Exposing corruption: How does electoral competition affect ...

Exposing corruption: How does electoral competition affect politician behavior?

Farzana Afridia Amrita Dhillonb

Eilon Solanc

June 30, 2016

Abstract

We build a simple game theoretic model to capture the effect of electoral competition on re-election concerns when there is public exposure of corruption through mandated audits of government expenditures. The setting is one where two parties compete in elections, however, the incumbent candidate is fixed to begin with. We show that in equilibrium, corruption has a non-monotonic relationship with electoral competition. If the election is safe for the incumbent (low competition) or if it is extremely fragile (high competition) then corruption is higher, but for intermediate levels of competition, corruption is low. We use the model’s predictions to test our hypotheses using data gathered on audit findings of a large public program from Indian villages in a southern state during 2006-10 and on elections to the village council headship in 2006. Our results largely confirm the theoretical results, but also suggest that the impact of electoral competition varies by whether theft is from the public or private component of a good.

Keywords: Corruption, Electoral Competition, Audit

JEL Classification: H4, H7, I3

_________________ The authors thank seminar participants at the ISI (Delhi), the NOPOOR conference (Delhi) for comments and suggestions. This paper has been supported by CAGE, PPRU, NOPOOR and the IGC. aEconomics and Planning Unit, Indian Statistical Institute, New Delhi. bDepartment of Political Economy, King's College London. cSchool of Mathematical Sciences, Tel Aviv University.

1. Introduction

Politicians can transform policies and institutions. Evidence suggests that if political

leadership is committed, corrupt institutions can change rapidly (Svensson, 2005; Acemoglu

and Jackson, 2011). Therefore, the persistence and high incidence of corruption in low

income democracies (Transparency International, 2015; Svensson, 2005) is puzzling: if

electoral competition is fair, and if voters care about honesty, voters can punish corrupt

incumbents by voting them out of office. Forward looking incumbents will then respond to

these incentives by lowering corruption. We investigate this question in a context where legal

sanctions for punishing the corrupt are not binding but there exists mandated exposure of

corruption through audits of public expenditures. Can mandated exposure of corruption, in

the presence of electoral competition, discipline politician behavior?

It is fairly well established that corruption is costly, both in terms of efficiency and

equity in the provisions of government services in developing countries (Pande and Olken,

2012). From a theoretical point of view, if we assume that an incumbent’s utility increases in

bribes taken and the length of time in office, then her effort will be a function of her inherent

ability and desire to control corruption, which itself depends on the link between the cost of

the effort (or the illegal gains from office) and the chances of being thrown out of office by

voters. The behavior of an incumbent politician between elections can be disciplined by the

threat of exposing corrupt behavior and voting her out, but too much discipline might lead to

a breakdown of the implicit contract between voters and incumbent. It is often pointed out

that being corrupt may itself lead to re-election benefits.

In this paper we investigate the link between political competition and corrupt

behavior when corruption is revealed publicly through mandated audits. We define the

governance as the effort that an incumbent leader puts into preventing illegal personal gains

by political or bureaucratic functionaries. The higher is this effort, the lower would be the

leakages from public funds and higher the quality of governance. In the data, therefore, we

can observe the politician’s effort in terms of the level of corruption in her constituency while

electoral political competition can be represented by how closely fought an election was or

the margin of electoral victory.

Ades and Di Tella (1997) provide evidence that shows that if there is low political

competition, then the quality of governance is low. On the other hand, more competition does

not necessarily lead to higher quality of governance. For example, Aidt et al.(2009) point out

that “In India, we appear to be witnessing the nearly wholesale takeover of large chunks of an

established democratic political system by persons with criminal records. Moreover, this is

occurring even as Indian party politics becomes increasingly competitive”. Moreover,

electoral competition may not reduce corruption but change its nature (Booth et al., 2011) -

with competition, vote buying becomes more attractive than other types of rent seeking.

There is thus little consensus in the literature on the link between quality of governance and

competition. As with economic competition, at first glance, welfare is expected to rise as

political competition increases. Highly competitive political environments lead to more

innovation and better economic outcomes (Acemoglu and Robinson, 2006). Besley et al.

(2010) show that for US municipalities, competition has a positive effect on economic

outcomes. Bannerjee and Pande (2009) point out that competition improves the quality of

governance when voters vote on multiple dimensions. Thus, there is agreement in the

literature that the disciplining factor for politicians is their incentives for re-election and thus

their self-perceived probability of winning again. Ferraz and Finan (2011), thus, find that

when the probability of winning the next election by incumbents is high corruption tends to

be lower in Brazilian municipalities.

However, benefits of ex-ante competition can be mitigated by various factors, such as,

resource constraint (release of funds from the central government especially if the local head

is not affiliated to the political party running the centre), asymmetric information regarding

costs of public projects and a legal system which provides immunity to politicians. For

instance, Bardhan and Yang (2004) develop a theoretical model that suggests that in the event

of asymmetric information between the politician and the electorate there exists an optimal

level of political competition beyond which aggregate welfare can fall due to reduced public

investments. Moreover, if the incumbent doesn’t deliver on a public program, a political

contestant can indulge in vote buying during elections and then try to recover that

expenditure by embezzlement of public funds during her term in office. For instance,

Persson, Roland and Tabellini (1997) refer to the need to allow some graft by the politician as

“power between elections”. Acemoglu and Robinson (2000 and 2002) investigate the

incentives to undertake investment when the short term effects of investment are

destabilizing, and political competition may exacerbate this problem (Aidt et al., 2009).

The innovation in our study is two-fold. First, unlike previous theoretical models

which assume asymmetric information – voters get information about incumbents from

signals such as actual spending on public good or from benchmarking with neighboring

jurisdictions (Besley and Case, 1995) – we study the effect of electoral competition in a

context where voters can observe the level of corruption by the incumbent. Recent literature

on disclosure of information on degree of corruption – audits make information available to

voters (Ferraz and Finan (2008), Bobonis et al. (2009). Second, we provide a general, but

nuanced framework for understand the underlying relationship between political competition

and governance by taking into account the possible non-monotonicity of this relationship and

the endogeneity of competition to corruption.

We construct a reduced form model to see what happens when a single agent

(incumbent) chooses how much to “steal” from a pie of size 1. In the next period there is an

election. Voters vote on the basis of ideology and the observed corruption level, relative to

the challenger who comes with a known ideology and an unknown reputation for corruption.

The trade-off is between stealing a lot now and getting voted out against any challenger, or

stealing less now and being able to stay in power for a longer time. We derive the stationary

equilibrium level of theft in office and show how it varies with the level of competition

measured by the closeness of the election. We can show that in this simple set up, even when

voters can perfectly observe corruption, too little or too much competition leads to higher

levels of theft, while a medium-level competition leads to lower levels of theft. Moreover, if

the electoral advantage is with the incumbent then as competition increases, corruption

decreases. If the electoral advantage is with the challenger, then increases in competition

have no impact on corruption. The story is thus more nuanced than it appears in the literature.

A critique of the reduced form model is that the challenger has a passive role: in

reality the challenger will propose platforms based on the competitiveness of the election. We

show that the same result can be obtained in a more realistic (and more complicated) set up

where there are both parties and candidates. While parties take into account the effect of

corruption on the probability of winning, they are composed of candidates who do not

internalize the effect of their own actions on the probability of winning and thus sometimes

the incentive constraint may be binding. Thus, we show that even though parties always

respond to higher competition by reducing corruption, the incentive compatibility constraints

imposed by candidates imply that corruption is also high when competition is too stiff.

We test our theoretical predictions using survey and secondary data on one of the

largest public sector projects in India: the National Rural Employment Guarantee Act

(NREGA) - a rights based program that aims to guarantee 100 days of annual work to rural

households willing to volunteer adult labor to rural public works. The major problem that

plagues studies on corruption is lack of objective measures of corruption. However, a novel

feature of the NREGA, distinct from previous workfare programs in India, is the mandatory

audits of projects implemented under the program. The Act envisages that competent audits

will be organized by the beneficiary households at regular intervals. The guidelines thus

make audits and stakeholder participation therein, the main mechanism to ensure

transparency and the accountability of those in charge of program implementation. As is true

with most public programs in developing countries, NREGA has also been besieged with

concerns about theft and leakage of public funds (Afridi and Iversen 2013).

Two sources of data are available for the empirical analysis. We use panel data on the

implementation of the program in the state of Andhra Pradesh during 2006-11, a pioneer in

institutionalizing audits of NREGA projects. We measure leakages of public funds using the

audit data as well as the program expenditures and employment generated. This is coupled

with data on village councils, responsible for planning and the subsequent execution of at

least 50 percent of all NREGA works in villages within its purview, and the direct elections

of village council heads for a five year term in 2006. In addition, we have also accessed data

on the candidates who stood for election and the closeness of the election at the sub-district

and district level for our sample. Finally, our data allow us to inquire whether electoral

competition affects the nature of corruption. All corruption may not be viewed as equally bad

from the voters’ point of view – the incumbent may prefer to pilfer from the public

component of the program rather than its private component when political competition is

perceived to be high (Olken, 2007).

The remainder of the paper is organized as follows. In section 2 we discuss our

theoretical model and its predictions which can be tested empirically. Section 3 provides the

contextual background for the empirical analysis. The data and methodology are discussed in

section 4. Section 5 examines the empirical findings and section 6 concludes.

Section 2: Model

In the one dimensional model of political competition, if voters are united against corrup-

tion and if corruption is observed, then parties should compete to be seen as less corrupt.

This logic breaks down when either of these assumptions is violated. E.g. if the level

of corruption is imperfectly observed then levels of corruption depend on the signalling

mechanism. The idea of asking whether electoral competition enhances accountability and

outcomes was raised quite early on in Besley et al (2008). Our model has imperfectly ob-

served corruption, audits as a monitoring tool, and competition between parties. In the

literature there are models that are similar in that they have both parties and candidates

(Besley (2008), Banerjee and Pande (2009)) but no incomplete information, models with

incomplete information but without competition (Ferraz and Finan (2011)), or models of

competition without audits (e.g. Svaleryd and Vlachos (2008)). None of the existing papers

combine the elements of party discipline and moral hazard with competition. There are also

papers that discuss how instability of a regime a↵ects corruption in a non monotonic way

(eg. Campante et al (2008)) but their driving force is quite di↵erent.

The model builds on the fact that elections at the gram panchayat (village) level are

run on party lines. The process involves selecting candidates to run for the local elections.

A large part of electoral success is based on the success of various public redistribution

programmes like NREGA. Parties are not individuals- they are composed of candidates who

may or may not follow what is in the party’s best interest - incumbent candidates do not

internalize the e↵ect of their own behaviour on the party reputation. We find that a result

of this incentive compatibility restriction on party’s reputation levels, sti↵ competition may

sometimes lead to worse outcomes in terms of corruption. The role of audits as a source to

prevent candidates from stealing even more is crucial- in the absence of audits candidates

could steal much more than what the party reputation reflects.

While we can get the non monotonicity very easily in a reduced form model where there

are only candidates, no incomplete information, it comes at the cost of assuming that there

is an asymmetry between incumbent and challenger. The challenger is assumed to have a

reputation that does not depend on �.

There are three sets of players in the game: parties, candidates and voters, and we

discuss each one separately.

Parties: There are two parties L and R, with fixed platforms located at opposite ends

of an ideology spectrum [�1, 1]. Parties compete in a large number of constituencies. We

consider a constituency i which is characterized by the position of the median voter �i

. We

present results for a representative constituency with median �, so from now on we drop

the subscript. We will assume throughout that � < 0, so that party L is the larger party.

Each party has two decision variables.

One decision variable of each party k is the corruption level xk

= x

k

(�) 2 [0, 1] at the

constituency as a function of its margin �. This variable is determined once and for all at

the outset of the game, and it represents the fraction of the fixed pot of money that can be

embezzled/stolen. Parties care about winning and rents, so that the party’s payo↵ is

1X

t=1

�

k�1x

k

P(Party k is in power at stage t).

In addition, at every stage the party may decide to change its current candidate in the

constituency.

Candidates: Each party k has a pool of candidates in the constituency, and at every stage

selects one of its candidates as its representative. If elected, the candidate determines his

own corruption level yk

. Candidates care about winning and rents, so that his payo↵ is

1X

t=1

�

k�1y

k

P(the candidate is in power at stage t).

Voters: There is a continuum of voters distributed on the ideology space according to a

CDF F with median �. Voters care about ideology and about corruption levels. Consider a

voter j with ideal point zj

2 [�1, 1]. Suppose candidates follow the party line. Then voters’

utility from having each party in power is given by:

U

L

j

= �x

L

� (zj

+ 1)2, U

R

j

= �x

R

� (zj

� 1)2.

If candidates do not follow the party line, voters will take adjust expected corruption ac-

cording to equilibrium beliefs.

Election Mechanism:

Each Voter votes according to his or her utility function. In addition, there is a random

(anti-incumbency) shock ⇠ before elections, which is uniformly distributed in the interval

[�", "], where " > 0. Therefore, Party L wins the elections if xL

+(�+1)2+⇠ < x

R

+(��1)2

while party R wins if xL

+(�+1)2+⇠ > x

R

+(��1)2. Since the random shock ⇠ has a density,

the probability that the election ends with a draw, that is, xL

+(�+1)2+⇠ = x

R

+(��1)2,

is 0. Thus, " can be interpreted as the level of uncertainty in the outcome of the elections.

Note that xL

+ (� + 1)2 + ⇠ < x

R

+ (� � 1)2 if and only if xL

+ 4� + ⇠ < x

R

. Therefore

the role of ⇠ is to change the swing voter in a random way. When " is small, the outcome of

elections is determined by the corruption levels, while if " is high, the outcome is random

even if the quantity x

L

� x

R

+ 4� is far from 0.

The Timeline of the Game:

The timeline of the game is as follows:

• At the outset of the game the two parties learn the median �, and set their corruption

levels xL

and x

R

.

• At every stage,

– The incumbent candidate, who is from party k, chooses a level of corruption y

k

.

– With probability 1� q an audit takes place and reveals yk

.

– Voters observe x

k

, the result of audit (if an audit took place), yk

. The election

takes place and a new incumbent is elected.

Remark 1 We assumed that the parties set their corruption level once and for all at the

outset of the game. The results would not be a↵ected if the parties could have changed their

corruption level.

Analysis

The solution concept that we study is perfect Bayesian equilibria. The game that we

presented has a multitude of such equilibria. For example, there is an equilibrium in which

both parties set x

k

= 1, and voters vote for Party L unless it or its incumbent deviated,

in which case they vote for Party R. There are also time-dependent equilibria, for example

equilibria in which voters expect the party in power to replace its incumbent in every odd

period.

As the game is stationary, we will be looking for a perfect Bayesian equilibrium which

is stationary. Moreover, we we will look for an equilibrium in which the decisions of both

parties and opportunist incumbents are optimal without punishment, and voters punish

deviations from this optimal behaviour.

The Parties’ Decision Problem

The volatility of election outcomes depends on the distribution of the aggregate shock ⇠.

Since we assume that it is uniformly distributed, what matters is the interval over which

the shock is distributed [�", "]. When " is very large then the average corruption may be

driven less by competition and more by the volatility of preferences.

Denote by P

L

(xL

, x

R

) = P(party L wins) the probability that the L party wins. The

next lemma describes the probability that Party L wins elections as a function of the parties’

levels of corruption.

Lemma 1 The probability P

L

(xL

, x

R

) is given by

P

L

(xL

, x

R

) =

8>><

>>:

1 " x

R

� x

L

� 4�,xR�xL�4�+"

2" �" x

R

� x

L

� 4� ",

0 x

R

� x

L

� 4� �".

(1)

Since the situation is stationary, in the absence of candidates each party k maximizes

the quantity x

k

(�)Pk

(xL

, x

R

). This generates the choice of xL

as below.

Lemma 2 If there is full discipline among candidates then the equilibrium corruption levels

x

⇤L

and x

⇤R

, and the probability that in equilibrium party L wins elections are given by the

following table:

• If �� 12" and �� 1+"

4 , then x

⇤L

= 1, x⇤R

= 0, and P

L

(x⇤L

, y

⇤L

) = 1.

• If �� 34" and �� 2�"

4 , then x

⇤L

= �4�+"

2 , x

⇤R

= 0, and P

L

(x⇤L

, y

⇤L

) = "�4�4" .

• If �� 3(1�")4 , �� "�1

4 , and �� 1+"

4 , then x

⇤L

= 1, x

⇤R

= 1+4�+"

2 , and

P

L

(x⇤L

, y

⇤L

) = �4��1+3"4" .

• If �� 34" and �� 3

4(1 � "), then x

⇤L

= " � 4�3 , x

⇤R

= " + 4�3 , and P

L

(x⇤L

, y

⇤L

) =�4�+3"

6" .

• If �� "�14 then x

⇤L

= x

⇤R

= p

L

(x⇤L

, x

⇤R

) = 1.

The figure below shows the various possible solutions we could have depending on ".

��

0 = 3(1� ✏)

��

0 = 1 + ✏

��

0 = 3✏

��

0

✏

x

⇤L = ✏� �0

3

x

⇤R = ✏+ �0

3

PL = ��0+3✏6✏

x

⇤L = ��

0 � ✏

x

⇤R = 0

PL = 1

x

⇤L = 1

x

⇤R = 1+�0+✏

2

PL = �1��0+3✏4✏

x

⇤L = 1

x

⇤R = 0

PL = 1

��

0 = ✏� 1

x

⇤L = 1

x

⇤R = 1

PL = ��+✏2✏

0 10

2

Region A

Region B

Region C

Region D

Region E

Figure 1: Corruption levels and probability of winning.

The interpretation of the result is as follows. When the noise is large and the margin

is not too large (Region E), the probability of being reelected is low, and therefore parties

will steal as much as they can (x⇤L

= x

⇤R

= 1) and both parties have positive probability of

being in power P

L

(x⇤L

, x

⇤R

) 2 (0, 1). When the margin is high (Region A), the large party

wins elections with probability 1 even if it steals all the pot. When the noise is large and the

margin is not too large (Region C), the large party prefers to steal all the pot, even though

this causes it not to be elected always. When the noise is small (Region B), the large party

will not steal all the pot and will always be in power. When the noise is medium and the

margin is not too large (Region D), both parties steal, and both may win elections.

Our interest in this paper is on the relationship of competition and corruption. Thus

we focus on an " that is relatively small.

Lemma 3 In the absence of candidates, the expected corruption level is monotonic nonde-

creasing in ��, if and only if " 34 .

The Candidates’ Decision Problem

In this section we analyze the decision problem of the candidates, assuming the parties

use the behaviour calculated above.

Recall that �� 0; that is, party L is the larger party.

Lemma 4 Assume (1) that the parties use the following strategies:

• The parties set corruption levels to x

L

= x

⇤L

and x

R

= x

⇤R

, respectively as long as

��

k

> ��

k

and x

k

= 1 when ��

k

��

k

.

(2) Voters have the following beliefs:

• If an incumbent who was audited and found to steal x

⇤k

voters believe that he will steal

x

⇤k

every period.

• If an incumbent who was audited and found to steal y

k

6= x

⇤k

then voters believe he

will steal 1 every period.

Then for every " that satisfies " < min(34 ,1�1

2 �

1�12 q�

) there exists � = �

k

(�, q, ") > 0 such

that

• If �� < ��

k

then the best response of an incumbent of party k is to steal y

k

= 1.

• If �� > ��

k


k

= x

⇤k

.

The lemma in particular proves that the best response of the incumbents of the larger

party is decreasing: when there is high competition (�� is small) and the party’s corruption

level is low, the incumbent steals 1, while when there is low competition (�� is large) and

the party’s corruption level is high, the incumbent steals x

⇤L

. Note that the probability of

winning for each party in the interval [0,��] is 1/2. Thus in this equilibrium, the two parties

alternate in power and steal the maximum amount, We would like to find an equilibrium

with full party discipline. Suppose the audit reveals y

K

6= x

k

, then voters believe that he

will steal 1. Given these beliefs the incumbent would either follow party discipline and steal

x

k

or steal 1. Then the party could o↵er the incumbent a slightly higher amount, x̃k

> x

k

that makes him indi↵erent between stealing 1 and following the party line. Of course the

probability of winning changes as well. We conjecture that this equilibrium implies that

corruption is first decreasing and then increasing with competition.

Proposition 1 Assume that � < 0 and " < " < min(34 ,1�1

2 �

1�12 q�

). There exists a Perfect

Bayesian Equilibrium where party k proposes the strategy: x

k

= x

⇤k

(�, ", �, q) for ��

��

k

(�, q, ") and x

k

= 1 for �� < ��

k

(�, q, "). If audit reveals y

k

6= x

k

voters believe that

the incumbent will steal 1 in future.

In this equilibrium, candidates follow the party discipline as long as competition is

not too sti↵. When competition is very sti↵ they will deviate. Given that when �� <

��

k

(�, q, "), both parties choose x

k

= 1, voters are indi↵erent between the two and the

probability of winning is 1/2,

The figures below present the corruption levels of the parties and the incumbent candi-

dates when " = 12 .

��

CorruptionLevel

✏ = 0.5

34

20

1 x

⇤L

x

⇤R

Figure 2: Corruption levels as a function of competition.

The straight lines in the figure show the corruption levels of the 2 parties, while the

curved line shows the average corruption. As we see, the average corruption is monotonically

decreasing with an increase in competition (�� ! 0).

�

yL

1 2�1�2�

Figure 3: Corruption level yL

at " = 12 .

The figure shows how the corruption level of an incumbent of an L party changes as the

bias towards him goes from being certain to win, to when he is certain to lose

3. Background – The National Rural Employment Guarantee Program

The National Rural Employment Guarantee Act (NREGA, 2005) mandates the

provision of 100 days of manual work on publicly funded projects to rural households

in India. The Act envisions a rights based approach – rural adults can demand work at

a mandated minimum wage. The program was initially implemented in the country‟s

poorest 200 districts in February 2006, with 130 additional districts added in the next

stage (2007) and national coverage thereafter (2008). In 2011-12, the Act had

provided employment to almost 40 million households at an annual expenditure of

more than $8 billion, making it one of the most ambitious poverty alleviation

programs in India to date.

While the primary objective of the program is provision of employment, it

also aims to create durable community assets, viz. irrigation canals, water

conservation, roads and other rural infrastructure. The NREGA, therefore, mandates

that at least 60 percent of the program expenditures be on wage payments and the

remainder on the materials used for the rural projects. A novel feature of the NREGA,

unlike all other public programs in India, is mandated audits of program expenditures

by local stakeholders at the village level, hence referred to as „social‟ audits.

Grass roots institutions, i.e. directly elected village, sub-district and district

level governments (or panchayats) under India‟s decentralized system of governance

(the Panchayati Raj system created by the 73rd amendment to the Indian constitution

in 1992) have a leading role in the planning and implementation of NREGA works.

In particular, the portfolio of projects to be implemented under the program has to be

prepared by and follow the priority expressed by an assembly of residents of a Gram

Panchayat (GP), a collection of 3-4 villages. The leader of the GP, directly elected by

its adult residents, is overall responsible for decisions made by the GP. Further, at

least 50 percent of the NREGA projects have to be implemented by the GP.

A. The administration of NREGA projects in Andhra Pradesh

This paper uses data from the state of Andhra Pradesh (AP).1 As of 2011 AP was

India‟s fifth largest state in terms of population (Census, 2011) and among the leading

states in NREGA implementation due to consistently high generation of NREGA

employment and its unique solution to the challenge of credible auditing of NREGA

projects. The state has conducted regular and systematic audits of NREGA projects

(unlike in other states of India where audits, are either not conducted or conducted in

an ad-hoc and unsystematic manner) by vesting its responsibility within an

autonomous arm of its Department of Rural Development. The state, therefore, claims

to maintain high levels of accountability and transparency in program

implementation.2

Although village councils in AP are typically less autonomous than in some

other states of India (Ban and Rao, 2008), they maintained a crucial role in managing

and executing NREGA projects during the period of our study.3 First, the Gram

1 In 2014 Andhra Pradesh was bifurcated into two separate states – Andhra Pradesh

and Telangana. The data in this paper are until 2010 and includes villages in both AP

and Telangana.

2 For a fuller account of the genesis and evolution of the AP „social audit model‟, see

Aiyar et al. (2013).

3 Through a Government of Andhra Pradesh order in December 2007, the

administrative functions relating to the implementation of all projects under the

NREGA were devolved to Panchayati Raj institutions (G.O. Ms. No. 571)

(www.rd.ap.gov.in).

Rozgar Sevak or the Field Assistant (FA), a resident of the GP who assists the village

council in NREGA implementation, was appointed on the recommendation of the

village council. 4 The FA represents the direct interface of beneficiary households

with the program. Second, the sarpanch selected suppliers of the material inputs to

projects implemented under the program and was therefore well positioned to fudge

material expenditures in connivance with the technical staff (viz., Assistant Engineers,

Technical Assistants and/or the suppliers) as suggested by anecdotal evidence from

the field. The village council and its leader, thus, are accountable for efficient

program implementation and the labor and material expenditures on the NREGA

projects. Given the scale of the NREGA program, funds at the grassroots level

increased manifold following the passing of the Act in 2005 (Afridi, 2008).

B. The audit process in AP5

4 Data on the process of FA appointments from our GP survey suggests that in the

appointment of both the first FA and the latest, the village council had the main say in

at least 80 per cent of the GPs. In about 40 per cent of appointments, village councils

controlled the entire selection process with no involvement of the sub-district

bureaucracy. The FA assists the GP in the following processes: registration of

households for job cards, verification of registration applications, distribution of job

cards to „registered‟ households, receipt of applications for employment and reporting

the demand to the MPDO, informing applicants to report for work, maintenance of

attendance records at work sites, verification and closure of labor records every week

for weekly wage payments. 5 See Aiyar et al. (2013) for a detailed discussion of AP‟s audit model.

The AP audit process is unique, combining a top-down approach with grass-roots,

beneficiary participation. The first step in conducting the audit of NREGA projects is

a notification to the relevant sub-district (or mandal) office with reference to the Right

to Information obligations and requesting unrestricted access to the records on wage

and materials payments by the state‟s independent audit body. A team, comprising

state and district auditors, along with residents of the mandal to be audited (but not of

the audited GP to prevent intimidation and ensure that audit outcomes are unbiased)

conduct a two-day workshop on NREGA rights and regulations and, how to conduct

the audits. Following the training audits are organized in all GPs of the mandal over a

week. In each GP, official labor expenses are verified by visiting labourers listed in

the worksite logs („muster-rolls‟). Complaints by individuals or groups of

beneficiaries and the audit team are recorded and attested using a standardized audit

report template.6 For verification of material expenditure, the audit team is mandated

to undertake worksite inspections. Since the verification of material expenditure is

typically perceived to be more complex and demanding, the works-site inspections

are typically carried out by state and district auditors of the team.

Once the audits of all GPs have been completed, a mandal level public hearing

to discuss the audit findings is organised with mandatory attendance for all

implementing officials. Those present typically should include „wage seekers from the

villages in the mandal, the social audit team, key implementing officials, members of

the vigilance cell, elected representatives and a district-level ombudsman (Aiyar et al.,

p. 261)‟. Complaints are read out, testimonies verified and accused officials given an

6 The auditors are expected to verify labour records for 100 percent of the

beneficiaries. This may not be true in practice.

opportunity to defend themselves. The scope for frivolous complaints is therefore

minimal, if at all.

After the “public hearing” a decision taken report is created by the officer

presiding over the hearing. In this report the responsibility for each confirmed

malfeasance is pinned on a programe functionary or, as the case may be, on multiple

functionaries.7 Research suggests, however, that the guilty often escape legal or

administrative penalties (Afridi and Iversen, 2013).

The audit process, therefore, is initiated and lead by the state but it assimilates

local stakeholders into the process of appraising program expenditures. Systematic

and standardized audits have been carried out in all 23 districts of the state with an

average of over two rounds of audits completed per mandal/GP between 2006 and

2010.

4. Data and metholodgy

A. Data

Our audit data were extracted and codified from the official and original, audit reports

for 100 randomly sampled mandals across eight districts of AP.8 In each randomly

7 The audit body has created checks and balances within the audit process such that

the programe functionaries who are being audited, do not corrupt the auditors

themselves. 8These eight districts were Mahbubnagar, Medak, Nizamabad, Warangal and

Khammam (north or Telangana region, now part of Telangana state), Anantpur and

Kurnool (south or Rayalseema region) and Guntur (west or coastal region). NREGA

was implemented in February 2006 in all these districts, except Kurnool and Guntur,

which implemented the program from April, 2007 onwards.

chosen mandal, three gram panchayats were selected based on the following criteria:

the GP which was the administrative headquarter of the mandal, one GP randomly

selected from all GPs reserved for a woman sarpanch and one randomly selected from

GPs not reserved for a woman sarpanch in that mandal in 2006.9 We, thus, have a

sample of 300 GPs from 100 mandals.

The NREGA audit reports contain data at the GP level. They have two

components: a standard audit report card which records the date of the audit along

with the demographic characteristics of the GP, and more importantly, audit team

impressions about process performance since the last audit including a financial

misappropriations estimate. These impressions and estimates are based largely on the

second component of the audit report – the list of complaints filed during the

verification process by individuals, groups or by audit team members. We use data

from the first round of audits that began in 2006 and until 2010.10 We construct a

panel of audit reports for each GP with an average of over two reports per GP for the

period 2006-10.11

9 1/3rd of all village council seats are randomly reserved for a woman sarpanch in AP

and across all states in India.

10Original audit reports that were missing were supplemented with abridged versions

of the audit reports available from the state auditor‟s website:

http://125.17.121.162/SocialAudit/ 11 Information in the audit reports were coded as follows: each complaint was first

classified into labor, material or worksite facilities related. The former two were

further categorized by type. For each complaint we recorded whether any

misappropriated amount was mentioned; if yes, the amount was mentioned, otherwise

the information is missing.

In addition to the official audit data, a survey was conducted in all sampled

GPs in 2011-12 to collect information on GP and elected sarpanch characteristics.

Retrospective data on the elections to the village council held in July 2006 following

which new village council members, including the sarpanch, assumed office for a

five-year term, were gathered. Specifically, data were collected on the sarpanch

elections from up to four respondents in each GP – the winning candidate (elected

sarpanch), closest losing contestant in terms of proportion of total votes received, a

worker of the losing political party and the GP secretary. Information was obtained on

votes received by each contestant and their party affiliation. It is worth noting that the

timing of this election overlaps with the phasing-in of the NREGA in February, 2006

in AP, giving us the opportunity to study program implementation up to 2010 with the

characteristics of the village council and the sarpanch unchanged. Finally, GP level

characteristics on infrastructure and availability of public goods, such as schools,

were obtained from the village census abstract for 2001.

The summary statistics are described in Table 1. Panel 1 describes the village

level characteristics from the village level census abstract (2001). Panel 2 describes

the characteristics of the elected sarpanch – age, gender, education and caste. The

village or GP level characteristics in Panel 1 suggest that the villages in our sample

are moderately developed in terms of availability of public facilities and are

physically well-connected to an urban area. 28 per cent of the sampled villages belong

to the GP which is the headquarter of the mandal. 42.7 per cent of the sampled GPs

were reserved for a woman sarpanch in 2006. More than 67 percent of the GPs were

reserved on the basis of caste for the head of the village council in the 2006 sarpanch

elections.

In Panel 2, we present the individual characteristics of the elected sarpanch.

44.4 percent of the elected candidates belong to the Indian National Congress (INC)

while 35.7 were affiliated with the Telugu Desam Party (TDP).12 The remaining,

approximately 20 percent of candidates, were distributed among regional or

communist parties (viz. Telangana Rashtra Samithi, Communist Party of India

(Marxist)) or were independent candidates. Thus, the two main political parties during

the 2006 elections were INC and TDP. Almost 20 per cent of the winning candidates

had prior political experience either as a political party worker or in a position in the

panchayat. The average number of prior terms in a political office was less than 1.

The summary statistics on the retrospective sarpanch election data are in Panel 3

of Table 1. We report data obtained from interview with the elected sarpanch.13 The

number of contestants in the sarpanch election was a little under 3, on average. The

winning candidate received 20 per cent more votes, of total votes polled than her

closest contestant. The margin of victory is our measure of electoral competition in

the empirical analysis.

In Table 2 we show the summary statistics for the audit data. The total number

of audits conducted over the period 2006-10 in our sampled GPs was 711 or 2.37

audits per GP. The average number of registered irregularities was 5.823, the majority

of which were labor related (86.9 percent). Non-labor and non-materials related

12 Note that candidates‟ political affiliation is not formally declared to the State

Election Commission in the sarpanch elections. However, each candidate belongs to a

political party and this is common knowledge amongst the electorate.

13 The coefficient of correlation between the margin of victory reported by the elected

sarpanch and the other respondents varies between 0.95 and 0.97 and is significantly

correlated at 1 per cent significance level.

irregularities (program facilities on the project site) were insignificant and are not

included in the analysis. The labor related irregularities are primarily related to

payment of wages to beneficiaries – a private good that the electorate is likely to care

deeply about. The materials related irregularities consisted of expenditures on the

materials used in the NREGA projects, viz. whether the sanctioned project was

executed or not, quality of materials used etc.. These expenditures can be termed as

providing the public component of the program – one that voters are more likely to

free ride on.

B. Methodology

Our outcome of interest is the number of each type of irregularity registered across all

audits for each GP over the period 2006-10. Specifically, we run the following

analysis:

Irregularityjkl = α0 + α1 competitionjkl + α2 competition2jkl + α3Xjkl, + ηk Dk +µjkl (1)

where Irregularityjkl is the total number of irregularities over the period 2006-10 in

GP j in mandal k in district l. The variable „competition jkl‟ is defined as 1 less than the

margin of victory in the sarpanch elections in 2006 (before the audits were

conducted). It is the difference between the percentage of votes polled in favor of the

winning candidate and her closest rival in the election. Hence, if the candidate is

unanimously elected the margin of victory is 1 and the competition variable equals 0.

Electoral competition is, therefore, increasing as the magnitude of this variable rises.

The square of this variable accounts for any non-linear impact of electoral

competition on our measure of corruption. Xjkl is a vector of GP level characteristics

that includes the characteristics of the sarpanch elected in 2006 (for a five year term)

such as gender, caste, education and age, the GP‟s access to health, education

facilities and its distance from the nearest town. It also includes a dummy variable for

whether the GP is the mandal headquarter. Dk is a dummy for mandal k to account for

mandal level variation in program implementation. Our theoretical model suggests a

U-shaped relationship between electoral competition and malfeasance in program

expenditures. We should, therefore, expect a negative coefficient (α1) on

competitionjkl, which would signify that at low levels of electoral competition the

number of program irregularities decline. A positive coefficient (α2) on

competition2jkl, would indicate that as electoral competition rises there is an increase

in irregularities related to program implementation.

The above specification identifies the effect of electoral competition on

program implementation by taking advantage of variation in the degree of

competition across GPs within the same mandal. However, there may be variation in

implementation of the program and the level of corruption over the period under study

as well as type of labor and materials related irregularities. Taking advantage of the

panel structure of our data by conducting the analysis at the GP-audit level would

give more power to the analysis due to the resulting larger number of observations.

Our empirical specification is then given by:

Irregularityjklmt = β0+ β1 Competitionjkl,t-1 + β2 Competitionjkl,t-1 + β3Xjkl + βt Yeart + δlt

(Dl*Yeart)+ δm Auditm + δk0 Dk+εjklm t -1 (2)

where the number of irregularities in GP j in mandal k in district l in audit m at time t

(Irregularityjklmt) is a function of electoral completion (Competitionjkl,t-1) prior to any

audits and Xjkl as outlined in equation (1) above. In addition, there may exist secular

time trends (Yeart) and district specific time trends (Dl*Yeart) that affect the level of

corruption in a GP. Furthermore, we include audit and mandal fixed effects to account

for unobservables such as auditor‟s capacity to detect malfeasance which may

improve with successive audit rounds and the mandal bureaucracy‟s preferences.

Both empirical specifications are used to analyse data for up to three rounds of

audits in each GP between 2006 and 2010.

5. Results

We first assess the impact of political competition on program leakages in Table 3.

Our main outcomes of interest are – the total number of irregularities, total labour

related irregularities and total materials related irregularities registered during 2006-

10 in each GP. To ensure that our outcome variable is not influenced by the variation

in the number of audits (due to missing audit reports) across GPs in a mandal, we

drop mandals in which data are not available for the same audit rounds for all three

sampled GPs. We are then left with a sample of 257 GPs. In columns 1,3 and 5 we

model a linear relationship between electoral competition and reported irregualrities.

In coulmns 2,4 and 6 we introduce a polynomial term for electoral competition.

Table 3 shows that the coefficient on „electoral competition‟ is positive and

insignificant overall (coulmn 1) and for labor related irregularities (column 3) but

significant for material related irregularities as shown in column 5. When we

introduce a quadtratic term for electoral competition, our U-shaped hypothesis from

the theoretical model holds up as suggested by the negative coefficient on „electoral

competition‟ and the positive coefficient on „electoral competition2‟, albeit

insignificant in columns 2 and 4 . The direction of the coefficients indicates that

electoral competition at low levels reduces reported program irrgularities but as

electoral competition rises there is an increase in the number of irrelgularities. This

non-linear effect of electoral competition is again significant for materials component

of the program in column 6.

In Table 4 we conduct the analysis at the GP-audit level. In columns 1,3 and 5

we include only mandal fixed effects to make the model comparable with Table 3.

Columns 2, 4 and 6 show the results of specification 2 discussed above. Our results

are remarkably consistent with those in Table 3 – there is a significant non-linear, U-

shaped relationship between electoral competition and irregularities for the materials

component. The coefficients are not significantly different between the more

parsimonious model which includes only mandal fixed effects and stricter model

which accounts for trends and audit fixed effects as well across all outcomes. This

suggests that secular or district specific trends and audit level unobservables were not

correlated with electoral competition and did not play a significant role in uncovering

program related malfeasance over time. In the remainder of our analysis, therefore,

we continue present the results from specification 2.

In Tables 3 and 4 our outcome included all reported irregualrites. However,

irregularities registered by the auditors are less likely to suffer from noise due to

personal biases. Beneficiary households who report program irregularities, as part of

the bottom-up approach of the audit design, are likely to be subject to threats and

intimidation , particularly due to the public announcement of the audit findings. This

may affect the irregularity reporting behavior of GP residents. Hence to reduce the

noise in the data we re run the analysis for irregularities reported by the auditors

alone. These results, reported in Table 5, suggest that our previous findings are robust

to this concern.

In the survey we gauged incumbent‟s perception of voters‟ bias for or against

her by asking the incumbent to rank her chances of re-election in the forthcoming GP

elections.14The response of the incumbents to this hypothetical question correlates

14The survey question was “Please rank the chances of your being re-elected in the

significantly with the degree of political entrenchment of the sarpanch‟s family.

Political entrenchment is a function of the number of relatives of the incumbent who

have held political positions and the number of years they held those positions. An

incumbent‟s family is considered to be more politically entrenched the higher the

average number of years (total years in political office/number of relatives who have

held political positions) her relatives were in a political office. The correlation

between re-election expectations and political entrenchment is positive and significant

at 5 per cent level in our data.

Using this self-reported measure of voter bias, we stratify the sample into

electoral bias against (reported re-election probability moderate or less) and for

(reported re-election probability more than moderate) the incumbent and re-run

equation 2. Our theoretical model suggests that the effect of electoral competition

would be significantly non-linear when bias is in favor of the left leaning incumbent.

This shows up quite starkly in the analysis in Table 6. The U-shaped, non-linear

effect of electoral competition is significant in the lower panel of the table but mostly

flat in the upper panel, as predicted by the model.

Our results, so far, highlight the significant effect of electoral competition on

irregularities in the materials component of the NREGA and an absence of any overall

impact on labour related irregularities. However, findings reported in Table 6 suggest

that there is significant variation in the total irregularities by its components. In panel

next sarpanch elections in this Gram Panchayat on a scale of 0 to 5: (0) No chance of

re-election; (1) Very low; (2) Low; (3) Moderate; (4) High; (5) Almost certain to be

re-elected. The average rank (excluding 12 non-responses) was 3.83 and 3.69 in

unreserved and reserved GPs, respectively.

1 we report the relationship between electoral competition and types of irregularities

when electoral bias is against the incumbent politician. We find an insignificant effect

of electoral competition on disciplining politican behavior across all outcomes.

However, when we restrict the reported irregularities to those filed by auditors alone,

we find a significant U-shaped relationship between electoral competition and the

number of materials related irregularities. In contrast, in the lower panel, when the

electoral bias is in favor of the incumbent, we find a strong non-linear relationship

between total and labour related irregularities in coulmns 1 and 2 as well as when we

study the irregularites reported by auditors in columns 4 and 5. These results suggest

that the nature of corruption in public programs may vary depending on the preceived

electoral bias of an incumbent when audits are mandated and its findings are made

public. When the incumbent perceives that electoral bias is in her favor (which may

be due to pilitical entrenchment), there is significant theft from the private component

of the program – i.e., in wage expenditures. On the other hand, when re-election is

perceived to be more difficult, there is a significant effect of electoral competition in

the public component of the program – materials expenditures. This suggests that at

low levels of electoral competition incumbents pilfer from the private component of

the program even though voters care more about this aspect of the program benefits

because re-election probability is high inspite of public exposure of corruption. When

re-election probability is low, then incumbents are more likely to steal from the

component of the program that voters care less about – the public expenditures on

materials. Our results echo Ferraz and Finan (2011) but also nuance the impact of

mandated disclosure of fraud in public expenditures on the nature of corruption.

The findings outlined in Table 6 are held up when we break-up the labor and

materials related irregularities into three main components: non-payment or delay in

payment of wages, impersonations or ghost workers and bribes paid to receive wages;

and ghost projects, poor qulaity of materials and bribes for materials related

irregularities. Consistent with the previous table, electoral competition significantly

affects the number of reported ghost projects, an irregularity in the public good

component of the program, when electoral bias is against the incumbent. Irregularities

in the private component of the program (columns 1 and 3) show up along with those

in the public component (column 5) when electoral bias is in favor of the incumbent

in the lower panel of Table 7.

The audit reports also provide information on the discrepancy in program

expenditures, in rupee amount, between reported and actual expenditures for an

irregularity as discussed above. However, this information is missing in several

instances. For those irregularities for which this information is available we calculated

the rupee amount per irregularity by the type of irregularity. Conditional on the data

being available, we find that the amount of potential pilferage per material related

irregularity is almost three times higher than a labor related iregularity (Afridi and

Iversen, 2013). This might explain the larger magnitude of the point estimates on

electoral competition for labor relative to material related malpractices in the lower

panel. The number of irregularities in the labor component has to be of comparable

value to theft in the materials component when the electoral bias is in favor of the

incumbent.15

15 we restrict the data to those mandals (GPs) which had conducted 2 audits over the

period 2006-10 to account for variation in the intensity of audits which may have

been systematically related to existing levels of corruption. Our results are

qualitatively unchanged.

6. Conclusions

In this paper we build a simple game theoretic model to capture the effect of electoral

competition on re-election concerns when there is public exposure of corruption

through mandated audits of government expenditures. We show that in a scenario

where the politician acts as the agent of the principal, the party, corruption has a non-

monotonic relationship with electoral competition. At low levels of electoral

competition the party‟s platform on corruption is high and the agent adheres to it. As

competition increases, the party lowers its threshold level of corruption which is

followed by the incumbent. However, when re-election is extremely fragile (high

competition) then although the party lowers the threshold level of acceptable

corruption, the incumbent deviates from the party‟s dictum and actual corruption is

high.

We use the model‟s predictions to test our hypotheses using official data on

mandated audits of the NREGA projects implemented by village councils in Andhra

Pradesh during 2006-10 and data on the elections to the headship of these same

village councils in 2006. Our results largely confirm the non-linear relationship

between electoral competition and corruption, but also suggest that the impact of

electoral competition varies by whether theft is from the public or private component

of a good.

Our findings suggest that mandated exposure of corruption may not always

lead to lower theft from public goods. This relationship is mediated by the extent of

electoral competition, and hence re-election incentives faced by an incumbent.

References

Acemoglu, D. and M. Jackson, (2013), “History, Expectations and Leadership in the Evolution of social norms”, MIT Department of Economics Working Paper No.11-10.

Acemoglu, D. and J. Robinson, (2000), “Political Losers as a barrier to economic development”, American Economic Review Papers and Proceedings, 90, pp.126-130

Acemoglu, D. and J. Robinson, (2002), “Economic Backwardness in Political Perspective”, NBER Working Paper No. 8831.

Ades, A. and R. Di Tella (1997), “The New Economics of Corruption: A survey and some new results”, Political Studies, XLV, pp. 496-515.

Aidt, T., M. Golden and D. Tiwari (2011), “Incumbents and Criminals in the Indian National Legislature”, Cambridge Working Papers in Economics 1157.

Bannerjee, A. and R. Pande (2009), “Parochial Politics: Ethnic Preferences and Politician Corruption”, Working Paper.

Bardhan, P. and T. Yang (2004), “Political Competition in Economic Perspective”, BREAD Working Paper 078.

Besley, T. , T. Persson and D. Sturm (2010), Political Competition, Policy and Growth: Theory and Evidence from the US”, Review of Economic Studies,77, pp. 1329-1352.

Booth, R., D. Filmer and J. Rigolini (2011), “Does Electoral Competition improve Accountability?” Working Paper.

Ferraz, C. and F. Finan (2011), “Electoral Accountability and Corruption: Evidence from the audits of local governments”, American Economic Review, 101(4), pp. 1274-1311.

Persson, R. and G. Tabellini (1997), “Separation of powers and political accountability”, Quarterly Journal of Economics, 112(4), pp. 1163-1202.

Svensson, J. (2005), “Eight Questions about Corruption”, Journal of Economic Perspectives, 19(3), pp. 19-42.

Table 1: Summary statistics

Notes: Election characteristics as reported by the elected sarpanch of the GP; proportion of votes received includes unanimous candidates; proportion of votes polled is 0 if the sarpanch election was unanimous.

Variable No. of GPs

Mean Standard Deviation

Min. Max.

GP characteristics Cultivated area (sq. km) 294 1643.24 1697.202 12 16578 Proportion of irrigated area 294 0.242 0.233 0 0.999 Population density (per sq. km) 296 3.431 3.726 0 39.306 Distance from town(km) 296 30.371 20.158 0 125 Medical facility available 294 0.829 0 .376 0 1 Communication facility available 294 0.918 0.274 0 1 Banking facility available 294 0.374 0.484 0 1 Middle school 296 0.709 0.454 0 1 Sarpanch election No. of contestants 299 2.916 1.767 1 16 Proportions of votes polled out of total voters

297 0 .757 0.259 0 1

Proportions of votes received by winner

297 0.565 0 .173 0.194 1

Winning margin 297 0.209 0.274 0.0007 1 Sarpanch Characteristics Age 299 44.685 9.957 25 75 Female 299 0.468 0.499 0 1 Illiterate 299 0.110 0.313 0 1 Literate but no schooling 299 0.117 0.322 0 1 Higher secondary & above 299 0.204 0.403 0 1 SC 299 0.157 0.364 0 1 ST 299 0.163 0.370 0 1 OBC 299 0.478 0.500 0 1 INC 299 0.444 0.497 0 1 TDP 299 0.357 0.480 0 1 Prior political experience 297 0.804 0.397 0 1

Table 2: Audit summary statistics Variable Number of

audits Mean Std. Dev. Min. Max.

All Complaints 711 5.823 5.299 0 43 Labour Complaints 711 5.062 4.594 0 30 Material Complaints 711 0.684 1.520 0 18 Other Complaints 711 0.077 0.307 0 3 Note: Data at GP-audit level for the period 2006-10. „Other complaints‟ refer to complaints related to facilities provided at the NREGA worksite.

Table 3: GP level sam

ple (mandal fixed effects)

A

ll irregularities Labor related irregularities

Material related

irregularities

(1) (2)

(3) (4)

(5) (6)

Electoral competition

5.254 -19.12

2.526 -4.683

2.973* -13.26*

(5.383) (35.73)

(4.750) (33.74)

(1.709) (7.384)


2

16.07

4.752

10.70**

(23.64)

(21.90)

(5.223) N

257

257 257

257 257

257 R

2 0.720

0.610 0.601

0.724 0.724

0.610 N

ote: Each observation is the sum of the num

ber of irregularities reported during 2006-10 at the GP level. Standard errors in parentheses.

*significant at 10%, ** 5%

and ***1%.

Table 4: Pooled sample of all audits, by G

P-audit

A

ll Labor

Material

(1)

(2) (3)

(4) (5)

(6) Electoral com

petition -4.597

-4.662 2.008

1.95 -6.167**

-6.187** (12.55)

(12.800) (11.91)

(12.130) (2.687)

(2.758) Electoral com

petition2

4.070 4.052

-0.851 -0.865

4.676** 4.680**

(8.143) (8.304)

(7.596) (7.742)

(1.853) (1.897)

Mandal FE

√

√ √

√ √

√ Y

ear FE

√

√

√ A

udit FE

√

√

√ D

istrict x Year FE

√

√

√

N

677 677

677 677

677 677

R2

0.361

0.430 0.244

0.459 0.395

0.291

Note: Each observation is the num

ber of irregularities reported at the GP level in each audit over 2006-10. Standard errors clustered at the G

P level in parentheses. *significant at 10%

, ** 5% and ***1%

.


P-audit (irregularities filed by auditors only)

A

ll Labor

Material

(1)

(2) (3)


-1.625 -4.088

-6.048** (11.82)

(2.896) (2.945)


2 0.0456

2.548 4.615**

(7.801) (1.857)

(2.032) M

andal FE

√ √

√ Y

ear FE √

√ √

Audit FE

√ √

√ D

istrict x Year FE

√ √

√ N

635

635 635

R2

0.421 0.323

0.291 N


ber of irregularities reported by auditors during 2006-10 at the GP level.

Standard errors clustered at the GP level in parentheses. *significant at 10%

, ** 5% and ***1%

.


P-audit

Irregularities filed by all

Irregularities filed by auditors

All

Labor M

aterial A

ll Labor

Material

(1)

(2) (3)

(4) (5)

(6) Electoral bias against incum

bent

Electoral C

ompetition

0.093 12.920

-11.600 -5.450

3.002 -13.32*

(24.05)

(18.84) (8.002)

(25.730) (5.231)

(7.721) Electoral C

ompetition

2 -2.048

-9.901 7.274

1.325 -1.672

8.225*

(14.960) (11.800)

(5.056) (15.830)

(3.320) (4.861)

R2

0.537 0.392

0.59 0.527

0.392 0.428

N

387 387

387 363

363 363

Electoral bias in favor of incumbent

Electoral Com

petition -105.6**

-113.1*** 7.519

-110.8** -35.64***

6.705

(44.250) (35.010)

-11.92 (46.79)

(8.743) (11.32)

Electoral Com

petition2

66.36** 68.60***

-2.535 73.03**

20.89*** -1.626

(26.600)

(21.130) -7.083

(28.03) (5.262)

(6.806) R

2 0.646

0.434 0.678

0.654 0.442

0.582 N

266

266 266

250 250

250 N




, ** 5% and ***1%

.


P-audit (irregularities filed by all)

Labor related irregularities

Materials related irregularities

N

on-provision,

nonpayment

of wages

Impersonations/

Ghost w

orkers

Bribes

Ghost

projects Poor quality

Bribes

(1)

(2) (3)

(4) (5)

(6) Electoral bias against incum

bent

Electoral C

ompetition

1.288 -1.714

1.737 -9.941**

0.314 -2.991

(7.322)

(12.010) (4.887)

(4.028) (1.486)

(4.004) Electoral C

ompetition

2 -1.343

-1.033 -0.658

6.152** -0.282

2.146

(4.513) (7.859)

(3.068) (2.568)

(0.965) (2.602)

R2

0.708 0.457

0.395 0.322

0.361 0.398

N

387 387

387 387

387 387

Electoral bias in favor of incumbent

Electoral Com

petition -45.11***

-25.13 -26.42***

4.04 -11.18***

11.64

(10.700) (20.250)

(9.687) (3.386)

(2.603) (7.976)

Electoral Com

petition2

27.24*** 15.7

15.18** -2.531

6.429*** -4.35

(6.668)

(12.200) (5.981)

(2.044) (1.510)

(4.733) R

2 0.659

0.583 0.53

0.487 0.474

0.356 N

266

266 266

266 266

266 N




, ** 5% and ***1%

.

Appendix

Lemma 1 The probability P

L

(xL

, x

R

) is given by

P

L

(xL

, x

R

) =

8>><

>>:

1 " x

R

� x

L

� 4�,xR�xL�4�+"

2" �" x

R

� x

L

� 4� ",

0 x

R

� x

L

� 4� �".

(1)

Proof. As mentioned above, party L wins if xL

+ ⇠ x

R

+ (� � 1)2. It follows that

P(party L wins) = P(xL

+ (� + 1)2 + ⇠ x

R

+ (� � 1)2) (2)

= P(⇠ x

R

� x

L

� 4�), (3)

= P(⇠ x

R

� x

L

� �

0). (4)

Lemma 2 If there is full discipline among candidates then the equilibrium corruption levels

x

⇤L

and x

⇤R

, and the probability that in equilibrium party L wins elections are given by the

following table:

• If �� 12" and �� 1+"

4 , then x

⇤L

= 1, x⇤R

= 0, and P

L

(x⇤L

, y

⇤L

) = 1.

• If �� 34" and �� 2�"

4 , then x

⇤L

= �4�+"

2 , x

⇤R

= 0, and P

L

(x⇤L

, y

⇤L

) = "�4�4" .

• If �� 3(1�")4 , �� "�1

4 , and �� 1+"

4 , then x

⇤L

= 1, x

⇤R

= 1+4�+"

2 , and

P

L

(x⇤L

, y

⇤L

) = �4��1+3"4" .

• If �� 34" and �� 3

4(1 � "), then x

⇤L

= " � 4�3 , x

⇤R

= " + 4�3 , and P

L

(x⇤L

, y

⇤L

) =�4�+3"

6" .

• If �� "�14 then x

⇤L

= x

⇤R

= p

L

(x⇤L

, x

⇤R

) = 1.

1

��

0 = 3(1� ✏)

��

0 = 1 + ✏

��

0 = 3✏

��

0

✏

x

⇤L = ✏� �0

3

x

⇤R = ✏+ �0

3

PL = ��0+3✏6✏

x

⇤L = ��

0 � ✏

x

⇤R = 0

PL = 1

x

⇤L = 1

x

⇤R = 1+�0+✏

2

PL = �1��0+3✏4✏

x

⇤L = 1

x

⇤R = 0

PL = 1

��

0 = ✏� 1

x

⇤L = 1

x

⇤R = 1

PL = ��+✏2✏

0 10

2

Region A

Region B

Region C

Region D

Region E

Figure 1: Corruption levels and probability of winning.

Proof. We will calculate the equilibrium in the decision problem of the parties, denoted

(x⇤L

, x

⇤R

). The expected payo↵ of the parties is:

u

L

(xL

, x

R

) = x

L

p

L

(xL

, x

R

), (5)

u

R

(xL

, x

R

) = x

R

(1� p

L

(xL

, x

R

)). (6)

Since ⇠ has a uniform distribution, and in particular, its distribution has a density, the

functions u

L

and u

R

are di↵erentiable. We will first look for an equilibrium in an interior

point, xL

, x

R

2 (0, 1). To this end we di↵erentiate the functions uL

and u

R

of the last case.

@u

L

@x

L

(xL

, x

R

) =x

R

� 2xL

� � + "

2", (7)

@u

R

@x

R

(xL

, x

R

) ="+ x

L

� 2xR

+ �

2". (8)

2

The first order conditions are:

@u

L

@x

L

(xL

, x

R

) = 0 () x

R

= 2xL

+ � � ", (9)

@u

R

@x

R

(xL

, x

R

) = 0 () x

R

= xL+�+"

2 . (10)

The two derivatives vanish at

bxL

= "� �

3 , bxR

= "+ �

3 . (11)

Since � is negative, to ensure that (x⇤L

, x

⇤R

) is an interior point we need to ensure that

bxL

1 and bxR

� 0, which amounts to �� 3(1� ") and �� 3". Denote

RD

= {(�, ") : � � 3(1� "),�� 3"}.

On RD

we have x

⇤L

= "� �

3 , x⇤R

= "+ �

3 , and by Lemma 1 we have P

L

(x⇤L

, x

⇤R

) = ��+3"6" .

If 0 � "+ �

3 or "� �

3 � 1, then the solution will be on the boundary of the unit square

[0, 1]2.

If �� 3" then we have x

⇤R

= 0. We identify when x

⇤L

is an interior point. By (12) we

deduce that in this case x

⇤L

= ��+"

2 . Since � is negative, we need to verify that x

⇤L

1,

which solves to �� 2� ". Denote

RB

= {(�, ") : � � 3(1� "),�� 3",�� 2� "}.

On this region we have x

⇤L

= ��+"

2 , x⇤R

= 0, and by Lemma 1 p

L

(x⇤L

, x

⇤R

) = "��

4" .

If �� 3(1�") then we have x⇤L

= 1. We first identify when x

⇤R

is an interior point. By

(13) this happens when x

⇤R

= 1+�+"

2 . We need to ensure that 0 x

⇤R

1. The condition

x

⇤R

� 0 is equivalent to �� 1 + ", and the condition x

⇤R

1 is equivalent to �� "� 1.

Denote

RC

= {(�, ") : � � � 3(1� "),�� 3",�� 1 + ",�� "� 1}.


⇤L

= 1, x⇤R

= 1+�+"

2 , and consequently p

L

(x⇤L

, x

⇤R

) = �1��+3"4" .

Assume now that �� 3(1� ") and �� 1 + ", and set

RE

= {(�, ") : � � "� 1}.


⇤L

= 1, x⇤R

= 1, and p

L

(x⇤L

, x

⇤R

) = ��+"

2 .

Assume now that �� 3(1 � ") and �� 1 + ". Then x

⇤L

= 1, x

⇤R

= 0, and

p

L

(x⇤L

, x

⇤R

) = 1. Assume finally that �� 3(1 � "), �� 3", and �� 2 � ". Then

x

⇤L

= 1, x⇤R

= 0, and p

L

(x⇤L

, x

⇤R

) = 1. Denote the union of these two regions by

RA

= {(�, ") : � � � 3(1� "),�� 3",�� 1 + "}.

3

Lemma 3 In the absence of candidates, the expected corruption level is monotonic nonde-

creasing in ��, if and only if " 34 .

Proof. The expected corruption level is x⇤L

P

L

(x⇤L

, y

⇤L

) + x

⇤R

(1�P

L

(x⇤L

, y

⇤L

)). The func-

tions x⇤L

, x⇤R

, and P

L

(x⇤L

, x

⇤R

) are continuous in � and ", hence to verify that the expected

corruption level is monotonic nondecreasing in �� we check whether it is monotonic non-

decreasing in each region in Figure 1.

• In Regions A and E in Figure 1 the expected corruption level is 1, hence constant in

��.

• In Region B in Figure 1 the expected corruption level is ��+"

2 , which is monotonic

nonincreasing in ��.

• In Region C the expected corruption level is

P

L

(x⇤L

, x

⇤R

) + x

⇤R

(1� P

L

(x⇤L

, x

⇤R

)) =�1� � + 3"

4"+

"+ 1 + �

4"⇥ 1 + � + "

2. (12)

This is a quadratic function in �. Ignoring the term 4" in the denominator and the

constant term �1+3"+ ("+1)"2 , which does not a↵ect the monotonicity of the function,

the right-hand side of Eq. (15) is equivalent to ��(�� 2"), which is monotone

increasing in ��, provided �� ". Note that for every " 34 , all �’s such that the

pair (",��) is in Region C satisfy �� ".

• In Region D the expected corruption level is

x

⇤L

P

L

(x⇤L

, x

⇤R

) + x

⇤R

(1� P

L

(x⇤L

, x

⇤R

)) =�� + 3"

6"

✓"� �

3

◆+

� + 3"

6"

✓"+

�

3

◆

= "+2�2

18",

which is monotonic nondecreasing in ��.

Lemma 4 Assume (1) that the parties use the following strategies:

• The parties set corruption levels to x

L

= x

⇤L

and x

R

= x

⇤R

, respectively as long as

��

k

> ��

k

and x

k

= 1 when ��

k

��

k

.

(2) Voters have the following beliefs:

4

• If an incumbent who was audited and found to steal x

⇤k

voters believe that he will steal

x

⇤k

every period.

• If an incumbent who was audited and found to steal y

k

6= x

⇤k

then voters believe he

will steal 1 every period.

Then for every " that satisfies " < min(34 ,1�1

2 �

1�12 q�

) there exists � = �

k

(�, q, ") > 0 such

that

• If �� < ��

k


k

= 1.

• If �� > ��

k


k

= x

⇤k

.

The lemma in particular proves that the best response of the incumbents is decreasing:

when there is high competition (�� is small) and the party’s corruption level is low, the

incumbent steals 1, while when there is low competition (�� is large) and the party’s

corruption level is high, the incumbent steals x⇤L

.

Proof. The payo↵ to incumbent k if he steals yk

= x

k

is

x

⇤k

1� �P

k

(x⇤L

, x

⇤R

).

The payo↵ to incumbent L if he steals yk

= 1 is

1

1� �qP

k

(x⇤L

, x

⇤R

).

Therefore, incumbent k steals yk

= 1 if

x

⇤k

1� �P

k

(x⇤L

, x

⇤R

)<

1

1� �qP

k

(x⇤L

, x

⇤R

).

In Region D we have p

k

(x⇤L

, x

⇤R

) =x

⇤k

2" for k 2 {L,R}, hence the incumbent steals yk

= 1 if

x

⇤k

1� �

2"x⇤k

<

1

1� �q

2"x⇤k

. (13)

Note that this equation holds for x⇤k

= 0, and does not hold for x⇤k

= 1. Eq. (16) solves to

�q

2"(x⇤k

)2 ��

�

2" � 1�x

⇤k

+ 1 > 0. (14)

This is a quadratic function, and since it is satisfied for x

⇤k

= 0 and not for x

⇤k

= 1, it is

satisfied in an interval. To complete the proof note that Eq. (17) is satisfied for � = 0

whenever " <1�1

2 �

1�12 q�

.

5

Exposing corruption: How does electoral competition affect ...

Documents