1

Accountability as a deterrent to corruption:

new data from Brazilian municipalities

By Bianca Vaz Mondo

PhD Candidate, Hertie School of Governance

Paper prepared for ECPR Joint Sessions - Warsaw, 29 March-02 April 2015

Version of 18 March, 2015

Abstract

Democracy has been highlighted in the comparative empirical literature as a strong

determinant of corruption, explained mainly by the expected effect of electoral accountability

as a deterrent of corrupt behavior. Despite some recent findings that the electoral

accountability mechanism is associated with lower corruption levels in the short term, there is

so far no evidence that this effect subsists over time. This paper exploits the setup of multiple

audit rounds conducted in randomly selected Brazilian municipalities to further assess the

effect of accountability on future corruption levels. Moreover, it contributes to this line of

research by considering also alternative forms of accountability as potential explanatory

factors of corruption. Based on insights from discussions on the dimensions of accountability,

the study integrates social and horizontal accountability into the analysis of a deterrent effect

of accountability on corruption. Consistent with previous findings, a preliminary analysis of the

data fails to support the claim that effective electoral accountability contributes to lower future

corruption. Only a deterrent effect by social accountability is partly supported by the data.

1. Introduction

Democracy has been highlighted in the comparative empirical literature as an important

determinant of corruption. In the past two decades, numerous studies have examined this

relationship from different angles and two broader patterns appear to emerge from those

analyses: (a) countries with longer democratic experience are consistently associated with

lower levels of corruption (Lederman et al. 2005; Pellegata 2012; Sandholtz and Koetzle 2000;

Serra 2004; Treisman 2000), and (b) at higher levels of democracy, stronger democratic

institutions are also associated with lower corruption (Bäck and Hadenius 2008; Pellegata

2012; Saha 2008; Sung 2004)1.

1 This finding refers to evidence that the association between democracy and corruption is non-linear and the effect of democracy on corruption varies with the level of democracy itself. At lower levels of democracy, corruption seems to increase as democracy improves. This is corroborated by qualitative evidence from earlier case studies on new democracies after the so-called Third Wave of democratization, which have shown that democratic transition was followed by an apparent increase in

2

One of the most prominent arguments on why we should expect a negative effect of

democracy on corruption refers to the role of elections in generating incentives for office

holders to refrain from corrupt behavior. This can be traced back to theories of democratic

representation, which portray representative democracy as a system where rulers are

systematically induced to act according to the interests of the citizenry through a mechanism

of electoral accountability (Przeworski et al. 1999). More specifically, democracy should induce

representation on the part of elected officials because voters can ultimately threaten to

remove them from office, should they act in ways detrimental to the public interest

(Przeworski et al. 1999). Corruption, as an example of action that benefits private interests at

the expense of the collective interest, should then also decrease with democratic

representation.

At the same time, however, other insights from the literature on electoral accountability raise

questions about how effective this mechanism can be in substantially lowering corruption in a

political system. Some theoretical discussions, for instance, put into question the role of

elections as a sanctioning mechanism more generally. It is argued that elections are a very

limited instrument for voters to effectively punish unresponsive incumbents (Maravall and

Sánchez-Cuenca 2008, l. 139; O’Donnell 1999, 30), since holding politicians accountable for

past performance is only one of several purposes that elections serve in a political system

(Persson, Roland, and Tabellini 1997) 2. Even under the most favorable conditions, when voters

have enough information about politicians’ behavior and can clearly attribute responsibility for

specific outcomes, each voter still has only one vote to make a judgment on hundreds of

decisions that the government has taken during the previous term. This contributes to weaken

the incentive mechanism expected to control moral hazard among office holders (Fearon

1999; Manin et al. 1999a). Additionally, an extensive empirical literature has found only limited

evidence that voters in fact punish corrupt incumbents: although the latter apparently face

some vote share loss after being associated with corruption scandals, a vast majority of them

still manages to retain office (e.g. Bågenholm 2013; Chang et al. 2010; Dimock and Jacobson

1995; Eggers and Fisher 2011; Jiménez and Caínzos 2004; Peters and Welch 1980).

An emerging literature has sought more direct evidence of whether this deterrence

mechanism indeed works as originally predicted, and whether the empirical findings linking

corruption in several countries (Geddes and Neto 1992; Harriss-White and White 1996; Mohtadi and Roe 2003; Moran 2001; Rock 2007; Weyland 1998; Whitehead 2002).

2 These include aggregating and representing voters' conflicting preferences, aggregating dispersed information about the correct political decisions, and allowing citizens to select the most competent individuals for public office (Persson, Roland, and Tabellini 1997, 1165).

3

stronger and more consolidated democratic institutions to lower corruption can be attributed

to this aspect of democratic regimes. Nevertheless, results are inconclusive: if, on the one

hand, some studies have found that the anticipation of future punishment by voters can be

linked to lower corruption outcomes in some cases (Bobonis et al. 2012; Ferraz and Finan

2011), on the other hand, this effect does not appear to be sustained over time (Bobonis et al.

2012; Crisp et al. 2014).

Building upon this line of research, this paper further explores whether the electoral

accountability mechanism proposed in the literature can be empirically associated with lower

levels of corruption. However, it aims to contribute to this literature by integrating the

perspective that elections are not the only source of accountability that may act as a deterrent

against corrupt behavior. Based on insights from the democracy and accountability literatures,

the study applies a framework of dimensions of accountability, examining simultaneously

electoral, social (as components of vertical accountability) and horizontal accountability as

relevant explanatory factors of corruption. Only few studies have taken the effect of different

dimensions of accountability on corruption into account, and the objective here is to

contribute to a more systematic analysis of how these dimensions independently and

simultaneously impact corruption. This angle of analysis becomes particularly relevant in light

of arguments regarding how different dimensions of accountability may affect one another

(Mainwaring and Welna 2003; Peruzzotti and Smulovitz 2006a), which raises questions about

potential confounding effects of the different dimensions of accountability if they are not

adequately modeled.

Additionally, the analysis follows recent studies focusing more specifically on whether effective

accountability affects future corruption levels, taking as background the overarching

discussions and empirical findings about the impact of democracy on corruption over time. The

empirical strategy exploits the setup of multiple audit rounds conducted in randomly selected

Brazilian municipalities, which provide a concrete measure of corruption for different points in

time for a subsample of 140 municipalities.

The paper is structured as follows: section 2 briefly presents a conceptual discussion about the

dimensions of accountability considered as main independent variables in the study; section 3

reviews the relevant theoretical and empirical literature on accountability and corruption;

section 4 describes the methodological approach and the data collection process and sources;

section 5 presents preliminary results of the empirical analysis; and section 6 concludes and

discusses the implications for further research.

4

2. The dimensions of accountability

As mentioned in the introduction, electoral accountability is the most ubiquitous causal

mechanism discussed in the literature on the effects of democracy on corruption. The

argument that elections enable voters to exercise some form of “control” over politicians by

generating incentives for their “good behavior” has been present in most studies throughout

the past 20 years of research on this relationship3. It relies on the basic notion of electoral

accountability as a deterrent to self-interested behavior in office, under the assumption that

politicians derive utility from staying in power and thus have an interest in behaving in a way

to maximize their probability of being reelected. In this context, the anticipation of

punishment by voters at the ballot box should ex ante induce incumbent politicians to be more

responsive towards voters’ interests and refrain from corruption.

Despite the predominance of this argument, a number of other potential causal mechanisms

for an effect of democracy on corruption are raised in the literature. Several studies refer to

the role of democratic freedoms of information, association and expression, also linked to the

emergence of a free press, in contributing to the monitoring of governments by the citizenry

and the consequent reduction of corruption in society. Furthermore, the importance of

institutional mechanisms of control to put a check on the government is discussed. Finally, a

normative dimension of democratization is also included in this debate, with regards to the

consolidation of democratic norms that crystallize the belief among the citizenry that

corruption is antithetical to democracy and the common interest.

The framework proposed here interprets these additional causal mechanisms as directly or

indirectly associated with the dimensions of accountability discussed in the literature on

democratic accountability. Accountability implies “[…] subjecting power to the threat of

sanctions; obliging it to be exercised in transparent ways; and forcing it to justify its acts”

(Schedler 1999, 14)4. Several types of accountability have been proposed in the literature

3 This idea is the basis of a long-standing tradition of political agency models that focus more broadly on analyzing the incentives for responsiveness on the part of a political actor in office, i.e. actions that correspond to voters’ preferences (Manin, Przeworski, and Stokes 1999a). The same logic has been applied to the specific case of corruption as an example of non-responsive behavior and how those incentives can constrain corrupt behavior.

4 It is important to make a distinction between this understanding of accountability and other meanings that appear in the literature. Accountability here refers to the existence of concrete mechanisms (formally institutionalized or not) through which public office holders can be called to provide an account of their actions and be sanctioned for misconduct. This differs from other more normatively

5

(Bovens 2006), but particularly relevant is the contribution of O’Donnell (1994, 1999) with his

discussion of the vertical and horizontal dimensions of accountability. According to O’Donnell,

“Elections, social demands that usually can be articulated without suffering state

coercion, and regular coverage by the media of at least the more visible of these

demands and of apparently wrongful acts of the public authorities are dimension of

[…] ‘vertical accountability’.” (1999, 30)

Horizontal accountability, on the other hand,

“[…] is the existence of state agencies that are legally enabled and empowered, and

factually willing and able, to take actions that span from routine oversight to criminal

sanctions or impeachment in relation to actions of omissions by other agents of the

state that may be qualified as unlawful.” (O’Donnell 1999, 38)

More specifically, these agencies undertake actions “[…] with the explicit purpose of

preventing, cancelling, redressing and/or punishing actions (or eventually non-actions) by

another state agency that are deemed unlawful, whether on grounds of encroachment or of

corruption.” (O’Donnell 2003, 35)

The vertical dimension of accountability can be divided into two components. When it is

exercised through elections, this represents electoral accountability (O’Donnell 1999).

However, O’Donnell also acknowledges the role of social demands and monitoring by the

media as part of vertical accountability. Peruzzotti and Smulovitz have called this component

“social accountability”, defined as “[…] a non-electoral yet vertical mechanism of control of

political authorities that rests on the actions of an array of citizens’ associations and

movements and the media” (Peruzzotti and Smulovitz 2006b, 10).

It becomes clear from this discussion that these three sources of accountability vary with

regards to their degree of institutionalization and formalization. In the case of electoral and

social accountability, the two components of vertical accountability, the answerability side of

accountability may take a more informal nature, in the sense that the demand for

transparency and justification for acts taken by the government can materialize not only

through formal channels of inquiry that may be available to individuals, media and civil society,

but also through popular pressure. With regards to the sanctioning aspect of accountability, on

the other hand, electoral accountability involves a formal channel through which voters may

collectively “punish” a corrupt incumbent, whereas social accountability disposes only of

loaded conceptualizations of accountability that portray it in association with a sense of individual responsibility or moral obligation, or even as a synonym of responsiveness (Mulgan 2000).

6

reputational sanctions, which may nonetheless be of considerable weight to politicians (Grant

and Keohane 2005; Philp 2009). In the case of horizontal accountability, on the other hand,

both answerability and sanctioning are linked to formal institutional channels and procedures.

Taking into consideration these dimensions of accountability, it is possible to directly link them

to the main causal mechanisms proposed in the literature as pathways for an effect of

democracy on corruption. The link between the electoral component of democracy and

electoral accountability is straightforward: elections provide the citizenry with a means of

sanctioning corrupt politicians by removing them from office. Democratically safeguarded

rights to expression, association and information are in turn connected to the social

component of vertical accountability, as they provide for the monitoring of government by an

organized civil society and a free press. Institutional mechanisms of control, in the form of

checks and balances and separation of powers, for instance, can be seen as a direct

manifestation of horizontal accountability, as they often institutionalize channels through

which the Legislative and the Judiciary may hold rulers accountable for misconduct. The fourth

mechanism, related to democratic norms, is interpreted here as indirectly associated to all

dimensions of accountability, in the sense that it contributes to foster and enable their

consolidation and effectiveness over time. Therefore, this dimension is not considered

independently from the others, but rather as a factor that feeds into the three dimensions of

accountability.

This framework is useful for the analysis of the effect of accountability on corruption because

it allows a more comprehensive study of this question, considering not only the electoral

component that is most often discussed in the literature, but also other analogous institutional

mechanisms that potentially create similar incentives against corrupt behavior. Moreover, this

perspective relates to the broader question of how democracy affects corruption through

different mechanisms.

3. Theoretical background and previous empirical evidence

The connection between accountability and corruption can be derived mainly from the

political agency literature. In general terms, this research tradition highlights especially how

electoral accountability creates incentives for political actors in office to be more responsive,

i.e. to take actions that correspond to voters’ preferences (Manin et al. 1999). Based on a

principal-agent framework, central assumptions behind models of political agency are that

office holders, as agents, seek to maximize their own utility and that this may often conflict

7

with the interest of their principal, the electorate. Given the information asymmetry between

the principal and the agent, office holders have the possibility of betraying their representative

mandate and pursuing their own interests without the knowledge of voters. In this setup,

elections give voters an opportunity to replace incumbent politicians when they perform

below their expectations. Therefore, politicians have incentives to perform well in office in

order to avoid losing their position in the next elections.

This model has been applied to the case of corruption as an example of non-responsive action.

In the particular case of corruption, the theoretical framework is similarly based on an agency

relationship between voters and elected officials, whereby the former, in the role of principal,

entrust the latter, as agents, with power to make decisions on behalf of their interests, in

order to achieve a set of preferred outcomes determined by the principal (Kitschelt et al. 2009;

Lancaster and Montinola 1997; Rose-Ackerman 1978). Corruption thus constitutes a violation

of the obligation taken by the agent to act according to the principal’s interests. Therefore, in

this framework corruption is commonly conceptualized as the misuse of public or entrusted

power for private gain (Lambsdorff 2007; Rose-Ackerman 1999).

What determines agents’ decisions to pursue their own interests instead of the electorate’s is

a calculation of the probability of facing the potential costs of corruption against the benefits

that it would entail (Andvig et al. 2000). Here is how accountability comes into play: in all its

forms, the sanctioning component of accountability mechanisms implies potential costs that

may offset the benefits generated by corruption. In the case of electoral accountability, as

already mentioned, politicians face the risk of losing office. Information on the malfeasance of

agents may eventually become available to their principal, i.e. voters, who may react and

punish corrupt incumbents by removing them from office (Andvig et al. 2001; Kolstad and Wiig

2011; Lambsdorff 2007; Rose-Ackerman 1999).

The logic of a deterrent effect through social and horizontal accountability mechanisms is very

similar. Although social accountability does not involve strict formal sanctions, the watchdog

role of civil society and the media may involve severe reputational costs for politicians.

Moreover, social accountability agents can also contribute to electoral accountability by

exposing information on wrongdoing that potentially influences voter behavior, and to

horizontal accountability by activating formal mechanisms and institutions to redress abuses

by office holders (Peruzzotti and Smulovitz 2006a, l. 148). Horizontal accountability in turn can

lead to formal penalties for office misconduct, ranging from administrative sanctions to

criminal sentences, depending on the institutional channels involved. Therefore, the costs

related to mechanisms of horizontal accountability can be quite severe, when effectively

8

implemented. Thus the anticipation of costs from these different mechanisms of

accountability, and an assessment of their respective probabilities, are expected to produce

incentives that serve to ex ante discipline politicians and constrain corruption.

In recent years some empirical studies have sought to identify more specifically the effect of

these mechanisms on corruption. Regarding electoral accountability, one such example is the

study by Ferraz and Finan (2011), who try to indirectly capture this mechanism by estimating

the effect of reelection incentives on corruption at the municipal level in Brazil. A common

prediction derived from political agency models is that the deterrent effect of electoral

accountability should be present only when reelection incentives exist. Therefore, incumbents

that face term limits should not have the same incentives as those that are still eligible for

reelection. The authors use data from Brazilian municipalities to compare corruption levels5

across municipalities with first-term and second-term mayors, as the latter cannot run for

reelection in the Brazilian system. As predicted by the model, they find that on average 27%

less corruption was found under first-term mayors than under second-term mayors, and these

results are robust to different specifications, indicators and to the test of alternative

explanations.

Another study taking a similar approach uses the release of information on corruption before

and after an election to capture the effect of electoral accountability. Bobonis et al. (2012) use

the setting of an audit program for municipalities in Puerto Rico to verify whether incumbent

mayors that have their municipality audited in the period before an election have incentives to

engage in less corruption than mayors whose municipalities are audited only after the

election6. The main logic behind this is that mayors anticipate that voters’ behavior will be

affected by the audit results only if they are disclosed before the election. They find that the

pre-electoral disclosure of audit results is associated with 67% less corruption in comparison to

the municipalities where the audit results were disclosed after the election.

These two findings relate to the ex ante effect described earlier, i.e. how the anticipation of

punishment by voters in the future affects incumbent behavior in the present. A different

approach tries to observe the effect of electoral accountability on future levels of corruption.

Although this perspective is not often explicitly discussed in the literature on political agency

5 Similar to a previous study (Ferraz and Finan 2008), they use original concrete corruption indicators derived from municipal audit reports.

6 The research design is made possible by the structure of the audit program, which relies on a pre-determined ordering of the municipalities that will be audited. This should allow mayors to predict whether they will be audited before the next elections or not.

9

models, we may interpret the dynamic effect of electoral accountability on corruption, valid

also for the other dimensions of accountability, as related to how political actors adjust their

behavior in the face of changes in the probability of sanctioning over time. By observing the

effective functioning of accountability in the present, these actors would then reassess their

expectation on being sanctioned in the future should they engage in corruption. In this way,

the same mechanism functions repeatedly over time, only with updates in the probability of

sanctions taking place. This perspective also has direct parallel to the broader picture of how

democracy affects corruption and the temporal component in this relationship that is

emphasized in the literature. After all, it is quite plausible that accountability mechanisms in a

democracy should become more effective as those institutions consolidate over time.

The study by Bobonis et al. (2012) also examines whether this is the case in the Puerto Rican

context, which provides them with an unique opportunity of analyzing longitudinal data for

their concrete measure of corruption drawn from the municipal audit results. This is possible

because the centralized program has been implemented since 1987, and each municipality in

the country has been audited several times since then. Interestingly, however, they find no

effect of electoral accountability on corruption levels in the administration term subsequent to

an election7. The authors interpret this as evidence that corrupt incumbents try to improve

their reputation by refraining from corruption before an election, thus mimicking the behavior

of non-corrupt politicians, in order to reap higher rents in the next term in office.

A more recent article by Crisp et al. (2014) looks at this effect also indirectly, by analyzing how

the impact of electoral accountability on corruption can be captured through a measure of

electoral volatility. Referencing the standard theoretical arguments from political agency and

electoral accountability models, these authors postulate that electoral accountability of

corrupt politicians should generate increased electoral volatility, once voters move their

support to other parties. This should in turn contribute to reducing corruption in the future, as

voters signal to politicians that their threat of sanctioning bad incumbents is real. Based on this

causal chain, they estimate a reciprocal relationship between electoral volatility and

perceptions of legislative corruption for 249 elections in 74 democracies and partial

democracies. The authors find that stronger perceptions of corruption do indeed lead to more

electoral volatility, but no statistically significant effect of electoral volatility on future

7 Their empirical approach for measuring the effect of electoral accountability on future corruption still captures electoral accountability indirectly, using the disclosure of audit results as a main independent variable. Therefore, they do not take fully into account the extent to which accountability has been effective in the previous election, that is, whether the mayor has been voted out and replaced by another one, or whether he/she has been reelected despite involvement in corruption.

10

corruption levels is found in their analysis. They interpret this result as possibly reflecting two

different responses: (a) high electoral volatility may lead the disciplining mechanism to break

down, as politicians interpret that as a sign that their chances of retaining office are slim

regardless of their behavior, thus having incentives to extract more rents while they can, and

(b) even if politicians do change their behavior as a result of electoral accountability, it would

take some time for voters to change their perception of corruption, and this could be the

reason why they fail to find the hypothesized effect.

A few other studies present valuable insights on the impact of the other accountability

dimensions on corruption outcomes. A very influential study by Olken (2007) takes an

innovative approach to examining the effect of increased oversight, both in the form of top-

down audits–representing a form of horizontal accountability–and bottom-up monitoring–an

example of social accountability–, on corruption levels in road construction projects in

Indonesian villages. In an experimental design, municipalities were randomly selected to be

informed that the implementation of the road construction project would be audited, thus

increasing the probability of audit from 4% to 100%. Additionally, two experiments were

conducted to assess the effect of increased social monitoring: (a) the distribution of invitations

to the village’s “accountability meeting”, where officials give an account of how public

resources have been used; and (b) the distribution of comment forms together with the

invitations, in order to allow villagers to anonymously give information on the account given at

the meeting. The author found that the amount of missing expenditures was lower by 8

percentage points in the villages that were informed about a future audit of the project. At the

same time, nepotism seemed to increase in jobs related to the project, suggesting that one

form of corruption may have been substituted for with another. The social monitoring

treatments, on the other hand, had no significant overall effect on missing expenditures, even

though they were successful in increasing participation in the accountability meetings.

The study by Ferraz and Finan (2011) also includes insights on the role of horizontal and social

accountability, as they consider alternatively to their original models that the presence of

prosecutors or radio stations may condition the effect of reelection incentives on their

corruption estimates. The presence of prosecutors would imply a higher probability of legal

procedures against the mayor, thus increasing the potential costs of corruption. Similarly, the

presence of local radio stations would represent a larger probability of revealing corruption to

voters. The authors then predict that the difference in corruption violations between

municipalities with first-term and second-term mayors should be lower in the presence of

either prosecutors or local radio. They find that the evidence supports these hypotheses, but

11

they only examine them in separate models. They also find a significant negative independent

effect of local media on corruption levels.

A few other studies have looked at the impact of media on corruption. Adsera et al. (2003), for

instance, examine the effect of political accountability on corruption considering not only the

level of democracy, but also a measure of free newspaper circulation, in order to capture the

degree of information available to voters. Using panel data from the International Country Risk

Guide (ICRG) for the period 1980-1995 (averaged for each five-year period) for more than 100

countries, they find that newspaper circulation has a significant positive effect on the

corruption estimate8. This result is robust to an endogeneity test and to alternative models

with a panel of US States, where corruption is measured as the number of convicted public

officials per 100 officials between 1977 and 1995.

Brunetti and Weder (2003) also argue that a free press is an important mechanism of control

against corruption, and look at the effect of press freedom on corruption levels across 125

countries. Their results show a significant negative association between press freedom and

corruption, robust to a number of different specifications. The authors also include a measure

of rule of law as a potential alternative control mechanism, in the form of checks and balances

and solid legal institutions. This variable, which broadly corresponds to the horizontal

accountability category discussed here, is found to have no statistically significant association

with corruption level when a range of other institutional, economic and cultural factors are

controlled for.

Mungiu-Pippidi (2011) analyzes determinants of corruption based on a model of resources

versus constraints, in which indicators of constraints can be associated to the mechanisms of

horizontal and social accountability discussed earlier. Judicial independence, for instance, is

considered as a form of legal constraint to corruption, and has parallel to the concept of

horizontal accountability. Indicators of normative constraints include internet access and

number of civil society organizations, which relate to information and mobilization aspects

associated with social accountability. A cross-national analysis with 114 countries shows

evidence of a significant positive effect of those three variables on a measure of control of

corruption.

8 The free newspaper circulation variable included in the analysis is already the product of a measure of democracy and of newspaper circulation, based on the argument that the effect of information is conditioned on the presence of democratic institutions. However, the models estimated do not specify this interaction hypothesis properly, as they do not include the original measure of newspaper circulation, which could represent a source of bias in the results.

12

Grimes (2013), on the other hand, finds no evidence of an universal effect of civil society

density on corruption levels, in a comparative setup with data for 133 countries. However,

once the conditioning effect of political competition, transparency and press freedom are

taken into consideration, the author finds that civil society density is associated with lower

corruption in contexts where those conditions are met.

One other study has also tried to estimate the effect of vertical and horizontal accountability

on corruption levels across different countries. Camaj (2013) focuses on the impact of media

freedom, but considers other proxies of accountability in the estimation procedures. Different

model specifications show a robust association between higher levels of press freedom and

lower levels of corruption, while controlling for political competitiveness and voter turnout–

taken as indirect measures of vertical (electoral) accountability–, freedom of association–as an

indicator of civil society strength–and judicial independence–an indirect measure of horizontal

accountability. Out of these other accountability indicators9, voter turnout and judicial

independence are also significantly associated with lower corruption.

As the review presented above shows, an increasing literature has sought to directly or

indirectly capture the effect of electoral, social and horizontal accountability, either

simultaneously or separately, on corruption indicators. In the case of electoral accountability,

current evidence offers a rather pessimistic outlook on its sustained deterrent effect on

corruption. For horizontal accountability, some evidence for an effect of independent judiciary

(Camaj 2013; Mungiu-Pippidi 2011) and increased external control (Olken 2007) exist, but this

may also be reduced in the long run if actors adapt their behavior to less detectable forms of

rent-seeking (Olken 2007). Finally, the case for a significant impact of social accountability

seems to be stronger in the case of media presence and freedom (Adsera et al. 2003; Brunetti

and Weder 2003; Ferraz and Finan 2011; Grimes 2013), but less conclusive regarding the

monitoring role of civil society (Olken 2007), which may also be conditional on the existence of

a favorable institutional context (Grimes 2013).

The analysis conducted here seeks to contribute to this existent knowledge by examining these

different dimensions more systematically, accounting for potential simultaneous effects of the

distinct mechanisms of accountability. Moreover, differently than some of the studies

reviewed above, this study seeks to implement more direct measures of accountability, which

are expected to capture more closely the extent to which the mechanisms in question are

9 The indicators used to measure different accountability features face validity issues, since some of them may only weakly capture the underlying mechanisms of accountability theorized in the literature. The author does acknowledge that pitfall in her concluding remarks (Camaj 2013, 37).

13

really effective and not just potential. A similar effort is made regarding the indicator for the

dependent variable, which follows on the contribution of other studies employing concrete

measures of corruption from audits, instead of measures of perceived corruption more

commonly used in comparative contexts. The details on the empirical strategy and the data

collected for the estimation procedures are described in the next section.

4. Empirical strategy and data collection

In the empirical analysis presented here, the goal is to estimate the effect of accountability on

future levels of corruption, at the same time taking into consideration the functioning of

electoral, social and horizontal accountability mechanisms. As discussed in the previous

section, the existence of accountability mechanisms is expected to produce a deterrent effect

on corrupt behavior based on the threat of sanctions. Therefore, corruption levels at a certain

point in time are influenced by politicians’ expectation of punishment in the future. At the

same time, the concrete functioning of accountability mechanisms informs politicians about

the probability of sanctioning, allowing them to update their expectations and adapt their

behavior accordingly. Therefore, in contexts where accountability mechanisms are seen to be

effective we should expect political actors to update the expected probability of sanctioning

upwards10 and refrain more from corruption then in contexts where the same mechanisms are

perceived to be less effective. Based on this dynamic, the general hypothesis considered for

the analysis is the following: future levels of corruption where accountability is effective will

be lower than where it is ineffective.

In order to test this hypothesis empirically, this study relies on a dataset of 140 Brazilian

municipalities. The choice of this setting for the empirical analysis was driven mainly by the

availability of unique corruption data from a randomized audit program conducted by the

Federal government since 2003. These municipalities constitute a sub-sample of all

municipalities audited until 2013, and they share one central common feature: they have been

targeted by at least two audits throughout the period in question. This makes it possible to

estimate corruption levels for these municipalities for at least two points in time. The measure

of corruption drawn from the first audit is used mainly to distinguish the municipalities where

corruption has been found from those where no corruption has been detected. This distinction

10 This argument would probably not apply in contexts where the probability of effective accountability is already very high. However, the analysis conducted here considers the context of a relatively young democracy where accountability mechanisms can be understood as undergoing a process of consolidation, which makes this assumption more plausible.

14

is important because the accountability indicators, at least in the case of electoral and

horizontal accountability, are conceptualized to reflect the concrete occurrence of sanctions.

Therefore, the occurrence of corruption in the first period audited is considered as a pre-

condition for a meaningful measure of accountability, and only the municipalities for which

this condition is met are included in the sample used. The measure of corruption from the

second audit is then used as the dependent variable for the regression analysis conducted. By

using the corruption estimate from the second period, we can examine whether there are

significant differences in the level of corruption observed in municipalities where electoral,

horizontal and social accountability are effective and where they are not.

The use of future corruption levels as a dependent variable also has the purpose of

circumventing some of the endogeneity problems that exist in analyzing an effect of

accountability on corruption. Since accountability is expected to affect corruption ex ante, and

accountability in the form of sanctioning can only happen if corruption has taken place, these

two factors are not independent from each other. More importantly, the occurrence of

corruption is affected by the expectation of sanctions in the future. The second measure of

corruption, on the other hand, is posterior to the occurrence of effective accountability, and it

arguably does not affect accountability ex ante.

The reasons to rely on data from these Brazilian municipalities are manifold. Firstly, the

random selection of the municipalities in the audit program creates important advantages for

the empirical analysis, avoiding some potential pitfalls associated with selection bias in the

conduction of the audits. Secondly, the fact that all audits are conducted by the same external

agency strengthens comparability across all units of analysis. Thirdly, the possibility of further

exploring a concrete indicator of corruption is an important resource for research in this field

as an alternative to other common perception-based indicators. The indicator developed for

this study also allows the analysis to focus on instances of corruption that are more relevant to

the relationship of interest, namely the impact of accountability on corruption. Broader

perception-based measures are likely to pool together perceptions of corruption both

involving higher political offices and low-ranking officials, which would not be ideal for the

analysis in question, since the latter are rarely subject to electoral accountability, and are also

less likely to be actively targeted by agents of social accountability. Finally, complementary

sources for other variables relevant to the study offer great richness of data for the sub-

national level in Brazil, thus providing an advantageous setup with enough information on a

number of potential covariates of corruption and the different dimensions of accountability, at

15

the same time controlling for other country-specific characteristics that would need to be

taken into account in a cross-national comparison.

The following sub-sections discuss in more detail the respective data sources used and, in the

case of the corruption data, the contextual information on the audit program and also the

coding procedure applied.

a. The Federal audit program: background information and criteria for the sample selection

In 2003, the Federal Comptroller’s Office (CGU), the equivalent to an anti-corruption agency in

Brazil, launched a large scale program targeted at auditing the application of federal funds

transferred to municipalities. The selection of the municipalities is conducted through a lottery

system. Since it was first implemented, the program has audited around 2000 municipalities in

a universe of about 5,600 Brazilian municipalities, selected in a total of 40 lottery draws. Each

lottery typically selects 60 municipalities with a maximum population of 500,000 inhabitants,

all state capitals excluded11, taking into account a pre-defined number of municipalities to be

selected from each of the 26 states12, in order to keep some degree of proportionality. Once a

municipality is selected, it is taken out of the pool for the three subsequent draws. The exact

timing of the lottery is not previously known and is announced by the CGU only shortly prior to

its taking place.

The analysis conducted here considered only the municipalities selected in audit rounds until

2013 for its sample selection. More specifically, it focused on a sub-group of 227 municipalities

that have been audited more than once since the program started, which are taken as the

potential sample for the study. However, since electoral accountability can materialize only at

specific points in time, the actual sample of analysis was further restricted to those cases

where it could have concretely occurred, given the following conditions: (a) an election for

municipal office took place between audits 1 and 2, and (b) the mayor in power during audit 1

ran for reelection–a narrow interpretation of electoral accountability–or, when the mayor was

11 Earlier lottery rounds had slightly different rules, and the number of municipalities selected gradually increased from only 5 in the first edition to 26 in the second, 50 from the third to the ninth rounds, and finally 60, which has been the standard since then. Also, the population threshold was 250,000 or 300,000 in earlier rounds, and was eventually increased as well.

12 An exception is made with regards to one group of five less populated northern states, from which a total of two municipalities from two different states is selected in each round.

16

not eligible to run again13, either a candidate from the same party, a relative of the mayor14 or

a member of the administration (typically a cabinet member) was presented as candidate for

succession–a broader interpretation of electoral accountability. Based on these criteria, 140

possible observations were identified15, with 91 cases in which mayors themselves were

running for reelection, and 49 cases with a succession candidate instead.

The adoption of both a narrow and a broad interpretation of potential electoral accountability

serves mainly the purpose of allowing for a larger alternative sample of cases, but it also

makes it possible to analyze the data according to two different sets of assumptions regarding

the incentives faced by incumbent politicians. Political agency models are often structured as a

finite two-period game where the incumbent takes an action in period 1, which is followed by

an election where voters can decide on whether to keep or replace the incumbent, and a

second period in which the election winner, either the reelected politician or an elected

challenger, decides on a new action (Ashworth 2012; Besley 2006; Fearon 1999). According to

this model structure, electoral accountability can only happen between periods 1 and 2, and

this would correspond to the narrower interpretation of reelection mentioned above.

However, although useful for analytical purposes, the assumptions behind this model structure

overlook the fact that politicians may face a longer time horizon, and even if incumbent

mayors themselves are not directly eligible for reelection, they still have a stake in their party’s

or political group’s remaining in power. This can be the case for well-intentioned politicians,

who have an interest in the continued implementation of the political and ideological platform

that they support, and even more so in the case of rent-seeking politicians, who would profit

from continued access to public resources through their political allies. Therefore, it is also

reasonable to consider electoral accountability from the perspective of an infinitely repeated

game where an incumbent party or political group subjects a candidate to the approval of the

popular vote for continuing in office (Ferejohn 1986), which corresponds to the broader

interpretation of electoral accountability discussed for the cases at hand.

13 In Brazil, mayors are elected for a four-year term and are allowed to run for only one consecutive term after that. They may run again only after a hiatus of four years in which another mayor has been in power.

14 It is not uncommon in Brazilian municipalities for local politics to be marked by the dominance of certain political dynasties that remain in power for longer periods of time. In such cases, incumbent mayors that cannot be reelected often present a relative or even their spouse as a candidate to continue their “legacy” in office.

15 A few other cases were excluded in which audit 1 took place before an election, but its results were released only shortly before or after election day, since the relevant information about corruption could not have reached voters. Cases in which the subsequent election had only one registered candidate were also disregarded.

17

b. A corruption measure derived from the audit reports

For the 140 municipalities selected, the next step was to extract measures of corruption from

the irregularities identified in the first and second audits16. The detailed audit reports for each

municipality selected in the program are readily available on the CGU website17. For each

audit, a number of service orders is issued corresponding to specific federal programs and

projects for which funds were transferred to the municipal administration. The time scope of

each of those service orders tends to cover mainly two to three years prior to the audit, but

there is some variation and the administrative period covered can be extended to include also

funds related to projects executed in earlier years. In terms of policy areas, the audits usually

cover a broad range of programs implemented under different ministries, in particular those

related to voluntary transfers for education, healthcare and social policies.

Each report describes the individual irregularities found by the auditors in the municipality.

These may be related to the lack of formal compliance with federal regulations in the

application of transferred funds, to unsatisfactory quality in public service provision or to the

ineffectiveness of institutionalized social monitoring mechanisms. What is of particular interest

for the development of a corruption indicator is the fact that auditors often uncover cases of

non-compliance with the legislation that signal favoritism or even outright fraud in the use of

public resources, which can be considered as signs of corruption.

Following previous work that has used concrete indicators of corruption retrieved from the

same sources (Ferraz and Finan 2008, 2011), the data constructed for this study is based on

the coding of irregularities related to procurement fraud, diversion of public funds, and over-

invoicing as instances of corruption. This puts an emphasis on corruption instances that may

be characterized as political corruption, understood as “[…] the behavior of public decision-

makers where preferential treatment is provided to individuals and where narrow interests are

advanced at the expense of the interests of broader segments of society” (Lambsdorff 2007,

82). As argued earlier, this conceptualization of corruption is considered as more relevant for

the study of a deterrent effect of accountability.

Cases of procurement fraud include in essence all situations in which procurement regulations

were circumvented in order to avoid the required competitive awarding procedures, restrict

16 Although a small group of municipalities was audited also a third or even a fourth time, only the first and second audits were considered. For one observation, the corruption estimates were taken from audits 2 and 3 instead, because there was no election between audits 1 and 2 and they were conducted with a relatively short interval between each other.

17 http://sistemas.cgu.gov.br/relats/relatorios.php

18

competition among potential bidders or favor particular suppliers and service providers.

Instances of diversion of funds encompass, besides instances of embezzlement by authorities,

also cases where the provision of purchased goods or hired services did not take place or could

not be confirmed, the destination of payments could not be identified, or evidence of fraud in

the documentation of expenditures was found (e.g. forged invoices). Over-invoicing refers

largely to cases where goods were purchased or services hired at prices above market level, or

when the municipal administration paid for higher amounts than what was in fact delivered18.

The main corruption indicator computed for each municipality in each audit period is simply a

count of all the individual instances of corruption identified in the audit reports, according to

the classification listed above19. As the reports often cover more than one administrative term,

some temporal distinctions were made to ensure conceptual consistency for the analysis. For

the reports from the first audit, the corruption estimate applied to the analysis of the narrow

sample includes only violations from the period in which the mayor running for reelection later

was in power. In the case of the broader sample, violations from previous terms are

considered only when there is administrative continuity from one term to the next, based on

the same criteria used regarding the potential of electoral accountability. For the corruption

indicator extracted from the second audit report, the number of violations includes only those

related to the post-electoral period.

The coding procedure applied here includes also an additional criterion related to the cases of

procurement fraud, namely that the instances counted are individual procurement procedures

associated with the kind of irregularities mentioned above, and not the occurrences as

presented in the report. This distinction is considered necessary because the entries associated

with procurement irregularities oftentimes describe flaws found in more than one

procurement procedure. Therefore, considering each entry in the report as one instance of

procurement fraud would introduce considerable measurement error in the corruption

18 The concrete situations that exemplify each of the cases considered are numerous and cannot be fully described here, but a complete list can be obtained from the author.

19 Previous studies (Ferraz and Finan 2008, 2011) have also explored the share of funds linked to corruption violations in the total amount of audited funds as an alternative measure of corruption. This was also calculated for the sample considered in this study, but was not used in the estimation procedures due to reliability concerns. As many of the reports fail to include this information while describing some of the irregularities, considerable discrepancy is observed between this indicator and the number of violations found. Therefore, the latter was considered as a more accurate measure of the extent of corruption in each municipality.

19

indicator. This aspect was thus taken into account in the coding process in order to more

accurately reflect the extent of corruption found in the municipality20.

c. Accountability measures and data sources

As mentioned earlier, the indicators for the three dimensions of accountability considered in

this study seek to capture, whenever possible, the extent to which those mechanisms of

accountability are effective in generating sanctions against corrupt politicians. However, there

are some challenges in finding measures that adequately reflect that aspect of how

accountability mechanisms function.

The operationalization of electoral accountability is the most straightforward and more closely

reflects whether mayors whose administration had corruption violations in the audit 1 were

effectively sanctioned by voters. Based on data from electoral records from the Superior

Electoral Court (TSE), the analysis identified the mayors in power during the administrative

term in which the first audit took place21. For the cases in which any corruption violation was

found in the first period, it was then checked simply whether the mayor running for

reelection–in the case of the narrower sample–or the candidate presented for succession of

the administration at the time–in the case of the broader sample–succeeded in securing

another term in the subsequent election, which occurred in 2004, 2008 or 2012, depending on

when the municipality was first audited. The resulting binary variable takes a value of 0 if the

incumbent mayor/administration was successful and a value of 1 if a challenger candidate won

the election.

In the case of horizontal accountability, the measure applied in the analysis also attempts to

reflect whether the related institutional mechanisms are effective in punishing mayors

potentially involved in corruption cases. For this purpose, the operationalization of this

variable takes into account three main sanctioning mechanisms that Brazilian mayors can face,

20 This introduces some differences between the indicator estimated here and similar indicators used in previous studies based on the same audit reports. Ferraz and Finan (2011), for instance, find the mean of irregularities per municipality to be around 2.5, and Brollo (2010) estimates 1.8 as the mean number of violations in her sample. The data collected for the present study, on the other hand, finds the mean number of corruption violations at 11.6 for the second audit period, which is the indicator used as main dependent variable.

21 This information was cross-checked with other official records in order to confirm whether the elected mayors remained in power throughout the term. For the cases in which there was a change in power (e.g. when the elected mayor died or resigned, or when the election results were annulled and a new election took place), the mayor in power on the election year was considered as the relevant incumbent for the analysis.

20

and which are considered to be particularly relevant for their calculations of risks associated

with corruption. The most clear-cut of these is the imposition of criminal sanctions for

malfeasance in office, when characterized either according to corruption-related provisions of

the Penal Code or as so-called “responsibility crimes”, applicable to holders of Executive office

at different levels of the public administration. Additionally, any politician in an Executive

office is subject to civil prosecution under the Administrative Improbity Law for actions

resulting in undue private advantage or enrichment through their office, either for themselves

or for other private actors. In this case, direct culpability or involvement is not necessary, as

they may also be considered liable by omission. This type of civil procedure can lead to

penalties such as the return of lost or diverted funds to municipal coffers and the suspension

of political rights for up to 10 years, which imposes a particularly high cost on politicians’

career prospects. Finally, mayors are required to submit a yearly report of municipal accounts

and expenditures to the respective State Court of Accounts, and to the Federal Court of

Accounts in the case of federal funds received. In case these institutions reject the accounts

due to grave administrative irregularities, office holders may also become ineligible to run for

political office for eight years22, which similarly represents a relevant threat to their career

ambitions.

However, there is one fundamental empirical challenge in devising an indicator that will fully

capture the occurrence of effective legal sanctions. If we employ a strict notion of effective

horizontal accountability in the form of a definitive sanction to capture the causal mechanism

of deterrence that is hypothesized, the mayor(s) in power in the period covered by audit 2

would need to have observed the conviction of the former mayor prior to that period, so that

the level of corruption observed in the second period can reflect a change in behavior as a

consequence of having observed that punishment. Nevertheless, for a majority of the

municipalities considered this is not a realistic expectation, given that the time frame between

the audits is too short for such a conviction to take place. Therefore, the measure used in the

analysis applies a less strict criterion, which considers as effective horizontal accountability

when the mayor in power during audit 1 has faced criminal or civil prosecution associated with

corruption-related charges, and prosecution has been initiated prior to the administrative

period covered in the second audit. Even though this does not necessarily reflect the

occurrence of definitive legal sanctions, it is taken as a sign that horizontal accountability is

22 This time period was defined by changes to this legislation introduced in 2010. Before that, the law established a period of five years.

21

more effective than where such investigative procedures have not been established23. With

regards to the sanctioning mechanisms of Courts of Accounts, this consideration does not

apply, and the cases taken as an example of effective horizontal accountability include simply

those where the accounts have been rejected before the period covered in audit 2.

For the coding of the horizontal accountability variable, a search with the names of the mayor

in power in the first audit period was conducted on the databases of both the Federal and the

respective State courts, in order to identify ongoing or closed criminal and civil proceedings

(restricted to those related to administrative improbity) in which they appear as defendants.

Complementary information from the associated court rulings was considered to ensure that

the lawsuits in question were related to corruption-related actions corresponding to the same

categories considered for the corruption indicator used in the analysis24, i.e. accusations

related to diversion of public funds, procurement fraud or over-invoicing in public

expenditures. In the case of Court of Accounts, a similar search was conducted on databases of

previous rulings by those instances, accompanied by a check on whether the accounts for any

of the years in the first audit period had been rejected, attributing responsibility to the mayor

at the time. Considering the information from these different sources, the observations for

which the mayor in question was prosecuted, or had the accounts rejected for any of the years

in power, received a value of 1, otherwise the variable was assigned a value of zero.

For social accountability, three separate indicators were considered to operationalize distinct

facets of this form of vertical accountability. The main limitation faced here is that none of the

available measures concretely reflects the extent to which civil society and media actors

effectively monitor and expose corruption in the local administration. The only available

indicators thus cannot completely fulfill the goal of capturing effective social accountability,

but are considered as important measures of whether certain conditions are in place for social

accountability to be effectively exercised.

The first indicator used is the number of private non-profit associations registered in each

municipality, as previous studies have also applied a similar measure for comparative purposes

23 It is important to acknowledge that the prosecution of individuals accused of corruption does not fully reflect the probability that a conviction will take place. In the Brazilian context, cases where such lawsuits are extinguished due to procedural flaws or statute of limitations are not uncommon, and the overall effectiveness of the judicial system in punishing corruption is rather low (de Alencar and Gico Jr. 2011). However, the indicator chosen should to some degree reflect the variation in effectiveness of horizontal accountability across the units of analysis, since the lack of such procedures would indicate an even less effective functioning of this mechanism of accountability.

24 In the cases in which no additional information on the substance of the case was available, the lawsuit was disregarded as evidence of effective horizontal accountability.

22

at the cross-national level (Grimes 2013; Mungiu-Pippidi 2011) This data is available from the

Brazilian Institute of Geography and Statistics (IBGE) with estimates for 2002, 2005, 2006, 2008

and 2010. Unfortunately, a change in the methodology of classification of non-profit

organizations since 2008 creates some problems for comparability across all data points.

Therefore, the dataset includes data on this variable only for the municipalities where the

election year after the first audit was either 2004 or 2008, for which the data points used were

those from 2002 and 2006, respectively, therefore from two years prior to the election of

interest. The indicator for the remaining five municipalities in the sample for which the

relevant election year was 2012 was considered as missing data.

The second indicator used reflects the presence of local media in each municipality. Following

previous related studies, emphasis was put in the presence of local radio stations. This has

been argued to be a more relevant source of political information than printed media in the

Brazilian context, given the relatively low level of education and literacy in the country (Brollo

2010; Ferraz and Finan 2008, 2011). This is also a binary variable that distinguishes between

municipalities where at least one local AM or FM radio station is present from those where

none are established. The coding of the variable was based on information from municipal

profile surveys published by IBGE. Between 2001 and 2012, the Institute has conducted almost

on a yearly basis (with the exceptions of 2003 and 2007) nation-wide surveys with municipal

administrations in order to collect a wide range of data on all municipalities. The surveys from

2001, 2005, 2006 and 2009 include information on the existence of some local media outlets in

the municipalities. In order to have data temporally equidistant from the election years of

reference for the municipalities in the sample, the data points from 2001, 2005 and 2009 were

selected for election years 2004, 2008 and 2012, respectively. Therefore, this data also related

to the pre-electoral period.

The third indicator for social accountability devised in the study was based on content from

the CGU audit reports as well. As mentioned earlier, the municipal audits also examine the

effectiveness of an institutionalized mechanism of social accountability in the decentralized

system of policy implementation in Brazilian municipalities, namely social oversight councils

that are legally required under federal guidelines for a number of policy areas. These councils

have the function of monitoring policy implementation both at the administrative level and at

the users’ end. Moreover, their existence is a condition for the continued transfer of funds for

regular federal programs in the areas of health, education and social assistance. Their

composition varies according to the specific regulations for the related program, but as a rule

includes representatives of civil society (usually about 50% of the councils’ members), the

23

municipal government, the local legislative council and sometimes also of the program’s direct

users or beneficiaries25. In order to avoid capture by the local Executive, government

representatives are precluded from presiding the council. All members have a two-year term.

Moreover, the councils are usually required to meet on a monthly basis to discuss issues

related to the respective policy area or program (e.g. school meals, healthcare services,

program against child labor) and verify the application of transferred funds and the conduction

of procurement procedures. Finally, the council is also required to discuss and issue a report

approving or not a yearly financial report submitted by the mayor for the respective subject

areas or programs.

Among the irregularities described in each audit report, there are also items that refer to non-

compliance to the regulations determining the functioning of these councils, or to factors that

lead to an ineffective exercise of their monitoring role. These are often associated with

insufficient physical and financial resources for the council, lack of training for its members,

over-representation of the municipal government (e.g. when part of the members that are

supposed to represent civil society are also municipal employees), or limited access to relevant

information on how program’s are managed and executed. Sometimes, councilors are also

pressured by the mayor to approve the program’s accounts without the chance of properly

assessing them.

Two coding systems were applied to produce an indicator of effectiveness of these councils.

First, a binary variable was composed distinguishing municipalities with no irregularities in the

functioning of any council from those in which any irregularity was found. However, this

method did not generate any meaningful variation, since it became clear that the institution of

the social oversight councils is not fully effective across all policy areas in virtually any of the

municipalities in the sample. An alternative indicator was then produced based on the number

of councils with irregularities mentioned in each audit report. However, this indicator is

essentially a measure of the ineffectiveness of councils. In order to allow for a more intuitive

interpretation, the scale of the measure was reversed, so that it expresses a degree of

effectiveness in the functioning of those bodies, even though the numbers lose their

substantive meaning with that procedure.

This indicatorwas also included in the analysis in an attempt to propose an alternative new

measure of social accountability that also captures the effective existence of participatory

monitoring channels in the local administration. However, one pitfall regarding this indicator is

25 http://www.portaldatransparencia.gov.br/controlesocial/ConselhosMunicipaiseControleSocial.asp

24

that it may not be completely exogenous to the corruption level observed in the municipality.

Indeed, the data collected shows a rather strong correlation between this indicator and the

number of corruption violations found in the same audit report. However, this may not

necessarily be an indication that more effective social councils successfully curb corruption,

but rather that more corrupt administrations have incentives to sabotage the functioning of

the councils. As a matter of fact, the types of irregularities observed by the auditors in their

functioning include examples of that. Therefore, in order to avoid inference errors based on

this information, the indicator for the effectiveness of social councils from the first audit report

was used as a potential explanatory factor for the level of corruption found in the second

audit, as to minimize endogeneity concerns. Although this does not fully ensure time

consistency between these two indicators, since the level of effectiveness observed in the first

period does not necessarily reflect whether the councils will function as (in)effectively in future

administrative terms, the expectation is that the indicator also captures underlying differences

in the disposition of civil society to use this institutional mechanism to check on the

government.

These indicators of different accountability mechanisms should, at least to some degree,

capture how effective the mechanisms of electoral, horizontal and social accountability de

facto work in the different municipalities. They are seek to make a contribution to new

approaches for measuring this characteristic of democratic regimes, in comparison to other

indicators of accountability that take into consideration only the formal existence of certain

institutional mechanisms. The data collected here for Brazilian municipalities indeed illustrate

how actual variation on the functioning of different dimensions of accountability may emerge

even in an environment where formal institutional uniformity based on national-level

regulation exists.

d. Additional factors: controlling for alternative explanations

In addition to these main independent variables of interest, the empirical analysis controls for

other factors that may exercise confounding effects on corruption. One important alternative

mechanism to be considered is the selection aspect associated with electoral accountability. If

voters can also use elections to select more competent politicians, based on the information at

hand, and corrupt behavior is correlated with politicians’ competence, then it is possible that

part of the effect of electoral accountability is in fact capturing this selection effect (Ferraz and

Finan 2011). For this purpose, previous political experience in office is used a proxy for the

25

competence of the mayor elected in the election after audit 1. This is measured as a binary

variable indicating whether the candidate in question has been a mayor before, based on

electoral results from the TSE database and also on a complementary research on their

political profiles on media reports. Although the winning candidate in that election was not

necessarily in office for the whole period covered by the second audit, this indicator should at

least partially capture a potential selection effect of electoral accountability.

A second mechanism related to models of political agency is the predicted effect of term limits

on corrupt behavior. Since term-limited mayors are expected to have fewer incentives to

refrain from corruption than first-term mayors, an indicator of whether the mayor in the

period of the second audit was in his/her first or second term is also included in the data

analysis for the narrow sample, i.e. the ones in which mayors themselves were running for

reelection. For the broader sample, this does not apply because no lame-duck effect is

assumed.

Yet a third mechanism that might affect both corruption and electoral accountability is

controlled for, namely the potential effect that other practices associated with elections lead

to an increase in corruption26. Electoral competition, through political finance, could also

strengthen the influence of private interests on the government, and thus motivate more

particularistic decisions on the part of incumbent politicians in benefit of their financial

supporters. We could thus expect candidates with higher amounts of campaign funds to be

potentially under more pressure to repay for campaign donations with political favors when in

power. At the same time, campaign funds have been shown as positively associated with the

reelection chances of candidates in Brazil (Rennó Jr. 2008), and are likely to affect voter

behavior by increasing the visibility of candidates. Therefore, the amount of campaign funds

raised by the winning candidate in the election of interest is included as a control variable. This

information is originally available from TSE as well, but the data used in the analysis was

extracted from a database published by the non-profit organization Transparência Brasil27,

which offers easier access to political finance data in the country.

26 The literature on democracy and corruption also includes arguments challenging the view that electoral competition has predominantly positive effects on corruption. It is claimed that electoral competition and the “uncertainty” associated with it may create both opportunities and incentives for politicians to subject to increasing pressures from business or to engage in electoral corruption through vote-buying, clientelism and illegal party-financing in order to maximize voter support (Blake and Morris 2009; Moran 2001; Rose-Ackerman 1999).

27 Their project Às Claras (http://www.asclaras.org.br/@index.php) includes data on campaign funds for all candidates in national and local elections from 2002 to 2012.

26

Other potential determinants of corruption are included in the analysis. Firstly, the amount of

rents available is discussed in political agency models as a relevant factor influencing

incumbent behavior: the more rents available, the more likely are incumbents to engage in

corruption and take advantage of potential rents in the present. As a proxy for this factor, the

share of intergovernmental transfers in municipal revenue is used, since additional resources

in the form of intergovernmental transfers are mentioned in the relevant literature as

impacting both the electoral performance of incumbents and the level of corruption observed

in Brazilian municipalities (Brollo et al. 2010; Litschig and Morrison 2010). This indicator was

calculated based on data available from the National Treasury Department (STN) on municipal

finances. As this information is only available until 2011, the data point included in the analysis

refers to the year before the election year of interest for each municipality. Another indicator

considered refers to the literature that finds an impact of revenue from the exploitation of

natural resources on corruption levels28. For the municipal level, the share of natural resource

royalties in municipal revenue is used as a proxy for that, and this is also calculated based on

STN’s financial data. Analogously, the availability of these additional resources in the budget

could also impact the reelection prospects of incumbent mayors.

Political factors that potentially reduce the constraints to corrupt behavior at the local level

are also taken into consideration. The level of electoral competition, for instance, may be

negatively related to corruption: mayors who face more competition should be more inclined

to refrain from corruption, since the risk of losing office in the next elections is higher than for

those who face less competition from challengers. In order to control for this effect, the

margin of victory for the winning mayor in the election of interest considered as a control. The

legislative support that the mayor enjoys in the local legislative chamber may also reduce the

constraints on corrupt behavior, and for this reason an indicator for the share of seats held by

the winning candidate’s party was calculated from data on the results of local legislative

elections.

A number of additional controls are considered in order to capture potential unobserved

differences across the units of analysis. Municipal characteristics such as the year the

municipality was founded, total population, share of urban population, indicators of education

level and municipal GDP are taken from data published by IBGE. Alternatively, a municipal

Human Development Index for Brazilian municipalities, published by the United Nations

28 The impact of natural resources has also been considered in some cross-national analyses of determinants of corruption (Montinola and Jackman 2002; Treisman 2000, 2007). Moreover, anecdotal evidence suggest that cases of corruption are common in oil-producing municipalities in Brazil (Caselli and Michaels 2009).

27

Development Program (UNDP) based on 2000 and 2010 Census data, is included in the

analysis. Regional dummies are also considered in some specifications to control for remaining

regional discrepancies. Additionally, the election year of interest is included as a control for

time-related factors that may simultaneously affect the functioning of accountability

mechanisms and corruption outcomes.

5. Preliminary results

This section presents preliminary results from the empirical analysis based on a portion of the

dataset produced for this study. For 20 of the 140 observations considered as the broader

sample for the analysis, the coding of the audit reports that serve as reference for both the

corruption data and the measure for effective social oversight councils has not yet been

completed, therefore the preliminary analysis presented here considers only the remaining

120 cases. Moreover, 14 observations for which no corruption violations were identified in the

first audit report were dropped, as the disclosure of corruption in the first period is an

important premise for the occurrence of electoral and horizontal accountability as

operationalized in the study. As a consequence, the estimation procedures described here

were based on data for 106 observations composing the broader sample considered, i.e.

including not only cases where the mayor was running for election, but also where a candidate

for succession was presented. This first examination of the data considers only the broader

sample, as the narrower sample consists of only 66 observations at this point, and it will be

examined at a later stage once the data for all the 140 cases has been collected.

For analytical and comparative purposes, Brazil is often divided into five main regions: North

(N), Northeast (NE), Southeast (SE), South (S) and Center-West (CO). The observations included

in the present analysis cover four of those regions, excluding the South. This distinction is

useful because it reflects a number of essential geographic and demographic differences that

also correlate with relevant social-economic indicators relevant in the analysis at hand. The

Northeastern region, for instance, includes some of the poorest areas of Brazil, partly

explained by the dry climate that characterizes this region. The Southeastern states, on the

other hand, show higher levels of economic and social development, and are also among the

most industrialized. Moreover, they may capture cultural differences that cannot be accounted

for by other indicators available.

Regarding the election year of reference for each municipality, for a slight majority of the cases

the relevant election occurred in 2004, and for most of the remaining cases in 2008, except for

28

four observations with the relevant election taking place in 2012. As mentioned earlier, the

data on civil society organizations was considered as not available for the latter observations,

and for this reasons the models including this variable include only 102 observations. Table 1

shows the distribution of observations across the four regions and the election years

considered.

[Table 1 here]

The sample used in the analysis presents substantial variation is all the main variables of

interest. This is true also for the additional variables considered as controls for alternative

explanations, and for the municipal characteristics that define some contextual aspects related

to the cases at hand. This constitutes another advantage of working with a sample of Brazilian

municipalities, because the within-country diversity in terms of social and economic

characteristics offers a setting that to a large extent mimics cross-national variation, therefore

offering considerable potential for external validity of the findings and applicability beyond the

sub-national level. Table 2 shows descriptive statistics for all the variables considered in the

analysis29, with the main control variables divided into different groups. Some of the variables,

such as number of non-profit organizations, municipal GDP and campaign revenue are

presented not only in absolute values, but also adjusted for differences in municipal

population.

[Table 2 here]

Two other sets of variables are included as controls. The first one refers to audit characteristics

related to the scope of the second audit, and which impact the corruption measure: the length

of the time period (in months) covered by the audit after the election of interest, and the

number of service orders analyzed by the auditors. Intuitively, it can be expected that audits

covering a longer period of time and a broader range of programs are more likely to find more

corruption violations. Indeed, both these factors have a moderate positive correlation with the

corruption measure. The other set of controls includes variables indicating whether the mayor

ran for reelection or a successor candidate was presented, and whether the second audit took

place in the term immediately after the election of interest or only after a hiatus of one

administrative term. These two variables are also positively correlated with the corruption

measure and, therefore, are taken into account in the estimated models.

29 A more detailed definition, with the data sources and respective time periods considered for each variable, can be found in the Appendix.

29

A first look at the bivariate relationships between the dependent and the independent

variables offers initial insights on the validity of the general hypothesis presented earlier,

namely that future levels of corruption where accountability is effective will be lower than

where it is ineffective. This can be broken down into specific hypothesis for each one of the

indicators of accountability considered:

H1: Corruption in the postelection period will be lower in municipalities where the previous

administration involved in corruption was voted out of office.

H2: Corruption in the postelection period will be lower in municipalities where the previous

mayor involved in corruption has faced civil or criminal prosecution for corrupt behavior,

or has had the municipal accounts rejected or disapproved by the State or Federal Court of

Accounts.

H3a: Corruption in the postelection period will be lower in municipalities where more

municipal social oversight councils function more effectively.

H3b: Corruption in the postelection period will be lower in municipalities where media

presence is stronger.

H3c: Corruption in the postelection period will be lower in municipalities with higher civil

society density.

Additionally, based on previous findings by Grimes (2013) related to a conditioning effect of

media freedom on the impact of civil society on corruption outcomes, a complementary

hypothesis is proposed to verify whether a similar interaction effect is observed with regards

to media presence:

H3d: The deterrent effect of organized civil society on future corruption levels is

conditional on media presence.

Table 3 presents the bivariate estimates for each of the accountability variables, and models

including measures for all three dimensions of accountability, first alternating the different

indicators for social accountability and then adding combinations of those as well. The

measure for civil society density used in these first iterations is weighted by 1000 inhabitants.

Similarly, one of the models includes the interaction term the combinations of the civil society

measure and radio presence. We can see that a significant bivariate association between

stronger accountability and corruption outcomes can be observed only in the case of civil

society density and radio presence, both when these variables are included separately as

predictors of corruption and when electoral and horizontal accountability are taken into

30

account. Once both these indicators of social accountability are included in the models, their

effect remains significant at the 90% confidence level. However, the hypothesized interaction

effect is not present. The third measure for social accountability in turn shows no significant

association with corruption levels in the post-election period. It is also interesting to notice

that, even though not statistically significant, the coefficient estimates for horizontal

accountability have a consistent positive sign, contrary to the expectation from the hypothesis

proposed.

[Table 3 here]

Additional models were estimated introducing the different sets of controls listed in table 2.

Election year, audit characteristics and group controls were included in all model

specifications. The different political factors were also added, but only political experience and

campaign revenue had an impact on the estimates for the independent variables of interest in

the different specifications tested. From the municipal characteristics, geographic area and

demographic density did not show any significant impact on the effect of the accountability

variables either. The remaining variables were found to be the most relevant controls for

estimating the effect of accountability on corruption and are thus included in the models

reported below. Additionally, regional fixed-effects are taken into account to capture potential

unobserved differences across the four regions in the data. As some of the controls variables

are strongly skewed towards lower values, log-transformed values were used in the models to

avoid distortions due to outliers and to produce more efficient estimates.

Table 4 shows the estimates also for each accountability variable individually and for the

various combinations taking into account the three dimensions of accountability and the

different indicators of social accountability. These models include a maximum of 100

observations, because values for the resource variables are not available for all municipalities

in the sample. Given a strong correlation between radio presence and population, the

specifications shown below have log population as a control variable, thus the variables

related to the number of non-profit organizations, campaign revenue and municipal GDP

include absolute values instead of the population-adjusted alternatives30. The inclusion of the

main control variables does not substantially alter the result for electoral accountability, which

still presents no statistically significant effect on future corruption levels. The same is true of

30 The population-adjusted measures, i.e. non-profit organizations per 1000 inhabitants, campaign funds per capita and GDP per capita, were included in other models without the population measure, but this specification appears to bias the coefficient estimates for civil society and radio presence. These results are not reported, but can be obtained from the author.

31

the horizontal accountability indicator, which keeps a positive sign across all specifications. The

civil society and radio presence variables still have statistically significant negative coefficients

when they are included in the models separately, but in the models in which both are included

as predictors, their effect loses statistical significance at conventional levels. The measure of

effective social oversight councils continues to show no significant effect on corruption, but

now the coefficients for this variable also display a positive sign, contrary to our expectation.

The estimates for an interaction effect, on the other hand, are statistically significant when the

relevant control variables are introduced. Results from different specifications with literacy

rate as an alternative measure of education and with municipal HDI replacing municipal GDP

as an indicator of development are quite similar to those reported in the table.

[Table 4 here]

A closer examination is necessary to substantively interpret the interaction between civil

society density and radio presence estimated in Model 10. The coefficient for the civil society

variable is not statistically significant, meaning that the effect of civil society on corruption in

the absence of radio is not statistically distinguishable from zero. However, a significant

negative effect of civil society density can be observed in the municipalities where radio is

present; the coefficient estimate for the marginal effect of civil society in that case is

approximately -6.52. Since the civil society variable is in log units, the coefficient can be

interpreted as a reduction of 0.065 corruption violations for a 1% increase in the number of

non-profit organizations in the municipality. More meaningfully, if the number of organizations

would increase by half, the impact on future corruption levels would amount to a reduction of

about 3.25 violations in municipalities where radio is present, with all other variables kept

constant. The difference in the marginal effect is illustrated in Figure 1.

[Figure 1 here]

Similarly, the deterrent effect of radio presence is conditioned by civil society density. In this

case, the statistical significance of the coefficient for the radio variable shows that it would

even have a positive effect on corruption in the absence of an organized civil society. However,

this is somewhat misleading, because an estimation of the marginal effect of radio presence

for varying levels of civil society density shows that, apart from the statistically significant in

the case of complete absence of non-profit organizations, radio presence significantly impacts

future corruption levels only for values equal or above 3.6 units in the civil society variable,

which represent an approximate number of 37 organizations, in a sample where the mean

number of non-profit organizations is 55, and the median is 30. Moreover, the estimated

32

effect is quite substantial: at this threshold point, the presence of local radio stations is

associated with a reduction of 7 violations in future corruption levels, and the effect gradually

increases with higher civil society density. The variation in the marginal effect of radio

presence for different degrees of civil society density is illustrated in Figure 2.

[Figure 2 here]

A couple of robustness checks were conducted in order to verify whether these results hold for

different specifications, or are excessively influenced by extreme values in some of the

variables. Firstly, an analysis of dfbeta values for all variables in the different models was

conducted, in order to identify influential observations. For two municipalities dfbeta values

were above the usually adopted cut-off point of one standard deviation for some of the

control variables and for radio presence. Based on this assessment, the models presented

above were estimated again, excluding both influential observations. These results reveal that

the previously identified effect of civil society density is largely driven by those influential

cases, as the coefficients for this variable are not longer statistically significant. The estimates

for an effect of radio presence, on the other hand, are consistently significant in models with

only this indicator of social accountability, and remain significant at the 90% confidence level

also when civil society and effective social councils are taken into account. Regarding the other

accountability indicators, the measure for effective social councils is still positive and not

statistically significant, as is the case with the horizontal accountability measure. For electoral

accountability, the coefficients are not statistically significant either, but now they take a

positive sign, contrary to the theoretical prediction. These results are summarized in table 5.

[Table 5 here]

Some evidence of an interaction effect remains, but only regarding the effect of radio as

conditioned by the existence of a minimal degree of civil society density. As illustrated in

Figure 3, the impact of civil society density in not distinguishable from zero regardless of radio

presence. Figure 4 illustrates the marginal effect of radio presence on corruption, given

different values of the civil society density variable. Here we see that radio presence has a

statistically significant negative effect on future corruption outcome for a value equal to or

above 2.9 units in the logged civil society variable, which de-transformed to the regular scale

represents about 18 non-profit organizations, therefore a value much below the mean and

median for the sample at hand. At this level of civil society density, the estimated effect of

radio presence is -3.9, thus somewhat weaker in magnitude than in the previous models, and it

also increases with higher numbers of non-profit organizations.

33

[Figures 3 and 4 here]

A second robustness check conducted was the estimation of the same models specifications

presented earlier, but with the log transformation of the dependent variable to correct for

skewness in the distribution and minimize the impact of outliers. In this iteration, there is

again a negative estimated effect of electoral accountability, but it is not statistically significant

in any of the models. Horizontal accountability still presents positive coefficients, but they are

not statistically significant at conventional levels. The effect of social councils is similar to what

was previously observed, with positive coefficients and not statistically significant at the 95%

confidence level. Regarding the effect of civil society density, we see that the negative effect

observed originally in the models displayed in table 4 is not fully confirmed by the models with

the alternative dependent variable, as the coefficients for this variable are not statistically

significant once radio presence is taken into account in the model. The effect of radio, on the

other hand, is statistically significant for the models in which the other indicators for social

accountability are not included, and remains significant at the 90% confidence level in the

complete models. In terms of magnitude, these coefficients can be interpreted as representing

a reduction of about 36% in future corruption levels, as the dependent variable is expressed in

log units. Additionally, the model with the logged dependent variable does not support the

hypothesis of an interaction between radio presence and civil society density. The results for

this set of regression models are displayed in table 6.

[Table 6 here]

6. Conclusion

The different estimation approaches described above present only limited preliminary

evidence for a deterrent effect of accountability on future corruption levels. Similar to

previous studies examining this for the case of electoral accountability (Bobonis et al. 2012;

Crisp et al. 2014), the analysis based on data from Brazilian municipalities does not support the

claim that punishment by voters against corrupt incumbents contributes to reduce corruption

in the long run. None of the estimated models shows a statistically significant effect for this

variable, and in some iterations the signs of the coefficients become unstable.

The case of horizontal accountability is somewhat puzzling: all of the estimated models show a

positive relationship between effective horizontal accountability and future corruption levels,

although not statistically significant at conventional levels. One possible explanation for this

pattern could be related to the operationalization of this variable. Given the empirical

34

limitations to measure effective sanctions against mayors accused of corruption, the data

collected for the horizontal accountability indicator reflects, in the vast majority of cases, only

the potential for sanctions through existing legal procedures. This is apparently not sufficient

to trigger a deterrent effect and, therefore, may not be an adequate measure to capture the

hypothesized mechanism. This may also be associated to contextual factors linked to general

deficiencies in the Brazilian judicial system and the oversight role exercised by the Courts of

Accounts (Abrucio and Loureiro 2005; de Alencar and Gico Jr. 2011).

The case for a deterrent effect of social accountability appears to be somewhat stronger,

mainly regarding the effect of media presence. The models estimated so far as part of this

study render support to a negative effect of radio presence on future corruption levels, and

this effect has both statistical (although only at the 90% level) and substantial significance in

most of the specifications tested. Part of the models suggests that this may be conditional on a

minimum degree of civil society density, but this is not robust to all of the tests conducted. In

the case of civil society, a negative effect on corruption is identified, conditioned on the

presence of radio, but this is not robust to different specifications and appears to be driven by

a few influential observations. With respect to an effect of social oversight councils, the

empirical evidence does not show any significant influence of this variable on future corruption

levels.

These preliminary empirical results emphasize the potential role of mechanisms of social

accountability in curbing political corruption in democratic contexts. Further research efforts

are necessary to shed light on the impact of this dimension of accountability on corrupt

behavior, and also to develop better ways to measure it empirically. At the same time, the

findings presented here further suggest that a deterrent effect of electoral accountability

should not be taken for granted in discussions about the effect of democracy on corruption.

Studies engaged with that research problem must consider alternative causal mechanisms,

related to different dimensions of accountability, as explored here, but also to other potential

explanations that may contribute to the understanding of why democratic regimes tend to be

associated with lower corruption.

The present analysis will be further developed with the full sample of municipalities devised

for the study, and also with the application of alternative estimation strategies. These shall

include estimations with both the narrower and the broader sample of municipalities, and also

an analysis of the data structured as an unbalanced panel of post-electoral terms as

observations, since the corruption indicator from the second audit covers more than one

35

administrative term for part of the municipalities. This would not only provide for a larger

number of observations and for two distinctive periods of time in the post-electoral period

examined, but would also contribute to improve time consistency for some of the variables

used in the analysis.

7. References

Abrucio, Fernando Luiz, and Maria Rita Loureiro. 2005. Finanças Públicas, Democracia E Accountability: O Debate Teórico E O Caso Brasileiro.

Adsera, Alicia, Carles Boix, and Mark Payne. 2003. “Are You Being Served? Political Accountability and Quality of Government.” Journal of Law, Economics, and Organization 19(2): 445–90.

De Alencar, Carlos Higino Ribeiro, and Ivo Gico Jr. 2011. “Corrupção E Judiciário: A (in)eficácia Do Sistema Judicial No Combate À Corrupção.” Revista Direito GV 7(1): 75–98.

Andvig, Jens C. et al. 2001. Corruption. A Review of Contemporary Research. Chr. Michelsen Institute.

Andvig, Jens C., Odd-Helge Fjeldstad, Inge Amundsen, and Tina Søreide. 2000. Research on Corruption. A Policy Oriented Survey.

Ashworth, Scott. 2012. “Electoral Accountability: Recent Theoretical and Empirical Work.” Annual Review of Political Science 15(1): 183–201.

Bäck, Hanna, and Axel Hadenius. 2008. “Democracy and State Capacity: Exploring a J‐Shaped Relationship.” Governance 21(1): 1–24.

Bågenholm, Andreas. 2013. “Throwing the Rascals out? The Electoral Effects of Corruption Allegations and Corruption Scandals in Europe 1981-2011.” Crime, Law and Social Change 60: 595–609.

Besley, Timothy. 2006. Principled Agents? New York: Oxford University Press.

Blake, Charles H., and Stephen D. Morris, eds. 2009. Corruption & Democracy in Latin America. Kindle edi. Pittsburgh: University of Pittsburgh Press.

Bobonis, Gustavo J., Luis R. Cámara Fuertes, and Rainer Schwabe. 2012. Working Papers, Banco de México The Dynamic Effects of Information on Political Corruption: Theory and Evidence from Puerto Rico. Mexico City.

Bovens, Mark. 2006. 13 European Governance Papers Analysing and Assessing Public Accountability. A Conceptual Framework.

36

Brollo, Fernanda. 2010. Boston University-Department of Economics Who Is Punishing Corrupt Politicians: Voters or the Central Government? Evidence from the Brazilian Anti-Corruption Program.

Brollo, Fernanda, Tommaso Nannicini, Roberto Perotti, and Guido Tabellini. 2010. The Political Resource Curse. Bonn, Germany.

Brunetti, Aymo, and Beatrice Weder. 2003. “A Free Press Is Bad News for Corruption.” Journal of Public Economics 87(7-8): 1801–24.

Camaj, Lindita. 2013. “The Media’s Role in Fighting Corruption: Media Effects on Governmental Accountability.” The International Journal of Press/Politics 18(1): 21–42.

Caselli, Francesco, and Guy Michaels. 2009. NBER Working Paper Series Do Oil Windfalls Improve Living Standards? Evidence from Brazil.

Chang, Eric C. C., Miriam A. Golden, and Seth J. Hill. 2010. “Legislative Malfeasance and Political Accountability.” World Politics 62(02): 177–220.

Crisp, Brian F., Santiago Olivella, Joshua D. Potter, and William Mishler. 2014. “Elections as Instruments for Punishing Bad Representatives and Selecting Good Ones.” Electoral Studies 34: 1–15.

Dimock, Michael A., and Gary C. Jacobson. 1995. “Checks and Choices: The House Bank Scandal’s Impact on Voters in 1992.” The Journal of Politics 57(04): 1143–59.

Eggers, Andrew, and Alexander Fisher. 2011. Electoral Accountability and the UK Parliamentary Expenses Scandal: Did Voters Punish Corrupt MPs?

Fearon, James D. 1999. “Electoral Accountability and the Control of Politicians: Selecting Good Types versus Sanctioning Poor Performance.” In Democracy, Accountability, and Representation, eds. Adam Przeworski, Susan C. Stokes, and Bernard Manin. Cambridge, UK: Cambridge University Press, 55–97.

Ferejohn, John. 1986. “Incumbent Performance and Electoral Control.” Public Choice 50: 5–25.

Ferraz, Claudio, and Frederico Finan. 2008. “Exposing Corrupt Politicians: The Effects of Brazil’s Publicly Released Audits on Electoral Outcomes.” The Quarterly Journal of Economics (May).

———. 2011. “Electoral Accountability and Corruption: Evidence from the Audits of Local Governments.” American Economic Review 101(June): 1274–1311.

Geddes, Barbara, and Artur Ribeiro Neto. 1992. “Institutional Sources of Corruption in Brazil.” Third World Quarterly 13(4): 641–61.

Grant, Ruth W., and Robert O. Keohane. 2005. “Accountability and Abuses of Power in World Politics.” American Political Science Review 99(01): 29–43.

37

Grimes, Marcia. 2013. “The Contingencies of Societal Accountability: Examining the Link Between Civil Society and Good Government.” Studies in Comparative International Development 48: 380–402.

Harriss-White, Barbara, and Gordon White. 1996. “Corruption, Liberalization and Democracy.” IDS Bulletin 27(2): 1–5.

Jiménez, Fernando, and Miguel Caínzos. 2004. “La Repercusión Electoral de Los Escándalos Políticos. Alcance Y Condiciones.” Revista Española de Ciencia Política 10: 141–70.

Kitschelt, Herbert, Kent Freeze, Kiril Kolev, and Yi-Ting Wang. 2009. “Measuring Democratic Accountability: An Initial Report on an Emerging Data Set.” Revista de Ciencia Política (Santiago) 29(3): 741–74.

Kolstad, Ivar, and Arne Wiig. 2011. CMI Working Paper Does Democracy Reduce Corruption? Bergen, Norway.

Lambsdorff, Johann Graf. 2007. The Institutional Economics of Corruption and Reform: Theory, Evidence and Policy. Cambridge: Cambridge University Press.

Lancaster, Thomas D., and Gabriella R. Montinola. 1997. “Toward a Methodology for the Comparative Study of Political Corruption.” Crime, Law and Social Change 27(3): 185–206.

Lederman, Daniel, Norman V. Loayza, and Rodrigo R. Soares. 2005. “Accountability and Corruption: Political Institutions Matter.” Economics and Politics 17(1): 1–35.

Litschig, Stephan, and Kevin Morrison. 2010. Government Spending and Re-Election.

Mainwaring, Scott, and Christopher Welna, eds. 2003. Democratic Accountability in Latin America. Oxford University Press.

Manin, Bernard, Adam Przeworski, and Susan C. Stokes. 1999a. “Elections and Representation.” In Democracy, Accountability, and Representation, eds. Adam Przeworski, Susan Stokes, and Bernard Manin. Cambridge, UK: Cambridge University Press, 29–54.

———. 1999b. “Introduction.” In Democracy, Accountability, and Representation, eds. Adam Przeworski, Susan C. Stokes, and Bernard Manin. Cambridge, UK: Cambridge University Press, 1–26.

Maravall, José María, and Ignacio Sánchez-Cuenca, eds. 2008. Controlling Governments: Voters, Institutions, and Accountability. Kindle e-b. Cambridge University Press.

Mohtadi, Hamid, and Terry L. Roe. 2003. “Democracy, Rent Seeking, Public Spending and Growth.” Journal of Public Economics 87(3-4): 445–66.

Montinola, Gabriella R., and Robert W. Jackman. 2002. “Sources of Corruption: A Cross-Country Study.” British Journal of Political Science 32(01): 147–70.

38

Moran, Jon. 2001. “Democratic Transitions and Forms of Corruption.” Crime, Law & Social Change 36: 379–93.

Mulgan, Richard. 2000. “‘Accountability’: An Ever-Expanding Concept?” Public Administration 78(3): 555–73.

Mungiu-Pippidi, Alina. 2011. Contextual Choices in Fighting Corruption: Lessons Learned.

O’Donnell, Guillermo. 1994. “Delegative Democracy.” Journal of Democracy 5(1): 55–69.

———. 1999. “Horizontal Accountability in New Democracies.” In The Self-Restraining State, eds. Andreas Schedler, Larry Diamond, and Marc F. Plattner. London: Lynne Rienner Publishers, 29–51.

———. 2003. “Horizontal Accountability: The Legal Institutionalization of Mistrust.” In Democratic Accountability in Latin America, eds. Scott Mainwaring and Christopher Welna. Oxford: Oxford University Press, 34–54.

Olken, Benjamin A. 2007. “Monitoring Corruption: Evidence from a Field Experiment in Indonesia.” Journal of Political Economy 115(2): 200–249.

Pellegata, Alessandro. 2012. “Constraining Political Corruption: An Empirical Analysis of the Impact of Democracy.” Democratization (June): 1–24.

Persson, Torsten, Gerard Roland, and Guido Tabellini. 1997. “Separation of Powers and Political Accountability.” The Quarterly Journal of Economics 112(4): 1163–1202.

Peruzzotti, Enrique, and Catalina Smulovitz, eds. 2006a. Enforcing the Rule of Law: Social Accountability in the New Latin American Democracies. Kindle edi. University of Pittsburgh Press.

———. 2006b. “Social Accountability.” In Enforcing the Rule of Law: Social Accountability in the New Latin American Democracies, eds. Enrique Peruzzotti and Catalina Smulovitz. University of Pittsburgh Press, 3–33.

Peters, John G., and Susan Welch. 1980. “The Effects of Charges of Corruption on Voting Behavior in Congressional Elections.” The American Political Science Review 74(3): 697–708.

Philp, Mark. 2009. “Delimiting Democratic Accountability.” Political Studies 57(1): 28–53.

Przeworski, Adam, Susan C. Stokes, and Bernard Manin, eds. 1999. 5 Democracy, Accountability, and Representation. Cambridge, UK: Cambridge University Press.

Rennó Jr., Lúcio Remuzat. 2008. “Rewarding the Corrupt? Reelection and Scandal Involvement in the Brazilian 2006 Legislative Elections.” Colombia Internacional 68: 98–106.

Rock, Michael T. 2007. DESA Working Paper Corruption and Democracy. Bryn Mawr, PA, USA.

Rose-Ackerman, Susan. 1978. Corruption: A Study in Political Economy. New York: Academic Press.

39

———. 1999. Corruption and Government. Causes, Consequences and Reform. New York: Cambridge University Press.

Saha, Shrabani. 2008. “Democracy and Corruption: An Empirical Analysis in a Cross-Country Framework.” In New Zealand Association of Economists Annual Conference, Wellington, 9-11 July,.

Sandholtz, Wayne, and William Koetzle. 2000. “Accounting for Corruption: Economic Structure, Democracy, and Trade.” International Studies Quarterly 44(1): 31–50.

Schedler, Andreas. 1999. “Conceptualizing Accountability.” In The Self-Restraining State, eds. Andreas Schedler, Larry Diamond, and Marc F. Plattner. London: Lynne Rienner Publishers, 13–28.

Serra, Danila. 2004. Empirical Determinants of Corruption : A Sensitivity Analysis.

Sung, Hung-En. 2004. “Democracy and Political Corruption: A Cross-National Comparison.” Crime, Law and Social Change 41(2): 179–93.

Treisman, Daniel. 2000. “The Causes of Corruption: A Cross-National Study.” Journal of Public Economics 76(3): 399–457.

———. 2007. “What Have We Learned About the Causes of Corruption from Ten Years of Cross-National Empirical Research?” Annual Review of Political Science 10(1): 211–44.

Weyland, Kurt Gerhard. 1998. “The Politics of Corruption in Latin America.” Journal of Democracy 9(2): 108–21.

Whitehead, Laurence. 2002. “High Level Political Corruption in Latin America: A ‘Transitional’ Phenomenon?” In Political Corruption: Concepts and Contexts, eds. Arnold J. Heidenheimer and Michael Johnston. New Brunswick, NJ: Transaction Publishers.

40

Tables and Figures

Table 1. Distribution of cases per region and election year of reference

Region Frequency Percent

CO 12 11.32

N 19 17.92

NE 59 55.66

SE 16 15.09

Total 106 100

Election year Frequency Percent

2004 54 50.94

2008 48 45.28

2012 4 3.77

Total 106 100

41

Table 2. Descriptive statistics

Variable N Mean Median St. Deviation Min Max

Dependent variable

N. of corruption violations in audit 2 106 11.6 9 10.63 0 59

Independent variables

EA: mayor voted out 106 0.42 0 0.50 0 1

HA: mayor prosecuted/accounts rejected 106 0.40 0 0.49 0 1

SA: N. of non-profit organizations 102 55.1 30 72.16 2 392

SA: N. of non-profit organizations per 1000

inhabitants

102 2.21 2.11 1.28 0.1 5.68

SA: radio presence 106 0.54 1 0.50 0 1

SA: effective social councils in audit 1 106 3.7 4 1.65 0 7

Audit 2 characteristics

Time period covered (in months) 106 29.3 28 19.92 2 79

Number of audit service orders 106 26.6 24 11.00 8 73

Group controls

Successor candidate 106 0.38 0 0.49 0 1

Term hiatus between election and audit 2 106 0.29 0 0.46 0 1

Political factors

Election winner margin of victory 106 15.20 10.76 15.61 0.26 89.84

Election winner previous office experience 106 0.54 1 0.50 0 1

Election winner campaign revenue 106 158459 86808 259920.80 0 2080263

Election winner campaign revenue per

capita

106 7.30 5.62 6.32 0 37.59

Election winner legislative support (% of

party seats)

106 26.6 22.20 15.74 0 77.8

Resources

% of transfers in municipal revenue 100 91.03 93.94 8.54 47.87 99.36

% of natural resource royalties in municipal

revenue

100 0.82 0.50 2.43 0 23.2

Municipal characteristics

% of urban population 106 57.9 58.4 21.31 15.5 100

% of literate population 106 71.6 71.4 12.17 47.5 93.1

% of population with at least 8 years of

education

102 18.0 15.8 8.8 3.51 42.8

Education component of HDI 106 0.30 0.30 0.11 0.102 0.609

Municipal HDI 106 0.48 0.47 0.09 0.293 0.717

Municipal GDP (1000 BRL) 106 245415.2 86323 436476.40 6092 2617476

Municipal GDP per capita 106 7410.73 5295 5587.23 2214 42602

Year of foundation 106 1961.85 1959 20.51 1933 1997

Population 106 27888.7 16136.5 34788.21 2019 186357

Municipal area 106 2085 587.5 4291.19 33 33595

Demographic density 106 64.5 26.8 127.72 0.7 906.9

42

Table 3. Bivariate estimates

DV: Number of corruption violations in 2nd audit

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

EA: Mayor voted out -0.216 -0.538 -0.0999 -0.387 -0.520 -0.583 -0.699

(2.050) (2.025) (1.971) (2.077) (1.984) (2.066) (2.033)

HA: mayor prosecuted/accounts rejected 1.406 1.670 2.211 1.401 2.071 2.045 2.024

(2.077) (2.017) (1.978) (2.077) (1.978) (2.028) (1.967)

SA: CSOs per 1000 inhabitants -2.373** -2.428*** -1.782* -2.038 -1.751*

(0.906) (0.909) (1.011) (1.752) (0.993)

SA: radio presence -5.709*** -5.994*** -4.518* -5.344 -4.298*

(2.050) (2.012) (2.299) (5.522) (2.449)

SA: effective social councils -0.468 -0.482 -0.407

(0.586) (0.590) (0.635)

CSOs x radio presence 0.404

(2.196)

Observations 106 106 102 106 106 102 106 106 102 102 102

Adjusted R-squared -0.010 -0.005 0.073 0.064 -0.004 0.061 0.056 -0.019 0.091 0.082 0.085

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

43

Table 4. Models with control variables

DV: Number of corruption violations in 2nd audit

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

EA: Mayor voted out -1.460 -1.809 -1.382 -1.102 -1.844 -1.738 -1.695

(2.220) (2.122) (2.160) (2.167) (2.198) (2.134) (2.240)

HA: mayor prosecuted/accounts rejected 2.395 1.763 2.815 2.367 2.211 2.036 2.325

(2.272) (2.128) (2.205) (2.220) (2.153) (2.051) (2.179)

SA: log CSOs -4.181** -4.211** -3.350 -0.692 -3.345

(1.762) (1.803) (2.038) (1.998) (2.052)

SA: radio presence -4.880* -5.282** -3.751 13.99** -3.769

(2.610) (2.476) (2.990) (6.878) (3.004)

SA: effective social councils 0.324 0.343 0.468

(0.581) (0.594) (0.638)

CSOs x radio presence -5.828**

(2.455)

Observations 100 100 96 100 100 96 100 100 96 96 96

Adjusted R-squared 0.245 0.254 0.276 0.285 0.243 0.272 0.290 0.240 0.286 0.351 0.281

Controls

Election year Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Audit characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Group controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Political factors Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Resources Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Municipal characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Region fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Election year distinguishes between observations for which the election after the 1st audit was in 2004, 2008, and 2012. Audit characteristics include (a) the length (in months) of the administrative period covered by the 2nd audit and (b) the total number of audit service orders. Group controls includes (a) whether a successor candidate was presented and (b) whether audit 2 occurred with a term hiatus after the election of interest. Political factors include (a) log campaign funds for the winning candidate in the reference election year (constant 2012 BRL) and (b) whether elected candidate had previous office experience. Resources include (a) log % of intergovernmental transfer in municipal revenue and (b) log % of natural resources royalties in municipal revenue. Municipal characteristics include (a) % of urban population, (b) education component of municipal HDI, (c) log GDP (constant 2012 BRL), (d) year of foundation and (e) log population. Regional fixed effects identify regions as N, NE, SE or CO.

44

Table 5. Models excluding influential observations

DV: Number of corruption violations in 2nd audit

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

EA: Mayor voted out 0.0597 0.00606 0.0517 0.184 0.101 0.0674 0.231

(1.817) (1.850) (1.792) (1.838) (1.844) (1.832) (1.875)

HA: mayor prosecuted/accounts rejected

0.869 0.894 1.225 0.905 1.239 1.157 1.305

(1.769) (1.752) (1.701) (1.741) (1.766) (1.742) (1.766)

SA: log CSOs -1.660 -1.617 -0.563 0.777 -0.557

(1.279) (1.310) (1.565) (1.756) (1.573)

SA: radio presence -4.757** -4.860*** -4.383* 5.637 -4.436*

(1.815) (1.832) (2.255) (5.268) (2.256)

SA: effective social councils 0.168 0.184 0.365

(0.452) (0.463) (0.533)

CSOs x radio presence -3.303*

(1.887)

Observations 98 98 94 98 98 94 98 98 94 94 94

Adjusted R-squared 0.320 0.322 0.311 0.383 0.321 0.295 0.372 0.306 0.333 0.359 0.328

Controls

Election year Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Audit characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Group controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Political factors Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Resources Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Municipal characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Region fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Election year distinguishes between observations for which the election after the 1st audit was in 2004, 2008, and 2012. Audit characteristics include (a) the length (in months) of the administrative period covered by the 2nd audit and (b) the total number of audit service orders. Group controls includes (a) whether a successor candidate was presented and (b) whether audit 2 occurred with a term hiatus after the election of interest. Political factors include (a) log campaign funds for the winning candidate in the reference election year (constant 2012 BRL) and (b) whether elected candidate had previous office experience. Resources include (a) log % of intergovernmental transfer in municipal revenue and (b) log % of natural resources royalties in municipal revenue. Municipal characteristics include (a) % of urban population, (b) education component of municipal HDI, (c) log GDP (constant 2012 BRL), (d) year of foundation and (e) log population. Regional fixed effects identify regions as N, NE, SE or CO. Influential observations 12 (Marituba-PA) and 21 (Alto Alegre do Pindaré-MA) excluded from the analysis.

45

Table 6. Models with logged dependent variable

DV: Log number of corruption violations in 2nd audit

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

EA: Mayor voted out -0.0674 -0.0703 -0.0562 -0.0268 -0.0736 -0.0710 -0.0598

(0.171) (0.172) (0.172) (0.172) (0.176) (0.177) (0.178)

HA: mayor prosecuted/accounts rejected 0.229 0.207 0.266 0.237 0.250 0.246 0.261

(0.166) (0.162) (0.166) (0.164) (0.166) (0.165) (0.165)

SA: log CSOs -0.243** -0.234** -0.151 -0.0868 -0.151

(0.106) (0.102) (0.119) (0.131) (0.118)

SA: radio presence -0.373** -0.408** -0.360* 0.0687 -0.362*

(0.168) (0.158) (0.186) (0.451) (0.184)

SA: effective social councils 0.0464 0.0502 0.0431

(0.0473) (0.0475) (0.0538)

CSOs x radio presence -0.141

(0.150)

Observations 100 100 96 100 100 96 100 100 96 96 96

Adjusted R-squared 0.419 0.434 0.394 0.454 0.425 0.395 0.463 0.429 0.420 0.418 0.418

Controls

Election year Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Audit characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Group controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Political factors Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Municipal characteristics Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Resources Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Region fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Election year distinguishes between observations for which the election after the 1st audit was in 2004, 2008, and 2012. Audit characteristics include (a) the length (in months) of the administrative period covered by the 2nd audit and (b) the total number of audit service orders. Group controls includes (a) whether a successor candidate was presented and (b) whether audit 2 occurred with a term hiatus after the election of interest. Political factors include (a) log campaign funds for the winning candidate in the reference election year (constant 2012 BRL) and (b) whether elected candidate had previous office experience. Resources include (a) log % of intergovernmental transfer in municipal revenue and (b) log % of natural resources royalties in municipal revenue. Municipal characteristics include (a) % of urban population, (b) education component of municipal HDI, (c) log GDP (constant 2012 BRL), (d) year of foundation and (e) log population. Regional fixed effects identify regions as N, NE, SE or CO.

46

Figure 1. Marginal effect of civil society density conditional on radio presence

Figure 2. Marginal effect of radio presence according to number of non-profit organizations

47

Figure 3. Marginal effect of civil society density conditional on radio presence, without influential observations

Figure 4. Marginal effect of radio presence according to number of non-profit organizations, without influential observations

48

Appendix – Description of variables

Variable Description Source and time frame

Future corruption Number of corruption violations in audit 2 (for post-election period)

Own calculations based on CGU audit reports

Electoral accountability

Whether incumbent mayor/administration during audit 1 was voted out of office in subsequent election

Electoral records (TSE)

Horizontal accountability

Whether incumbent mayor during audit 1 faced civil or criminal prosecution for corrupt behavior, or had the municipal accounts rejected by the State or Federal Court of Accounts

Federal and State Court records; court of accounts' records

Social accountability: civil society density

Number of non-profit entities (missing data for municipalities with 2012 as election year)

National Statistics office (IBGE) - data from 2 years before the election in question

Social accountability: civil society density (adjusted for population)

Number of non-profit entities per 1000 inhabitants (missing data for municipalities with 2012 as election year)

Own calculation based on National Statistics office (IBGE) - data from 2 years before the election in question.

Social accountability: radio presence

Whether an AM or FM radio station was present in the municipality

Municipality profiles from National Statistics office (IBGE). Data from 3 years before the election year in question (2001, 2005 and 2009, respectively).

Social accountability: effective social councils

Reversed scale of number of ineffective councils found in the 1st audit

Own calculations based on CGU audit reports

Audit time period Number of months covered by audit 2 Own calculation based on time frame covered by CGU audits

Audit service orders

Number of service orders audited in audit 2 CGU reports

Successor candidate

Mayor in period 1 running for reelection = 0; Successor candidate for the administration in period 1 = 1

Own assignment based on electoral records (TSE) and additional research

Term hiatus Whether audit 2 occurred with a term hiatus after the election of interest (No = 0; Yes = 1)

Own assignment based on timing of audits

Election winner margin of victory

Margin of victory of winning candidate (difference between share of votes of winner and 1st runner-up candidate) in election of interest

Own calculation based on electoral records (TSE)

Election winner previous office experience

Whether candidate who won in election of interest (after 1st audit) served any previous term as mayor

Search on official publications, biographical research

Election winner campaign revenue

Total campaign funds of candidate who won in election of interest (after 1st audit)

Electoral campaign finance data (TSE), compiled by Project Às Claras (Transparência Brasil)

Election winner campaign revenue per capita

Per capita campaign funds of candidate who won in election of interest (after 1st audit)

Own calculations based on electoral campaign finance data (TSE), compiled by Project Às Claras (Transparência Brasil)

49

Election winner legislative support

Share of municipal councilors from the same party as the mayor elected in election of interest

Own calculation based on electoral records (TSE)

Share of transfers Current transfers (including all intergovernmental transfers and earmarked grants - 'convenios') as share of total current revenue

Own calculations based on data from National Treasury (STN) - data from year prior to electoral year in question. Data available only until 2011

Share of natural resource royalties

Compensation funds for natural resource exploitation as share of total current revenue

Own calculations based on data from National Treasury (STN) - data from year prior to electoral year in question. Data available only until 2011

Share of urban population

Share of urban population National Statistics office (IBGE) - data from 2000 census for election years 2004 and 2008; data from 2010 census for election year 2012

Share of literate population

Share of literate population National Statistics office (IBGE) - data from 2000 census for election years 2004 and 2008; data from 2010 census for election year 2012

Share of population with at least 8 years of education

Share of population with 10 years or older with at least 8 years of education (equivalent to basic education in Brazil)

National Statistics office (IBGE) - data from 2000 census for election years 2004 and 2008. Missing data for 2012.

Education component of HDI

Education component of Municipal Human Development Index. Divergence regarding global HDI: no data on average schooling years for population of 25 or older is available, so they consider share of population of 18 and older with basic education (8 years)

UNDP Brazil, based on census data from 2000 and 2010. Therefore, data from 2000 for election years 2004 and 2008; data from 2010 for election year 2012

Municipal HDI Municipal Human Development Index (data from 2000 census for election years 2004 and 2008; data from 2010 census for election year 2012)

UNDP Brazil, based on census data from 2000 and 2010. Therefore, data from 2000 for election years 2004 and 2008; data from 2010 for election year 2012

Municipal GDP Municipal GDP estimate in 1000 BRL, current 2012 prices

National Statistics office (IBGE) - data from the year prior to the first year covered by the 2nd audit

Municipal GDP per capita

Municipal GDP per capita, current 2012 prices Own calculations based on data from National Statistics office (IBGE) - data from the year prior to the first year covered by the 2nd audit

Year of foundation

Year in which municipality was founded IBGE Municipal Profiles 2004

Population Population estimate National Statistics office (IBGE) - data from the year prior to the first year covered by the 2nd audit, referent to July 1st). Exceptions: for audits beginning in 2008, data from 2006 was taken, and for municipalities with audit beginning in 2011, data for 2009 was used - 2007 population count and 2010 Census data left out to keep the same methodology

Municipal area Municipal geographic area in km2 IBGE basic municipal data

Demographic density

Number of inhabitants per km2 Own calculations based on IBGE population estimate