante 2011) or political representation (Anduiza 1999; Bartels 2009; Gingerich 2013), undermines pro-redistributive coalitions, and alters the
incentives of political parties to target low-income voters (Anderson and Beramendi 2012; Gallego 2014).
The Journal of Politics, volume 82, number 3. Published online May 6, 2020. https://doi.org/10.1086/707397
q 2020 by the Southern Political Science Association. All rights reserved. 0022-3816/2020/8203-0004$10.00 859
860 / Democracy under High Inequality Francesc Amat and Pablo Beramendi
higher involvement by low-income citizens and, as a result, persist (Robinson and Verdier 2013). As such, we offer a
higher levels of turnout. But, as capacity increases, the incen- genuinely political mechanism behind the persistence of
tives to rely on targeted efforts toward the poor, including perverse accountability (Stokes 2005), bad development
clientelism, decline.2 equilibriums, and the self-reproduction of inequality, both
We make three contributions. First, theoretically, we take economic and political. In high-inequality contexts, elites
the choice of strategic targeting of spending back to first magnify their political influence by conditioning the po-
principles and focus on its determinants without assuming a litical voice of the poor as opposed to excluding them al-
given ideological space ex ante. Previous contributions to the together from the political process, ultimately reproducing
literature on clientelism and turnout buying (Hidalgo and and anchoring economic inequality over time.
Nichter 2016; Larreguy, Marshall, and Querubin 2016; Our third contribution speaks to a central issue in the
Nichter 2008) assume a policy space where the ideological political economy of development: the impact of interven-
distance between parties and voters is part of the latter’s tions designed to undermine the sort of persistent political
utility function. Assuming a policy space implies the exis- underdevelopment described above. Empirically, we analyze
tence of programmatic politics, that is, a style of political the relationship between inequality, capacity, and partici-
competition in which targeting low-income voters explicitly pation from two perspectives: a characterization of the status
is far less frequent, which is part of what we aim to explain.3 quo equilibrium and a thorough study of the consequences
Our second contribution lies in how we approach the of a truly exogenous capacity shock on preexisting political
analysis of strategic spending decisions. While a large litera- relationships. To this end, we leverage microlevel informa-
ture has focused on the connection between development and tion in a case, Brazil, where the status quo equilibrium of
modes of political competition (Kirchheimer 1965; Kitschelt high economic inequality and low capacity is prevalent.
and Wilkinson 2007; Stokes et al. 2013), the link between in- Exploiting information about turnout, economic inequal-
equality, capacity, and politicians’ incentives to spend strate- ity, and spending in Brazilian municipalities, we show how
gically is far less understood. Our model studies how inequal- the status quo patterns reflect our model’s implications: (1) in
ity and capacity, jointly, determine the use of local targeted contexts of high inequality and low capacity, incumbents do
goods as a mobilization strategy. In particular, we study how, resort to higher levels of targeted spending; (2) these deci-
given inequality, the institutional capacity of the state to mon- sions, in turn, translate into higher levels of turnout. To ex-
itor its elites shapes the electoral incentives to resort to tar- plore the use of targeted goods, we focus on primary educa-
geted mobilization strategies. Low levels of monitoring ca- tion spending and the patterns of public employment at the
pacity reinforce and facilitate elites’ strategies to provide local local level. These are policy realms where local authorities
targeted goods for political purposes and contribute to the have large amounts of discretion to alter the local labor force
political engagement of low-income voters. In examining this (via wages or employment; Calvo and Murillo 2004), privilege
connection, our argument complements recent contributions schools’s resources by area (thus benefiting different pools of
on the relationship between redistributive capacity and turn- voters), and even manipulate access through the matching of
out inequality (Kasara and Suryanarayan 2014). facilities to specific subareas within the municipality (de
More generally, a better understanding of the connec- Oliveira and Adrião 2007). This form of inefficient redistri-
tion between economic inequality, party strategies, and bution (Robinson and Verdier 2013), we argue, is part of a
capacity helps illuminate when bad equilibriums (with high political strategy that yields higher levels of turnout.
inequality and low state capacity) are likely to emerge and In turn, the randomized anticorruption audits of Brazilian
municipalities, introduced in the early 2000s by the federal
government, provide an opportunity to analyze how an ex-
ogenous capacity shock alters preexisting politico-economic
2. These targeted efforts can take many forms, including clientelistic, equilibriums. We model the impact of these reforms as an
interpersonal exchanges or the manipulation of budget allocations to what
exogenous change in the state’s effort to monitor elites. The
otherwise are labeled as targeted goods. In this sense, we see clientelism as a
specific type of a more general phenomenon. For a more detailed analy-
audits were directly designed as an effort from the federal
sis of different forms of nonprogrammatic politics consistent with the ap- government to curb down corruption and the misuse of
proach in this article, see Stokes et al. (2013). federal funds by Brazilian municipalities (on average 70% of
3. Our focus on the link between structural conditions and party the local budget rests on federal transfers). By exploiting the
strategies also enhances our understanding of elites’ mobilization strate-
gies (Diaz-Cayeros, Estevez, and Magaloni 2016) and contributes to the
randomized nature of the audits and their timing (before or
study of the politics of turnout buying (Hidalgo and Nichter 2016; after the 2004 election), we show that the exposure to an audit
Larreguy et al. 2016; Nichter 2008). before the election depresses turnout rates (especially in rural
Volume 82 Number 3 July 2020 / 861
contexts with preeminent low levels of education), leads to a how much to devote to local targeted goods for low-income
reduction of local targeted goods in the subsequent period, voters (bP), how much to targeted goods for high-income
and, by implication, reduces the incumbent’s likelihood of voters (bR), and how much to general public goods (g). Within
reelection. No such process is observable in the municipalities this framework, we study how inequality shapes the conditions
that were not audited before 2004. under which elites resort to the targeted mobilization of low-
Our analysis provides a detailed case study of a major re- income voters.
form effort to undermine corruption and budgetary manip-
ulation, and it contributes to a growing literature on the Model
impact of randomized audits as an instrument for political We assume politics to be an activity initiated by elites at all
and economic development (Ferraz and Finan 2008; Olken ends of the ideological spectrum. In other words, we assume
2007). Analytically, we predict that an increase in such efforts a sequential setup in which the elites (the rich) move first and
reduces elites’ ability to politically manipulate targeted goods. the low-income voters move afterward and in which the rich
And rather counterintuitively, we show that, under high in- have perfect information. Accordingly, mobilization is a
equality, exogenous increases in the state’s monitoring ca- choice by different groups of rich citizens. The fundamental
pacity lead to a reduction of budgetary efforts to lure low- problem for any party is to maximize the utility of its base
income voters. Our results illustrate the complex effects of such that it attracts the support of low-income voters. That
interventions aimed to sever this kind of strategy. Using pri- is, the rich will optimize their policy selection in such a way
mary education policy at the local level as a relevant case that they (1) meet their budget constraint and (2) at least
study, we show that exogenous attempts to control political leave the poor indifferent between their policy offering and
machines may actually reduce the welfare of low-income the offering that the poor would consider optimal.
citizens in the short and medium run, especially in contexts in To incorporate inequality into the analysis, we define d
which weak fiscal and state capacity prevents new incumbents and (1 2 d) as the fraction of, respectively, rich and poor
from replacing targeted benefits with programmatic public citizens in any given society. Similarly, we define f and
good provision (Nathan 2016). (1 2 f) as the share of income of, respectively, the rich and
the poor. Using these simple definitions we can express the
THEORY: INEQUALITY, CAPACITY, AND SPENDING income of the rich (wR) and the poor (wP) as a function of
Our central argument is as follows: observable patterns of inequality:
turnout, in particular that of low-income voters, reflect stra-
w
f (1 2 f)
w
tegic decisions by governments about how to spend. These wR p and wP p :
decisions shape electoral participation. The main analytical d 12d
contribution of this article is to theorize this choice as jointly
determined by inequality and capacity. But, we assume elites face a standard budget constraint
In designing their budgets, political parties prioritize the defined by t w p bP 1 bR 1 g. To capture variation in terms
expected electoral returns of their spending priorities. Their of state capacity, we impose the behavioral assumption that
strategic choice focuses on the scope of targeting different a share, l, of the income of the rich is nontaxable by low-
income groups. At one end of the continuum, effort in public income voters. Thus, l allows us to capture the role played by
goods such as national defense, strictly defined as nonrivalrous the state’s ability to monitor and tax its citizens. It also allows
and nonexcludable, implies no targeting; at the other, the most us to analyze the predictions emerging from exogenous
personalistic version of clientelistic exchanges of political changes in such capacity. Essentially it implies that the poor
support for transfers of consumables or money to individuals had limited ability to tax the elite, but the elite has full ca-
implies full targeting (Kitschelt and Wilkinson 2007). In be- pacity to tax itself. Accordingly, the budget constraint is
tween, tax-funded, loosely defined public goods vary on their defined as
expected distributive incidence of the policy (Stigler 1970).
Some policies, such as higher education, benefit dispropor- (1 2 lf) p bP 1 bR 1 g for the share of citizens (1 2 d)
tw
tionately the top end of the income distribution (Ansell 2010); p bP 1 bR 1 g for the share of citizens d:
and t w
others, such as primary education or universal primary health
care, are relatively more redistributive. On the basis of these premises, we model the problem as
In this section, we model the manipulation of policy and a strategic interaction in which low-income voters decide
spending choices as a mobilization device to attract low- whether to vote, and the elite parties choose which policy
income voters. Parties have limited resources and must choose tool to concentrate their efforts on. Critically, we assume
that the poor will vote if their utility threshold is satisfied by citizens shapes turnout, especially turnout among low-
the offerings made by the party of the rich. Therefore, income voters. Recall from the setup above that we proxy
solving the model requires taking two steps, sequentially. inequality from two angles: the proportion of low-income
citizens in society (1 2 d) and the share of income owned by
Step 1. Identify the optimal values for the poor of taxes (t*), high-income citizens (f). The model yields the following
targeted goods for low-income voters (b＊P ), and general public comparative statics between these two aspects of inequality
goods ( g*) given the budget constraint. These values define and the choice of targeted (b＊P ) goods:
the indifference threshold for the poor to turnout to vote. The
problem for low-income voters is defined as follows: ∂ ln(b＊p ) l
2tmax w
p ≤ 0: ð4Þ
∂(f) a
maximize U i (t; bp ; g) p (1 2 t)wP 1 a ln(bP ) 1 g
t;bp ;g ð1Þ
subject to (1 2 lf) p bP 1 bR 1 g;
tw ∂ ln(b＊p ) tmax wP l
p ≥ 0: ð5Þ
where a capture the sensitivity of low-income voters to ∂(1 2 d) a
targeted goods. As detailed in the appendix (apps. A–I are
available online), this yields the following results: b＊P p a, Inequality shapes the optimal choice of b＊P by the elite
b＊R p 0, t ＊ p t max ≤ 1, and g ＊ p t w
(1 2 lf) 2 a. These in through the interaction of two mechanisms: one economic,
turn allow definition of the poor voter’s utility threshold for one political. The economic mechanism concerns both the
voting. Poor voters will vote under any combination of t, bP, size of the pool of targetable voters (1 2 d) and the economic
and g that generates levels of utility at least similar to those ability of the elite to finance such efforts (f). On the one
defined by hand, equation (4) implies that as the elites become wealthier
they need fewer resources to meet the mobilization con-
U P p (1 2 t max )wP 1 a ln(a) 1 t w
(1 2 lf) 2 a: ð2Þ straint of the poor since low-income voters are more easily
This expression defines the level of reservation utility of the mobilized.4 On the other, equation (5) means that the opti-
poor that the elites must meet so that the latter turn out to mal level of targeted goods toward the poor (b＊P ) increases
vote. Importantly for our subsequent analysis, note that this with the number of poorer voters, especially in low-capacity
implies that the reservation utility of the poor declines as contexts (high l).
capacity decreases (l increases): ∂ UP =∂l p 2t w
f. The political mechanism concerns the incentives of the
elites to meet the low-income citizens’ reservation utility
Step 2. Identify the optimal values for the elite of taxes (t*), constraint. Equation (5) suggests that as the absolute income
combined targeted goods (b＊P , b＊R ), and general public goods of the poor increases (wP), the level of targeted goods neces-
( g*). The elites, irrespective of their ideological leanings, need sary to meet the poor’s reservation utility also rises. In addi-
to choose a portfolio of targeted goods, public goods, and taxes tion, the incentives to meet such a constraint, and therefore
that meets two constraints: (1) a budget constraint (recall that get the poor to vote, are also affected by the ability of the state
the elite has full capacity to tax itself ) and, crucially, (2) a to monitor and tax its citizens and, accordingly, the ability of
political constraint driven by the need to meet the mobilization the elite to hide away part of their wealth or engage in mis-
threshold of low-income voters defined previously in (2). management for political purposes (l).
Accordingly, its maximization problem can be defined as Remember that a lower l implies greater capacity and
therefore an increase in the reservation utility of the poor to
maximize U i (t; b; g) p (1 2 t)wR 1 b ln(bR ) 1 g actually turnout to vote.5 Substantively, this implies that as
t;b ;g
r
subject to p bP 1 bR 1 g and to (1 2 t)wP ð3Þ

tw the capacity of the state increases it becomes more expensive
1 a ln(bP ) 1 g ≥ U P ; to acquire the support of the low-income voters. Accord-
ingly, the extent to which elites offer targeted goods in re-
where b captures the sensitivity of high-income voters to sponse to increases in their income or increases in the share
targeted goods for the elite and UP captures the low-income of poor voters is moderated by the level of the state capacity.
voters’ utility threshold as defined above. In the context of equation (5), this implies that, as the state's
Comparative statics and hypotheses

4. This effect is stronger, the higher the average income of society (
w),
Solving the model (see the appendix for details) allows us to
reflecting the well-known intuition that development undermines clien-
explore how the relationship between economic inequality telism (Scott 1969; Stokes et al. 2013).
and the elite’s choice of local targeted goods for low-income 5. Recall that this is the case as, working from (2), ∂ UP =∂l p 2t w
f.
monitoring capacity declines, elites are better off using more In addition, a corollary follows from hypothesis 2: external
targeted goods to meet the low-income voters’ reservation interventions that limit the ability of elites to manipulate
utility constraint. By contrast, as l tends to 0 and capacity spending to their electoral advantage should translate to a re-
increases, the connection between economic inequality duction in the probability of reelection of the incumbent, as her
(whether captured through the number of poor or the share ability to secure the support of a large pool of voters declines.
of income by the rich) and the optimal level of targeted goods Overall, hypothesis 1 concerns equilibrium levels of turnout
provided by the elite weakens. Until the point at which, when and targeted spending under the status quo. Hypothesis 2
l p 0, it actually disappears. This result uncovers a channel speaks to the implications of exogenous manipulation in the
through which interventions to increase capacity, thus re- levels of capacity.
ducing l, crucially affect elites’ mobilization strategies.
The analysis above suggests that elites do not only react by EMPIRICAL STRATEGY: DESIGN AND MEASUREMENT
mobilizing against the increasing revenue-raising power of Research design
the state (as in Kasara and Suryanarayan 2014). Rather, under To evaluate them, we join a recent stream of scholarship
conditions of high inequality and low state capacity, they exploiting Brazilian municipalities to identify mechanisms
strategically mobilize low-income voters to secure their po- driving the interaction between voters and politicians in
litical position. These are the circumstances that render contexts with a strong incidence of corruption, clientelism,
targeted spending toward the poor (including clientelism) and inequality (Brollo 2012; Brollo et al. 2013; Hidalgo and
both rational and self-enforcing.6 These results yield two sets Nichter 2016). Our specific strategy focuses primarily on the
of empirical expectations that will structure the empirical random audits by the Brazilian government on its munici-
analysis in the rest of the article. The first one concerns palities (Ferraz and Finan 2008, 2011).
equilibrium patters in the relationship between inequality, Beyond the obvious advantage of holding constant po-
capacity, strategic spending, and turnout; the second one tential cross-national sources of heterogeneity, three reasons
concerns the political implications of capacity shocks through render Brazil a suitable case for our purposes. First, in 1997
monitoring interventions. Brazilian authorities successfully promoted a constitutional
change to allow reelection of incumbents at the local level, a
Status Quo provision implemented from 2000 onward (Ferraz and
Finan 2008, 2011). Second, despite being a democracy in
H1a. Under high inequality and low capacity, elites
which voting is legally compulsory for individuals between
should resort to a higher use of local targeted goods as
18 and 70 in all elections, there remains considerable vari-
a mobilization strategy of low-income voters.
ation in the levels of turnout across localities. For the lo-
H1b. Under high inequality and low capacity, we calities in our sample, the range was between 65% and 96% in
should observe higher rates of political participation 2000 and 2004. In both instances the distribution was ap-
especially in areas with more low-income voters. proximately normal (see app. C). This reduced variation due
to institutional constraints makes Brazil a harder case to test
Capacity Shocks via Monitoring Interventions. Given a our hypotheses.
status quo of high economic inequality and low ca- Third, the launch of a major anticorruption initiative in
pacity, institutional reforms that increase the state’s 2003, led by the Controladoria General da União (CGU) to
monitoring ability on incumbents (i.e., reduce l in the scrutinize the use of federal funds by local authorities, offers
model) imply: an opportunity to identify the impact of increases in the state’s
monitoring capacity. The audits are themselves a partisan
H2a. A reduction in the provision of local targeted
effort by Lula’s Partido dos Trabalhadores (Workers’ Party) to
goods toward lower income voters.
undermine the dependency of many of its potential voters on
H2b. A reduction in the level of turnout, as incum- established clientelistic networks: they were part of a multi-
bents are less capable of meeting the low-income dimensional policy strategy that included Bolsa Familia and
voters’ reservation utility to participate. major infrastructural investments such as the program Luz
para Todos. The goal was to reduce voters’ dependencies on
the demand side (Bobonis et al. 2017) and, through the audits,
limit the ability of local incumbents to manipulate the budget
6. For evidence consistent with this theoretical contention in the on the supply side. Interestingly, when it became apparent
Brazilian case, see Timmons and Garfias (2015). that auditing all municipalities was unfeasible, the CGU
turned to randomization to select its targets, thus creating a Linking model and empirical analyses:
natural experiment. The audits are de facto an exogenous Measurement
capacity shock on the ability by local incumbents to steer the Before discussing our estimation and specification choices, it
budget to their electoral advantage.7 is important to be clear about the link between our model
The implementation of the audits works as follows. parameters and the available measures in the sample of
Through a sequence of lotteries, the CGU chose randomly Brazilian municipalities. We proxy the targeted goods toward
about 8% of a total of 5,500 Brazilian municipalities, in- low-income voters (bP) by local primary education spending.
cluding state capitals and coastal cities. Once a municipality Its incidence on lower income groups relative to other forms
is chosen, the CGU gathers information on all federal funds of spending is well understood (Ansell 2010) and, in the
received and sends a team of auditors to examine them. context of Brazil, is a policy in which incumbents have ample
Auditors get information from the community and the local margin to engage in inefficient forms of redistribution via
council members about any form of malfeasance or misuse part-time employment (Bursztyn 2016; Colonnelli, Teso, and
of funds, as well as from the local documentation available. Prem 2017; Robinson and Verdier 2013). Importantly, for
Immediately, after the inspection, a detailed report is sent robustness, we also use the log of public employees at the
back to the CGU, which in turn forwards it to the federal municipality as an additional indicator of targeted spending
accounting auditor (Tribunal de Contas da União), the toward low-income voters.
judiciary, and all members of the local council. A summary The number of low-income voters (1 2 d) and the wealth
with the key findings for each audited municipality is made share of high-income voters (f) capture, respectively, the
publicly available. demand and the supply sides of political exchanges between
Critically, we have information (thanks to Ferraz and elites and low-income citizens. Conveniently, they are, by
Finan 2008, 2011) on the date when the reports were released construction, two key dimensions of standard inequality
to parties and voters. As a result we can exploit the contrast measures. The shape of the distribution of income depends
between those municipalities where the audit results were on the relative share of low- versus high-income citizens and
released before the 2004 election and those where they were the relative share of income that accrues to each of them.
not.8 The combination of the random selection of munici- This match between our theoretical model and the con-
palities and the discontinuity around the 2004 election struction of the measure is what leads to using the Gini co-
determines the nature and composition of the treatment and efficient at the level of the municipality as our key inde-
control groups. Since all the municipalities included in our pendent variable of interest.9
sample have been investigated, the treatment is purely in- Note that l is defined in the model as the inverse of the
formational. The treatment group includes municipalities state capacity to monitor its elites. A high l implies an ability
that have been audited where the results of the investigation of elites to hide their wealth, which gives them a lot of political
have been released before the 2004 election, and the control discretion. Our empirical measure, taken directly from the
group includes all the municipalities where the investigation audit reports, follows the same logic. We employ a measure
took place and its results were released after the 2004 elec- of audited local mismanagement, which is defined as “the
tion. For most of the covariates, there are not significant number of violations divided by the number of service items
differences across the two groups, except for some munici- audited” (Ferraz and Finan 2011, 1284). These violations in-
pality size variables (e.g., population and number of legis- clude the performance of uncompetitive bidding for local
lators per voter; see table B.2). contracts and various ways of turning public goods into tar-
geted goods, most prominently the misuse of resources ear-
marked for other purposes (i.e., using resources intended for
7. This is important because endogenous relationships abound in the health to boost teachers salaries or, as typically recorded in
literature. Fergusson, Larreguy, and Riano (2015) show how parties with a individual municipal reports, to hire a larger pool of part-time
strategic advantage in clientelistic politics will oppose investments in state
teachers). The higher the reported audited mismanagement,
capacity, thus limiting pro-equality politics. Debs and Helmke (2010)
show that the left fares better under equality because voters are more likely to
the stronger the ability of incumbents to manipulate the use of
cling to pro-redistributive coalitions that in turn help contain inequality. federal funds for political gain. By implication, the higher the
Bursztyn (2016) focuses in turn on voter’s demand: it is the voters themselves monitoring capacity of the state, the lower the chances that
who may not want more public goods.
8. Given the short time span between selection, visit, and release,
randomization determines both which particular municipality is selected 9. For completion purposes, we show in app. I that our main em-
and when the information is released. There is no room for strategic ma- pirical results regarding the capacity shock are robust to substituting the
nipulation of the information by parties or the federal government. Gini measure for a poverty measure.
local incumbents can engage in such practices.10 Accordingly, Table 1. Spending under the Status Quo
audit exposure is a measure of a genuinely exogenous capacity
shock on l. Public Primary Education
Employees Spending
We match the key measures of our parameters of interest
to census-based sociodemographic and economic informa-
(1) (2) (3) (4)
tion at the local level, as well as to detailed political infor-
mation obtained from the Tribunal Superior Electoral, in-
Inequality 2.84 2.83 2.53* 2.49
cluding the level of turnout in local elections and data on
(.60) (.68) (.30) (.30)
budgetary choices by local governments. The latter capture
Mismanagement 2.31** 2.31* 2.19* 2.18*
how local governments use policies as part of their electoral (.15) (.15) (.09) (.10)
strategy (Bursztyn 2016; Calvo and Murillo 2004). These Inequality #
three features allow us to test whether in equilibrium mu- mismanagement .55** .55* .33** .30*
nicipalities with a higher incidence of inequality and low (.27) (.27) (.15) (.16)
capacity are associated with larger budgetary commitments Constant 2.25 2.23 21.22** 21.30**
toward policies that can be targeted toward low-income (.66) (.77) (.48) (.54)
Municipality controls Yes Yes Yes Yes
constituents.
Incumbent
characteristics
EMPIRICAL ANALYSIS controls Yes Yes Yes Yes
The status quo: Turnout rates and targeted goods Regional state FEs Yes Yes Yes Yes
Targeted mobilization under the status quo. We begin by Party FEs No Yes No Yes
analyzing the conditions under which local political elites are R2 .88 .89 .97 .97
more likely to resort to targeted mobilization. According to N 366 366 306 306
our theoretical expectations (hypothesis 1a), targeted mobili-
Note. Ordinary least squares. Targeted goods spending in 2004, all
zation should be higher when inequality is high and state ca-
incumbents. Standard errors (in parentheses) are clustered at the state
pacity is low. Recall that we assume that the higher the levels of level. FEs p fixed effects.
local mismanagement (as captured by Ferraz and Finan 2011), * p ! .1.
the lower the capacity at status quo. As outlined above, to ** p ! .05.
capture incumbents’ budgetary effort during the 2000–2004 *** p ! .01.
legislature toward low-income voters, we use two measures:

first, the (log) local spending in primary education. Specifi-
cally, we employ the data collected recently by Bursztyn (2016) and other electoral controls.11 Columns 2 and 4 also include
on the amount of local public spending on primary education party FEs, since spending priorities might be responsive to
at the municipality level. And second, we analyze the log of parties’ ideological concerns. The sample includes all the
total municipal public employees as our proxy for the provision available observations, including mayors in their first and
of targeted goods at the local level. To explore the determinants second term, with information regarding the level of audited
of municipal spending in targeted goods toward the poor under mismanagement at the local level. All standard errors are
the status quo, we estimate the following specification: clustered at the regional state level.12
LogSpending p a 1 b1 Ineqm;s 1 b2 Mismm;s

1 b3 Ineqm;s # Mismm;s 1 hX m;s 1 vs 1 εm;s :
11. Municipality controls include the following local characteristics in
ð6Þ table 1 cols. 1 and 2: population, gross domestic product (GDP), education
levels, fraction of urban population, size, and judicial town. Models in
Regarding potential confounders, we follow a similar cols. 3 and 4, instead, control for average municipality budget and median
specification to the one in Bursztyn (2016) and Ferraz and household income as municipality controls. Incumbents’ controls are age,
gender, level of education, and past experience as a mayor or council
Finan (2008). In addition to state-level fixed effects (FEs; vs), all
member. Electoral controls are also included in cols. 1 and 2, and they are
the models in table 1 also include controls at the local level for the share of council members from the same party as the mayor, whether
basic municipality characteristics, incumbents’ characteristics, the mayor was from the same party as the governor, the effective number
of parties in the 2000 election, and the margin of victory and the change in
the electoral census.
10. Note that the measure does not include clerical or accounting 12. The clustering at the state level is designed to account for the
errors or fiscal adjustments in year-to-year budgets. geographical distribution of the units of observation.
Figure 1. Marginal effects of audited mismanagement on spending decisions: A, local primary education spending; B, local public employees
Consistent with our theoretical expectations, all models in A potential concern about these analyses, however, is that
table 1 report a positive and significant interaction between the use of local mismanagement as a proxy for capacity (or to
economic inequality and state capacity, which, again, is be more precise, its inverse) conflates the capacity of the state
proxied by the level of mismanagement recorded during the to monitor elites’ misbehavior and other potential motives to
audits. Regardless of the specific indicator of targeted spending engage in mismanagement. To the extent that mismanage-
used, our findings provide strong support to hypothesis 1a: ment is a good proxy for capacity, it should be the case that
both local primary education spending and the log of mu- local elites exploit the capability to manipulate spending
nicipal employees reach the highest level when inequality is when they need it most. In table 2 we use the margin of
high and capacity is low. Figure 1 illustrates the marginal ef- victory in the 2000 elections as a moderator in the rela-
fects (based on the models in cols. 1 and 3 in table 1) of ca- tionship between inequality, capacity, and spending. As the
pacity, inversely proxied by audited mismanagement, on ei- margin tends to zero, local incumbents’ incentives to exploit
ther type of spending. Unequivocally, low capacity enhances the lack of capacity and further increase targeted mobiliza-
targeted mobilization precisely at high levels of inequality.13 tion grow stronger. We take this evidence to validate our
Subsequent explorations of the interaction between mis- strategy to use the past levels of local mismanagement as a
management and economic inequality in two subsamples of proxy for local capacity under the status quo.15 Crucially, we
urban and rural municipalities confirms that this effect is can also show that this electoral exploitation of the lack of
dominant in rural areas and completely absent in urban capacity occurs only among mayors that are in their first
areas.14 This additional evidence further reinforces the no- term (i.e., when they are facing reelection incentives) but not
tion that strategic spending in primary education and public among second-term mayors (see also app. E.5).
employment functions, in part, as an instrument to mobilize
low-income voters. Turnout under status quo. Having established that targeted
spending is more pervasive under high inequality and low
13. See app. D for a robustness check of the linear interactions in

which we provide a graphical illustration of both the bin and kernel 15. Note that all the models in table 2 include the lag measure of
estimates (Hainmueller, Mummolo, and Xu 2019). public employees in 2002. Therefore, the models that employ the log of
14. See, for more details, the additional tables in the app. E exploring public employees in 2004 as an outcome variable can be interpreted as
the heterogeneous effects. being dynamic models estimating the increase in public employment.
Table 2. Spending under the Status Quo: Capacity and Incentives
Public Employees Primary Education Spending
(1) (2) (3) (4)
Inequality 21.95* 21.92* 21.25 21.48*

(.98) (.95) (.85) (.73)
Mismanagement 2.55** 2.51** 2.40** 2.44***
(.23) (.24) (.16) (.14)
Inequality # mismanagement .97** .89** .65** .71***
(.42) (.43) (.28) (.23)
Win margin 27.72*** 27.89*** 24.36* 25.02*
(2.20) (2.23) (2.49) (2.62)
Win margin # inequality 14.10*** 14.30*** 7.40* 8.58*
(4.18) (4.26) (4.28) (4.52)
Win margin # mismanagement 2.93** 2.74** 2.22** 2.36**
(1.12) (1.26) (1.03) (1.12)
Win margin # inequality # mismanagement 25.29** 24.94** 23.69** 23.95*
(2.03) (2.29) (1.79) (1.95)
Constant 23.27* 24.05** 21.98 22.33
(1.74) (1.81) (1.28) (1.40)
Lag public employees Yes Yes Yes Yes
Incumbent characteristics controls Yes Yes Yes Yes
Electoral controls No Yes No Yes
Regional state FEs Yes Yes Yes Yes
Party FEs Yes Yes Yes Yes
R2 .96 .96 .98 .98
N 200 200 169 169
Note. Ordinary least squares. Targeted goods spending in 2004, all incumbents. Standard errors (in parentheses) are clustered at the state
level. FEs p fixed effects.
* p ! .1.
** p ! .05.
*** p ! .01.
capacity, we now turn to hypothesis 1b to explore whether introduce a similar set of major specific controls and two
political participation is higher under such conditions. In additional sets at the municipality and party levels.16 All
table 3 we explore how economic inequality and capacity models include a dummy that takes the value 1 if the in-
jointly affect turnout rates at the local level. Specifically, we cumbent faces reelection incentives and 0 otherwise. Finally,
estimate the following equation: the model in column 4 includes party FEs to account for
potential behavioral heterogeneity among political parties.
TurnoutRates04 p a 1 b1 Ineqm;s 1 b2 Mismm;s
1 b3 Ineqm;s #Mismm;s 1hX m;s 1vs 1 εm;s :
ð7Þ
16. Municipal-level characteristics now include the area, the log of the
population, the share of the urban population within the municipality, the
As before, all models in table 3 include regional state FEs (log) local GDP per capita, the change in the level of the population be-
(vs) and standard errors clustered at the state level. Impor- tween censuses, the share of the population over 18 with at least a sec-
tantly, now the models in columns 3 and 4 include controls ondary education, whether the municipality is new, and the number of
active public employees. We also add controls for specific political and
for the levels of fiscal transfers, whereas the other two do not
judicial institutions at the municipality level: presence of a judicial district,
include them since differences in transfers received by the use of participatory budgeting during 2001–4, and the seats-to-voters ratio
municipalities might be an important confounder. We also within each municipality.
Table 3. Turnout Rates under the Status Quo higher. To do so, we simply divide the sample according to
municipalities that are above or below the median share of
(1) (2) (3) (4) urban population. Additional results show that, as expected,
the positive and significant effect of mismanagement on turn-
Inequality 2.26*** 2.26*** 2.26*** 2.27*** out is mostly prevalent in rural areas with high inequality.19
(.07) (.07) (.06) (.06)
Mismanagement 2.05*** 2.06*** 2.06*** 2.05**
(.02) (.02) (.02) (.02)
Capacity shock
Inequality # Audit exposure and changes in local targeted goods. We
mismanagement .09*** .10*** .10*** .10** now turn to study exogenous changes in the monitoring ability
(.03) (.03) (.03) (.04) of the state and their consequences. We begin evaluating the
Constant 1.15*** 1.17*** 1.09*** 1.09*** consequences of audits released before 2004 on the provision
(.08) (.08) (.08) (.08) of local spending in primary education (hypothesis 2a). As
Municipality argued before, we expect an exogenous shock that increases
controls Yes Yes Yes Yes
the monitoring ability at the municipality level to be associated
Electoral controls No Yes Yes Yes
with a decline in the provision of targeted goods toward low-
Fiscal controls No No Yes Yes
Regional state FEs Yes Yes Yes Yes income voters. Given this, we model the determinants of the
Party FEs No No No Yes change in local primary education spending as a function of
R2 .53 .54 .54 .55 the interaction between economic inequality, capacity, and a
dummy capturing whether the municipality belongs to the
Note. Ordinary least squares. Turnout rates in 2004, all incumbents. treatment or the control group (exposure before vs. after 2004).
Standard errors (in parentheses) are clustered at the state level. FEs p
Specifically, we study the determinants of the change in
fixed effects. N p 366.
* p ! .1.
the average levels of local spending in primary education by
** p ! .05. comparing the (log) averages between the 2000–2004 and
*** p ! .01. the 2004–8 legislature terms:20
The results in table 3 lend support to our theoretical DSpendingEducp LogSpending08204Leg 2 LogSpending04200Leg
expectations. All models report a positive and significant p a 1 b1 Ineqm;s 1 b2 Mismm;s 1 b3 Ineqm;s
coefficient for the interaction term between the Gini measure
and the audited mismanagement at the municipality level.17 # Mismm;s 1 b4 Expm;s 1 b5 Expm;s
The estimated coefficients remains highly stable with the
gradual introduction of the controls, even with the inclusion # Ineqm;s 1 b6 Expm;s # Mismm;s
of party FEs. Figure 2 shows the marginal effect (based on
1 b7 Expm;s # Ineqm;s # Mismm;s
col. 3 in table 3) of the audited mismanagement on turnout
rates conditional on the levels of economic inequality at the 1 hX m;s 1 vs 1 lt 1 εm;s :
municipality.18 The effects are substantively important.
ð8Þ
Under high inequality, the marginal effect of audited mis-
management is associated with between 1 and 2 more per-
centage points in turnout rates, whereas under very low in- All the models in table 4 control for the log change in the
equality the marginal effect of audited mismanagement total municipality budget as well as the median household
amounts to a 1.5 percentage point reduction in participation. income level during the 2000–2004 legislature. Also im-
Unfortunately we do not have individual-level data available, portant, all models except the one in column 1 include re-
but we can explore the heterogeneous effects depending on gional FEs (vs) to account for unobserved heterogeneity. The
the share of urban population as a measure of urban versus columns reported gradually incorporate controls for the
rural areas, where the prevalence of low-income voters is incumbent and party characteristics during the 2004–8 leg-
islature: mayor in previous reelection, mayor characteristics,
17. Interestingly, the levels of audited mismanagement are also asso-
ciated with higher levels of turnout rates when inequality is high and 19. See additional results in app. E exploring the heterogeneous effects.
capacity is low in the preceding local elections in the year 2000. 20. The data come from Bursztyn (2016) and refer to the averaged
18. In app. D we provide a graphical illustration of the interaction and deflated spending levels across 2001–4 (for the first legislature) and
with bin and kernel estimates. 2005–8 (for the subsequent legislature).
capacity, and provision of targeted goods is expected to dis-

appear. Thus, our findings suggest that efforts to curb down
clientelism may have constraining effects on outcomes such
as spending on primary education, thus contributing to an in-
cipient literature on the potential detrimental effects of anti-
corruption programs in Brazil (Lichand, Lopes, and Medeiros
2016). We resume this discussion in the conclusion.
Audit exposure and changes in political participation.

We turn now to study the impact of randomized audits on
turnout (hypothesis 2b). Again, we model the determinants
of the change in the levels of turnout between 2000 and 2004
as a function of the interaction between economic inequality,
Figure 2. Marginal effects of audited mismanagement on turnout
capacity, and a dummy capturing whether the municipality
belongs to the treatment or the control group (exposure be-
fore vs. after 2004).23 And as before, to keep the comparison as
and party FEs.21 Column 6 in table 4 includes lottery FEs (lt).
sharp as possible, we limit the sample to mayors who seek
And as usual, all models report clustered standard errors at
reelection for the first time.24 By restricting the sample this
the state level. The novelty is that now we only include mayors
way, we avoid the confounding effect of the term in office.
who were in their first term when localities were audited. We
Thus, we estimate the following equation for the sample of
do so to sharpen the comparison and to make sure that sub-
audited municipalities with mayors in their first-term man-
sequent changes in spending priorities are not driven by ex
date during the 2000–2004 legislature:
ante differences in reelection incentives.22
To directly illustrate the results, figure 3 plots the marginal DTurnout p LogTurnout2004 2 LogTurnout2000
effects of audited mismanagement on the log change in pri-
p a 1 b1 Ineqm;s 1 b2 Mismm;s 1 b3 Ineqm;s
mary education spending for nonexposed (control) and ex-
posed (treated) municipalities. As expected, for the control #Mismm;s 1 b4 Expm;s 1 b5 Expm;s # Ineqm;s ð9Þ
group the status quo persists, and mismanagement continues
to exert a positive effect on the dynamics of spending in pri- 1 b6 Expm;s # Mismm;s 1 b7 Expm;s # Ineqm;s
mary education when inequality is high. Interestingly, and in
full alignment with our theoretical predictions, this relation- # Mismm;s 1 hX m;s 1 vs 1 lt 1 εm;s :
ship completely vanishes for the treated municipalities (i.e., ð9Þ
municipalities exposed to the shock of an audit’s release). In All models now include lottery FEs (lt) to account for
the latter, the levels of audited mismanagement bear no effect different timing in the audit release. Table 5 columns 1 and
on the dynamics of primary education spending at the local 2 do not include regional FEs (vs), whereas all the other col-
level. umns include them. Since we know that clientelism is geo-
The fact that the relationship between audited misman- graphically concentrated among certain areas, the inclusion of
agement and the midterm changes in local spending in pri- regional FEs is important. Finally, columns 5 and 6 exclude
mary education vanishes for exposed municipalities is highly those municipalities in which the mayor was member of the
consistent with the model results. According to our theoret- Partido do Movemento Democrático Brasileiro (PMDB). In
ical expectations, as the monitoring capacity of the state in- contexts in which clientelistic parties are hegemonic, incum-
creases (i.e., lower values of l) elites are better off relying less bents have the potential to activate compensating mechanisms
on local targeted goods to meet the low-income voter’s res- that mute the political consequences of the federal audits. Since
ervation utility. And as l tends to 0, arguably what happens
under exposure, the relationship between economic inequality,
23. Note, however, that now we focus on the short-term change
(between 2000 and 2004).
21. The controls for the mayor characteristics include gender, age, 24. Recall that we have motivated the exclusion of the second-term
age2, marriage, and education. mayors on the basis of reported differences in audited mismanagement.
22. Importantly, in app. H.1 we provide evidence that mayors in their However, the results do not depend on excluding the second-term mayors.
first mandate tend to resort to less mismanagement compared to second- In app. F.5, we show that the results hold when including both first- and
term mayors, especially under high levels of inequality. second-term mayors (i.e., all the incumbents).
Table 4. Audit Exposure and Changes in Local Targeted Goods
(1) (2) (3) (4) (5) (6)
Inequality 2.31 21.12*** 21.14*** 21.14** 21.74*** 22.14***

(.51) (.39) (.37) (.46) (.43) (.42)
Mismanagement 2.33** 2.42*** 2.46*** 2.44*** 2.51** 2.52**
(.15) (.15) (.14) (.14) (.19) (.20)
Inequality # mismanagement .54** .69*** .76*** .74*** .87** .90**
(.24) (.23) (.22) (.24) (.32) (.34)
Exposed 2.33 2.64** 2.64** 2.57 2.96** 21.00**
(.30) (.25) (.25) (.33) (.39) (.41)
Exposed # inequality .56 1.07** 1.07** .96 1.63** 1.87***
(.53) (.44) (.43) (.57) (.67) (.66)
Exposed # mismanagement .34* .45** .48*** .45** .56** .55**
(.17) (.18) (.17) (.18) (.23) (.23)
Exposed # inequality # mismanagement 2.56* 2.73** 2.78*** 2.75** 2.94** 2.93**
(.28) (.28) (.27) (.30) (.39) (.39)
Constant .19 .63** .64*** .74 1.07* 1.15**
(.31) (.23) (.21) (.50) (.58) (.50)
Municipality budget controls Yes Yes Yes Yes Yes Yes
Reelected incumbent control No No Yes Yes Yes Yes
Incumbent’s controls No No No Yes Yes Yes
Party FEs No No No No Yes Yes
Lottery FEs No No No No No Yes
Regional state FEs No Yes Yes Yes Yes Yes
R2 .66 .75 .75 .77 .80 .82
Note. Ordinary least squares. Primary education spending change, first-term incumbents. Standard errors (in parentheses) are clustered at the state level.
FEs p fixed effects. N p 145.
* p ! .1.
** p ! .05.
*** p ! .01.
the PMDB is widely recognized as one of the parties with Audit information releases have a negative effect on turnout,
powerful clientelistic machines, we want to assess how sensi- especially when the levels of both audited mismanagement
tive our findings are to the inclusion/exclusion of municipal- and economic inequality are high. Interestingly, though, the
ities under its control. results are especially strong once the incumbents who are
We have also checked whether the correlation between the members of the PMDB are excluded in columns 5 and 6 of
levels of audited mismanagement and inequality is the same table 5. This may reflect the fact that mayors in areas where
across exposed (treatment) and nonexposed (control) mu- clientelism is hegemonic have a wider array of exonerative
nicipalities. Interestingly, the correlation between audited mis- strategies against the impact of the audits at their disposal.
management and inequality is negative and significant in the In appendix F we provide two additional robustness checks
treatment group but insignificant in the control group. This that are important regarding the effects of the capacity shock
is because under high inequality the levels of audited mis- on changes in turnout: (i) we show that the demobilization
management are lower in exposed municipalities. If any- effects were especially significant in municipalities where the
thing, this negative correlation runs contrary to the possi- incumbent was not reelected, and (ii) we also illustrate that,
bility of finding significant results in our main specification. if we substitute the mismanagement variable for the log of
To test the robustness of the results, we have rerun the models public employees in 2002 at the municipality level as an alter-
excluding extreme upper values for the mismanagement var- native proxy for local capacity, the results are exactly the same.
iable, and the results do not change (see app. F.1). The first check is reassuring in terms of providing evidence
The findings reported in table 5 are robust to different that the electoral demobilization went hand in hand with the
specifications and consistent with our theoretical expectations. loss of credibility of the incumbent. The second check is key
Figure 3. Marginal effect of audited mismanagement on education spending change: A, nonexposed; B, exposed
since employing an alternative measure of capacity—the pothesis 2b). While our theoretical model does not directly
stock of public employment before the audits—yields similar theorize electoral survival, a direct implication from these two
results.25 effects is the reduction in the incumbent’s electoral survival.
To facilitate the interpretation of the results, figure 4 If the incumbent’s ability to meet the low-income voters’ po-
compares the marginal effect of audited mismanagement on litical constraints declines, because of an increase in the mon-
changes in municipal levels of turnout in the control and itoring capacity of the state, ceteris paribus so should her
treatment groups at various levels of inequality. Given high probability of survival.
levels of inequality, in those municipalities where the ex- To evaluate this empirical corollary of our argument, we
ternal audits were not released, the more incumbents misuse estimate the reelection probability in the 2004 local elections
federal funds, the higher the increase in turnout. By contrast, in of incumbents in their first term as a function of the inter-
those municipalities where the audit took place and was re- action between the audited levels of mismanagement and
leased before the 2004 election, the same strategy triggers a the shock associated to the exposure of the audit results be-
reduction in electoral participation of a similar magnitude. In fore the 2004 elections.26 We anticipate that when there is no
other words, the status quo persisted in the control group but exposure, greater mismanagement should be associated with
was radically transformed in the treatment group. Consistent a higher reelection probability. In contrast, when the audit
with Hidalgo and Nichter (2016), the findings here contribute reports were released before the 2004 elections, the reelection
to an emerging agenda on the impact of audits on turnout. Our probability should decline sharply. To explore this prediction,
results, though, emphasize that the negative effects of external we estimate the following equation:
audits on turnout changes were highly conditional on the pre- Reelected p a 1 b1 Expm;s 1 b2 Mismm;s 1 b3 Expm;s
existing levels of economic inequality and capacity. ð10Þ
# Mismm;s 1 hX m;s 1 vs 1 lt 1 εm;s :
Corollary: Audit exposure and reelection probability.
Our analysis of the capacity shock implies a stop in the use We run the models in table 6 with a logit specification
of targeted mobilization toward low-income voters (hypoth- since the dependent variable is dichotomous (1 if reelected
esis 2a) and an attendant reduction in turnout rates (hy- and 0 otherwise), but the results are the same if instead we
implement a linear probability model. All models reported in
25. Recall that before we showed that the increase in the number of
employees might be itself understood as a targeted good, whereas here we 26. Note that again we limit the sample to municipalities with mayors
employ the stock of public employment as a proxy for capacity. in their first term.
Table 5. Audit Exposure and Changes in Turnout Rates
(1) (2) (3) (4) (5) (6)
Inequality 2.19 2.24 2.13 2.19 2.33* 2.47**

(.14) (.16) (.11) (.15) (.17) (.17)
Mismanagement 2.05 2.06 2.03 2.04 2.06 2.09*
(.03) (.04) (.03) (.04) (.05) (.05)
Inequality # mismanagement .09 .11 .04 .07 .11 .15*
(.06) (.06) (.05) (.07) (.08) (.08)
Exposed 2.17* 2.20* 2.15 2.18 2.28* 2.35**
(.10) (.11) (.10) (.11) (.14) (.14)
Exposed # inequality .30 .35* .27 .33* .54** .67**
(.18) (.19) (.17) (.19) (.25) (.24)
Exposed # mismanagement .10** .11** .08* .10* .14* .17**
(.05) (.05) (.05) (.05) (.07) (.07)
Exposed # inequality # mismanagement 2.19** 2.21** 2.16* 2.19* 2.26** 2.31**
(.08) (.09) (.08) (.09) (.12) (.12)
Constant .33** .35** .08 .09 .21 .27
(.15) (.15) (.14) (.14) (.20) (.21)
Municipality controls Yes Yes Yes Yes Yes Yes
Electoral and fiscal controls Yes Yes Yes Yes Yes Yes
Incumbent characteristics controls No Yes No Yes No Yes
Lottery FEs Yes Yes Yes Yes Yes Yes
Regional state FEs No No Yes Yes Yes Yes
PMDB incumbents excluded No No No No Yes Yes
R2 .54 .55 .68 .70 .62 .66
N 203 203 203 203 165 165
Note. Ordinary least squares. Turnout change, first-term incumbents. Standard errors (in parentheses) are clustered at the state level. FEs p fixed
effects; PMDB p Partido do Movemento Democrático Brasileiro.
* p ! .1.
** p ! .05.
*** p ! .01.
table 6 include lottery FEs (lt), since the proximity to the 2004 Figure 5 provides a graphical representation of the scale
elections might be associated with unobserved heterogeneity of the effects. The reelection probability declined dramati-
in the incumbents’ ability to circumvent the audit results. As in cally in those municipalities in which the release of the audit
the previous analyses, the reported standard errors are clus- results occurred before the 2004 elections. For this particular
tered at the state level, and all models include regional FEs. The subgroup, the predicted drop in the probability of reelection
model in column 1 does not include the fiscal transfers con- is 50% or more, depending on the existing levels of misman-
trols, whereas the models in columns 2–4 do. The models in agement. The contrast with the incumbency advantage pro-
columns 3 and 4 incorporate further controls and account for vided by the continuation of preexisting practices in the
the incumbent characteristics. Finally, the model in column 4 control group is striking. This result is in line with previous
includes party FEs. contributions on the negative effect of the audits on the prob-
Table 6 provides strong evidence consistent with our ex- ability of reelection. At the same time, it offers a novel per-
pectations. Low capacity (high mismanagement) enhances spective since it shows that the negative effect was especially
the likelihood of reelection in nonexposed municipalities. By severe at very low levels of capacity—specifically, under high
contrast, in localities where the audit results were released
FEs models, following Bursztyn (2016), with the entire sample of Brazilian
before the elections, higher levels of audited mismanagement
municipalities that show that party reelection probability was indeed higher in
actually lead to a much lower reelection probability.27 municipalities with both greater spending in primary education and higher
inequality. Specifically, in app. H.3 we show that spending in primary education
27. To provide further evidence of the relationship between targeted exerts a positive effect on party reelection probability, under mid and high
spending and expected reelection probability, we have also run municipality development levels, when inequality is high.
Figure 4. Marginal effect of audited mismanagement on turnout change: A, nonexposed; B, exposed
levels of mismanagement. Importantly, we have also explored reduced the effectiveness of targeted mobilization in low-
the heterogeneous effects of the drop in reelection probabilities. capacity contexts. If the mechanism operates in this way, the
In appendix H.2 we show that, as expected, the drop in reelec- impact of audits on the relationship between economic in-
tion rates was especially high in exposed municipalities with equality and turnout changes should be more apparent in
both high levels of mismanagement and economic inequality. areas with a higher share of low-income voters. To the extent
that the effects are stronger in these areas, this would suggest
Exploring the mechanism a stronger demobilization impact in localities with a larger
Heterogeneous effects. Our empirical strategy rests on the presence of low-income voters. Figure 6 reports the hetero-
premise that in contexts such as Brazilian municipalities geneous effect of audits in urban versus rural areas, whereas
turnout rates provide relevant information on the behavior figure 7 compares the heterogeneous effects of audit expo-
of low-income voters. Unfortunately, there are no microdata sure in areas with a low and a high prevalence of education.28
available within municipalities to fully validate this premise. Taken altogether, figures 6 and 7 lend considerable sup-
In support of our approach, recent findings by Cepaluni and port to the claim that the federal audits worked to undermine
Hidalgo (2016) suggest that the type of voters affected very the effectiveness of clientelism as a mobilization strategy and
much depends on the type of intervention (and associated triggered an exogenous change in the linkage between eco-
penalties) being evaluated. When the penalties associated with nomic inequality and turnout. The disruption of the pre-
the intervention affect services with access primarily reserved existing political equilibrium took place most visibly in rural
to middle- and upper-income groups, changes in turnout areas with a higher density of low-education, low-income
rates will reflect the elasticity of these groups to the inter- voters, where the ex ante capacity was particularly low. These
vention (in their case, age-related enforcement of compulsory are the areas where the demobilization effect triggered by
voting). Yet when the intervention affects instruments such the audits emerges most strongly.
as mismanagement of cash funds or access to basic social
services, the expectation is that aggregate turnout rates trace Precinct-level results. To further overcome concerns
in large part responsiveness by lower income strata. about a potential ecological fallacy, we take one additional
The data, however, allow us to go further in support of step. Using data on turnout and educational levels of the
the notion that the demobilization effects concentrate pri-
marily around areas with a higher share of low-income voters. 28. The corresponding tables for the exploration of the heterogeneous
The theory suggests that the randomized audits exogenously effects are reported in app. G.
Table 6. Reelection Probability and Audit Exposure
(1) (2) (3) (4)
Exposed .90 .80 1.79 3.59*

(1.04) (1.10) (1.50) (1.99)
Mismanagement .64 .82* .99* 1.33**
(.42) (.44) (.53) (.61)
Exposed # mismanagement 2.85** 21.01** 21.12** 21.47**
(.41) (.47) (.57) (.66)
Constant .64 23.98 1.56 10.04
(2.76) (10.97) (14.63) (21.18)
Electoral and fiscal controls No Yes Yes Yes
Incumbent characteristics controls No No Yes Yes
Party FEs No No No Yes
Regional state FEs Yes Yes Yes Yes
Lottery FEs Yes Yes Yes Yes
R2 .159 .260 .421 .484
N 200 200 200 194
Note. Logit. Incumbent reelection in 2004, first-term incumbents. Standard errors (in parentheses) are clustered at the state
level. FEs p fixed effects.
* p ! .1.
** p ! .05.
*** p ! .01.
population at the precinct level (seção eleitoral), we reesti- demobilization effect is particularly strong precisely in those
mate the relationships of interest. We use the distribution of precincts where the political payoffs of targeted mobilization
educational qualifications within precincts as a proxy for were at the highest level before the audits.
income distribution at each precinct, which allows us to come
one step closer to having income data within the munici- CONCLUSION
pality. Thus, we can model how the impact of audits is Democracy works differently under very high inequality.
moderated by inequality, the scope of mismanagement, and Contrary to the standard view based on rich democracies,
the distribution of education (our proxy for income here) we have shown how high economic inequality and low ca-
within municipalities. We perform both a precinct-level pacity jointly facilitate a self-enforcing equilibrium in which
analysis with municipality FEs (table 7 cols. 1 and 2) and a both turnout and budgetary commitments to targetable poli-
series of hierarchical linear models (cols. 3 and 4). Again, we cies emerge as an optimal strategy. In this equilibrium high
limit the analysis to first-term mayors and include several
specifications, the more demanding ones including all rele-
vant municipality controls from previous models. Figure 8
reports the main insights from the analysis.
To simplify the discussion of the key results, figure 8 pres-
ents the marginal effect of audited mismanagement across the
share of citizens with low education at the precinct level for
both control (nonexposed) and treatment (exposed) groups,
given a high level of inequality (Gini p 0:7). In line with the
theoretical predictions and with earlier results, a higher share
of uneducated citizens in the control group enhances the
mobilization effect of audited mismanagement. By contrast, in
treated precincts the relationship reverses. Indeed, the rela-
tionship switches more strongly, the higher the share of un-
educated voters at the precinct level. In other words, the Figure 5. Audit exposure, mismanagement, and reelection probability
Figure 6. Turnout change, rural versus urban municipalities: A, rural nonexposed; B, rural exposed; C, urban nonexposed; D, urban exposed
formal (but not factual) political equality is achieved through benefit from the low monitoring capacity of the state. We have
elite’s strategic spending directed toward low-income voters also studied the consequences of exogenous disruptions of
with malleable targeted goods. In this type of political equi- such equilibrium: as external interventions increase capacity,
librium a pervasive form of accountability prevails in which both the budgetary use of targeted goods and turnout decline,
economic inequality is linked to greater turnout of low-income and with them so does the incumbents’ likelihood of reelec-
voters, thanks to the strategic behavior of incumbents who tion. Perhaps counterintuitively, then, in the context of high
Figure 7. Turnout change, low-education versus high-education municipalities: A, low education nonexposed; B, low education exposed; C, high education
nonexposed; D, high education exposed.
Table 7. Turnout Change at the Precinct Level
(1) (2) (3) (4)
Low education # Gini 2.11** 2.10** 2.11** 2.10*

(.05) (.05) (.05) (.05)
Low education # mismanagement # Gini .04* .04** .04* .04*
(.02) (.02) (.02) (.02)
Exposed # low education # Gini .10* .10* .10* .10*
(.06) (.06) (.06) (.06)
Exposed # low education # mismanagement # Gini 2.07** 2.07** 2.07** 2.07**
(.03) (.03) (.03) (.03)
Change log registered voters 2.03*** 2.03*** 2.03*** 2.03***
(.00) (.00) (.00) (.00)
Relative to municipality low education .01** .01**
(.01) (.01)
First term Yes Yes Yes Yes
Other interaction terms Included Included Included Included
Municipality controls No No Yes Yes
Model Mun FEs Mun FEs HLM HLM
Note. Standard errors in parentheses. Mun FEs p municipality fixed effects; HLM p hierarchical linear model. N (precincts) p 7,451; N (localities) p 163.
* p ! .1.
** p ! .05.
*** p ! .01.
economic inequality such interventions might be associated monitoring ability of the state against clientelism may re-
with negative welfare consequences for low-income voters. duce subsequent budgetary efforts in targeted goods whose
Our findings point to two complementary lines of research. budgets can be manipulated politically, particularly under
The first one concerns welfare effects of capacity-enhancing conditions of high inequality and low capacity. In addition
interventions. Our findings suggest that an increase in the to our findings, recent evidence suggests that local politicians
Figure 8. Marginal effects of audited mismanagement on turnout under high inequality: A, nonexposed precincts; B, exposed precincts
adjust to both changes in their institutional constraints sharing data that allowed us to perform the quasi-experimental
(Cheibub Figueiredo 2005; De La O and Garcıa 2015) and analysis of Brazilian municipalities, and Fernando Limongi for
modifications in their budget constraint. Taken together, these sharing precinct-level data on educational distributions.
analyses point to a discussion about the short- and long-run
effects of reforms in contexts of low institutional capacity, REFERENCES
about what works and does not across democracies with Amat, Francesc, and Pablo Beramendi. Forthcoming. Economic and Po-
varying levels of institutional capacity (Harding and Stasavage litical Inequality: Competition, Spending, and Turnout. Cambridge:
Cambridge University Press.
2014). The results in this article suggest that targeted mobili-
Anderson, Christopher J., and Pablo Beramendi. 2012. “Left Parties, Poor
zation (including clientelism) emerges as a second-best strat- Voters, and Electoral Participation in Advanced Industrial Societies.”
egy in low-capacity contexts: some people incur welfare losses Comparative Political Studies 45 (6): 714–46.
as a result of efforts to transition toward more programmatic Anduiza, Eva. 1999. Individuos o sistemas? Las razones de la abstención en
politics.29 These are, however, short-run effects. More work is Europa occidental. Madrid: Centro de investigaciones sociológicas.
Ansell, Ben W. 2010. From the Ballot to the Blackboard: The Redistributive
needed to establish the length of the valley of tears associated Political Economy of Education. Cambridge: Cambridge University
with such a transition. Press.
The second line for future research concerns the political Bartels, Larry M. 2009. Unequal Democracy: The Political Economy of the
conditions under which capacity itself evolves over time. Ca- New Gilded Age. Princeton, NJ: Princeton University Press.
Bobonis, Gustavo J., Paul Gertler, Marco Gonzalez-Navarro, and Simeon
pacity enhancing would appear suboptimal for local incum-
Nichter. 2017. “Vulnerability and Clientelism.” Working paper 23589.
bent elites under conditions of high inequality (see, e.g., National Bureau of Economic Research, Cambridge, MA.
Hollenbach and Silva 2019). Our analysis suggests that these Bond, Robert M., Christopher J. Fariss, Jason J. Jones, Adam D. I. Kramer,
local elites have much to gain from preventing large-scale ca- Cameron Marlow, Jaime E. Settle, and James H. Fowler. 2012. “A 61-
Million-Person Experiment in Social Influence and Political Mobili-
pacity investments. Challengers, however, have the opposite
zation.” Nature 489 (7415): 295–98.
incentives. Understanding the conditions under which this Brollo, Fernanda. 2012. “Why Do Voters Punish Corrupt Politicians?
contest results in more programmatic, tax-compliant de- Evidence from the Brazilian Anti-corruption Program (May 30, 2012).”
mocracies is a fruitful avenue for future research. SSRN. https://papers.ssrn.com/sol3/papers.cfm?abstract_idp2141581.
Brollo, Fernanda, Tommaso Nannicini, Roberto Perotti, and Guido
Tabellini. 2013. “The Political Resource Curse.” American Economic
ACKNOWLEDGMENTS Review 103 (5): 1759–96.
Previous versions were presented at the 2015 American Po- Bursztyn, Leonardo. 2016. “Poverty and the Political Economy of Public
litical Science Association meetings and workshops at Wash- Education Spending: Evidence from Brazil.” Journal of the European
ington University in St. Louis; University of Wisconsin, Economic Association 14 (5): 1101–28.
Calvo, Ernesto, and Maria Victoria Murillo. 2004. “Who Delivers? Parti-
Madison; Duke University; Princeton University; the National
san Clients in the Argentine Electoral Market.” American Journal of
School of Development in Beijing; Harvard University; the Political Science 48 (4): 742–57.
Paris School of Economics; and the University of Chicago. Campante, Filipe R. 2011. “Redistribution in a Model of Voting and
We thank participants in all these events as well as Atila Campaign Contributions.” Journal of Public Economics 95 (7): 646–56.
Cepaluni, Gabriel, and Fernando D. Hidalgo. 2016. “Compulsory Voting
Abdulkadiroglu, Michael Albertus, Victor Araujo, Marta
Can Increase Political Inequality: Evidence from Brazil.” Political Analysis
Arreche, George Avelino, Carles Boix, Ana de la O, Betul 24 (2): 273–80.
Demirkaya, Ruben Enikolopov, German Feiherd, Tim Cheibub Figueiredo, Argelina, Haroldo da Gama Torres, Fernando
Feddersen, Héctor Galindo, Guy Grossmann, Kimuli Kasara, Limongi, Marta Arretche, and Renata Bichir. 2005. Rede de Pesquisa e
Herbert Kitschelt, Fernando Limongi, Eddy Malesky, Monika Desenvolvimento de Políticas Públicas. São Paulo: Centro Brasileiro de
Análise e Planejamento.
Nalepa, Maggie Penn, Didac Queralt, Pia Raffler, Guillermo
Colonnelli, Emanuele, Edoardo Teso, and Mounu Prem. 2017. “Patronage
Rosas, David Rueda, Jeremy Spater, Susan Stokes, Pavithra in the Allocation of Public Sector Jobs.” SSRN. https://papers.ssrn.com
Suryanarayan, and Erik Wibbels for their comments on pre- /sol3/papers.cfm?abstract_idp2942495.
vious versions of this manuscript. We thank Asli Cansunar, Dahl, Robert Alan. 1991. Democracy and Its Critics. New Haven, CT: Yale
University Press.
Haohan Chen, and Diego Romero for excellent research
Debs, Alexandre, and Gretchen Helmke. 2010. “Inequality under De-
assistance. Finally, we thank Herbert Kitschelt for sharing mocracy: Explaining the Left Decade in Latin America.” Quarterly
his data on types of political competition, Claudio Ferraz for Journal of Political Science 5 (3): 209–41.
De La O, Ana, and Fernando Martel Garcıa. 2015. “Can Intrastate Ac-
countability Reduce Local Capture? Results from a Field Experiment in
Mexico.” Unpublished manuscript.
29. In an analysis complementary to the one in this article, Amat and de Oliveira, Romualdo Portela, and Theresa Adrião. 2007. Organização do
Beramendi (forthcoming) show that, following the introduction of audits, ensino no Brasil: Níveis e modalidades na constituição federal e na
poverty actually increases in the treated municipalities. LDB. São Paulo: Xamã.
Diaz-Cayeros, Alberto, Federico Estevez, and Beatriz Magaloni. 2016. The Krishna, Anirudh. 2008. Poverty, Participation, and Democracy. Cambridge:
Political Logic of Poverty Relief: Electoral Strategies and Social Policy in Cambridge University Press.
Mexico. Cambridge: Cambridge University Press. Larreguy, Horacio, John Marshall, and Pablo Querubin. 2016. “Parties,
Fergusson, Leopoldo, Horacio Larreguy, and Juan Felipe Riano. 2015. Brokers and Voter Mobilization: How Turnout Buying Depends upon
“Political Constraints and State Capacity: Evidence from a Land Al- the Party’s Capacity to Monitor Voters.” American Political Science
location Program in Mexico.” Working paper 764, CAF Development Review 110 (1): 160–79.
Bank of Latin America. Lichand, Guilherme, Marcos Lopes, and Marcelo Medeiros. 2016. “Is
Ferraz, Claudio, and Frederico Finan. 2008. “Exposing Corrupt Politicians: Corruption Good For Your Health?” Unpublished manuscript.
The Effect of Brazil’s Publicly Released Audits on Electoral Outcomes.” Nathan, Noah L. 2016. “Does Participation Reinforce Patronage? Policy
Quarterly Journal of Economics 123 (2): 703–45. Preferences, Turnout and Class in Urban Ghana.” British Journal of
Ferraz, Claudio, and Frederico Finan. 2011. “Electoral Accountability and Political Science 49 (1): 229–55.
Corruption: Evidence from the Audits of Local Governments.” American Nichter, Simeon. 2008. “Vote Buying or Turnout Buying? Machine Poli-
Economic Review 101 (4): 1274–311. tics and the Secret Ballot.” American Political Science Review 102 (1):
Gallego, Aina. 2014. Unequal Political Participation Worldwide. Cambridge: 19–31.
Cambridge University Press. Olken, Benjamin A. 2007. “Monitoring Corruption: Evidence from a Field
Gingerich, Daniel W. 2013. Political Institutions and Party-Directed Cor- Experiment in Indonesia.” Journal of Political Economy 115 (2):
ruption in South America: Stealing for the Team. Cambridge: Cambridge 200–249.
University Press. Przeworski, Adam. 2010. Democracy and the Limits of Self-Government.
Hainmueller, Jens, Jonathan Mummolo, and Yiqing Xu. 2019. “How Much Cambridge: Cambridge University Press.
Should We Trust Estimates from Multiplicative Interaction Models? Simple Robinson, James A., and Thierry Verdier. 2013. “The Political Economy of
Tools to Improve Empirical Practice.” Political Analysis 27 (2): 163–92. Clientelism.” Scandinavian Journal of Economics 115 (2): 260–91.
Harding, Robin, and David Stasavage. 2014. “What Democracy Does (and Scott, James C. 1969. “Corruption, Machine Politics, and Political Change.”
Doesn’t Do) for Basic Services: School Fees, School Inputs, and African American Political Science Review 63 (4): 1142–58.
Elections.” Journal of Politics 76 (1): 229–45. Solt, Frederick. 2008. “Economic Inequality and Democratic Political
Hidalgo, F. Daniel, and Simeon Nichter. 2016. “Voter Buying: Shaping the Engagement.” American Journal of Political Science 52 (1): 48–60.
Electorate through Clientelism.” American Journal of Political Science Stigler, George J. 1970. “Director’s Law of Public Income Redistribution.”
60 (2): 436–55. Journal of Law and Economics 13 (1): 1–10.
Hollenbach, Florian M., and Thiago N. Silva. 2019. “Fiscal Capacity and Stokes, Susan C. 2005. “Perverse Accountability: A Formal Model of
Inequality: Evidence from Brazilian Municipalities.” Journal of Politics Machine Politics with Evidence from Argentina.” American Political
81 (4): 1434–45. Science Review 99 (3): 315–25.
Kasara, Kimuli ,and Pavithra Suryanarayan. 2014. “When Do the Rich Vote Stokes, Susan C., Thad Dunning, Marcelo Nazareno, and Valeria Brusco.
Less than the Poor and Why? Explaining Turnout Inequality across the 2013. Brokers, Voters, and Clientelism: The Puzzle of Distributive
World.” American Journal of Political Science 59 (3): 613–27. Politics. Cambridge: Cambridge University Press.
Kirchheimer, Otto. 1965. “Confining Conditions and Revolutionary Break- Timmons, Jeffrey F., and Francisco Garfias. 2015. “Revealed Corruption,
throughs.” American Political Science Review 59:964–74. Taxation, and Fiscal Accountability: Evidence from Brazil.” World
Kitschelt, Herbert, and Steven I. Wilkinson. 2007. “Citizen-Politician Development 70:13–27.
Linkages: An Introduction.” In Patrons, Clients, and Policies: Patterns Verba, Sidney, Kay Lehman Schlozman, and Henry E. Brady. 1995. Voice and
of Democratic Accountability and Political Competition. Cambridge: Equality: Civic Voluntarism in American Politics, vol. 4. Cambridge:
Cambridge University Press, 1–49. Cambridge University Press.
Copyright of Journal of Politics is the property of The Southern Political Science Association
and its content may not be copied or emailed to multiple sites or posted to a listserv without
the copyright holder's express written permission. However, users may print, download, or
email articles for individual use.

Democracy Under High Inequality: Capacity, Spending, and Participation

Uploaded by

Copyright:

Available Formats

Democracy Under High Inequality: Capacity, Spending, and Participation

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Democracy Under High Inequality: Capacity, Spending, and Participation

Uploaded by

Copyright:

Available Formats

Democracy under High Inequality: Capacity, Spending,

I nequality is known to bias political inﬂuence, under-

subject to p bP 1 bR 1 g and to (1 2 t)wP ð3Þ

Comparative statics and hypotheses

legislature toward low-income voters, we use two measures:

LogSpending p a 1 b1 Ineqm;s 1 b2 Mismm;s

13. See app. D for a robustness check of the linear interactions in

Table 2. Spending under the Status Quo: Capacity and Incentives

Public Employees Primary Education Spending

(1) (2) (3) (4)

Inequality 21.95* 21.92* 21.25 21.48*

capacity, and provision of targeted goods is expected to dis-

Audit exposure and changes in political participation.

Table 4. Audit Exposure and Changes in Local Targeted Goods

(1) (2) (3) (4) (5) (6)

Inequality 2.31 21.12*** 21.14*** 21.14** 21.74*** 22.14***

Table 5. Audit Exposure and Changes in Turnout Rates

(1) (2) (3) (4) (5) (6)

Inequality 2.19 2.24 2.13 2.19 2.33* 2.47**

Figure 4. Marginal effect of audited mismanagement on turnout change: A, nonexposed; B, exposed

Table 6. Reelection Probability and Audit Exposure

(1) (2) (3) (4)

Exposed .90 .80 1.79 3.59*

Table 7. Turnout Change at the Precinct Level

(1) (2) (3) (4)

Low education # Gini 2.11** 2.10** 2.11** 2.10*

You might also like

Inequality 2.31 21.12* 21.14* 21.14 21.74* 22.14***

Low education # Gini 2.11 2.10 2.11** 2.10*