Pani2011 Article HoldYourNoseAndVoteCorruptionA

Public Choice (2011) 148: 163–196
DOI 10.1007/s11127-010-9651-4
Hold your nose and vote: corruption and public decisions

in a representative democracy
Marco Pani
Received: 29 May 2009 / Accepted: 30 April 2010 / Published online: 15 May 2010
© Springer Science+Business Media, LLC 2010
Abstract This paper analyzes how corruption alters policy decisions in democracy, and
examines whether this distortion can result in a long-term persistence of corruption even
when the voters are well informed and rational. By applying a citizen-candidate model of
representative democracy, the paper analyzes how corruption distorts the allocation of re-
sources between public and private consumption, altering the policy preferences of elected
and nonelected citizens in opposite directions. The outcome is a reduction in real public
expenditure and, if the median voter’s demand for public goods is sufficiently elastic, a re-
duction in taxes. In this case, some citizens benefit indirectly from corruption. The paper
also presents some empirical evidence that, in democratic countries, corruption results in
lower tax revenue, and proceeds to show that, when this occurs, citizens anticipating a shift
in preferences in favor of public expenditure may support institutions that favor corruption.
This result complements the findings of other studies that have attributed the persistence of
corruption in democracy to some failure on the part of the voters or the electoral system. It
also bears implications for developing effective anticorruption strategies and for redefining
the role that can be played by the international community.
Keywords Corruption · Political equilibrium · Public goods · Public expenditure · Taxation
1 Introduction
Curbing corruption remains a challenge in democratic countries. In principle, public con-

trol and accountability should induce public officials in a democracy to be honest (Treis-
man 2000), but historical experience shows that democracy alone does not guarantee the
absence of corruption. Well-known examples of democracies that have experienced high
DISCLAIMER: The views expressed herein are those of the author and should not be attributed to the
IMF, its Executive Board, or its management.
M. Pani ()
International Monetary Fund (IMF), Washington DC, USA
e-mail: mpani@imf.org
164 Public Choice (2011) 148: 163–196
levels of corruption over long periods include Italy, Japan, India, and the United States be-
tween the Civil War and the Great Depression. The importance of the problem cannot be
underestimated: corruption is socially wasteful, harmful to growth (Mauro 1995; Tanzi and
Davoodi 1998; Del Monte and Papagni 2001), diverts resources to unproductive rent-seeking
efforts, distorts incentives, may increase inequality and poverty (Chong and Calderon 2000a;
Gupta et al. 2002), and prevents an effective management of public expenditure. A country
that fails to control corruption suffers substantial welfare losses. Aware of these problems,
multilateral organizations like the IMF and the World Bank are intensifying their efforts to
promote governance and combat corruption.
The political economy of corruption has been widely discussed in the literature, but some
issues remain to be fully resolved. A central question that deserves further investigation is
how corruption distorts democratic policy choice. How does corruption affect decisions on
key policy variables such as public expenditure and taxation in a representative democracy?
The persistence of corruption in some democratic countries also poses intriguing questions;
most explanations provided in the literature focus on some imperfections in the democratic
mechanism, such as voters’ failure to exert appropriate controls as a result of apathy, asym-
metric information, or of some shortcoming of the electoral system; but can corruption per-
sist also in a well-functioning democracy in which voters are interested and fully informed?
A deeper understanding of these issues could enable us to design more effective anticorrup-
tion strategies.
This study contributes to research on these topics by examining how corruption alters de-
cisions concerning public expenditure and taxation in a representative democracy in which
citizens vote rationally with a full understanding of the incentives faced by their represen-
tatives once elected. Using a simplified version of Osborne and Slivinski’s (1996) “citizen-
candidate” model, we analyze how corruption distorts the allocation of resources between
private and public goods; we then proceed to assess empirically whether corruption has a
significant impact on public expenditure and revenue in democratic countries. On the basis
of the results thus obtained, we discuss how, and under what conditions, the citizens of a
democracy may deliberately support institutions that encourage corruption.
Our analysis shows that, while corruption induces elected representatives to expand pub-
lic expenditure, well-informed citizens respond to this distortion by electing candidates that
prefer, ceteris paribus, lower public expenditure than the majority. As a result, corruption
reduces the physical supply of public goods, while its impact on the nominal amount of
public expenditure and taxes depends on the citizens’ demand elasticity. When demand for
public goods is elastic, corruption also reduces taxes, to the benefit of some citizens who
are not themselves corrupt. The empirical results suggest that this may actually be the case:
corruption appears to have a significant negative impact on tax revenue in democratic coun-
tries.
When this occurs, if the current majority expects a future change in preferences in favor
of more taxes and more public expenditure, corruption can be strategically used to induce
the future majority to keep taxes low. In this sense, decisions concerning law enforcement,
accountability, and other institutional variables that determine whether elected representa-
tives will engage in corruption are in some way comparable to decisions on constitutional
amendments, in the sense that they set long-standing parameters that condition future short-
term policy decisions, and are made, to some extent, taking these effects into consideration.
These results show that corruption can arise even in a well-functioning democracy. In
other circumstances, the threat of corruption can alter the rational policy choice stemming
from free and fair elections even if corruption does not actually take place.
The rest of the paper is organized as follows: Sect. 2 discusses how this paper builds on
three different strands of research that have examined the political economy of corruption,
Public Choice (2011) 148: 163–196 165
the principal-agent relation inherent in an act of corruption, and the impact of corruption
on public expenditure. Section 3 introduces a model of political choice in which corrup-
tion distorts the decision of the elected representatives in a way that is fully predictable and
discounted by the voters; it shows how, once elected, candidates respond to corruption ac-
cording to their policy preferences and the risk of punishment; the section also discusses
the anticipatory response of the nonelected voters. Section 4 presents an empirical assess-
ment of the impact of corruption on public expenditure and revenue in democratic countries.
Section 5 extends the model discussed in Sect. 3 to incorporate a collective decision on the
appropriate degree of law enforcement, which has a predictable impact on future policy deci-
sions; it shows that under certain conditions citizens vote in favor of a weak law enforcement
regime that encourages corruption. Section 6 draws some conclusions and suggestions for
further research.
2 Relations with previous studies
This study builds on three different strands of research that have analyzed corruption from
the standpoint of political economy, agency theory, and its impact on public expenditure.
A series of studies have analyzed the political economy of corruption in a representative
democracy. These studies have generally linked corruption to failures in the functioning of
the democratic system such as informational asymmetries between voters and candidates,
lack of interest (“apathy”) on the part of the voters (Rose-Ackerman 1978), or a lack of hon-
est candidates with policy preferences close to those of the majority (Rundquist et al. 1977;
see also Kurer 2001). Several studies have traced corruption to shortcomings of the electoral
system such as the political fragmentation that emerges under proportional representation
(Galeotti and Merlo 1994; Kunicová and Rose-Ackerman 2005), the coordination failures
that occur under plurality voting (Myerson 1993), or other features such as the size of elec-
toral districts (Persson et al. 2003). The literature on lobbying and interest groups (Becker
1983; Grossman and Helpman 1996) has analyzed how corruption can be used to distort the
democratic process to serve the interests of particular groups. Some studies (Easterly 2007)
have noted that foreign aid can facilitate the permanence in power of corrupt governments.
However, these studies do not fully explain how corruption can persist when it is
widely perceived to be causing severe welfare losses. Under pressing needs, in democ-
racy, voters should eventually become more interested in the political process and over-
come initial informational asymmetries; new candidates with a reputation for honesty and
more varied policy preferences should challenge the corrupt incumbents; and more force-
ful action should be taken to counter the pressure of special interest groups. If corrup-
tion were caused by inadequacies in the electoral law, the law itself could be changed,
as occurred, for instance, in Italy in the early 1990s. Some studies (Treisman 2000;
Calderon and Chong 2006) have actually found evidence that democracy curbs corruption
in the long run.1 Foreign aid, in turn, may explain the persistence of corruption in some
recipients of external assistance (typically developing and low-income countries) but not in
other countries.
1 A negative impact of democracy on corruption is also found in Sandholtz and Koetzle (2000), Montinola
and Jackman (2002) and (more weakly) in Paldam (2002). Sung (2004) identifies empirically a cubic relation
between the degree of democracy and corruption, which suggests the possibility of a temporary surge in
corruption during intermediate phases in the process of democratization.
166 Public Choice (2011) 148: 163–196
Our model contributes to this line of research by further exploring the causes of corrup-
tion in a democracy but, unlike previous studies, focuses on a well-functioning democracy
in which citizens are interested and well informed and candidates of all political inclinations
are always available. We assume that the electoral system aggregates efficiently the prefer-
ences of the citizens, and model it by using a simplified “citizen-candidate” model (Osborne
and Slivinski 1996) in which any citizen can stand as a candidate and the candidate who is
elected is the one who cannot be defeated in a pairwise contest by any other candidate. As in
Rose-Ackerman (1978), we feature an elected representative who trades off bribes against
some risk (in her case, losing votes; in ours, being punished by law). Voters also face a
trade-off, between honest candidates with less-preferred policy preferences and corrupt can-
didates that are politically closer to their ideal points. Unlike in Peters and Welch (1980),
however, this trade-off does not arise because the set of candidates is limited but because
the propensity to engage in corruption is correlated with a candidate’s policy preferences.
Contrary to models of special interest groups, bribes are not aimed here at influencing policy
decisions but are used for mere personal gain. This allows us to put in evidence some other
aspects of corruption: corruption ultimately affects the policy outcome and can also be used
deliberately for this purpose by a large group of citizens who are not themselves corrupt.
The theory of agency has shown in some detail how corruption can be tolerated as a result
of an informed, rational choice when controlling it is costly (Basu et al. 1992; Chander and
Wilde 1992; Hindricks et al. 1999). These studies, based on Becker’s seminal works on
the economics of crime (Becker 1968; Becker and Stigler 1974), generally assume a well-
defined objective function (such as maximizing tax revenue net of collection costs) and
abstract from the problems of the aggregation of preferences.
Arguably, the high degree of corruption observed in some democracies is hard to rec-
oncile with an informed welfare-maximizing decision on the part of the voters, even taking
into account the high costs of controlling corruption; over time, improvements in the anti-
corruption “technology” (including a better understanding of its institutional aspects) should
reduce these costs, and corruption should decline. The outcome can change, however, if indi-
viduals have different policy preferences and try to use corruption to impose their preferred
policy choice on the rest of the community. Our model examines this case by applying the
incentive structure discussed in previous studies to a setting in which individual preferences
are aggregated by representative democracy.
The impact of corruption on public expenditure has been discussed from various angles
in previous studies. Barro (1973) analyzed how corruption induces elected public officials to
increase public expenditure, while other authors (such as Dudley and Montmarquette 1987)
discussed how corruption alters the policy preferences of nonelected citizens. Mauro (1998)
found empirical support for the argument that corruption diverts public expenditure toward
items “on which it is easier to levy large bribes and maintain them secret.”2
This is a central theme in our model because the citizens’ response to corruption depends
critically on how corruption alters the allocation of resources between public and private
uses (although our model can be extended to include the impact of corruption on a more
general set of policies). Our model bears some similarities to Barro’s, with the notable dif-
ferences that, in our model, citizens do not have a tendency to reelect the incumbent but vote
2 A series of studies have explored the relations between corruption and foreign development aid (which
contributes to financing public expenditure). Knack (2001) found that foreign aid has a significant negative
impact on the quality of governance of the recipient; as noted above, Easterly (2007) remarked how aid can
facilitate the permanence in power of corrupt governments. Conversely, Chong and Gradstein (2008) found
that corruption has a significant impact on the provision of aid (a component of public expenditure) by donor
countries. Somewhat weaker evidence of this effect is also found in Alesina and Weder (2002).
Public Choice (2011) 148: 163–196 167
to maximize their expected utility, and that their policy preferences depend on whether or
not they expect corruption to occur. While corruption induces the elected representatives to
increase expenditure in order to get more bribes, the citizens respond by electing candidates
with different policy preferences (if not, also, more honest). The net impact on the volume
of public expenditure is negative, but the impact on its value is ambiguous. Applied to the
problem discussed in Mauro (1998), our model would yield somewhat different predictions.
3 A model of public expenditure with corruption
There are two types of goods, a pure private “consumer good” (unspecified) used individ-
ually by each citizen, and a pure public good, “roads,” 3 which can be used jointly by all
citizens without rivalry or congestion. Both goods are produced by a competitive sector of
enterprises with a common technology that exhibits constant marginal costs (normalized to
one). The consumer good is sold directly by the enterprises to consumers, but roads cannot
be purchased by individual consumers directly owing to free-rider problems; instead, the
producers sell them to a common representative (the “minister”), who makes them avail-
able to all citizens. Roads can be supplied in different amounts, but, once purchased by the
minister, they are available to everybody in their entirety.4
The minister pays for the roads with funds collected from the citizens through in-
come taxes. For simplicity, it is assumed that all citizens receive the same gross income
(normalized to one) and pay the same taxes. Similar (and generally equivalent) results
can also be obtained by assuming that each citizens receives a different gross income,
but shares the same quasi-linear utility function with the other citizens and pays propor-
tional income taxes.5 These assumptions allow one to abstract from the impact of tax-
ation on income distribution. It should be noted, however, that income inequality may
play a major role in determining how corruption distorts policy decisions in a democ-
racy. Various empirical studies have shown that corruption and income inequality are in-
terdependent; corruption tends to increase income inequality (for instance, Li et al. 2000;
Chong and Calderon 2000b; Gupta et al. 2002; Dincer and Gunalp 2008),6 and in turn
a higher degree of income inequality encourages more corruption (Glaeser et al. 2003;
You and Khagram 2005).7 While these effects can be taken as given in the static model
discussed in this section, when long-term effects are considered (Sect. 5) this interrelation
may become critical. In this paper we abstract from these effects, leaving their investigation
3 The word “roads” is only used to simplify the narrative. Roads are not, of course, “pure” public goods; they
are excludable (through tolls) and subject to congestion. Nevertheless, they are frequently built by private
contractors under government contract and they are generally open to all citizens; roads can also be supplied
in different amounts.
4 The network of roads, for instance, can vary in extension and coverage, but it is open to all users in its
entirety. Each potential user benefits to some extent from the size of the network even if he necessarily uses
only part of it.
5 A variant of the model based on these assumptions, hereafter defined as the version “with income inequality
and proportional income taxes,” will be discussed in the text and footnotes where appropriate.
6 Li et al. (2000) and Chong and Calderon (2000b) identify an “inverse U”-shaped relation, whereby income
inequality (measured by the Gini coefficient) increases with corruption (or, more generally, with institutional
quality) when corruption is low and decreases when corruption is high.
7 Paldam (2002), in contrast, does not find a significant impact of income inequality on corruption.
168 Public Choice (2011) 148: 163–196
to future research. We further assume that all tax revenue is used to pay for road construc-
tion, and that all income net of taxes is spent on the purchase of consumer goods: there are
neither credit nor savings, neither public nor private.8
The citizens have different preferences concerning the allocation of resources between
roads and consumer goods. Their utility function is quasi-linear in consumer goods, and
differs by a preference parameter θ , which captures the intensity of each citizen’s utility
from roads; θ identifies the “type” of each citizen:
U (x, g; θ ) ≡ x + θ u(g),
where x is the individual consumption of consumer goods, g is the quantity of roads avail-
able to all citizens, θ > 0 is the citizen’s type and u(•) is a continuous, concave, and twice
differentiable function that is common to all citizens (in Sect. 5, to enhance tractability, it
is further assumed that u(•) is three times differentiable with u ≤ 0 for all g > 0).9 Types
are distributed among the population according to a cumulative distribution function F (θ ),
which is common knowledge. It is assumed that the set of types is dense, but the results of
this paper can be generalized to nondense (and even discrete) sets of types.
Each citizen pays a tax equal to t; total tax revenue is equal to tn, where n > 1 is the
size of the population.10 The public budget constraint requires that tax revenue equals public
expenditure, which is in turn equal to g; hence, t = g/n.11 The allocation of income between
consumer goods and roads (through taxes and the public budget) is the same for all citizens
and is hereafter called a “policy.” Since n is given and the public budget constraint is binding,
a policy can be identified either by the tax rate t or by the supply of roads g (but using the
latter yields simpler formulas). Given g (and t), the private budget constraint requires that
x = 1 − t = 1 − g/n, which yields the indirect utility function:
g
V (g; θ ) ≡ 1 − + θ u(g).
n
V (g; θ ) measures the utility derived by a citizen of type θ when the amount of roads is equal
to g and the public and private budget constraints are binding. The value g θ that maximizes
this function, hereafter defined as the “ideal policy” of a citizen of type θ ,12 is identified by
8 In addition, income is exogenous, inelastic to the tax rate, and unaffected by the consumption of either type
of goods. This situation may apply, for instance, to a comparatively small province of a federal state, where the
income of citizens depends on country-wide macroeconomic conditions and is not affected by the fiscal policy
of the provincial government (which is constrained to run a balanced budget). In more general situations, the
decisions concerning the allocation of resources between private and public consumption would have an
impact on production, employment and income, and the effects of corruption should be analyzed in a model
of general equilibrium.
9 Different utility functions have been applied in the literature on spatial voting, depending on the type of
problem being analyzed. For the purposes of this study, the quasi-linear specification provides a good balance
of tractability and generality.
10 Most of the ensuing discussion can be simplified by normalizing n to unity. That n is larger than 1 becomes
relevant when discussing the decision problem of an elected candidate (see Sect. 3.1 below).
11 In the version of the model with income inequality and proportional income taxes, all citizens share the
same utility function U (x, g) = x + u(g) and each citizen perceives a different gross income, exogenously
determined and equal to y ≡ 1/θ , where θ , the reciprocal of gross income, is called the citizen’s “type.” The
average gross income is normalized to unity, so that total aggregate gross income is equal to n. Each citizen
pays a tax ty that is proportional to his income, the tax rate t being determined by the public budget constraint
as above (t = g/n).
12 In the literature on spatial voting, ideal policies are sometimes called “ideal points.”
Public Choice (2011) 148: 163–196 169
the first-order condition (FOC) Vg (g; θ ) = 0, equivalent to13
1 14
u (g θ ) = . (3.1)
nθ
Since g θ is strictly increasing in θ , the ranking of citizens according to their ideal policy is
the same as their ranking by type. Under the public budget constraint, the optimal tax rate
for a citizen of type θ is equal to t θ = g θ /n.
Since the citizens have different policy preferences, the policy that is actually imple-
mented (the amount of roads g and the tax rate t) is chosen by the minister, who is in turn
elected by the citizens in free elections in which each citizen casts one vote. In order to
abstract from lobbying effects, we assume that enterprises play no political role: compe-
tition and free entry drive any rents to zero so that the enterprises are indifferent between
producing either type of goods or even going out of business.15 Enterprises act only as a
conduit that enables the minister to effectively steal part of the funds that are allocated to
the construction of roads—for instance, by paying bribes in exchange for overinvoicing.16
Once elected, the minister implements the policy that maximizes his own personal util-
ity, unconcerned about the impact of his decision on the utility of the other citizens, and
unbound by any pre-electoral commitments. All citizens are aware of this and vote accord-
ingly: voting for a candidate is equivalent to voting for the policy that maximizes his own
utility once elected, which may differ not only from any policy announced before the elec-
tions but also from the policy that he would prefer as a nonelected citizen (this distinction
becomes relevant in the presence of corruption). 17 All citizens can stand as candidates, and
the election is won by the citizen/candidate who survives a hypothetical infinite sequence
of pairwise contests.18 The outcome of this model is identified by the type of candidate that
is elected minister and by the policy that candidate implements once elected. An equilib-
rium of the model is an outcome such that that (i) the elected minister cannot be defeated
in a pairwise contest by any other candidate; (ii) the elected minister cannot increase his
13 Since V (g; θ) is concave in g, the ideal policy of each citizen is unique.

14 With income inequality and proportional income taxes, the indirect utility function becomes: V ∗ (g; y) =
y(1 − g/n) + u(g), which is also maximized by u (g) = y/n = 1/(nθ).
15 This assumption becomes relevant when corruption is introduced, since corruption may induce the owners
of the enterprises to support a larger supply of roads. In general, if the entrepreneurs have the right to vote,
corruption alters the distribution of preferences. For the purposes of this study, this effect can be ignored.
16 This assumption allows us to put in evidence the consequences of the minister’s personal interest in ex-
panding public expenditure. In real life, the lobbying activity of special interest groups interacts with the
factors analyzed here and affects the equilibrium outcome.
17 Similar assumptions are found in “citizen-candidate” models of political equilibrium (for instance, Osborne
and Slivinski 1996; Besley and Coate 1997). In our model, however, citizens do not have to register and do
not incur costs to stand as a candidate; each citizen can vote for any citizen, including himself; all citizens are
thus candidates “by default.”
18 This simplifying assumption approximates the equilibrium outcome that occurs, under some circumstances,
under various electoral systems. With plurality voting, for instance, when the costs of standing as a candidate
are sufficiently low, equilibria with a single candidate or with candidates having the same policy preferences
are equivalent to the equilibrium outcome that would result in an infinite sequence of pairwise contests (Os-
borne and Slivinski 1996). With majority voting, a representative assembly reflecting the policy composition
of its electorate would tend, on average, to elect a government whose policy preferences are equivalent to
those of a single candidate that survives infinite pairwise contests. The equilibrium outcome of particular
electoral systems may of course be different, but its analysis is excluded from the scope of this paper.
170 Public Choice (2011) 148: 163–196
utility by implementing a different policy or changing his decision about corruption (ac-
cepting/rejecting the bribe); and (iii) if the elected minister has credibly committed before
the election to accept or reject the bribe, he cannot improve his utility by reneging on this
commitment. In equilibrium, the citizen/candidate that is elected is therefore a citizen whose
ideal policy (once elected) is preferred by a majority to the ideal policy of any other citizen
(Condorcet winner). In the absence of corruption, the only Condorcet winner is the median
voter,19 whose type μ lies at the median of the distribution.20 In the absence of corruption,
the median voter is elected and sets g = g μ and t = g μ /n.
3.1 How ministers respond to corruption: honest and corrupt citizens
How does this outcome change in the presence of corruption? Corruption alters the policy
choice of the elected minister, who spends more on road construction than he would other-
wise since by doing so he can get more bribes. Aware of this fact, the citizens respond by
electing a different candidate, who could be either a citizen motivated to act honestly, or a
corrupt citizen who is less inclined to spend on roads. This subsection discusses the impact
of corruption on the policy choice of the minister. The citizens’ response is discussed in the
next two subsections.
What is meant by “corruption?” Corruption takes many forms. In this model, corruption
is defined as the acceptance, by the minister, of a bribe from the enterprises that build the
roads, for example, in exchange for paying them an inflated price (out of the government
budget). There are several real-life examples of this type of corruption; a typical case is the
manipulation of public procurement auctions.21 This type of corruption effectively enables
the minister to steal part of the taxes paid by the citizens, and distorts the relative price of
consumer goods and roads, altering the allocation of resources.
The minister raises the price paid by the government for each mile of roads from 1 to p >
1, which exceeds production costs (p is assumed to be exogenous: for instance, it could be
the maximum price that can be paid to contractors without triggering a serious investigation).
In exchange, the enterprises that are selected as government suppliers pay the minister a
bribe of amount b for each mile of roads. This agreement is, of course, illegal and involves a
positive risk of punishment. Free entry and competition among enterprises bidding for public
contracts push the value of the bribe up to the artificial increase in the price of road granted
by the minister: b = p − 1: the minister extracts all the rent from corruption.22 The minister,
19 Although the singular is used to simplify the narrative in the rest of this paper, in general there may be more
than one “median voter” (all with the same preferences). In such cases, the identity of the winner among the
set of median voters would be decided by extrapolitical criteria such as random extraction, seniority, or
educational qualifications. These aspects are not relevant for the present discussion.
20 When the policy choice is unidimensional and preferences are single-peaked (as occurs as a result of the
concavity of the indirect utility function V (g; θ)), the median voter is pivotal. The only Condorcet winners
are therefore candidates who implement his ideal policy. Since each candidate, once elected, implements his
own ideal policy, the Condorcet winner in the absence of corruption can only be the median voter himself.
With corruption, as discussed below, the ideal policy of the elected candidate differs from that of a nonelected
citizen of the same type, and the median voter ceases to be the Condorcet winner.
21 See for instance (Rose-Ackerman 1975, and Beenstock 1979). Ample evidence of this type of corruption
in the State of Mississippi has been provided by an FBI investigation on county purchasing activities in the
mid 1980s (Karahan et al. 2006a, 2006b).
22 This assumption ensures that the enterprises are not acting as an interest group to alter the alloca-
tion of resources in a particular direction. This assumption allows us to focus on how corruption al-
ters the electoral choice of the voters. Some studies (Basu et al. 1992; Besley and McLaren 1993;
Mookherjee and Png 1995) assume that the rent is shared between the agents that pay and receive the bribe
according to the Nash bargaining solution.
Public Choice (2011) 148: 163–196 171
however, retains only a fraction a < 1 of the bribe; the rest (1 − a) is spent on legal fees
and other deadweight costs aimed at reducing the risk of punishment.23 The total amount of
illegal gains perceived by the minister is thus equal to abg, equivalent to a fraction a of the
bribe b raised on g miles of roads. The associated expected value of punishment, measured
in units of the consumer good, is fixed and equal to π . We assume that the enterprises run
no risk of punishment.24 The minister’s net expected gains from corruption are thus equal to
abg − π ,25 which he entirely spends on consumer goods.26 The minister pays taxes like any
other citizen on his legal income but does not pay taxes on the bribes.
When corruption occurs, total public expenditure becomes equal to pg = (1 + b)g, and
the amount of taxes paid by each citizen becomes (1 + b)g/n. Hence, the indirect utility
function of a nonelected citizen-taxpayer becomes V C (g, b; θ ) ≡ 1 − (1 + b)g/n + θ u(g) =
V (g; θ ) − bg/n, which is maximized by the policy ĝ θ that solves u (ĝ θ ) = (1 + b)g/(nθ ) =
pg/(nθ ).27 The indirect utility function of the minister is somewhat different, as it includes
the net illegal gains from corruption, abg − π :
(1 + b − nab)g
Ṽ C (g, b, π; θ ) ≡ V C (g, b; θ ) + abg − π = 1 − + θ u(g) − π (3.2)
n
which is maximized by the policy g̃ θ that satisfies the FOC u (g̃ θ ) = p ∗ /(nθ ), where p ∗ =
1 + b − nab < 1 is the effective tax rate that a corrupt minister pays for each unit of roads.28
Note that ĝ θ and g̃ θ , like g θ , are increasing in θ . Moreover, ĝ θ < g θ < g̃ θ for each type
θ : corruption alters the policy preferences of elected and nonelected citizens in opposite
directions: nonelected citizens demand less roads (that become more costly for the taxpayer),
whereas elected citizens engaging in corruption want more roads (to collect more bribes).29
The fact that bribes are offered does not imply that the minister accepts them. Aside
from any moral restraints, an elected citizen could decide to act honestly out of merely
23 Alternatively, one could assume that a fraction 1 − a of the price increase is retained by the enterprises so
that the minister only receives a bribe of amount ab for each mile of roads (as in a Nash bargaining model). As
mentioned above, however, this would give the enterprises a stake in corruption, making them a potentially
influential interest group.
24 This assumption simplifies the analysis by making the value of the bribe independent of the risk of detec-
tion. Alternatively, one could assume that the enterprise also runs a positive risk of punishment π , in which
case the value of the bribe determined below would be equal to p − 1 − π .
25 A necessary condition for corruption to be profitable for the minister is that a > 1/n, which assures that
the gross gains from corruption, abg, are larger than the additional tax burden arising from an inflated price
of the public good, bg/n. This condition is easily satisfied for sufficiently large values of n. This seems to
suggest that corruption should ceteris paribus be less frequent in small countries and that, when it occurs,
it would involve larger bribes. However, the minister in a large country may also be able to retain a smaller
share of the rent. Note that this condition would not be satisfied with n = 1; the requirement n > 1 has thus
become relevant.
26 π could also measure the amount of consumer goods spent by the minister ex ante to purchase insurance
against the risk of punishment.
27 With income inequality and proportional income taxes the indirect utility function of a nonelected citizen
becomes V C∗ (g, b; y) ≡ y(1 − pg/n) + u(g), which is also maximized by u (ĝ θ ) = ypg/n = pg/(nθ).
28 This result is slightly modified under income inequality with proportional income taxes. The indirect utility
function of the minister becomes equal to Ṽ C∗ (g, b, π ; y) ≡ y(1 − pg/n) + abg + u(g) − π , which is
maximized by u (g̃ θ ) = yp/n − ab = p/(nθ) − ab.
29 Note, too, that V C (g, b; θ) < V (g; θ) for all θ and for all values of g; other things being equal, nonelected
citizens suffer from the presence of corruption because they pay more taxes for given amounts of roads.
As discussed below, however, in equilibrium “other things” are not equal: the citizens vote for a different
candidate that sets a lower value of g.
172 Public Choice (2011) 148: 163–196
Fig. 1 The minister’s expected utility with and without corruption
economic considerations: corruption entails a risk of punishment, and a minister engages

in it only if the maximum expected utility that can be obtained by engaging in corruption
exceeds the maximum utility that can be earned by being honest. If the minister accepts
the bribe, he maximizes his utility by setting g equal to g̃ θ , which yields Ṽ C (g̃ θ , b, π; θ ).
If the minister remains honest, he sets g = g θ and obtains V (g θ ; θ ). An elected citizen
unrestrained by moral constraints thus engages in corruption if and only if Ṽ C (g̃ θ , b, π; θ ) >
V (g θ ; θ ), which yields the following lemma (Fig. 1):
Lemma An elected citizen engages in corruption if and only if
g θ − p ∗ g̃ θ
(θ ) ≡ + θ [u(g̃ θ ) − u(g θ )] > π 30 (3.3)
n
Hereafter, a citizen who, if he is elected minister, engages in corruption will be called

“corrupt,” while a citizen who acts honestly even once elected will be called “honest.” This
definition has no reference to a citizen’s moral standards but merely describes the outcome
of a rational, self-interested choice. All citizens in this model are assumed to be corruptible
and unrestrained by moral considerations. From this lemma, whether a citizen is honest or
corrupt depends only on his type θ and on the expected value of punishment. For any given
value of π , the set of citizens can thus be partitioned into two subsets according to their type:
a set H (π) grouping all honest citizens (for whom (θ ) < π ), and a set C(π) including all
corrupt citizens (for whom (θ ) > π ). Citizens for whom (θ ) = π are indifferent between
being honest and corrupt; we assume that in this case they can publicly commit to either type
30 With income inequality and proportional income taxes this lemma yields ∗ (y) ≡ y(g θ − p g̃ θ )/n −
u(g θ ) + u(g̃ θ ) + abg̃ θ > π .

Public Choice (2011) 148: 163–196 173
Fig. 2 Honest and corrupt candidates by type (given p)
of conduct before the election, once the value of π is known. This commitment is credible
since, once elected, these citizens do not have any incentive to deviate.
Since (θ ) is increasing in θ ,31 for a given value of π the set H (π) includes all citizens
whose type lies below some threshold θ ∗ (π) = −1 (π), while the set C(π) includes all
citizens with θ > θ ∗ (π). Citizens whose type is equal to θ ∗ (π) are indifferent (Fig. 2).
Citizens’ types and the expected value of punishment are common knowledge; therefore
at the time of the elections all citizens know who among them is honest and who is corrupt
and vote accordingly.32
3.2 The political response of nonelected citizens33
How do nonelected citizens respond to corruption? In this subsection, we discuss the citi-
zens’ choice when all candidates are corrupt (which occurs whenever the risk of punishment
31 (θ) is equal to Ṽ C (g̃ θ , b, π ; θ) − V (g θ ; θ) + π ; its first total derivative with respect to θ is equal to
ṼθC (g̃ θ , b, π ; θ) − Vθ (g θ ; θ) + ṼgC (g̃ θ , b, π ; θ)d g̃ θ /dθ − Vg (g θ ; θ)dg θ /dθ , which is equal, by the En-
velope Theorem, to ṼθC (g̃ θ , b, π ; θ) − Vθ (g θ ; θ) = u(g̃ θ ) − u(g θ ) > 0 (since g̃ θ > g θ ). In the same
way, the first total derivative of ∗ (y) with respect to y is equal to ṼyC∗ (g̃ θ , b, π ; y) − Vy∗ (g θ ; y) +
ṼgC∗ (g̃ θ , b, π ; y)d g̃ θ /dy − Vg∗ (g θ ; y)dg θ /dθ = ṼyC∗ (g̃ θ , b, π ; y) − Vy (g θ ; y) = (g θ − pg̃ θ )/n < 0; hence
∗ (y) is diminishing in y and increasing in θ .
32 The fact that citizens are perfectly informed of which candidates are corrupt, and that, in spite of this,
corruption is not always punished should not sound surprising. Citizens may be unable to prosecute their
representatives even when they know that they are acting against the law, for instance because the evidence
available to the general public is not sufficient to secure conviction in court.
33 The discussion in this subsection also applies to the version of the model with income inequality and
proportional income taxes.
174 Public Choice (2011) 148: 163–196
Fig. 3 Utility of a nonelected citizen as a function of g
is sufficiently low, or equal to zero).34 The choice when some candidates are honest is dis-
cussed in the next subsection.
The indirect utility that a nonelected citizen of generic type θ perceives if an amount of
roads g is supplied with corruption is equal to V C (g, b; θ ) and peaks at a lower value of g
than the indirect utility V (g; θ ) that the same citizen perceives if the same amount of roads is
supplied in the absence of corruption (Fig. 3). Since this holds for all citizens, including the
median voter, when all candidates are corrupt the supply of roads diminishes in equilibrium,
as the citizens elect a candidate that has weaker preferences for roads than the median voter:
Proposition 1 When all candidates are corrupt, the citizens elect a candidate who ceteris
paribus prefers a lower supply of roads than the majority of citizens; this candidate, once
elected, chooses a lower supply of roads than preferred by a majority of citizens in the
absence of corruption.
Proof See the Appendix.
Note that the median voter would not elect a citizen of his own type, because, once
elected, this citizen would not implement the policy preferred by a nonelected median voter
(ĝ μ ), but a policy g̃ μ > ĝ μ that entails a larger supply of roads (Fig. 4).
In summary: the citizens respond to the inflated cost of road construction by reducing
their demand for roads. Since corruption induces elected candidates to expand road con-
struction, the citizens elect a candidate who values roads less than the majority of citizens.
If citizens’ types are represented on a straight line running from right to left (with types on
34 All candidates are corrupt if and only if π ≤ (θ), where θ is the lowest citizen type (θ = min{θ : F (θ) >
0). With income inequality and proportional income taxes this occurs if and only if π ≤ ∗ (y H ), where y H
is the largest gross income perceived by any citizen.
Public Choice (2011) 148: 163–196 175
Fig. 4 Median voter’s utility as a function of the minister’s type
the left valuing roads more than types on the right) the citizens respond to a shift “to the
left” of the policy preference of elected candidates by shifting their votes “to the right.”35
The net effect on taxation depends on the demand elasticity of the median voter. Corrup-
tion yields lower taxes if and only if (1 + b)ĝ μ < g μ , which is equivalent to the following
condition, due to Allen and Lerner (1934):
Elasticity Condition (Allen and Lerner) The arc price elasticity of the demand for roads of
the median voter is larger than unity:

ĝ μ − g μ 1 + p
− > 1. (3.4)
b g μ + ĝ μ
When this condition holds, some citizens may benefit indirectly from corruption, because,
by inducing a tax reduction, corruption brings the choice of the community in equilibrium
closer to their own preferred policy:
Proposition 2 When all candidates are corrupt and the Elasticity Condition holds, a mi-
nority of citizens may benefit indirectly from the shift in policy induced by corruption.
35 The words “right” and “left” are not used here in their usual political connotation but only to highlight the
ordering of citizens’ preferences by type. Left-wing political parties frequently exhibit a comparatively strong
preference for public goods, and right-wing parties are more inclined to reduce taxes, but there are exceptions
(the political right, for instance, might spend more than the left on national security, which is also a public
good). In the version of the model with income inequality, the “right” side defined above would correspond
to the citizens with the largest gross income and the “left” side to the citizens with the smallest gross income.
176 Public Choice (2011) 148: 163–196
3.3 The choice between honest and corrupt candidates36
Assume now that the risk of punishment is sufficiently high to induce some citizens to be
honest (π > (θ )). Under what conditions would one of them be elected?
Let us prove first of all that the median voter is pivotal in this choice. Consider the
choice between two candidates of generic type θ1 and θ2 . If both of them are either honest
or corrupt, the citizens are confronted with a unidimensional policy choice under a single-
peaked utility function, and the median voter is pivotal by the median voter theorem. If,
instead, one of the candidates (say, θ1 ) is honest and the other is corrupt, a generic citizen of
type θ prefers the former to the latter if and only if V (θ1 ; θ ) > V C (θ2 , b; θ ), which yields:
g θ1 − (1 + b)g̃ θ2 g θ1 − (1 + b)g̃ θ2
θ> or θ< (3.5)
n[u(g θ1 ) − u(g̃ θ2 )] n[u(g θ1 ) − u(g̃ θ2 )]
depending on whether or not g θ1 > g̃ θ2 . Hence, any majority in the choice between the two
candidates includes the median voter, who remains pivotal.
Obviously, if the median voter is honest (μ ≤ θ ∗ (π)), he is elected minister: by revealed
preferences, a majority of citizens prefer him to any corrupt candidate.37 We therefore re-
strict the discussion to cases in which the median voter is corrupt (μ > θ ∗ (π)). In this
situation, the citizens have to trade off the benefits of electing an honest candidate for the
disadvantage of electing a candidate that implements less preferred policies. In some cases,
the citizens elect an honest candidate, even if his policy preferences are different from those
preferred by the majority. In other cases, they elect a corrupt candidate that, once elected,
implements the policy that most closely resembles the ideal policy of the median voter.
Proposition 3 When the median voter is corrupt, corruption alters the equilibrium policy
outcome by inducing either the election of a corrupt candidate or the election of an honest
candidate with policy preferences different from the median.
When an honest citizen other than the median voter is elected, the alteration of the equi-
librium induced by corruption always benefits a minority of citizens, because taxes diminish
while the price paid for each mile of roads does not change.
4 Empirical evidence
Proposition 1 above highlights how corruption reduces the physical supply of roads (more
generally, real public expenditure) in equilibrium. Its impact on nominal public expenditure
36 The discussion in this subsection also applies to the version of the model with income inequality and
proportional income taxes.
37 A corrupt candidate of generic type c ∈ C(π ) would set g = g̃ c and t = (1 + b)g̃ c /n; this is strictly
dominated, for all citizens, by the policy g = g̃ c , t = g̃ c /n, which would be achieved by electing a citizen
of type c = c/p∗ if the latter were honest. To this policy, however, the median voter (and hence a majority)
prefers the policy g = g̃ μ and t = g μ /n that is achieved by electing the (honest) median voter. Hence, a
fortiori, the median voter (and with him a majority) prefers this last policy to the policy implemented by any
corrupt candidate.
Public Choice (2011) 148: 163–196 177
and taxation depends however on the Elasticity Condition; when the latter holds, nominal
public expenditure and government revenue also decline as a result of corruption.
This section presents an empirical assessment of the impact of corruption on these two
variables. Using fiscal data from the World Economic Outlook database of the IMF and in-
dicators of corruption provided by Transparency International and by the World Bank, we
examine (i) whether corruption has a significant impact on public expenditure and govern-
ment revenue; (ii) whether this impact is stronger in countries with more solid democratic
institutions; and (iii) whether the sign of this impact is consistent with the Elasticity Condi-
tion, suggesting by Proposition 2 that a minority may indirectly benefit from corruption.
Methodology
In this cross-country analysis, the dependent variable is regressed against an index of cor-
ruption and a set of control variables, in order to highlight the marginal effect of corruption
on the variable of interest.
The dependent variable is, in different specifications, total government expenditure, gov-
ernment revenue, and government tax revenue. In the model discussed in Sect. 3, these three
variables are equivalent as a result of simplifying assumptions concerning the public budget
constraints; in practice, however, government revenue and government expenditure differ
by the amount of the government deficit (or surplus), which can be financed by borrowing
domestically and externally, spending (accumulating) government deposits at banks, bor-
rowing from the central bank, and through the purchase or sale of nonfinancial assets (for
instance, through privatizations). Insofar as the citizens fully discount the future flows asso-
ciated with these financial operations (Ricardian equivalence),38 public expenditure reflects
the citizens’ preferences on the allocation of resources more closely than government rev-
enue. Both variables can however be affected by external aid, in the form of grants (counted
in both variables) and concessional loans (counted as public expenditure but not as revenue).
Tax revenue, on the other hand, more closely reflects what the citizens have to pay to obtain
a given supply of public goods and services, and can be a better indicator of their preferences
in the failure of Ricardian equivalence or if external assistance accounts for a large share of
public expenditure.
For each specification, the regression has been run over three different samples of coun-
tries, one including the entire set of 107 countries for which complete data are available,
the other two including only countries that have been classified as “democratic.” This clas-
sification is based on the Index of Democracy (ID) published by the Economist Intelligence
Unit,39 which measures the degree of democracy of a country’s institutions on a scale from
0 to 10 (10 representing the strongest degree of democracy) on the basis of sixty indicators
grouped by five institutional categories, ranging from the electoral process to civil liber-
ties. On the basis of this index, the Economist Intelligence Unit classifies countries as “full
democracies” (whose index is above 7.95), “flawed democracies” (with a value of the index
between 6 and 7.95), “hybrid regimes” (with the index ranging from 4 to 6), and “authori-
tarian regimes” (with an index below 4). Only the latter are not considered democracies. In
line with this classification, in this study countries are considered “democracies” if the value
of the ID index is equal or larger than 4, and “strong democracies” if the value of the index
is equal or larger than 6.40
38 Including any inflation tax associated with excessive central bank financing.
39 Available on the Internet at the page http://graphics.eiu.com/PDF/Democracy%20Index%202008.pdf.
40 Corruption appears to be lower, on average, in countries with stronger democratic institutions. The average
value of the CPI index (discussed below) in 2007 increases from 4.01 over the whole sample to 4.44 in the
178 Public Choice (2011) 148: 163–196
Aside from the corruption index, which is the explanatory variable of interest, a set of
control variables has been included to take account of the impact of other factors that also
affect public expenditure and taxation. This set includes three macroeconomic variables
(nominal GDP, per capita GDP, and inflation), the old-age dependency ratio, and the Gini
coefficient of income distribution. The share of population living in rural areas, initially in-
cluded in the regression, turned out not to have a statistically significant impact and has been
expunged from the final specification. Nominal GDP is intended to capture any economies
(or dis-economies) of scale in the provision of public services associated with the size of the
economy. GDP per capita captures the relation between public expenditure and the level of
development; the share of resources allocated to public consumption may decline with the
average income level, as a result of changes in preferences and of the reduced need to redis-
tribute resources to the poor. Inflation captures the impact on preferences of macroeconomic
stability; a more unstable macroeconomic environment (reflected in higher inflation levels)
may alter the citizens’ preferences between public and private consumption. The old-age de-
pendency ratio (the share in the population of citizens aged 65 or more) is the only variable
with a highly significant impact on all the dependent variables. In the regression, nominal
GDP and per capita GDP are expressed in logarithms, and all other variables are in levels.
Unsurprisingly, the corruption indexes used in these regressions appear to be strongly
collinear with the other explanatory variables.41 To reduce multicollinearity, these indexes
have been replaced, in the OLS equations, with the residuals of their regression on the other
explanatory variables over the same sample of countries. This transformation does not alter
the coefficient (and t statistics) of the corruption index or the statistics of the entire regres-
sion, but yields more accurate results for the other explanatory variables. Collinearity among
the other regressors does not appear to be very strong.42
To control for possible simultaneity bias arising from reverse causality from the depen-
dent variables on corruption, the equations have been estimated using both the ordinary least
squares (OLS) and two-stage least squares (2SLS) method; in the 2SLS estimation, inflation
and the index of democracy have been used as instruments for the corruption variable.43
Potential reverse causality from the dependent variables to inflation has been addressed by
using a five-year inflation average; the inflation variable appears to be only weakly nega-
tively correlated with the dependent variables,44 and its exclusion from the equations does
not substantially alter the results.
Data
Data on total government expenditure and net lending and on government revenue and
grants, expressed in percent of GDP and used as dependent variables in two specifications
set of democracies and 4.93 in the set of strong democracies. The corresponding averages for the CCI index
are −0.061, 0.150 and 0.383.
41 The R 2 of the regression of these indexes on the other explanatory variables range form 0.61 to 0.76,
depending on the index (CPI or CCI, discussed below), on the sample (all countries / democracies / strong
democracies), and on the year (2006 or 2007).
42 The R 2 of the regression among the other explanatory variables range between 0.32 and 0.42 depending
on the index, sample, and year.
43 Both these variables appear to have a significant impact on corruption. The R 2 of the regression of the
corruption index on these two variables plus a constant ranges from 0.55 to 0.85 depending on the index,
sample, and year. For empirical evidence of the impact of inflation on corruption see Braun and di Tella
(2004).
44 The correlation coefficients range from −0.14 to −0.36 depending on the sample and the year.
Public Choice (2011) 148: 163–196 179
of the equation, are taken from the World Economic Outlook (WEO) database of the Inter-
national Monetary Fund; data on tax revenue in percent of GDP (used as dependent variable
in the third specification) have been compiled from various internal IMF sources.
Two different indicators of corruption have been used in the analysis; both are based on
the perceptions of professionals that are well informed about the countries being assessed.
The Corruption Perception Index (CPI), published by of Transparency International and
available on the Internet at the page http://www.transparency.org/policy_research/surveys_
indices/cpi, is a composite index calculated from 14 different indexes provided by 12 dif-
ferent institutions, measuring corruption in the “public and political sectors.” The source
indexes, in turn, are based on surveys of “business leaders” operating in the country and of
resident and nonresident experts. The CPI index ranges from 0 to 10 and diminishes with
the degree of corruption, 10 representing a complete absence of corruption. The Control of
Corruption Index (CCI) is computed by the World Bank Institute using a similar methodol-
ogy, but from a larger number of sources. The CCI index is also diminishing with the degree
of corruption, but ranges approximately on a scale from −2.5 to 2.5.45 Possibly owing to
a partial overlap of their sources, the two indexes are strongly correlated (the correlation
coefficient is between 97% and 98% in all samples).
Data on three other macroeconomic variables (nominal GDP, per capita GDP and infla-
tion) are also taken from the World Economic Outlook database of the IMF. Nominal and
per capita GDP are expressed in dollars at purchasing power parity. The old age dependency
ratio and the Gini coefficient are taken from the World Development Indicators database of
the World Bank.
In regressions using government expenditure or revenue and grants as the dependent vari-
able, 2007 data have been used for all variables except for inflation and the Gini coefficient;
in regressions using tax revenue as the dependent variable, all variables, except inflation and
the Gini coefficient, are for 2006. Inflation has been defined as the average increase in the
consumer price index over the period 2003–2007, to derive a more significant indicator of
the general degree of macroeconomic stability in the period of interest, purified from the
short-term impact of temporary shocks and cyclical fluctuations in prices. For the Gini coef-
ficient, owing to a lack of data, the most recent measure since 2000 has been taken for each
country.
Results
The results of the regressions do not show a significant impact of corruption on public ex-
penditure. The coefficient of the corruption index has a positive sign in all regressions but is
not significantly different from zero even at a 10% confidence level. The coefficient does not
vary much with the degree of democracy, although it is somewhat higher among strongly
democratic countries (Table 1). The variables included in the regression explain more than
50% of the variance of public expenditure, which appears to be significantly affected by the
old-age dependency ratio and to a lesser extent by per capita income and income inequality
(the Gini coefficient); the coefficients of the other variables are not significantly different
from zero.
Government revenue and grants and government tax revenue (in percent of GDP) appear
instead to be significantly affected by corruption in democratic countries (Tables 2 and 3).
While the corruption coefficients are not significantly different from zero when the regres-
sion is run over the entire sample, they become significant at the 10% confidence level (and
45 The actual observations for 2006 range from −1.83 to 2.56; for 2007, they range from −1.88 to 2.62.
180 Public Choice (2011) 148: 163–196
Table 1 OLS estimates of the impact of corruption on public expenditure in percent of GDP (2007)
Sample All Democracies Strong All Democracies Strong

countries democracies countries democracies
Constant 18.054∗∗ 27.746∗∗∗ 52.874∗∗∗ 17.866∗∗ 27.667∗∗∗ 52.862∗∗∗

(2.57) (3.18) (3.44) (2.54) (3.17) (3.43)
CCI 1.213 1.195 1.401 – – –
(1.08) (0.91) (0.80)
CPI – – – 0.461 0.479 0.503
(0.84) (0.74) (0.60)
Per-capita income 1.742∗ 0.479 −1.988 1.785∗ 0.494 −1.980
(1.71) (0.39) (−1.21) (1.75) (0.40) (−1.20)
GDP −0.584 −0.459 −0.557 −0.596 −0.455 −0.565
(−1.33) (−0.89) (−0.97) (−1.35) (−0.90) (−0.98)
Inflation 0.181 −0.012 −0.490 0.179 −0.002 −0.491
(1.42) (−0.01) (−1.59) (1.41) (−0.01) (−1.58)
Old-age dependency ratio 0.862∗∗∗ 1.020∗∗∗ 1.189∗∗∗ 0.853∗∗∗ 1.018∗∗∗ 1.188∗∗∗
(3.61) (3.66) (3.63) (3.57) (3.65) (3.61)
Gini coefficient −0.201∗ −0.186 −0.193 −0.202∗∗ −0.186 −0.193
(−2.07) (−1.64) (−1.35) (−2.08) (−1.64) (−1.34)
R2 0.520 0.537 0.577 0.518 0.535 0.575
Adjusted R 2 0.491 0.499 0.523 0.489 0.497 0.521
Number of observation 107 80 54 107 80 54
t -statistics in parentheses. ∗ = significant at the 10% confidence level; ∗∗ = significant at the 5% confidence
level; ∗∗∗ = significant at the 1% confidence level
Data sources: IMF and World Bank data; CPI is the corruption perception index of Transparency Interna-
tional. The classification of countries is taken from the Economist Intelligence Unit (EIU)
in one case at the 5% level) when the sample is restricted to the set of democracies, and at the
5% confidence level (and in one case at the 1% level) when the sample is restricted to strong
democracies. The value of the coefficient is also larger, the higher the degree of democracy
in the sample, suggesting that corruption has a stronger (negative) impact on government
revenue and taxes in countries with more democratic institutions.
These results are robust to the use of a 2SLS estimation method in place of ordinary least
squares (Table 4), suggesting the absence of a significant simultaneity bias. Since the results
are very similar for both corruption indexes, only results obtained using the CCI index are
reported in Table 4.
Altogether, these results suggest that, in democratic countries, corruption has a significant
impact on the allocation of resources between private and public consumption, although this
impact is more evident on the revenue than on the expenditure side.
This impact is significant, however, only in democratic countries; in the sample as a
whole, when countries not classified as “democratic” are included, the relation between
corruption and the level of public expenditure or revenue appears much weaker. Conversely,
when the sample is restricted to the set of countries with the strongest democratic institution,
the coefficient of the corruption index increases and becomes more significant. This suggests
that corruption affects the allocation of resources mainly through the democratic aggrega-
tion of preferences; where democracy is weak, the citizens cannot respond to corruption by
Public Choice (2011) 148: 163–196 181
Table 2 OLS estimates of the impact of corruption on government revenue and grants in percent of GDP
(2007)

countries democracies l countries democracies
Constant 9.078 22.334∗∗ 55.185∗∗∗ 8.903 22.139∗∗ 55.034∗∗∗

(1.05) (2.12) (3.13) (1.03) (2.10) (3.10)
CCI 1.909 2.785∗ 4.136∗∗ – – –
(1.39) (1.74) (2.07)
CPI – – – 1.010 1.373∗ 1.880∗
(1.50) (1.76) (1.95)
Per-capita income 2.717∗∗ 1.469 −1.397 2.756∗∗ 1.497 −1.364
(2.17) (0.98) (−0.74) (2.21) (1.00) (−0.72)
GDP −1.268∗∗ −1.285∗∗ −1.689∗∗ −1.277∗∗ −1.289∗∗ −1.702∗∗
(−2.35) (−2.09) (−2.55) (−2.37) (−2.10) (−2.56)
Inflation 0.253 −0.086 −0.914∗∗ 0.254 −0.083 −0.910∗∗
(1.62) (−0.36) (−2.57) (1.62) (−0.34) (−2.55)
Old-age dependency ratio 0.863∗∗∗ 0.978∗∗∗ 1.133∗∗∗ 0.855∗∗∗ 0.974∗∗∗ 1.126∗∗∗
(2.94) (2.90) (3.01) (2.92) (2.89) (2.97)
Gini coefficient −0.119 −0.151 −0.166 −0.120 −0.151 −0.167
(−0.99) (−1.10) (−1.01) (−1.01) (−1.10) (−1.01)
R2 0.430 0.468 0.562 0.432 0.468 0.558
Adjusted R 2 0.396 0.424 0.506 0.398 0.424 0.501
inducing a change in the policy adopted by the government; although corruption could affect
government policies through other channels (for instance, as noted above, corrupt ministers
tend to expand public expenditure), these effects are either statistically insignificant or offset
one another.
The sign of the coefficients of the corruption index deserves particular attention. In all
regressions, the corruption coefficient is positive, implying that corruption reduces the share
of resources allocated to public consumption. In the model discussed in Sect. 3, this implies
that the Elasticity Condition holds; which, in turn, implies, by Proposition 2, that a minority
of citizens benefits indirectly from corruption. In other words, in democratic countries, cor-
ruption induces the majority of citizens to reduce taxes and (indirectly) public expenditure,
and this effect benefits a minority who would have preferred lower taxes in any case.
The different impact of corruption on revenue and expenditure is somewhat unexpected.
Having used observations for only one year, it is not surprising that the difference between
these two variables should vary widely across countries, that would find themselves at dif-
ferent phases in the business and policy cycle; nevertheless, there is no reason to expect
that these differences should be correlated with the level of corruption. In the data, however,
corruption appears to have a significant impact on the government deficit in democratic
182 Public Choice (2011) 148: 163–196
Table 3 OLS estimates of the impact of corruption on tax revenue in percent of GDP (2006)

countries democracies countries democracies
Constant −20.024∗∗ −3.001 19.095 −19.929∗∗ −3.610 21.640

(−2.26) (−0.33) (1.11) (−2.24) (−0.39) (1.21)
CCI 1.206 3.502∗∗ 5.951∗∗∗ – – –
(0.86) (2.42) (2.80)
CPI – – – 0.467 1.128∗ 2.063∗∗
(0.71) (1.71) (2.22)
Per-capita income 3.926∗∗∗ 1.566 −0.039 3.913∗∗∗ 1.633 −0.077
(3.04) (1.13) (−0.02) (3.02) (1.15) (−0.04)
GDP −1.229∗∗ −0.705 −0.896 −1.221∗∗ −0.685 −0.911
(−2.09) (−1.22) (−1.22) (−2.08) (−1.15) (−1.20)
Inflation 0.192 −0.279 −0.523 0.196 −0.261 −0.560
(1.18) (−1.19) (−1.54) (1.20) (−1.09) (−1.60)
Old-age dependency ratio 0.302 0.647∗∗ 0.589 0.301 0.628∗ 0.523
(0.98) (2.02) (1.41) (0.98) (1.91) (1.22)
Gini coefficient 0.185 0.183 0.085 0.184 0.185 0.057
(1.34) (1.39) (0.43) (1.33) (1.37) (0.28)
R2 0.318 0.422 0.310 0.316 0.396 0.263
Adjusted R 2 0.268 0.366 0.201 0.266 0.337 0.147
countries, and this effect is even stronger in the subset of “strong democracies.”46 One pos-
sible explanation, that deserves to be investigated further, may relate to the link between
corruption and external aid: if more corrupt countries also receive more aid, aid could al-
low their governments to maintain a high level of spending despite the revenue reduction
induced by the citizens’ response. Corruption may also affect the citizens’ propensity to de-
lay tax payments to the future through the accumulation of public debt. This decision could
stem, for instance, from the hope that corruption will decline over time, producing a fairer
tax system and reducing the cost of the public provision of goods, thus creating fiscal space
for debt repayment. Corruption could also be used as a means of conditioning future policy
46 The correlation coefficients between the “government deficit” variable (defined here as the difference be-
tween expenditure and revenue and grants, in percent of GDP, as computed from the data) and the two
corruption indexes is weak; the correlation coefficients range from −0.09 to 0.05 depending on the index,
sample, and year. However, controlling for other explanatory variables, corruption appears to have a signif-
icant positively impact on the government deficit in democratic countries. Regressions of the government
deficit variable on the same set of explanatory variables used for expenditure and revenue yields a negative
coefficient for the corruption index (implying that corruption tends to widen the deficit) which is significant
at the 5% confidence level in the set of democratic countries, and at the 1% confidence level (with a larger
absolute value, implying a stronger impact of corruption on the deficit) among strong democracies.
Table 4 2SLS estimates of the impact of corruption on public expenditure, revenue, and tax revenue in percent of GDP (2006)
Dependent variable Expenditure and net lending Revenue and grants Tax revenue
Sample All Democracies Strong democracies All Democracies Strong democracies All Democracies Strong democracies
Constant 19.25∗ 40.19∗∗∗ 64.34∗∗∗ 6.74 43.68∗∗∗ 93.62∗∗∗ −26.96∗∗ 14.67 53.92∗∗
Public Choice (2011) 148: 163–196
(1.84) (3.18) (3.05) (0.52) (2.84) (3.74) (−2.11) (1.06) (2.53)

CCI 0.090 2.798 3.405 −0.811 5.018∗∗ 8.255∗∗∗ −1.955 4.827∗∗ 7.041∗∗∗
(0.05) (1.35) (1.54) (−0.36) (1.99) (3.15) (−0.88) (2.11) (2.97)
Per-capita income 1.576 −1.504 −4.098 3.088* −1.807 −7.189∗∗ 4.698∗∗∗ −1.157 −5.267∗∗
(1.09) (−0.80) (−1.60) (1.72) (−0.78) (−2.36) (2.72) (−0.55) (−1.98)
GDP −0.583 −0.226 −0.211 −1.371∗∗ −0.965 −1.088 −1.202∗∗ −0.490 −0.530
(−1.26) (−0.42) (−0.36) (−2.38) (−1.47) (−1.55) (−2.06) (−0.87) (−0.79)
Old-age dependency ratio 0.861∗∗∗ 1.107∗∗∗ 1.256∗∗∗ 0.853∗∗∗ 1.082∗∗∗ 1.255∗∗∗ 0.346 0.707∗∗ 0.764∗
(3.62) (3.96) (3.78) (2.87) (3.18) (3.18) (1.13) (2.24) (1.90)
Gini coefficient −0.165* −0.108 −0.143 −0.088 −0.039 −0.069 0.209 0.251∗ 0.207
(−1.69) (−0.90) (−0.94) (−0.72) (−0.26) (−0.39) (1.55) (1.91) (1.09)
R2 0.507 0.526 0.550 0.397 0.451 0.503 0.268 0.411 0.300
Number of observation 105 78 52 105 78 52 89 69 45
t -statistics in parentheses. ∗ = significant at the 10% confidence level; ∗∗ = significant at the 5% confidence level; ∗∗∗ = significant at the 1% confidence level
Data sources: IMF and World Bank data; CPI is the corruption perception index of Transparency International. The classification of countries is taken from the Economist
Intelligence Unit (EIU)
183
184 Public Choice (2011) 148: 163–196
decisions (as discussed, in a different setting, in the next section). Further investigation of
these issues could be useful.
5 Voting on law enforcement
Thus far, the expected value of punishment, π , has been treated as exogenous. In fact, this
variable is affected by a variety of institutional decisions concerning the accountability of
public officials, the mandate and resources of law enforcement agencies, the power of the
judiciary, the definition of crimes and the punishment that can be imposed, and the con-
stitutional rights of individual citizens and public officials. These decisions, which we can
indicate in aggregate as the “law enforcement regime,” jointly determine the probability that
an official guilty of corruption can be detected, successfully prosecuted, and punished, as
well as the value of the punishment he would suffer upon conviction.47 In democracy, de-
cisions concerning the reform of the law enforcement regime are ultimately made by the
citizens.
The law enforcement regime cannot be changed rapidly (although some parts of it, such
as the budget of the police, can) and its reforms produce their effects with a long time lag;
it is therefore set (or reformed) taking a long-term view of its impact on the variables of
interest. By affecting the incentives to engage in corruption, the law enforcement regime
determines which candidates are honest or corrupt and thus has an indirect impact on the
policy choice that is adopted in equilibrium. It is then conceivable that, when deciding on
law enforcement, the citizens take this impact into account. In this sense, decisions on law
enforcement are comparable to decisions on constitutional reform: just like constitutions,
which “are not really made behind a Rawlsian veil of ignorance” (van Winden 1999: 20),48
the law enforcement regime is chosen taking into consideration its impact on the subsequent
political equilibrium. Citizens who stand to benefit (indirectly) from the impact of corruption
on policy choice will tend to support institutional arrangements that are more favorable
to corruption, emphasizing, for instance, the personal right to privacy, protection against
intrusive investigations and disclosure requirements, and defense in court; whereas citizens
who stand to lose from corruption will vote in favor of giving more power and resources to
the judiciary, police, and public prosecutors. Citizens who stand to suffer from corruption
will also be less favorable to policies that create opportunities for illegal rent-seeking, such
as protectionism.
In this model, the collective decisions on law enforcement are formalized as a one-
dimensional choice on a single parameter, the expected value of punishment π , that sum-
marizes the effects of all decisions concerning the enforcement of laws against corruption.
The value of π is decided at an initial “constitutional” stage of the game, when the citi-
zens delegate this choice to an elected representative—the “constituent”—who decides only
on the basis of his own personal preferences, with no possibility of a credible precommit-
47 In this paper, “law enforcement” refers only to the enforcement of rules prohibiting corruption. There is, of
course, a large area of law enforcement that is not related to corruption, that is not discussed here. Empirical
evidence of how the “effectiveness of the legal system” affects corruption can be found in Herzfeld and Weiss
(2003).
48 Similar considerations may apply more generally to all institutional decisions bearing long-term effects on
future policy choices. The shift toward central bank independence has been explained, for instance, on the
basis of long-term electoral interests of elected politicians (Bernhard 2002; McGregor 2007).
Public Choice (2011) 148: 163–196 185
ment.49 The constituent chooses only the value of π; he cannot make decisions on taxes or
public expenditure, which remain the prerogative of the minister that is elected at a subse-
quent “political” stage. The constituent cannot directly influence the election of the minister,
but he can stand as a candidate in this election. The set of citizens (and hence the distribu-
tion of preferences) can vary between the constitutional and the political stage according to
a probability distribution that is common knowledge: with probability q, the distribution of
preferences does not change, while with probability 1 − q new citizens enter the community
at the political stage, shifting the median type toward “the left.”50 Two “states of the world”
are thus possible at the political stage: if the distribution of preferences is the same as in the
constitutional stage, the state is “conservative,” while if the distribution changes to the left
the state is “progressive.” 51 A majority of citizens present at the constitutional stage will
be called a “constitutional majority.” Moreover, the citizen(s) whose type ν lies at the me-
dian of the distribution at the constitutional stage will be called the “constitutional median
voter” (CMV), while the citizen(s) whose type μ lies at the median at the political stage
will be called the “political median voter” (PMV). Hence, μ = ν with probability q and that
μ = ξ > ν with probability 1 − q.
Consider the policy expectations of the citizens at the constitutional stage. They know
that, with probability 1 − q, the distribution of preferences at the political stage will shift in
favor of a higher supply of roads. For most of them, the policy choice that is adopted in this
case in the absence of corruption involves too many roads and excessive taxes. The citizens
are aware, however, that corruption can alter this choice in favor of less roads and—if the
Elasticity Condition holds—lower taxes. Under certain conditions, this outcome would be
preferable for a constitutional majority, who can thus use the law enforcement regime as a
tool to alter the political equilibrium in their favor. We assume here that is the case. Formally,
this implies that V (g ξ ; ν) < V C (ĝ ξ , b; ν): in the progressive state of the world, the CMV
prefers the policy outcome that occurs when all candidates are corrupt (whereby a corrupt
minister sets g = ĝ ξ ) to the outcome that occurs in the absence of corruption (when the
PMV is elected and sets g = g ξ ). This, of course, happens only if the Elasticity Condition
holds in the progressive state.
In the ensuing discussion, we abstract from the costs of law enforcement. Normally, a
more effective law enforcement regime (a higher value of π) involves higher costs that
are borne by the citizens through taxes (for instance, recruiting more police officers raises
the public wage bill). These costs are taken into account when choosing the appropriate
level of law enforcement. Here, we assume that law enforcement is costless, to emphasize
how corruption can arise for mere political reasons. The model can easily be generalized to
include positive, increasing enforcement costs.52
49 This is, of course, a simplified representation of the process. In most cases (the founding of a new country
being a rare exception), the institutional environment is not set once and for all at a separate constitutional
stage, but evolves through a gradual and ongoing process of institutional innovation and reform.
50 With income inequality, this situation would occur if new citizens entered the community at the lowest end
of the income distribution.
51 We assume that the number of citizens is the same in both states: in the conservative state, the preferences
of the citizens that enter the community at the political stage are distributed like those of the citizens that are
already present at the constitutional stage.
52 The discussion in the rest of this section also applies to the version of the model with income inequality
and proportional income taxes.
186 Public Choice (2011) 148: 163–196
5.1 Law enforcement and policy choice
At the political stage, when π has already been set, the citizens elect the minister from
among one of the two sets of citizens, corrupt and honest. As shown in the previous section,
the median voter (the PMV) is pivotal. We have also shown that the citizens do not always
end up electing an honest candidate and that if, at the given value of π , all honest candidates
have policy preferences too different from those of the PMV, the citizens elect a corrupt
candidate who, as minister, implements a policy that is closer to the preferences of the PMV
and hence of the majority.
To enhance tractability, we assume here that u(g) is thrice differentiable with u (g) ≤ 0
for all g. The implications of this condition will become clear below,53 but the results can
be generalized to a wider set of utility functions.
Note that (i) when π ≥ π μ the PMV is honest and is elected; (ii) when π < π μ , the PMV
prefers, among all honest candidates, the candidate with the highest value of θ in H (π), who
is a candidate of type θ ∗ (π) that has (credibly) signaled his intention to remain honest once
elected; (iii) when π is below some threshold π A ≡ (α), candidates of type α = (p ∗ /p)μ
are corrupt and the PMV prefers them to any other candidate in C(π);54 (iv) when π > π A
citizens of type α are honest and the PMV prefers, among all corrupt candidates, those
with the lowest value of θ , who are citizens of type θ ∗ (π) that have failed to manifest their
intention to act honestly (and are thus correctly deemed to be corrupt).
Hence,
(i) when π ≥ π A , the PMV prefers an honest candidate of type θ ∗ (π) to all other honest
candidates and a corrupt candidate of the same type θ ∗ (π) to all other corrupt candi-
dates. The citizens elect a minister of type γ ≡ θ ∗ (π), who is honest or corrupt de-
pending on whether the utility of the nonelected PMV is higher if he acts honestly or
53 The restriction u ≤ 0 ensures that the derivative of g̃ θ with respect to θ for a corrupt minister is larger
than the derivative of g θ with respect to θ for an honest minister of the same type. Let po ∈ {1, p∗ } be the
net unit price of roads paid by the minister (which is equal to 1 if the minister is honest and to p∗ < 1 if
he is corrupt); let g o ∈ {g θ , g̃ θ } be the policy choice of the minister (equal to g θ if he is honest and to g̃ θ
if he is corrupt). The FOC for maximization requires that u (g o ) = po /(nθ); differentiating with respect
to g and to θ yields u (g o )dg o = −po /(nθ 2 )dθ = −u (g o )/θdθ , whence dg o /dθ = −u (g o )/(u (g o )θ).
Differentiating dg o /dθ with respect to po yields:
o 2
d 2 go [u (g )] θ − u (g o )u (g o )θ dg o 1 u (g o )u (g o ) dg o
=−
=− 1−
.
dθdp o [u (g )θ]
o 2 dp o θ [u (g )]
o 2 dpo
Since dg o /dpo < 0 for po > 1, a sufficient condition for this derivative to be positive in the interval (1, p∗ ) is
that u (g) ≤ 0 for all g. When this occurs, dg o /dθ is larger when the minister is honest (po = 1) than when
he is corrupt (po = p∗ < 1). With income inequality and proportional income taxes, redefining po ∈ {1, p}
as the unit price of roads paid by the government, the FOC for maximization becomes u (g o ) = ypo /n − ab,
which yields: u (g o )dg o = po /ndy = [u (g o ) + ab]/ydy, whence dg o /dy = [u (g o ) + ab]/(u (g o )y).
Differentiating dg o /dθ with respect to po and recalling that b = po − 1 yields:
o
d 2 go {[u (g o )]2 y − u (g o )u (g o )y − abu (g o )y} dp
dg o
o + au (g )y
=
dydp o [u (g )y]
o 2

1 [u (g ) + ab]u (g o ) dg o
o a
= 1− + o .
y [u (g o )]2 dpo u (g )y
Since dg o /dpo < 0 for po > 1, a sufficient condition for this derivative to be negative (and hence for
d 2 g o /dθdpo to be positive) in the interval (1, p∗ ) is that u (g) ≤ 0 for all g.
54 See the proof of Proposition 1 in the Appendix.
Public Choice (2011) 148: 163–196 187
corruptly—formally, whether or not V (θ ∗ (π); μ) > V C (θ ∗ (π), b; μ), which is equiva-

lent to
g γ − p g̃ γ
μ[u(g γ ) − u(g̃ γ )] − > 0, (5.1)
n
where g γ and g̃ γ are both functions of θ ∗ (π), and hence of μ and π . The left-hand
side (LHS) of (5.1) is increasing in θ ∗ (π),55 hence in π , which implies that a majority
prefers the honest candidate if π lies above some value π B < π μ .56 Since the LHS is
also increasing in μ, π B is a decreasing function of μ, implying that π B (ν) > π B (ξ ).
(ii) if π < π A , the PMV still prefers an honest candidate of type θ ∗ (π) to all other honest
candidates, but his preferred corrupt candidate is now a citizen of type α. The citizens
elect the honest candidate of type θ ∗ (π) if and only if V (θ ∗ (π); μ) > V C (α, b; μ),
which is equivalent to
g γ − p g̃ α
μ[u(g γ ) − u(g̃ α )] − > 0; (5.2)
n
otherwise, they elect a corrupt citizen of type α. Since the LHS is increasing in θ ∗ (π)
(hence in π) for θ ∗ (π) < μ,57 this inequality holds only if π lies above some threshold
πC > πB.
Note that, while θ ∗ (π) does not depend on μ but only on π , the values of α, π A , π B and
π C change with μ and will be hereafter indicated when appropriate as functions of μ. It is
easily seen that π A (ν) < π A (ξ ), since α(ν) < α(ξ ), and that π B (ν) < π B (ξ ) (the LHS of
55 The first derivative of the LHS of (5.1) with respect to γ ≡ θ ∗ (π ) is equal to:
dg γ − p d g̃ γ
dg γ d g̃ γ
μ u (g γ ) − u (g̃ γ )
dγ dγ
−
dγ dγ n

μ dg γ μ p d g̃ γ
=γ − 1 u (g γ ) − − ∗ u (g̃ γ ) .
γ dγ γ p dγ
Since p > p∗ , given that u (g γ ) > u (g̃ γ ), a sufficient condition for the first derivative to be positive is that
dg γ /dθ > d g̃ γ /dθ , which is assured by the restriction u (g) ≤ 0. With income inequality and proportional
income taxes, (5.1) becomes: u(g γ ) − u(g̃ γ ) − (g γ − pg̃ γ )/(nμ) > 0 (where 1/μ is the gross income of the
median voter), and the first derivative of its LHS with respect to γ becomes:
dg γ − p d g̃ γ
γ dg γ γ d g̃ γ dγ dγ
u (g ) − u (g̃ ) −
dγ dγ nμ

1 dg γ p d g̃ γ
= u (g γ ) − − u (g̃ γ ) −
nμ dγ nμ dγ
γ γ
γ dg γ γ d g̃
= u (g γ ) 1 − − u (g̃ γ ) 1 − − ab .
μ dγ μ μ dγ
Since u (g γ ) > u (g̃ γ ), a sufficient condition for the first derivative to be positive is that dg γ /dθ > d g̃ γ /dθ ,
which is assured by the restriction u (g) ≤ 0.
56 Since the PMV is always elected when he is honest, the hypothetical case in which π B > π μ is irrelevant.
Given that the PMV always prefers an honest citizen of his type to a corrupt citizen of his type, then π A < π μ
implies π B < π μ .
γ γ
57 The first derivative with respect to θ C (π ) is equal to (μu (g γ ) − 1 ) dg = 1 ( μ − 1) dg ,which is positive
n dγ n γ dγ
when γ ≡ θ ∗ (π ) < μ.
188 Public Choice (2011) 148: 163–196
Fig. 5 Median voter’s utility as a function of π
(5.1) is increasing in both π and μ). If, in addition, π A < π B < π C both when μ = ν and
when μ = ζ > ν,58 in each state of the world four different law enforcement regimes are
possible (Fig. 5):
• Permissive: π ≤ π A (μ). Corruption is effectively permitted. Law enforcement is too weak
not only to prevent corruption but also to contain its distortion of the policy choice. The
citizens elect a corrupt candidate of type α(μ), who sets g = g̃ α(μ) and t = (1 + b)g̃ α(μ) /n.
An extreme case of permissive regime—“full permission”—occurs when π = 0 and cor-
ruption is nominally forbidden but remains unpunished.
• Tolerant: π A (μ) < π < π B (μ). Corruption is tolerated. The enforcement of anticorrup-
tion laws has an observable impact on policy outcomes, but corruption still occurs in
equilibrium. The citizens elect a corrupt candidate of type θ ∗ (π), who sets g = g̃ γ and
t = (1 + b)g̃ γ /n. Note that the utility of the PMV and of a majority of the citizens at the
political stage is lower under a tolerant regime than under a permissive one.
• Strong: π B (μ) ≤ π < π μ . Corruption is repressed but lingers under the surface. It does
not occur in equilibrium, but its threat has an impact on policies by reducing the set of
honest candidates. The citizens elect an honest candidate of type γ ≡ θ ∗ (π), who sets
g = g γ and t = g γ /n.
• Effective: π ≥ π μ . Corruption is completely prevented. Law enforcement is sufficiently
strong to induce the PMV to be honest; the latter is therefore elected, and sets g = g μ and
t = g μ /n.
The citizens elect a corrupt candidate only under a permissive or tolerant regime (when
π < π B ). Since π B (ν) < π B (ξ ), when π < π B (ν), the citizens elect a corrupt candidate in
58 π C is always larger than π B because, when π > π C , the PMV prefers an honest candidate of type θ ∗ (π )
to a corrupt candidate of type α while, when π < π B , he prefers a corrupt candidate of type θ ∗ (π ) to an
honest candidate of the same type. Since, by definition, the PMV prefers a corrupt candidate of type α to a
corrupt candidate of any other type, π B cannot be larger than π C . The case in which π A > π C > π B is less
interesting: the PMV prefers either an honest candidate of type θ ∗ (π ) (if π > π C ) or a corrupt candidate of
type α (if π < π C ). As discussed below, in this case, π is set either equal to π ν or to zero.
Public Choice (2011) 148: 163–196 189
both states of the world. If π is larger than π A (ξ ), in both states of the world the citizens
elect a corrupt candidate of type θ ∗ (π). Otherwise, the citizens elect candidates of different
types depending on the state of the world. When π B (ν) < π < π B (ξ ), the citizens elect an
honest candidate of type θ ∗ (π) in the conservative state59 and a corrupt candidate (of the
same type) in the progressive state. When π > π B (ξ ), the citizens elect an honest candidate
in both states of the world, which is of the same type θ ∗ (π) if π < π ν , and of different types
otherwise.
5.2 Deciding on law enforcement
This relation between the value of π and the candidate that is elected (hence, the policy that
is adopted) in equilibrium is taken into account by the citizens at the constitutional stage of
the game. Note that setting π at any level 0 ≤ π < π A (ν) always results in the election of a
corrupt candidate of type α(μ); in the same way, setting π at any level π ≥ π ξ induces the
election of the (honest) PMV. In these cases, we assume that the citizens at the constitutional
stage will choose the lowest level of π within these intervals (0 and π ξ respectively), as
would occur if the costs of law enforcement increased even modestly in π .
Let z(π, μ) be the type of the candidate that is elected, at the political stage, under a given
law enforcement regime (identified by π) in a given state of the world (identified by μ). Let
ι(π, μ) be an indicator equal to 1 if z is honest (which depends, of course, on π and on μ
as discussed above) and to 0 otherwise. Let ι∗ = 1 − ι, and let W (π, μ; θ ) be the indirect
utility that a citizen derives when z(π, μ) is elected (whether honest or corrupt depending
on ι(π, μ)) and implements his own preferred policy. W (π, μ; θ ) is equal to

gz p g̃ z ιg z + ι∗ p g̃ z
ι θ u(g z ) − + ι∗ θ u(g̃ z ) − = θ ιu(g z ) + ι∗ u(g̃ z ) −
n n n
= θ (π, μ) + (π, μ),
where (π, μ) = ιu(g z ) + ι∗ u(g̃ z ) and (π, μ) = −(ιg z + ι∗ p g̃ z )/n depend on ι, g z and
g̃ z , hence on π and μ. The function W (π, μ; θ ) is common knowledge.
Since the value of μ at the political stage is a random variable at the constitutional stage,
the expected value of the indirect utility that a citizen of type θ derives for a given value
of π before the realization of μ is known is equal to W E (π, q; θ ) ≡ qW (π, ν; θ ) + (1 −
q)W (π, ξ ; θ ) = θ ∗ (q, π) + ∗ (q, π), where ∗ (q, π) = q(π, ν) + (1 − q)(π, ξ ) and
∗ (q, π) = q(π, ν) + (1 − q)(π, ξ ).
The choice of π is delegated to the constituent elected at the constitutional stage, who
sets it at the level that maximizes his own expected utility. It is easy to see that the CMV
is pivotal at this stage: in a pairwise choice between two values of π (say, π1 and π2 ), the
CMV always votes with the majority.60 Therefore, the constitutional stage of the game has
a unique equilibrium in which the CMV is elected as constituent and sets π at the level that
maximizes W E (π, q; ν).
The characteristics of this equilibrium depend on the relative values of some key pa-
rameters of the model, including π B (ξ ), π ν and π ξ , as well as W (π B (ξ )), ξ ; ν) and
59 Unless π ν < π < π B (ξ ), in which case, in the conservative state, they elect the PMV.
60 Given the value of q, without loss of generality, let us order the two alternatives so that ∗ (π , q) >
1
∗ (π2 , q). A citizen at the constitutional stage prefers π1 to π2 if and only if W E (π1 , q; θ) > W E (π2 , q; θ),
which yields: θ > θ ∗∗ (π1 , π2 ) ≡ [∗ (π2 , q) − ∗ (π1 , q)]/[ ∗ (π1 , q) − ∗ (π2 , q)]. Hence, the majority
always includes the CMV.
190 Public Choice (2011) 148: 163–196
W (π A (ξ ), ξ ; ν). Hereafter, we discuss the case in which π B (ξ ) < π ν and the case in which
π B (ξ ) > π ν and W (π B (ξ )), ξ ; ν) < W (π A (ξ ), ξ ; ν). The case in which π B (ξ ) > π ν but
W (π B (ξ )), ξ ; ν) ≥ W (π A (ξ ), ξ ; ν) will not be discussed because it yields a qualitatively
similar outcome to the case in which π B (ξ ) < π ν .61
When π B (ξ ) < π ν , corruption is always prevented in equilibrium in both states of the
world:
Proposition 4 When π B (ξ ) < π ν , corruption does not occur in any state of the world. In the
conservative state it is completely prevented, while in the progressive state it is repressed and
alters the equilibrium policy choice in a direction favorable to the constitutional majority.
In other words, the constitutional majority implements a law enforcement regime which
is sufficiently strong to prevent any policy distortions in the conservative state and to prevent
corruption in the progressive state, but it is also sufficiently weak to induce the election of
the CMV instead of the PMV in the progressive state. This is obtained by ensuring that all
other candidates that would be preferred to the CMV by a majority in the progressive state
would be corrupt.
This outcome is certainly preferred by a constitutional majority to all other possible out-
comes but is not always feasible. If π B (ξ ) > π ν and W (π B (ξ )), ξ ; ν) < W (π ν , ξ ; ν) in
particular, the CMV’s expected utility can only be maximized by allowing corruption to
occur in the progressive state, which of course would leave him worse off.
Proposition 5 When π B (ξ ) > π ν and W(π B (ξ )), ξ ; ν) < W(π ν , ξ ; ν), corruption always
occurs in the progressive state. For sufficiently large (in some cases for all) values of q, it
is prevented in the conservative state, whereas for smaller values of q and under certain
conditions, it is fully permitted in both states.
6 Conclusions
This paper has examined how corruption distorts democratic policy decisions and how this
distortion may explain the long-term persistence of corruption that has been observed in
some democratic countries. It has found that corruption distorts the allocation of resources
away from public consumption, which, under certain conditions, can benefit indirectly a mi-
nority of citizens who prefer to pay lower taxes. If the citizens are aware of this distortion
and expect a future change in preferences in favor of public consumption, they may sup-
port the establishment of institutions that favor corruption in order to alter future collective
decisions in line with their preferences. In such cases, corruption would not stem from de-
ficiencies in the democratic system, but it would cause them in the long run, by effectively
61 When π B (ξ ) > π ν but W (π B (ξ )), ξ ; ν) ≥ W (π A (ξ ), ξ ; ν), the CMV prefers to prevent corruption in both
states, setting π = π B (ξ ) > π ν . The outcome is qualitatively similar to the case discussed in Proposition 4,
with the difference that in the progressive state the citizens elect a minister that constructs more roads than
preferred by the CMV and less roads than preferred by the PMV.
Public Choice (2011) 148: 163–196 191
limiting the set of choices available to the citizens in the future. Corruption thus has an im-
portant political dimension that must be taken into account in the design of anticorruption
strategies.
The paper has also highlighted some other important features of the political economy of
corruption in a representative democracy. Somewhat counterintuitively, corruption tends to
reduce public expenditure when measured in real (physical) terms, as the citizens respond to
corruption by reducing their demand for publicly provided goods and services, whose price
is inflated by bribery. If this also results in lower taxes, a minority can benefit indirectly from
corruption even if it does not actually take part in it. The empirical evidence discussed in this
study suggests that this situation may actually have occurred in many countries. The paper
has also shown that in some cases corruption is effectively prevented but its threat remains
sufficiently strong to alter the democratic political equilibrium and the associated allocation
of resources.
Further research is needed to test the robustness of these results to some critical assump-
tions. In this model, gross income is exogenous, and is not affected by the tax rate. It would
be interesting to examine whether similar results could emerge in a general equilibrium
model in which income, public expenditure, and taxes are all endogenously determined.
The role played by the distribution of income also deserves to be examined in depth; corrup-
tion affects, and is in turn affected, by income inequality, and the latter also affects the policy
choice, including the tax regime. Further research could explore, for instance, how the results
of this model could change under non-proportional (e.g., progressive) tax regimes, where the
overall tax pressure affects the distribution of net income and—hence—preferences about
public expenditure and taxes; it would also be useful to examine how the results would
change when the marginal utility of income (the “consumer goods” in the model) is not
constant. Other studies could also examine how the results discussed here would change if
public and private goods were complements, and if private enterprises played a more active
role in influencing the political equilibrium.
The different impact of corruption on revenue and expenditure highlighted by the data
raises new questions that deserve to be more closely examined. Does external aid enable
corrupt governments to maintain comparatively high spending levels (as suggested in some
studies) even as citizens respond by lowering taxes? Are higher levels of corruption associ-
ated with higher levels of public debt? In this case, does this link stem from the absence of
Ricardian equivalence? Do the citizens expect that the tax system will be fairer in the future?
Or is the increase in debt another way of conditioning future policy decisions?
Further empirical research could also be useful to confirm, and possibly qualify, some
of the results of this paper. Additional cross-country (and panel data) studies could extend
the empirical analysis performed in Sect. 4; in particular, it would be interesting to assess
the impact of corruption, in democratic countries, on some indicator of the real value of
public expenditure: does corruption actually reduce the physical supply of public goods, as
predicted in this paper? This would however require the computation of a deflator of pub-
lic expenditure that should be comparable across countries; a possible alternative could be
to use some physical indicators of the provision of public goods and services in a more re-
stricted area of public expenditure, such the number of beds in (public) hospitals, the number
of teachers per pupil in (public) primary schools, or the extension of the road network. The
relation between trends in political preferences and the intensity of corruption could also be
verified empirically.
New research could better relate our results with some findings of previous studies. No-
tably, Mauro (1998) found evidence that corruption tends to reduce public expenditure on
health and education in favor of items on which bribes can be more easily extracted; our
192 Public Choice (2011) 148: 163–196
theoretical results suggest that the citizens, in principle, would respond to corruption by re-
ducing real expenditure in the most affected areas. These results are not mutually exclusive,
but deserve further investigation.
If confirmed by future research, the results obtained in this study could bear impor-
tant implications for the anticorruption efforts of the international community, including
the mandate and policy advice of multilateral organizations such as the IMF and the World
Bank. This paper suggests, in particular, that in some situations resistance to effective ac-
tion against corruption might not be posed only by a comparatively small group of elected
or appointed public officials but could arise from a much larger number of nonelected cit-
izens, most of whom are not directly involved in corruption but benefit indirectly from its
impact on policy decisions. In such situations, technical improvements in specific legal or
institutional areas might not be sufficient to curb corruption and stronger action on incen-
tives might be necessary. The international community could then be called to play the much
more delicate—and controversial—role of providing external economic and political incen-
tives to reduce corruption.
Acknowledgements I am grateful to Reza Baqir, Amrita Dhillon, Domenico Fanizza, Clive Fraser, Gio-
vanni Ganelli, Francesco Luna, Carlo Perroni, Peter Sinclair, participants at internal IMF seminars, and two
anonymous referees for useful comments and suggestions. I also wish to thank Mohamed El-Harrak for his
precious research assistance, and Estefanía Fallas and Margaret Griggs for their valuable editorial support. I
am however solely responsible for the content of the paper, including any remaining errors.
Appendix: Proofs of propositions
Proof of Proposition 1 The citizens rank the corrupt candidates according to the policy g̃
that they implement once elected, which, to a nonelected citizen of type θ , yields an indirect
utility equal to V C (g̃, b; θ ) (Fig. 3). A citizen of generic type θ thus prefers a candidate
of type i to a candidate of type j < i if and only if V C (g̃ i , b; θ ) > V C (g̃ j , b; θ ), which
implies θ > p ∗ (g̃ i − g̃ j )/{n[u(g̃ i ) − u(g̃ j )]}. The median voter is therefore pivotal in any
pairwise contest among candidates. The candidate that is elected is therefore the candidate
that, once elected, implements the policy choice of a nonelected median voter, that is, a
candidate of some other type α, such that g̃ α = ĝ μ . This type is identified by u (g̃ α ) =
u (ĝ μ ) = p ∗ /(nα) = p/(nμ), whence α = (p ∗ /p)μ < μ.62
Since α < μ, ceteris paribus, the elected candidate prefers a lower supply of roads than
the median voter (and hence than a majority of citizens): ĝ α < ĝ μ . Once elected, he sets
the supply of roads equal to g̃ α = ĝ μ < g μ , hence smaller than the supply preferred by a
majority in the absence of corruption.
Proof of Proposition 2 A citizen of type θ prefers the equilibrium with corruption if and
only if V C (g̃ α , b; θ ) > V (g μ ; θ ), which yields:
g μ − (1 + b)g̃ α
θ < θ O (b) ≡ (A.1)
n[u(g μ ) − u(g̃ α )]
(note that u(g μ ) > u(g α ) because g̃ α = ĝ μ < g μ ). Some citizens therefore benefit from
corruption if and only if θ 0 (b) > θ > 0. If the Elasticity Condition fails, θ 0 (b) is negative
62 With income inequality and proportional income taxes α is identified by the relation u (g̃ α ) = p/(nα) −
ab = u (ĝ μ ) = p/(nμ), yielding α = μp/(p + abnμ) < μ.
Public Choice (2011) 148: 163–196 193
and therefore (by assumption) no citizen exists who prefers the equilibrium with corruption.
If the Elasticity Condition holds, θ 0 (b) is positive and the share of citizens who prefer the
equilibrium with corruption is equal to F (θ 0 (b)) ≥ 0, which is strictly positive if and only
if θ 0 (b) > θ . By revealed preferences, these citizens are a minority “on the right” (θ 0 (b) <
μ).63
Proof of Proposition 3 When the median voter is corrupt, the latter (and hence a major-
ity) prefers, among the set of honest candidates, the candidate whose type is closest to the
median, which is a candidate of type θ ∗ (π) who has credibly signaled his honesty.64 If candi-
dates of type α defined above are corrupt (α ≤ θ(π)), by definition, the median voter prefers
them to any other corrupt candidate; if they are honest, the median voter prefers, among the
set of corrupt candidates, the candidate whose type is closest to α, which is a candidate of
type θ ∗ (π) who has signaled that he is corrupt.65
Let g γ denote the supply of roads preferred by an honest candidate of type θ ∗ (π) and
assume first that α is corrupt. If V (g γ ; μ) < V C (g̃ α , b; μ), the median voter (and hence a
majority) prefers to elect the corrupt candidate α, because the policy preferences of the pre-
ferred honest candidate θ ∗ (π) are too different from his own. In this case, a corrupt candidate
is elected even if there are honest candidates available. If instead V (g γ ; μ) ≥ V C (g̃ α , b; μ),
the median voter (and with him a majority) prefers to elect the honest candidate θ ∗ (π) that
implements the policy g γ . In this case, corruption does not actually take place in equilibrium,
but its threat—reflected in the reduction of the set of honest candidates H (π)—reduces the
supply of roads from g μ to g γ < g μ . Assume now that α is honest and let g̃ γ be the policy
choice adopted by a corrupt minister of type θ ∗ (π). In this case, the citizens elect either
an honest candidate of type θ ∗ (π) or a corrupt candidate of the same type, depending on
whether or not V (g γ ; μ) ≥ V C (g̃ γ , b; μ).
Proof of Proposition 4 If π B (ξ ) < π ν , the CMV maximizes his utility by setting π = π ν ,

hence θ ∗ (π) = ν. This decision ensures that honest candidates with the CMV’s policy pref-
erences exist at the political stage. Under this condition, one of these candidates is elected
both in the progressive state (where they implement the PMV’s ideal policy) and in the con-
servative state (where they are preferred to any corrupt candidate with policy preferences
closer to the PMV’s since, by assumption, π B (ξ ) < π ν ). In the conservative state, however,
the candidate elected is not the PMV and does not implement the ideal policy of the PMV
63 If they were a majority, in the absence of corruption they could elect a candidate of type ζ = μ/p ∗ , whose
ideal policy g ζ is equal to g̃ α . This candidate would supply the same amount of roads that would be chosen
in the equilibrium with corruption, but would impose lower taxes; hence, at least one outcome achievable
without corruption is preferred by a majority of citizens to the equilibrium outcome with corruption. Since
this outcome is not chosen in equilibrium, a fortiori the equilibrium outcome without corruption is preferred
by a majority to the equilibrium with corruption.
64 This follows from the fact that the utility perceived by the median voter when an honest candidate of type
θ is elected—equal to V (g θ , μ)—increases in θ when θ < μ:

θ
1 dg θ 1 μ dg
Vθ (g θ ; μ) = μu (g θ ) − = −1 ,
n dθ n θ dθ
which is positive if and only if θ < μ since dg θ /dθ > 0. If the median voter is corrupt, θ ∗ (π ) < μ and hence
θ < μ for all honest candidates. The median voter’s utility is thus maximized by electing the honest candidate
with the largest value of θ , which is an honest candidate of type θ ∗ (π ).
65 This follows from V (g̃ θ ; μ) being negative if and only if θ > α.
θ
194 Public Choice (2011) 148: 163–196
but that of the CMV. Hence, in the conservative state, corruption distorts the policy choice
in favor of the CMV’s preferences.
Proof of Proposition 5 Let π Z ∈ [π B (ν), π ν ] denote the value of π such that π ≥ π B (ν) and
W (π Z , ν; ν) = W (0, ν; ν) (which implies that, in the conservative state, the citizens elect
an honest candidate which leaves the CMV weakly better off than in a corrupt equilibrium
in which π is set to zero), and let π W = max(π Z , π A (ξ )) be the minimum value of π that
ensures the above outcome in the conservative state and the honesty of citizens of type
α(ξ ) in the progressive state. The CMV would not set 0 < π < π W , since his utility in
both states of the world would be at least equally large if he set either π = 0 or π = π W .66
In addition, the CMV would not set π > π ν since W (π, ν; ν) = W (π ν , ν; ν) for all π >
π ν in the conservative state (the citizens always elect the honest PMV), but W (π, ξ ; ν) <
W (π ν , ξ ; ν) for all π > π ν in the progressive state.67
Hence, the CMV sets π either at a level equal to 0 (in which case corruption is fully
permitted in both states, and the citizens elect a corrupt candidate of type α(ν) in the con-
servative state and of type α(ξ ) in the progressive state) or at some positive level in the
interval [π W , π ν ].
In this interval, the CMV’s utility increases in π in the conservative state and diminishes
in π in the progressive state. The value of π that maximizes the CMV’s expected utility
varies continuously in q.68 If π A (ξ ) > π Z , for sufficiently low values of q, the CMV prefers
to set π = 0 (and elect a corrupt candidate in both states of the world) than to set π Z <
66 In the conservative state, the CMV’s utility is equal to W (π, ν; ν) ≤ W (π Z , ν; ν) = W (0, ν; ν)—by
definition—for all positive π ≤ π Z , and is at least as large if π is set instead equal to 0. In the progres-
sive state the CMV’s utility is equal to W (π, ξ ; ν) = W (π A (ξ ), ξ ; ν) = W (0, ξ ; ν) whenever π ≤ π A (ξ )
(in which case the citizens elect a corrupt candidate of type α(ξ ), as they do when π = 0) and is equal to
W (π, ξ ; ν) < W (π A (ξ ), ξ ; ν) whenever π A (ξ ) < π < π Z < π B (ξ ) (in which case the citizens elect a cor-
rupt candidate of type θ ∗ (π ) > α(ξ ) that is less preferred by the CMV than a candidate of type α(ξ )). The
CMV’s expected utility is therefore larger if he sets π = 0 than if he sets π at any positive level below π Z .
Note that π Z < π B (ξ ) follows from the assumption that π ν < π B (ξ ).
67 In the progressive state, the CMV’s utility is diminishing in the interval [π A (ξ ), π B (ξ )] (which includes
π ν ), where higher levels of π induce the election of a corrupt citizen whose type is more distant from
α(ν) and who therefore implements a policy less preferred by the CMV. The CMV’s utility also diminishes
in π in the interval [π B (ξ ), π ξ ], where higher levels of π induce the election of an honest citizen whose
type is more distant from the CMV’s. At π = π B (ξ ), the CMV’s utility increases discontinuously, from
∗ B ∗ B
V C (g̃ θ (π (ξ )) , b; ν) < V C (g̃ ν , b; ν) ≡ W (π ν , ξ ; ν) to V (g θ (π (ξ )) ; ν) ≡ W (π B (ξ ), ξ ; ν). Since by as-
sumption W (π B (ξ ), ξ ; ν) < W (π ν , ξ ; ν), then W (π, ξ ; ν) < W (π ν , ξ ; ν) in the interval [π ν , π ξ ] in the
progressive state.
68 When π W < π < π ν , in the conservative state, the citizens elect an honest candidate of type θ ∗ (π ) < ν
and the CMV’s utility is equal to W (π, ν; ν) = V (g γ ; ν), which is continuous and differentiable in π , with
Wπ (π, ν; ν) = Vπ (g γ ; ν)dg γ /dθ ∗ (π )dθ ∗ (π )/dπ continuous and positive (since V (g γ ; ν) is increasing in
θ ∗ (π ) for θ ∗ (π ) < ν). In the progressive state, the citizens elect a corrupt candidate of type θ ∗ (π ) > α(ξ )
and the CMV’s utility is equal to W (π, ξ ; ν) = V C (g̃ γ , b; ν), which is continuous and differentiable in π

with Wπ (π, ξ ; ν) = VπC (g̃ γ , b; ν)d g̃ γ /dθ ∗ (π )dθ ∗ (π )/dπ continuous and negative (since V C (g̃ γ , b; ν)
is diminishing in θ ∗ (π ) for θ ∗ (π ) > α). Hence, W E (π, q; ν) = qW (π, ν; ν) + (1 − q)W (π, ξ ; ν) =

qV (g γ ; ν) + (1 − q)V C (g̃ γ , b; ν) is continuous and differentiable in q and π , with WqE (π, q; ν) =

V (g γ ; ν) − V C (g̃ γ , b; ν) > 0 and WπE (π, q; ν) = qWπ (π, ν; ν) + (1 − q)Wπ (π, ξ ; ν). Let π ∗ (q) be the

value of π that maximizes W E given q (solving WπE (π ∗ , q; ν) = 0); this value is continuous in q since
E
Wπ (π, q; ν) is continuous in both π and q.
Public Choice (2011) 148: 163–196 195
π < π A (ξ ) and risk a less favorable outcome if the progressive state occurs.69 If instead
π A (ξ ) < π Z (and hence π W = π Z ), then the CMV does not set π = 0, as it is always
convenient for him to set π at least equal to π A (ξ ). If π = π A (ξ ) the CMV obtains exactly
the same utility as if π = 0 in the progressive state, but in the conservative state he gets
a larger utility, because π A (ξ ) > π Z . In this case, law enforcement is always sufficiently
strong to prevent corruption in the conservative state, irrespective of the probability of the
latter.
References
Allen, R. G. D., & Lerner, A. P. (1934). The concept of arc elasticity of demand. Review of Economic Studies,
1(3), 226–230.
Alesina, A., & Weder, B. (2002). Do corrupt governments receive less foreign aid? American Economic
Review, 92(4), 1126–1137.
Barro, R. J. (1973). The control of politicians: an economic model. Public Choice, 14(1), 19–42.
Basu, K., Bhattacharya, S., & Mishra, A. (1992). Notes on bribery and the control of corruption. Journal of
Public Economics, 48(3), 349–359.
Becker, G. S. (1968). Crime and punishment: an economic approach. Journal of Political Economy, 76(2),
169–217.
Becker, G. S. (1983). A theory of competition among pressure groups for political support. Quarterly Journal
of Economics, 98(3), 371–400.
Becker, G. S., & Stigler, G. J. (1974). Law enforcement. malfeasance, and compensation of enforcers. Journal
of Legal Studies, 3(1), 1–18.
Beenstock, M. (1979). Corruption and development. World Development, 7(1), 15–24.
Bernhard, W. (2002). Banking on reform: political parties and central bank independence in the industrial
economies. Ann Arbor: The University of Michigan Press.
Besley, T., & Coate, S. (1997). An economic model of representative democracy. Quarterly Journal of Eco-
nomics, 112(1), 85–114.
Besley, T., & McLaren, J. (1993). Taxes and bribery: the role of wage incentives. Economic Journal,
103(416), 119–141.
Braun, M., & di Tella, R. (2004). Inflation, inflation variability, and corruption. Economics and Politics, 16(1),
77–100.
Calderon, C., & Chong, A. (2006). Do democracies breed rent-seeking behavior? Journal of Policy Reform,
9(4), 247–260.
Chander, P., & Wilde, L. (1992). Corruption in tax administration. Journal of Public Economics, 49(3), 333–
349.
Chong, A., & Calderon, C. (2000a). Institutional quality and poverty measures in a cross-section of countries.
Economics of Governance, 1(2), 123–135.
Chong, A., & Calderon, C. (2000b). Institutional quality and income distribution. Economic Development
and Cultural Change, 48(4), 761–786.
Chong, A., & Gradstein, M. (2008). What determines foreign aid? the donors’ perspective. Journal of Devel-
opment Economics, 87(1), 1–13.
Del Monte, A., & Papagni, E. (2001). Public expenditure, corruption, and economic growth: the case of Italy.
European Journal of Political Economy, 17(1), 1–16.
Dincer, O. C., & Gunalp, B. (2008). Corruption, income inequality, and poverty in the United States. Fon-
dazione ENI Enrico Mattei Working Paper, 54.2008.
Dudley, L., & Montmarquette, C. (1987). Bureaucratic corruption as a constraint on voter choice. Public
Choice, 55(1–2), 127–160.
69 If π Z (ν) < π < π A (ξ ), the CMV’s expected utility is equal to W E (π, q; ν) ≡ qW (π, ν; ν) +
(1 − q)W (π, ξ ; ν) = qV (g γ , ν) + (1 − q)V C (g̃ γ , b; ν). If π = 0, the CMV’s utility is equal to

W E (0, q; ν) ≡ qW (0, ν; ν) + (1 − q)W (0, ξ ; ν) = qV (g̃ α(ν) , b; ν) + (1 − q)V C (g̃ α(ξ ) , b; ν). Both
W E (π, q; ν) and W E (0, q; ν) are continuous and diminishing in q, with limq→0 W E (π, q; ν) =
V C (g̃ γ , b; ν) < limq→0 W E (0, q; ν) = V C (g̃ α(ξ ) , b; ν). If q is sufficiently low, therefore, the CMV’s ex-
pected utility is larger when π is set equal to 0.
196 Public Choice (2011) 148: 163–196
Easterly, W. (2007). Was development assistance a mistake? American Economic Review, 97(2), 328–332.
Galeotti, G., & Merlo, A. (1994). Political collusion and corruption in a representative democracy. Public
Finance–Finances Publiques, 49(Suppl.), 232–243.
Glaeser, E., Scheinkman, J., & Shleifer, A. (2003). The injustice of inequality. Journal of Monetary Eco-
nomics, 50(1), 199–222.
Grossman, G. M., & Helpman, E. (1996). Electoral competition and special interest politics. Review of Eco-
nomic Studies, 63(2), 265–286.
Gupta, S., Davoodi, H. R., & Alonso-Terme, R. (2002). Does corruption affect income inequality and
poverty?. Economics of Governance, 3(1), 23–45.
Hindricks, J., Keen, M., & Muthoo, A. (1999). Corruption, extortion, and evasion. Journal of Public Eco-
nomics, 74(3), 395–430.
Herzfeld, T., & Weiss, C. (2003). Corruption and legal (in)effectiveness: an empirical investigation. European
Journal of Political Economy, 19(3), 621–632.
Karahan, G. R., Coats, R. M., & Shughart, E. F. II (2006a). Corrupt political jurisdictions and voter partici-
pation. Public Choice, 126, 87–106.
Karahan, G. R., Razzolini, L., & Shughart, E. F. II (2006b). No pretense to honesty: county government
corruption in Mississippi. Economics of Governance, 7(3), 211–227.
Knack, S. (2001). Aid dependence and the quality of governance: cross-country empirical tests. Southern
Economic Journal, 68(2), 310–329.
Kunicová, J., & Rose-Ackerman, S. (2005). Electoral rules and constitutional structures as constraints on
corruption. British Journal of Political Science, 35(4), 573–606.
Kurer, O. (2001). Why do voters support corrupt politicians? In Jain, A. K. (Ed.), The political economy of
corruption (pp. 63–86). London: Routledge.
Li, H., Xu, L. C., & Zou, H.-F. (2000). Corruption, income distribution, and growth. Economics and Politics,
12(2), 155–182.
Mauro, P. (1995). Corruption and growth. Quarterly Journal of Economics, 110(3), 681–712.
Mauro, P. (1998). Corruption and the composition of government expenditure. Journal of Public Economics,
69(2), 263–279.
McGregor, R. R. (2007). Federal reserve transparency: the more things change, the more they stay the same?
Public Choice, 133(3–4), 269–273.
Montinola, G. R., & Jackman, R. W. (2002). Sources of corruption: a cross-country study. British Journal of
Political Science, 32(1), 147–170.
Mookherjee, D., & Png, I. P. L. (1995). Corruptible law enforcers: how should they be compensated? Eco-
nomic Journal, 105(428), 145–159.
Myerson, R. B. (1993). Effectiveness of electoral systems for reducing government corruption: a game theo-
retic analysis. Games and Economic Behavior, 5(1), 118–132.
Osborne, M. J., & Slivinski, A. (1996). A model of political competition with citizen-candidates. Quarterly
Journal of Economics, 111(1), 65–96.
Paldam, M. (2002). The cross-country pattern of corruption: economics, culture and the seesaw dynamics.
European Journal of Political Economy, 18(2), 215–240.
Persson, T., Tabellini, G., & Trebbi, F. (2003). Electoral rules and corruption. Journal of the European Eco-
nomic Association, 1(4), 958–989.
Peters, J. G., & Welch, S. (1980). The effects of charges of corruption on voting behavior in congressional
elections. American Political Science Review, 74(3), 697–708.
Rose-Ackerman, S. (1975). The economics of corruption. Journal of Public Economics, 4(2), 187–203.
Rose-Ackerman, S. (1978). Corruption: a study in political economy. New York: Academic Press.
Rundquist, B. S., Strom, G. S., & Peters, J. G. (1977). Corrupt politicians and their electoral support: some
experimental observations. American Political Science Review, 71(3), 954–963.
Sandholtz, W., & Koetzle, W. (2000). Accounting for corruption: economic structure, democracy, and trade.
International Studies Quarterly, 44(1), 31–50.
Sung, H.-E. (2004). Democracy and political corruption: a cross-national comparison. Crime, Law and Social
Change, 41(2), 179–194.
Tanzi, V., & Davoodi, H. R. (1998). Corruption, public investment and growth. In Shibata, H., & Ihori, T.
(Eds.), The welfare state, public investment, and growth: selected papers from the 53rd congress of the
International Institute of Public Finance (pp. 41–60). Tokyo: Springer.
Treisman, D. (2000). The causes of corruption: a cross-national study. Journal of Public Economics, 76(3),
399–457.
van Winden, F. (1999). On the economic theory of interest groups: toward a group frame of reference in
political economics. Public Choice, 100(1–2), 1–29.
You, J.-S., & Khagram, S. (2005). A comparative study of inequality and corruption. American Sociological
Review, 70(1), 136–157.

Pani2011 Article HoldYourNoseAndVoteCorruptionA

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Pani2011 Article HoldYourNoseAndVoteCorruptionA

Uploaded by

Copyright:

Available Formats

Public Choice (2011) 148: 163–196

Hold your nose and vote: corruption and public decisions

Keywords Corruption · Political equilibrium · Public goods · Public expenditure · Taxation

Curbing corruption remains a challenge in democratic countries. In principle, public con-

2 Relations with previous studies

3 A model of public expenditure with corruption

the first-order condition (FOC) Vg (g; θ ) = 0, equivalent to13

13 Since V (g; θ) is concave in g, the ideal policy of each citizen is unique.

3.1 How ministers respond to corruption: honest and corrupt citizens

Fig. 1 The minister’s expected utility with and without corruption

economic considerations: corruption entails a risk of punishment, and a minister engages

Lemma An elected citizen engages in corruption if and only if

Hereafter, a citizen who, if he is elected minister, engages in corruption will be called

u(g θ ) + u(g̃ θ ) + abg̃ θ > π .

Fig. 2 Honest and corrupt candidates by type (given p)

3.2 The political response of nonelected citizens33

Fig. 3 Utility of a nonelected citizen as a function of g

Proof See the Appendix.

Fig. 4 Median voter’s utility as a function of the minister’s type

Proof See the Appendix.

3.3 The choice between honest and corrupt candidates36

Proof See the Appendix.

Sample All Democracies Strong All Democracies Strong

Constant 18.054∗∗ 27.746∗∗∗ 52.874∗∗∗ 17.866∗∗ 27.667∗∗∗ 52.862∗∗∗

Sample All Democracies Strong All Democracies Strong

Constant 9.078 22.334∗∗ 55.185∗∗∗ 8.903 22.139∗∗ 55.034∗∗∗

Sample All Democracies Strong All Democracies Strong

Constant −20.024∗∗ −3.001 19.095 −19.929∗∗ −3.610 21.640

(1.84) (3.18) (3.05) (0.52) (2.84) (3.74) (−2.11) (1.06) (2.53)

5 Voting on law enforcement

5.1 Law enforcement and policy choice

corruptly—formally, whether or not V (θ ∗ (π); μ) > V C (θ ∗ (π), b; μ), which is equiva-

Fig. 5 Median voter’s utility as a function of π

5.2 Deciding on law enforcement

Proof See the Appendix.

Proof See the Appendix.

Appendix: Proofs of propositions

Proof of Proposition 4 If π B (ξ ) < π ν , the CMV maximizes his utility by setting π = π ν ,

θ is elected—equal to V (g θ , μ)—increases in θ when θ < μ:

(1 − q)W (π, ξ ; ν) = qV (g γ , ν) + (1 − q)V C (g̃ γ , b; ν). If π = 0, the CMV’s utility is equal to

You might also like