PERSISTENT ELECTORAL SUCCESS WITH ENDOGENOUS RENTS:
CAN POLITICIANS EXTRACT RENTS AND STILL STAY IN POWER?
In the class of standard political agency models, most of them fail to account for the
fact that incumbent politicians in democracies tend to stay in power for long periods of
time, without having to trade-oﬀ rents for holding oﬃce. This paper examines under which
conditions this frequent scenario occurs by laying out a new theoretical and empirical direc-
tion of political agency models focused on endogenously determining rents within the public
good expenditure function. The paper tests the theory using United States gubernatorial
and state legislature elections from 1992 to 2008. It ﬁnds that for positive economic shocks
a patient incumbent anticipating more future rents may stay in power for a long period of
time and keep extracting rents with respect to the given constraints. For negative shocks
the rent-extracting decision will depend on the magnitude of the shock. The paper ﬁnds the
cut-oﬀ level of wasteful spending the politicians need to respect in order to maintain power.
JEL Classiﬁcations: D72, H11, H41, C73
Keywords: Political agency, endogenous rents, re-election, wasteful spending
Politicians in power have strong incentives to misuse that power for their own personal gain.
However, political accountability in front of voters (principals) prevents the politicians (agents)
from fully expropriating the public budget, even though due to lack of transparency and an
informational advantage politicians often do get away with allocating a fraction of public funds
to their private beneﬁt. Political agency models describe a general setting in which a rational
agent’s maximization problem is to capture this fraction of public funds, called rents. The
meaning of rents diﬀers slightly from the classical Tullock (1967) and Krueger (1974)
∗Corresponding author: Vuk Vukovic, Department of Economics, Zagreb School of Economics and Manage-
ment, Jordanovac 110, Zagreb 10000, Croatia. Email: firstname.lastname@example.org. The author would like to thank Dr
Josip Glaurdic, Dr Pierro Stanig, Dr Boris Podobnik, participants at the 50th Public Choice Society Conference
in New Orleans, participants of the economic seminars at the University Autonoma Barcelona and participants
at the ZSEM economic seminars for their useful comments and suggestions.
deﬁnition to include excess payments (bribes) extracted through public good expenditures on
various pork-barrel and white elephant projects obtained by an incumbent politician1. The voters
are unable to observe the budgetary allocation process directly, creating the problem of electoral
accountability of politicians (the monitoring problem).
Uncertainty and asymmetric information give further incentives to politicians to misrepresent
themselves and pursue their own interests. Due to such behavior of agents there exists a trade-
oﬀ between voter utility (policies appealing to voters) and rent-extraction (policies appealing to
politicians in power) (Brennan and Buchanan, 1980; Besley, 2006; Persson and Tabellini,
2000). The central issue is whether or not electoral competition and the discipline eﬀect of the
voters will induce the politicians to announce voter optimal policies or rent-maximizing policies.
There is thus an inherent trade-oﬀ between staying in power and extracting rents. The main
motivation of this paper is to show that this need not be the case, as politicians can successfully
balance between staying in power for long periods of time and optimize their rent-extraction
In general, political agency models are often characterized by a two period setting in which
a politician’s term ends in the second period (the standard term limit assumption in Besley
and Case, 1995a, 1995b; Alt, Bueno de Mesquita and Rose, 2011; Ferraz and Finan,
2011). In order to stay in oﬃce and reach the second period an incumbent politician should
limit his rent-extraction in t= 1 since retrospective voters will reward congruent behavior. The
re-election incentive should improve the discipline of politicians. However, in the second and ﬁnal
period (t= 2), a moral hazard problem arises since bad politicians are free to divert the entire
budget towards their private means. Newer models introduce adverse selection (Austen-Smith
and Banks, 1989; Banks and Sundaram, 1993; Besley and Case, 1995a; Rogoﬀ, 1990;
Persson and Tabellini, 2000; Besley and Smart, 2007) concerning how good politicians
should distinguish themselves from bad ones, where the ﬁrst period behavior of bad politicians
implies ‘mimicking’ the behavior of good politicians and sacriﬁcing ﬁrst period rents in order to
remain in oﬃce and expropriate the entire budget for rents in t= 2.
The main issue with such assumptions is the oversimpliﬁcation of rent-maximizing behavior.
Voters tend to punish politicians only when they see that they’ve taken maximum rents (i.e.
the entire budget) for themselves. In a democratic political order, regardless of whether it’s
a well-functioning or a poorly deﬁned democracy, these events almost never occur, even on a
local level. There are simply too many checks and balances for a politician to extract rents
directly from the budget which implies that rents must be hidden from the public. Taking this
crucial assumption into consideration implies that there need not exist a trade-oﬀ between rent-
extraction and re-election because politicians can present themselves to the voters as honest
1For example, while building a road or a bridge a politician can conceal his rent-extraction by presenting one
price to the public while charging a diﬀerent (lower) price to the contractor, thus taking the diﬀerence for him or
herself. Ferraz and Finan (2011) recognize such corruptive activities as frauds in public procurement, diversion
of public funds (expenditures without proof of purchase) and over-invoicing (buying goods above market price).
without having the need to cut down on their rent-extraction. In other words, politicians can
steal and deplete resources as long as they want, constrained by some formal rules, without this
having any adverse eﬀect on their re-election chances.
The focus of this paper is to uncover how is it possible that despite their persistent corrupt ac-
tivities, incumbent politicians, particularly at a local level of governance, manage to stay in power
for long periods of time. The paper lays out a new theoretical and empirical direction of political
agency models by specifying the distinction between rents and public goods and by eliminating
the standard trade-oﬀ between rents and re-election. Politicians operating in a democratic sys-
tem will never risk openly extracting the entire budget (as assumed by Brennan and Buchanan’s
(1980) Leviathan scenario) nor will they openly engage in corruption, but will always attempt
to hide their rent-extraction within speciﬁc types of public spending, making rents endogenous.
According to Mauro (1998) diﬀerent types of government expenditures provide diﬀerent op-
portunities for corruption, where large public infrastructure projects or high-technology goods
provided by specialized oligopolies (defense spending) are more suspect to collecting bribes and
rents than individualized social transfers or, for example, education spending. The necessity to
determine rents endogenously and dependent on other budgetary expenditures is driven by the
fact that in well-functioning democracies budget transparency signiﬁcantly diminishes the scope
for misuse of public funds, but politicians still very often do get away with taking rents during
their long-lasting mandates. Furthermore, oversimpliﬁed assumptions of general political agency
models imply quasi-linear preferences which make the public good function independent of rents,
implying that the preferred level of public goods is only an increasing function of its cost shock.
This paper shifts the focus from the cost shock onto budget size, as this partially explains the
desire of politicians to always have higher budgets.
An unavoidable consequence of hidden rents is higher public good spending and higher taxa-
tion, thus increasing the overall size of government. The vast empirical evidence on the increasing
size of governments in the past ﬁfty years (Maddison, 2001; Tanzi and Schuknecht, 2000)
veriﬁes this intuition, although the paper disregards the possible eﬀects of intrinsic voter prefer-
ences towards more redistribution, or other factors recognized by Higgs (1987), and focuses
solely on rent-extraction and the moral hazard problem as partial explanations for growth in
government size. By tying rents with re-election probabilities the paper attempts to show that
rents in the form of political income from holding oﬃce will ultimately lead to higher than voter
optimal overall taxation and public spending. This harmful relationship between higher spend-
ing and corruption was implied by Buchanan (1975) and Acemoglu and Verdier (2000),
and empirically tested in Goel and Nelson (1998) and Dzhumashev (2014). Future re-
search should go entirely in this direction where rents are hidden, thus opening scope to directly
empirically testing rent-extraction by assigning diﬀerent proxies of possible corrupt behavior to
measure their impact on political survival and economic performance.
The paper makes a key contribution to the existing literature in empirically ﬁnding the cut-oﬀ
value of wasteful spending the politicians need to respect in order to stay in power. Any level
of spending above the cut-oﬀ, meaning that politicians have extracted too much rents, implies
an electoral defeat. Furthermore, persistent electoral success is possible for patient politicians
under the condition that they face favorable economic shocks each period. They reduce current
rent-extraction as they anticipate better future rent-extracting opportunities. During times of
negative economic shocks an incumbent politician will increase the amount of wasteful spending
in order to capture more rents now, knowing he is facing less rents in future periods. Depending
on the magnitude of the shock his strategy will resemble that of the classical term limit con-
straint. Testing these assumptions empirically the paper ﬁnds that politicians increase all types
of spending (even wasteful) in times of economic downturns, which is hardly surprising, however
only spending on potentially wasteful public goods will sway their re-election chances. The paper
uses the data on US gubernatorial and state assembly elections from 1992 to 2008 to test the
After deﬁning the model’s main assumptions, the paper speciﬁes voter and political strategies
and decision rules, upon which the equilibrium levels of public good spending, rent-extraction,
and the state of the economy are determined. The empirical part tests the theoretical proposi-
tions, while the conclusion opens up space for potential future research in uncovering the trade-oﬀ
between rent-extraction and re-election.
The model is deﬁned as a repeated game with an inﬁnite horizon (Ferejohn, 1986, Banks and
Sundaram, 1993, Smart and Sturm, 2013) between the voters and incumbent politicians. It
rests upon the assumptions of asymmetric information over the allocation of rents and public good
spending, but not over politicians’ types. All politicians are assumed to be of the same type, non-
benevolent rent-seekers with a common goal of extracting rents and maximizing their probability
of staying in power (Buchanan and Tullock, 1962; Brennan and Buchanan, 1980; Besley,
2004; Caselli and Morelli, 2004; Bueno de Mesquita et al, 2005), thus adhering to the
moral hazard implication of political agency models (Barro, 1973, Ferejohn, 1986).
2.1 Budget constraint
In each period an incumbent politician (or political party in power) has to make budgetary
decisions on the allocation of social transfers (f), public sector wages (w) and public good
expenditures (g), after which it receives a payoﬀ deﬁned as rents r∈[0,br]. Rents are endogenous,
meaning they are hidden within the public good expenditure function, as they cannot be taken
directly from the budget, but rather allocated through diﬀerent public good projects. The
incumbent faces the following budget constraint in each period:
(1 + βt−1)τy =g(θ0, r) + T+V(1)
i=1 fiare aggregate transfers to the public (social and unemployment beneﬁts,
pensions etc.) while V=Pn
i=1 wiare aggregate public sector wage expenditures of the govern-
ment. The term on the left is total revenue (tax rate τ, times aggregate income y) multiplied
by the eﬀect of a previous period economic shock βt−1. Taxation is proportional to the level of
income and there is a balanced budget every time (no budget deﬁcits or public debts).
Economic shock βis speciﬁed as a random stochastic shock, uniformly distributed on h−1
where a positive shock (with probability p) implies higher future government revenues, while a
negative shock implies the opposite. It presents the crucial signal an incumbent politician re-
ceives upon which it bases his budgetary allocation decisions as well as his rent-extraction. Many
political agency papers use a similar random noise variable that depicts either a productivity
parameter transferring resources into public goods (Persson et al, 1997), a public good cost
shock (Persson and Tabellini, 2000, and Besley and Smart, 2007) or any exogenous oc-
currence that will determine the eﬀort of a politician (Ferejohn, 1986). Both politicians and
voters observe βwith certainty each period before they make their decisions.
The ﬁrst term on the right of Eq 1 (g=Pn
i=1 gi) are total public good expenditures which
depend on the realization of rents (r) and actual costs of all public goods (θ0). A single public
good (gi) expenditure function is deﬁned as:
i, ri) = θiGi= (θ0
Expenditure for a single public good equals its total unit costs (θi) as presented to the public
times the total quantity of the good (Gi= 1). The term θ0
irepresents the actual cost of a public
good which is known only to the politician and is never observed by the public. By concealing
the true costs of a good from the public, politicians can create rents (ri) as a bribe collected
from the diﬀerence between total and actual costs of a good. Rents are therefore being extracted
based on the asymmetry of information between voters and politicians.
The way rent per single public good (bribe) is deﬁned in Eq 3 implies that an incumbent
politician assigns a ﬁxed weight (λ) from every public good it produces to rent-extraction2. The
factor λ∈[0,1] can be interpreted as political preferences towards budget misappropriation (cor-
2Imagine a political party demanding a commission for any procurement it allows. This commission (a per-
centage of costs of a good that goes directly into the politicians’ pockets) stays the same in relative terms for
any project, but increases in absolute terms as more government revenue is allocated to public good expenditures
each period. So λ= 0,2 then 20% of spending on a single public good is allocated towards rents.
ruption) and wasteful spending3. It is an exogenous, cultural shock, drawn by nature speciﬁcally
for each politician. The political and institutional environment in which the incumbent operates
along with his intrinsic preferences towards rents will determine the total amount of wasteful
spending (similar to Bueno de Mesquita et al, 2005).
It can be inferred from Eq 3 that rents depend on how much a single public good actually
i, for 0 ≤λ < 1/2. Since λis always ﬁxed for a single agent, implying that the
relative diﬀerence between individual total and actual costs (θi−θ0
i) will always be the same for
every public good provided, higher rents can only be achieved by diverting more budget funds
towards public good expenditures (g). This means that rents don’t depend on the cost shock,
but on budget size, where the larger the budget, the larger the scope for rents.
2.2 Aggregate rents
Not all public good expenditures are subject to rent-extraction. Rents (bribes) can only be
collected from white elephant projects and pork-barrel spending the incumbent creates. This
implies that rents and public goods are characterized by a quasi-linear preference relation where
rent-extraction begins after a certain point, once the initially desired level of public goods and
services are provided. Accordingly, Eq 2 can be rewritten into an aggregate public good expen-
gi= (1 −λ)
Gj+λ H (θ0, r) (4)
for all i∈N , and for all j∈M , where i6=j,
Where Gjis some initially desired and provided number of public goods (for which the total
amount of public good expenditures is g), while H(·) is a quasi-convex function depicting the
total amount of wasteful spending upon which rents are created. Public good expenditures are
an increasing function of total rent-extraction and the propensity to extract rents (which diﬀers
from one party to another), and a decreasing function of actual costs. It is easy to see from
Eq 4 that higher spending allocated towards public good expenditures (as a budget item) is the
only way to increase rent-extraction via more wasteful spending, with λkept ﬁxed. The size of
wasteful spending within the public good expenditure function depends on the given value of λ4.
3In stable democracies λis likely to be low, as political preferences towards corruption and budget misappro-
priation are relatively smaller, but not nonexistent.
4Similar to the single public good expenditures function 2, a value of for example λ= 0,2 would imply 20%
of public good spending going towards white elephant projects and 80% towards voter preferred public goods.
2.3 Voter re-election threshold
Voters expect incumbent politicians to determine some intrinsically optimal level of spending
and taxes, ψv(gv, τ v), which is diﬀerent from the optimal level desired by politicians5. The
voters update their optimal desired levels with respect to the observed βshock. The politician
will always have an incentive to determine a combination of taxes and spending higher than
the voter optimum, partially in order to satisfy various special interest groups necessary for its
re-election6and partially to maximize his rents:
τp|β)> ψv(gv, τ v|β) (5)
Voter dissatisfaction with higher spending and taxes is purely due to wasteful spending,
corresponding partially to Peltzman’s (1992) voters as ﬁscal conservatives, where despite the
voters’ negative reaction to higher spending, politicians can still get away with higher budgets
Due to the existence of uncertainty and the consequential problem of political accountability,
voters cannot prevent the incumbents from determining higher than optimal taxes and spending,
but can punish them ex-post. Voters will punish any behavior of incumbents that sets the level
of taxes and spending above some control level ψ(g, τ ), which is higher (and thus worse oﬀ) than
the voter optimum, but still lower than the maximum level desired by the incumbent party:
τp)> ψ(g, τ )> ψv(gv, τ v) (6)
The control level of ψrepresents the voter re-election threshold above which the incumbent
party will be voted out of oﬃce. According to Ferejohn (1986) or Persson et al (1997)
this threshold is a level of the politician’s eﬀort determined by voters, which shouldn’t be set too
high to encourage rent-extraction, nor too low to encourage shirking. Instead of observing size
of eﬀort, this paper models the re-election rule as a set of voter determined boundaries of public
policy. The role of voting is to achieve a higher level of discipline and hence lower rent-extraction.
According to the assumptions of the re-election threshold the probability of winning for the
incumbent can be determined as a deterministic function of ψ:
0,if ψp> ψ.
Another way to look at the threshold is to determine the desired optimal values of ψthat
5Persson et al (1997), among others, recognize the conﬂicting interests over the composition of government
spending between voters and politicians. Their choice variable encapsulates this assumption.
6The paper doesn’t model transfers to special interests, but works on the ﬁndings of other political agency
papers such as for example Coate and Morris, 1995 where because of special interest groups, the level of
spending by politicians will always be higher than the optimum desired by the voters.
satisfy an aggregate voter utility function within a set of plausible outcomes in which the upper
boundary of the set would be the control level ψ. The re-election threshold would be deﬁned
within a positive, increasing set of diﬀerent choices on budgetary redistribution Ω ∈ψ, ψ. Voter
optimal provision of taxes and spending, ψv(gv, τ v) is necessarily equal to ψ7.
3 Voter and political utilities
3.1 Voter utility
Voters make decisions based on signals of political behavior and actions of politicians. They
evaluate whether an incumbent deserves to remain in oﬃce depending on how he sets taxes
and distributes public spending. They are unable to prevent rent-extraction but can punish the
incumbent ex-post, implying that the re-election threshold is ex-post optimal. Their punishment
threats are perceived to be credible by the politicians. Voters cannot observe any rents, nor the
actual costs of public goods, but can observe the shock β, and update their threshold accordingly.
There is one median, undecided voter group8consistent of voters homogenous in their pref-
erences over the re-election threshold. The voter expected utility function is then:
δtu(σt, ψt) (8)
where 0 < δ < 1 is the discount factor, u(σt, ψt) is a quasi-concave utility function monoton-
ically increasing in σt, while σtrepresents the state of the economy, a perception signal of the
voters on economic performance, based on which the voters make their inferences on the in-oﬃce
performance of incumbents9. Any level of public good provision that satisﬁes Ω ∈ψ, ψwould
send a signal of positive in-oﬃce performance and consequentially a good state of the economy:
σ∈σψ, σ ψ.
The maximization of the voter utility function in Eq 8 will give the voter optimal combination
of public spending and taxes, as deﬁned in Eq 5.
δtu(σt, ψt) (9)
7According to Eq 6 c
τp)> ψ(g, τ)> ψv(gv, τv), politicians always have an incentive to set taxes and
spending higher than the voter optimal distribution. Even if they behave completely congruent, they would aim
to satisfy the ψvthreshold but never go below it, as this would jeopardize both theirs and the voters’ utilities.
8One can easily assume a large number of groups, however in each case the median, undecided group will be
crucial for political re-election. The median group is the one with the highest density and most swing voters (as
in Persson and Tabellini, 2000).
9The state of the economy doesn’t necessarily imply economic performance; it represents signals sent in-
between voters on the perception of economic performance. Any level of spending and taxes that will break the
delicate balance of budgetary expenditures will result in losing voter support from those aﬀected. For example if
public sector wages would cease to grow at their predetermined level, this would result in discontent from public
sector workers, creating a distorted picture of the government to the median undecided voters leading to a lack
of political support for the incumbent.
s.t. f (σt, ψt) = c
Where cis some arbitrary constant, depicting the quasi-linear relationship between σtand ψt.
Lemma 1. Solving the voter maximization problem yields the optimal level of σv,∗
which determine the optimal level of budgetary redistribution and the state of the economy desired
by the voters: P∞
Proof : see Appendix.
Lemma 1 shows that voters intrinsically always pick an optimal re-election threshold to
optimize their perception on the state of the economy, however this doesn’t mean that politicians
will always satisfy this threshold. We will see for which cases this doesn’t occur once we deﬁne
the optimal strategies of the politicians.
As stated earlier voters tend to update their preferences over the upper control level of the
re-election threshold each period with respect to the βshock. They apply Bayesian updating
over expected values of ψv
To see this more clearly we can assign probabilities over βand ψ. Let shock βtbe either
positive β > 0. . . por negative β < 0. . . 1−p. Upon observing the shock, the re-election
threshold (in terms of desired levels of taxation and spending) can be updated either downwards
Lwith probability q, or upwards ψv
Hwith probability 1−q. The intuition is that upon observing
a positive shock, voters desire lower taxes and lower public good spending and hence update
their desired level of ψvdownwards to the new level of ψv
L, while negative shocks will imply
the opposite10. There is also a possibility that the shock updates the threshold in a diﬀerent
direction. Deﬁne µas the probability that β > 0 will cause the voters to update the threshold
upwards to ψv
H|β > 0) = µ(11)
Using the Bayes rule, the posterior probability that for a positive shock (β > 0) the threshold
gets updated downwards (ψv
L|β > 0) = pq
pq + (1 −p)(1 −q)µ(12)
This will occur if P(ψv
L|β > 0) ≥q, the probability that the updated threshold will be ψv
is easy to see that this holds for every µ≤p
1−p, which is always true for every p≥1
2, i.e. for
any positive βshock. In the same way we can calculate the posterior probability that a negative
10In times of crises (which would be an example of a negative shock) the majority of voters expect more
intervention from the government, as shown by Higgs (1987) on the US case. In addition, Goel and Nelson
(1998) ﬁnd that corruption increases in times of economic downturns.
shock (β < 0) causes voters to update the threshold upwards, for which we get that it holds
for every π≤1−p
p, which is true for every p < 1
2(i.e. for any negative βshock) where πis
deﬁned equivalently to µin Eq 11, representing the probability that a negative shock will cause
the voters to update the threshold downwards, or P(ψv
L|β < 0) = π. According to Lemma 1,
any chosen level of the threshold is always intrinsically optimal. Therefore any updated level of
L,H is also voter optimal.
In other words, for a positive shock voters update their threshold downwards for any p≥1
while for every negative shock they update their threshold upwards for any p < 1
2, signaling an
inverse relationship between βand ψ:
β > 0,· · · ψv
β < 0,· · · ψv
However, there is also a probability that the shocks update the threshold in a diﬀerent direction,
where a positive shock would lead to a higher level of spending (P(ψv
H|β > 0) = µ), while a
negative shock would result in a lower level of spending (P(ψv
L|β < 0) = π). Intuitively, this
can only happen in period t+ 1, where a positive shock enables more revenues (and hence more
spending) in the next period, while a negative one implies less revenues (and hence less spending)
in the next period.
Finally, we can easily propose the following relationship between βand ψ:
Which is true for both positive and negative βt. For positive economic shocks (good times)
voters demand a lower cut-oﬀ ψv, driven mostly by lower taxes. However, higher economic
activity in the current period will increase budget revenues in the next period and hence raise
t+1, mainly through higher spending, g. For a negative economic shock (bad times) voters
demand more spending in the current period to oﬀset the shock, however due to its negative
eﬀects there will be less budget revenues available in the future period.
3.2 Incumbent utility
An incumbent politician is a rational utility maximizer seeking to win elections in every period
in order to have an option of extracting rents. Since the position of holding oﬃce is primary
attractive because of possible rent-extracting opportunities, the optimal strategy of the incum-
bent is to keep this position as long as they are able to maximize the ﬂow of rents in the current
period and expected rents from future periods. In order to stay in power it needs to choose a
level of spending and taxes ψp≤ψaccording to the re-election constraints in Eqs 5 and 6.
The incumbent’s utility is a combination of ego rents from holding oﬃce and rents that can be
extracted once in oﬃce. In t= 0 this utility is achieved with certainty (since he is already in
oﬃce), while in every subsequent period it depends on the probability of winning oﬃce. The
previous period βt−1shock determines the scope for current period rents, meaning that every
current period βtwill determine higher or lower expected future rents11:
I=R0+ (1 + βt−1)H(r). . . t = 0 (15)
I= (R1+ (1 + βt−1)H(r)) pIψp
t−1. . . t = 1 (16)
In every period t= 1, . . . , n the incumbent decides on a new combination of taxes, spending, and
consequently rents from an aﬀordable set of white elephant projects. Therefore in each period
the incumbent faces a budget constraint and a choice of whether or not to satisfy the voter
re-election threshold and set ψp
t−1≤ψt−1. The incumbent politician maximization problem is
I= (R1+ (1 + βt−1)H(r)) pIψp
s.t. (1 + βt−1)τy = (1 −λ)Gj+λH (r) + T+V
Under the condition that ψp
t−1≤ψt−1, for which the deterministic probability of winning is
pI= 1 according to Eq 7. The choice of the level of ψpis purely determined by the incumbent,
and it will necessarily constrain the amount of rents an incumbent can extract. To see this
consider the solution to the politician maximization problem. Solving Eq 17 yields an optimal
amount of rents:
r=Fτy −(1 −λ)Gj−T−V−1
Rents thus depend on every budgetary category. For higher total revenues (τ y), the scope for
rents increases, while it decreases for a larger number of voter-desired public goods (Gj), and
the size of social transfers (T) and public sector wages (V). It also has a negative relationship
with the parameter λ(recall that λis assumed to stay ﬁxed within a political party, and is
institutionally determined), which makes sense since too high corruption incentives will deplete
too much resources. Dictatorships are faced with this particular problem.
Deﬁning rents this way yields a more realistic approach to political decision-making, taking
into consideration a multitude of factors when considering the decision to respect the re-election
threshold and be able to extract rents at the same time. However, to understand why a politician
will sometimes decide to violate the ψp
t−1≤ψt−1condition, we must observe the expected future
utilities of a politician and his reaction with respect to the anticipated βeconomic shock.
11It is important to include βdirectly into the incumbent utility function since it accounts for the fact that in
each period, for positive economic shocks, there will be more rents available, not less. It, in a way, oﬀsets the
An incumbent’s ex ante utility (expected utility at the start of term t= 0) can be deﬁned as:
I(r|g, λ)] + pI(ψ0) (1 + β0)
I(r|g, λ)] + (1 −pI(ψ0)) E[Ut
The ﬁrst term denotes expected utility in the actual period t= 0 as deﬁned in Eq 15; the utility
an incumbent will receive at the end of his ﬁrst term in oﬃce, when total rents are realized. The
second term is the sum of all future discounted expected utilities when in oﬃce12, from period
t= 1 onwards, if he wins re-election with probability pI(ψ0) depended on satisfying the re-
election threshold in period t= 0. The incumbent’s future rents will depend on β0in the current
period t= 0 as it will signal how big expected rents might be in all subsequent periods starting
from t= 1. The ﬁnal term denotes the probability of losing the election if the politician doesn’t
respect the re-election threshold and the utility he will get if the challenger, the opposition party,
is now in oﬃce13.
The incumbent plays the same inﬁnite horizon game each period. A cooperative strategy
implies adapting to voter expectations and respecting the re-election threshold every period in
order to remain in oﬃce. Any defection from this strategy, even though it will ensure higher
immediate rents, will induce a (credible) punishment from the voters in terms of electoral loss,
and will disable the incumbent from extracting further rents. The game can is therefore a tit-
for-tat game where any deviation from a cooperative strategy is met with immediate punishment
from the voters (a trigger strategy). Even though the agent does change after the voters imply a
punishment strategy, from the voters’ perspective they repeatedly play a tit-for-tat game where
they punish the agent’s defection and reward cooperation.
The incumbent politician compares the defection and cooperation strategies starting from his
ﬁrst term in oﬃce, t= 0. He plays a cooperative strategy if and only if the expected utility from
the cooperative strategy is higher than the expected utility from the defection strategy:
I(r|g, λ)] + (1 + β0)
I(r|g, λ)] ≥E[U0
I(br|bg, λ)] + E[Ut
The term on the right of the equation presents expected utility from taking maximum rents
(br, ∀r∈bg=τy) and the utility the incumbent gets from a challenger in power, achieved with
certainty for a defection strategy. When he defects he does so to maximize rent-extraction but
is faced with no immediate future payoﬀs in terms of rents. Utility in t= 0 will either be
cooperative (with (r, g )) or defective (with (br, bg)), and will depend on the level of βt−1observed
in the previous period, before holding oﬃce. However, the incumbent’s decision is based on
anticipating what future rents will be. He observes β0in the current period, and bases his
12For simplicity ego rents are normalized to zero in all future periods.
13This utility for the incumbent might even be negative once the opposition party is in oﬃce, as too much
rent-extraction may be subject to additional punishment (such as a corruption trial).
decision of current period rent-extraction on anticipated future rents. He chooses his strategy
with respect to β0and defects only when the βshock is suﬃciently low so that he might ﬁnd
himself in a better position now with maximum rents than with future lower rents.
Proposition 2. An incumbent politician will form his strategy on rent-extraction and conse-
quently his chances of re-election based on the realization of the current period shock β0. For any
I(br|bg, λ)] + E[Ut
−1 = β∗(21)
the incumbent plays a cooperative strategy and chooses his level of rent-extraction and public good
expenditures with respect to the voter re-election threshold, while for any
I(br|bg, λ)] + E[Ut
−1 = β∗(22)
the incumbent defects and by extracting too much rents is voted out of oﬃce. These sets of
strategies solved for β0are a unique subgame-perfect Nash equilibrium of the incumbent politi-
cian’s repeated game.
Proof: See Appendix.
One way to interpret this result with respect to the discount factor are varying levels of
patience. For example, a political party is by deﬁnition much more patient that an individual
politician, which is why their discount factor is always higher, i.e. suﬃciently closer to 1. A
patient incumbent (δ→1) has a lower cut-oﬀ value of β∗for which it chooses defection, meaning
that even for negative economic shocks it is willing to cooperate, while an impatient one (δ→0)
requires a much higher economic shock every period to stay in power and limit rent-extraction.
3.3 Equilibrium strategies
The intuition is as follows. During a positive shock β > 0 iﬀ β0≥β∗, politicians anticipate
more rents tomorrow (via higher expected revenues, according to Eq 1), however their current
spending and taxes will be lower ψp≤ψin order to stay in power and seize higher next
period rents. Positive shocks imply that patient incumbents adjust current rent-extraction for
higher expected rent-extraction. The voters also expect lower current taxes and less public good
spending as they adjust to a lower cut-oﬀ ψv
Lfor a positive growth shock (as speciﬁed under
Eq 13), but they also expect higher future tax revenues (and higher next period ψ), since better
economic opportunities will raise revenues in t+ 1, with probability µ.
During a negative shock β < 0 iﬀ β0< β∗, politicians anticipate less rents tomorrow (lower
revenues and hence lower spending) but their current spending and taxes will be higher since
they choose to take more rents now. If the incumbent wants to stay in oﬃce he needs to limit
his rent-extraction even further in order to get re-elected (more spending towards redistribution
programs, or programs that are aimed at a short-run boost to the economy, imply less scope for
wasteful spending14, according to Eqs 1 and 4). The incumbent in this case decides it will be
too costly for him (in terms of lower rents) to maintain the current threshold. When this occurs,
the situation is similar to reaching a term limit in the standard political agency framework when
incumbents extract maximum rents in this period knowing they will be removed from oﬃce with
certainty in the next one.
For β0< β∗, an incumbent deviates with probability P=1
2−φβ∗15 . However, not every low
economic shock aﬀects the politicians the same way. Sometimes they still ﬁnd it more favorable
to stay in future oﬃce and extract rents (for a low enough cut-oﬀ level of β∗). Since voters
rationally adjust their threshold, during negative shocks a politician has more leeway to increase
spending and taxes (↑g, ↑τ) as a policy response (or even ↑g, ↓τ, where ∆g > ∆τ).
However, if ∆β0>∆gR∆τ, meaning that if the negative shock is larger than it is feasible
to change government spending or taxes, then regardless of what the incumbent does he will lose
oﬃce. His only feasible strategy is to defect, i.e. take bgand brnow and lose oﬃce. This further
implies that for the defection strategy to occur, two conditions must be met:
1. β0< β∗(as stated under Proposition 2)
2. ∆β0>∆gR∆τ, i.e. φ≤1
Where the ﬁrst condition is necessary and the second is suﬃcient. This implies that politicians
will go above the voter re-election threshold if they observe strong negative shocks (the smaller
the value of parameter φ, the wider the distribution of the βshock). For every negative shock
politicians increase taxes and spending, which the voters observe and expect, but they only
go above the voter threshold for φ≤1, i.e. when the negative shock is too large to make it
proﬁtable for them to stay in oﬃce. If they are able to ﬁx the shock with their policy response
then it would be ex post obvious that β0wasn’t lower than their cut-oﬀ value of β∗. The
empirical implication is that there must be some optimal level of taxation and spending each
period for which the politicians adjust their levels of rent-extraction. Sometimes when they face
a large enough negative economic shock politicians will go above this optimal threshold and as
a consequence lose oﬃce.
The incumbent’s allocation strategies in each period can be summarized in Figure 1. The
ﬁrst graph on the lower left depicts the quasi-linear relationship between rents and public good
14Even though ‘bridges to nowhere’ tend to be an often used short-run stimulus mechanism.
15The probability of defection is calculated based on Proposition 1; P[β0< β∗]=1−
, for β∼
2φi. The intuition is that if the cut-oﬀ level of the shock is larger, it will take a higher value for which β
must be satisﬁed in order to make it proﬁtable for an incumbent to cooperate.
Figure 1: Relationship between public good expenditures, rent-extraction and re-election
production (as described in Eq 4). For a level of public good expenditures less than or equal to g
rents are zero. Any increase of public good expenditures above gsubstantially increases rents, as
here is where the wasteful spending kicks in (λis realized – it determines the slope of the curve).
With the realization of wasteful spending voter welfare starts decreasing: ∂W
∂g >0, ∂2W
since wasteful public goods satisfy partial interests (pork-barrels that beneﬁt certain interest
groups). After the level of rents r, the public goods produced inﬂict more harm than good to
the majority of voters, meaning that the incumbent is extracting more rents for himself (or for
special interest groups) than the amount of useful public goods he creates. It is important to note
that voters don’t react negatively to more government spending, but they do react negatively to
more targeted special interest group spending (Coate and Morris, 1995).
Proposition 3. If the incumbent is a rational rent-maximizer, he has no desire to choose any
level of public goods lower than or equal to g(and no ψplower than or equal to ψ). The chosen
level of public good expenditures will always be:
g > g (r)and ψp> ψ (23)
Proof: See Appendix.
The intuition is clear. Any ψp≤ψ, meaning that g≤g, implies rents to be r= 0. It
wouldn’t be proﬁtable for a rent-maximizing incumbent not to produce any wasteful spending,
as this would imply zero rents. The ﬁnding in Proposition 3 enables us to focus only on the
eﬀect after ψg.
The ﬁnal graph is a quasi-concave curve depicting the relationship between ψand σ. For
rising initial levels of public good expenditures and overall spending and taxation, the state of
the economy variable increases at a decreasing rate, as voter preferences for public goods and
other forms of spending are being satisﬁed. After ψgfurther public good expenditures start
including wasteful spending. The fact that rents can only be created after an initially provided
level of public goods gentails the discontinuous eﬀect they have on the state of the economy
curve. A decreasing state of the economy is a mere consequence of negative voter perception on
signals of political satisfaction of personal and partial interests.
The deteriorating state of the economy caused by higher rent-extraction will leave more and
more voters dissatisﬁed, who will if ψp> ψ, for which the state of the economy would be σ < σ,
elect an incumbent out of oﬃce. The threshold level ψwill present the point above which
further public good expenditure gains disproportionally more to the incumbent in rents than to
the voters in public goods16.
Proposition 4. Assume the incumbent observes β0≥β∗. If the incumbent maximizes rents via
the public good expenditures function, and if the re-election probability depends on staying within
the desired re-election set Ω∈ψ, ψ, he will always choose the voters’ higher threshold level ψ
for the observed positive β0shock. The equilibrium levels of public good expenditures and the
re-election threshold ψpare then:
g∗=g and ψp,∗=ψ(24)
The incumbent will choose the optimal equilibrium level of g∗from which it can extract the optimal
amount of rents, r∗=r. In other words, politicians respect the re-election threshold just enough
to stay in oﬃce.
Proof: See Appendix.
If gwould be the total ﬁnal amount of public good spending, then the area from gto g
depicts total wasteful spending, while rto rdepicts the total amount of rents. By choosing the
equilibrium g∗and ψp,∗, for a high enough shock β0, an incumbent is able to maximize both his
rent-extraction (r∗=r), within the allowed boundaries, and his chances of re-election, since the
voter threshold for the current period is respected, ψp≤ψ.
Proposition 5. If the equilibrium public good expenditure is g∗=g, and the equilibrium threshold
is ψ∗=ψaccording to Proposition 4, and under the assumption of the incumbent observing
16Note here how an update of the threshold ψupwards by the voters increases the scope for re-election.
β0≥β∗, the equilibrium level of the state of the economy is then always:
The state of the economy σis optimal σ∗=σ(ψ∗(g∗)), for any ψ∗and g∗chosen that satisfy
A possible normative implication would be that rent-extraction leads to a misappropriation
of resources which implies a worse oﬀ state of the economy and lower voter utility. Instead of
achieving a higher state of the economy σ, the equilibrium revolves around a lower σ, which
always implies some level of wasteful spending. In addition, Proposition 4 implies higher than
voter optimal equilibrium taxation and government spending (since ψp,∗=ψ > ψv) thereby
possibly explaining some of the growth of government size in the past century.
4 Empirical evidence
The empirical implication is that upon observing a suﬃciently negative economic shock, the
re-election threshold will be disturbed via more wasteful spending leading to the electoral defeat
of the incumbent. The crucial eﬀort in proving this proposition is to quantify the eﬀect of the
threshold ψon the probability of re-election. The paper tests the following propositions derived
from the model: (i) an increase of ψ(which is approximated by capital outlay spending per
capita) decreases the probability of re-election after a certain level; and (ii) a decrease of β
(approximated by a negative GDP growth shock) one year before the election will lead to an
increase of ψ, i.e. higher spending on potentially wasteful public goods.
4.1 Data and empirical strategy
A panel dataset is collected for gubernatorial and state legislature elections (both upper and
lower house) for 48 continental U.S. states over the period from 1992 to 2008. The database
contains state elections for every two years17 which includes 9 elections for both governor and
the state legislature. Using U.S. states oﬀers a number of attractive features in terms of common
methodology and data availability, and more importantly the stability of its electoral institutions
and rules. In addition, all states are accountable to the same constitutional boundaries and
long-lasting democratic order, not to mention the prevalence of democratic informal institutions
and a roughly similar perception towards corruption across the states (the λparameter). A
panel dataset allows the paper to account for such cultural factors, corruption perceptions, and
electoral institutions as ﬁxed both across states and over time. Data on state and local spending
17Five U.S. states (AL, LA, MA, MI, NB) are only holding legislature elections for the lower house every 4
years, while Nebraska has a unicameral and a non-partisan state legislature. All other states hold lower house
legislature elections every two years.
is collected for each state observed, along with the variables of economic performance proven
to have an eﬀect on re-election of incumbents according to the literature18. Summary statistics
of all variables used in the model are presented in Tables 1 and 2, along with the sources and
explanations of electoral data, budget spending and all other variables used.
[Tables 1 and 2 about here]
The empirical strategy estimates the following binary response model, predicting the eﬀect
of changes in ψon the electoral success of the incumbent:
P(Iit = 1|ψit, it ) = Gγ0+γ1ψit +γ2ψ2
it +ξXit +ϑDit +it(26)
Where Gis the standard cumulative distribution function (c.d.f.) deﬁned strictly between
zero and one, 0 < G(z)<1, for all real numbers z, ensuring that the estimated response
probabilities fall between zero and one.
The dependent variable Iit for state iand time tis the dummy indicator that takes the value
1 if the incumbent governor is (re-)elected or if the party stays in majority in the state legislature
and 0 if the incumbent governor loses elections or the party loses its majority. For a Republican
governor in power if the Republicans lose the local assembly elections in the middle of his term,
the value assigned is 0. If the Republicans win this implies that they retain majority (or have
won the majority in a previously Democratic held assembly), so the value assigned is 119.
The explanatory variable is the threshold ψit, or more precisely public good spending. De-
composing public good spending into white elephant projects and spending on voter-desired
public goods is a daunting task. The fact that politicians conceal their corruption and rent-
extraction within the budget allocation process makes this task even more diﬃcult. This is why
the paper assigns a proxy to try and evaluate the eﬀect of rents on re-election probabilities. As
assumed in the theoretical part, the only way to increase rent-extraction is via higher public
good spending, in particular higher wasteful spending (see Eq 4). To capture this the paper will
observe growth of public good spending deﬁned as capital outlays (deﬁnition given under Table
2), since this budgetary category is most usually subject to misappropriation in terms of fraud-
ulent procurements and diversion of public funds. Mauro (1998) recognized the existence of
such corrupt practices being more frequent for large infrastructure projects that generally fall
under the capital outlays category. Capital outlays are presented in per capita terms for each
state, to make it comparable across states.
Parameters γ1and γ2measure the eﬀects of capital outlay spending on incumbent re-election.
The squared value (ψ2
it) should be able to indicate the concavity of the voters’ preferences over the
threshold as presented in Figure 1 (provided that γ2turns out negative). The control variables
18See e.g. Brendner and Drazen (2008) or Besley and Case (2003)
19If the governor and the legislature are from two diﬀerent parties then a governor defeat is counted as zero,
since executive power surpasses the legislative one.
are divided into a vector of economic (Xit) and demographic (Dit) variables that may aﬀect the
likelihood of incumbent re-election. Economic controls include measures of economic performance
such as GDP growth in the election year, revenue and expenditure growth, unemployment rate,
personal income, and deﬁcit to GDP. Demographic controls include total state population, share
of population under 15 (young) and share of population over 65 (old), implying that states with
high shares of old or young people will have higher levels of targeted social spending.
4.2.1 Negative economic shocks and wasteful spending
Before testing the eﬀect of wasteful public good spending on re-election, it is necessary to estimate
whether there is a link between a negative economic shock and higher spending on white elephant
public goods, as assumed in Proposition 2. This could be diﬃcult to prove since politicians
could simply be applying countercyclical measures to combat a negative economic shock, thus
making the ﬁnding trivial. In order to distinguish between which eﬀect is more likely, the paper
contrasts the negative growth eﬀect on the proxy for wasteful spending (capital outlays per
capita) with how the negative economic shock aﬀects total expenditures. Furthermore the paper
also separates the two diﬀerent types of spending; capital outlays (spending on public goods) and
current expenditures which include social spending, public sector wages, unemployment beneﬁts,
education and health spending, etc. If an incumbent facing a negative shock is actually using
countercyclical measures to combat the shock, then we should expect to see a signiﬁcant negative
eﬀect between last year economic growth and both total and current expenditures. If however
a negative shock only aﬀects public good spending then this would, albeit partially, conﬁrm the
intuition presented in Proposition 2 of the model.
The following ﬁxed eﬀects panel data regression is estimated:
E(ψit|βit , µit) = αi+ηit βit +ξXit +ϑDit +µit (27)
The dependent variable, ψit is ﬁrst deﬁned and reported as capital outlay spending per capita
in regressions (1) and (2) in Table 3. Regressions (3) and (4) observe current expenditures per
capita as the dependent variable, while regressions (5) and (6) observe total expenditures per
capita. βit represents the main explanatory variable – an economic shock of state ione year
before the election, approximated by real GDP growth. In columns (2), (4) and (6) instead of
last year’s economic growth, a two year average growth rate has been used to take into account
a longer decision-making time span. Parameter ηit measures the total eﬀect of previous year(s)
GDP growth on the explanatory variable of interest. Xit and Dit represent vectors of economic
and demographic controls, while αiis the unobserved heterogeneity, containing all the possible
unobserved state characteristics, assumed to be ﬁxed across states and over time. Standard
errors are robust to heteroskedasticity and clustered by state.
The results are presented in Table 3. Column (1) and (2) show that for a lower GDP growth
rate one year before the election (or during the entire 2 year term), states tend to have higher
values of capital outlays per capita in the election year. In a given state, for a 1 percentage point
lower rate of GDP growth in the previous year, capital outlays per capita are predicted to be
higher in the current year by 0.19, controlling for all other time-invariant factors. Given that
the average value of capital outlays per capita being 0.753 for the entire sample, this represents
a rather strong eﬀect. In terms of the two year average growth levels, the eﬀect is much stronger
(as expected due to a longer decision-making horizon), but still in the same direction.
[Table 3 about here]
The control variables show expected directions; an increase of total expenditures results in
higher capital outlay spending, an increase in income taxes as well, while a higher unemployment
rate and a larger share of young and old in a state all predict a negative eﬀect on capital outlays
per capita. This makes sense since they all imply higher expenditures on social transfers, thus
lowering the amount of funds available for public good creation. Finally, the term limit eﬀect
signals that as the end of the ﬁnal term for the governor approaches, even though he has an
increasing likelihood to extract more rents20, the party as a whole will try to decrease public
good spending in order to remain in power. It makes sense that parties react diﬀerently to the
term limit rule than individual politicians.
In order to test the robustness of this initial result, the paper examines how the growth shock
aﬀects public spending in general. Hence in columns (3) and (4) the paper ﬁrst tests the eﬀect
of a negative growth shock on current expenditures (spending on social security, wages, health
and education) in per capita terms. In both cases there is a similar relationship as before – a
negative growth shock one year before the election increases current spending p/c, even though
some control variables lose their statistical signiﬁcance. In the ﬁnal two columns, the growth
eﬀect was tested for total spending per capita, and again the same result has been found. In
terms of the control variables in the ﬁnal four regressions the negative eﬀect of old and young
in the population is somewhat counterintuitive, even though it could probably be explained by
speciﬁc state idiosyncrasies.
Overall the ﬁndings in Table 3 point to a positive relationship between higher spending
on capital outlays and a negative growth shock, however lower GDP growth also causes total
spending to increase. It increases public spending on a state level across all categories. This
still leaves us unsure whether politicians use a negative growth shock to increase their rents or
to ensure their preservation in power, or is it in fact both, where their reaction depends on the
magnitude of the shock ∆β0>∆gR∆τ. The ﬁndings in Tables 4 and 5 could shed more light
20As empirically proven by many term limit models such as Alt, Bueno de Mesquita and Rose (2011), Besley
and Case (1995b), Ferraz and Finan (2011) and Smart and Sturm (2013).
In testing the diﬀerent models a Hausman test has been used every time to diﬀerentiate
between using ﬁxed eﬀects or random eﬀects. In every case the Hausman test suggested the use
of ﬁxed eﬀects. The Chi squared values and the corresponding p-values for the Hausman test
are reported under each column.
4.2.2 Wasteful spending and re-election
The results of the main prediction of the model — the eﬀect of capital outlay spending on the
probability of re-election – are presented in Table 4. Three limited dependent variable models
are compared; a probit random eﬀects panel data regression (columns 1 and 2), a logit random
eﬀects panel data regression (columns 3 and 4), and the standard linear probability model (LPM)
(column 5). Columns (2) and (4) present the average marginal eﬀects of the subsequent probit
and logit estimates, the reason for which was primarily to make all three models comparable in
terms of interpretation.
According to the aggregate results in Table 4 it can be inferred that over time the increasing
levels of capital outlays per capita increase the probability of re-election for the incumbent and
imply higher public good spending each period. As the population increases, the tax base is
larger, revenues are higher and so are the expenditures. The ﬁnding goes in line with the pre-
diction in Proposition 3, where the threshold chosen would always be the higher level. However,
the negative value of γ2(from Eq 26), signiﬁcant at a 5% level in the ﬁrst four columns and at a
1% in the ﬁnal column, implies the concavity of voter preferences where too high levels of capital
outlay spending lead to a decrease of voter utility that can cause the incumbents to lose oﬃce.
[Table 4 about here]
The total eﬀect of capital outlays on re-election must be calculated by jointly observing bγ1and
bγ2, where we can calculate the cut-oﬀ point using the estimated coeﬃcients with the following
In column (2) if bγ1= 1.405 and bγ2=−0.497, then the lower cut-oﬀ value of ψis b
1.405/2(0.497). This implies that after the average level of capital outlays per capita exceeds
1.41 it lowers the probability of winning. At that point of spending the incumbent party can
maximize its probability of staying in power. For example, the cut-oﬀ level of 1.41 will result in
a probability of winning of P(I)=0.74. Any value above the cut-oﬀ decreases the probability of
winning, holding all other parameters constant (see Figure 1). From this one can easily calculate
the upper cut-oﬀ level of ψ, above which politicians get thrown out of oﬃce. For the entire
sample, the average value of the upper cut-oﬀ (for which the probability of winning is lower than
0.5) would be around 2.11.
This can be seen by plugging in diﬀerent cut-oﬀ values and summing up the product of the
mean of the control variables with the resulting coeﬃcient from column (2). For example, for
close to extreme values in the sample a high level of capital outlay per capita of 2.5 will result
in a probability of winning of only 0.15, while the lowest value in the sample of 0.26 will yield
a probability of winning of only 0.08, controlling for all other factors. The average value of
capital outlays p/c for the entire dataset was 0.753, which yields a probability of winning of
0.485. An increase of capital outlays p/c from 0.83 to 1.4 (a two standard deviation increase
up until the lower cut-oﬀ) increases the probability of winning by 0.167, whereas an increase of
capital outlays from 1.4 to 1.97 (again a two standard deviation increase), decreases probability
of winning by 0.155. A one standard deviation increase or decrease from the cut-oﬀ value only
aﬀects the probability of winning by 0.04. However a one standard deviation increase of capital
outlays from the average sample value increases the probability of winning by 0.18.
If we compare the eﬀect across individual states, for example in California a one standard
deviation increase of capital outlays p/c from the average value (0.823) increases the probability
of winning by 0.12, while an increase above the cut-oﬀ level by one standard deviation decreases
the probability of winning by 0.15. In Alabama for example, an increase of capital outlays p/c
by one standard deviation from the average (0.647) will increase the winning probability by 0.17,
while a one standard deviation increase from the cut-oﬀ level will lower the probability of winning
Columns (4) and (5), as expected, show almost identical results in terms of size and magnitude
of the eﬀects for the other two models, the logit and the LPM. However the cut-oﬀ levels are
slightly diﬀerent (1.417 for the logit, but 1.865 for the LPM), as are the calculated probabilities
(for the average sample value of capital outlays at 0.753, the logit predicts a probability of
winning of 0.476, whereas the LPM predicts the probability of 0.607, controlling for all other
factors). In each case the percent correctly predicted is reported (the standard 0.5 threshold was
used) as a viable goodness-of-ﬁt measure, as is the pseudo R-squared (in case of the LPM it is a
regular R-squared), and the log-likelihood value for the ﬁrst two models. In each case the model
correctly predicts over 60% of the cases, while the pseudo R-squared is around 0.20 for logit and
probit, and slightly lower for the LPM.
The inclusion of the term limit variable signals an expected negative relationship in each
model tested, implying that if the party’s governor is reaching a term limit, the likelihood of
the party remaining in oﬃce will decrease. This is probably why the results in Table 3 yielded
the opposite of the standard term limit eﬀect – parties, unlike individual politicians, will try to
improve their winning probabilities by decreasing capital outlay spending (i.e. decreasing their
rent-extraction) in periods of pre-observed poor growth when facing a term limit. As stated in
the theoretical part of the paper, parties tend to be a more patient agent.
Most economic performance indicators across all models in Table 4 seem to show weak and
non-signiﬁcant eﬀects on the probability of re-election. Only deﬁcit to GDP, revenue growth
and population growth exhibit some signiﬁcant eﬀect, with an expected direction of each of the
variables according to the standard economic literature. This could be explained by the fact that
economic performance of states matters less in local elections than it does on a national level.
In local politics budgetary redistribution and public goods play a much more important role.
However, what if the voters respond to all categories of spending this way, not just capital
outlays? Table 5 tests the inclusion of other potential explanatory variables instead of capital
outlays, in a similar way as presented in Table 3. It shows that none of the alternative categories
of spending exhibit the same eﬀect capital outlay spending does. Columns (1) and (2) use total
expenditures p/c and current expenditures p/c (the same parameters as in Table 3), and even
though in the case of total spending p/c there is a positive eﬀect of total expenditures on the
probability of winning (as anticipated earlier), neither of the two variables report a comparative
eﬀect to that of capital outlay spending. Other potential variables used such as the ratio of
capital outlays to current spending (column 3) and current to total spending (column 4) also
show no signiﬁcant eﬀect on the probability of winning. In addition to the ones reported, many
other variables of spending have been used (including aggregate total and current spending,
and spending to GDP), neither of which showed any signiﬁcant eﬀect to the extent that capital
outlays per capita did. The implication is that politicians in local elections can only aﬀect their
re-election chances via manipulating public good spending, while current and social spending
seem to be ineﬀective vis-a-vis re-election probabilities.
[Table 5 about here]
Another possible concern may be the diﬀerences in agency dynamics between states that
have term limits and those that don’t. Fourteen states do not have any term limits in their
state electoral law, while the remaining thirty six have diﬀerent regimes and rules, however
they all apply the term limit electoral rule to some extent. Furthermore it also makes sense to
diﬀerentiate between gubernatorial and state legislature elections as diﬀerent incentives may be
driving individuals and parties in their reelection pleas.
Table 6 reports results only for the capital outlays per capita variable and the election year
growth rate, but it includes the same controls as in Table 4. The ﬁrst two columns separate the
sample into states with and without term limits, while the last two columns separate the sample
into governor and legislature elections. According to these results, it can be inferred that the
total eﬀect reported in Table 4 concerning capital outlays per capita is being driven by electoral
results in term limited states and for legislature elections (although the estimates in column 4,
as in column 2, fail the Wald test). The implication of this is that political parties respond quite
well to the term limit electoral rule. Only for term limited states can we conﬁrm the overall story
linking potentially wasteful spending to electoral chances. The same however cannot be inferred
for individual politicians running at gubernatorial elections. Even though the direction of the
eﬀect is the same, it isn’t signiﬁcant, meaning that for governors other things tend to be more
important in determining their electoral chances. In particular, economic growth in the election
year seems to be the most important factor for governors, which makes sense as the voters can
clearly place blame on individual governors for poor economic performance (and is also in line
with the incumbency hypothesis (Kramer, 1971)).
[Table 6 about here]
Finally, if we connect the ﬁndings of Tables 4 and 5 with the results reported in Table 3,
it would appear that for pre-observed negative shocks political parties in power opt to increase
all forms of public spending as an initial reaction to the adverse shock. However if they divert
too much of their spending towards public good production, there is a danger that this type of
spending is used for rent-extraction rather than as a way to help the economy recover. If this is
the case, the voters will punish them.
Other types of spending fail to oﬀer similar results with respect to re-election probabilities.
Intuitively, higher public spending on various social expenditures will hardly throw a politician
out of oﬃce, but higher spending on capital outlays will. Why is this so? One of the possible
explanations could be the implications vested in the model – capital outlays represent a budgetary
category most easily subject to misappropriation, so when politicians increase this category
too far (extract too much rents) voters punish them. It is far from conclusive that politicians
become more corrupt after a negative shock, but it is possible that higher rent-extraction throws
politicians out of oﬃce and that this rent-extraction can indeed be preceded and incentivized by
a negative economic shock.
In order to prove this relationship with more precision, one should perhaps use a better
proxy for political corruption and rent-extraction at the local state level. The availability of such
data is extremely scarce, even though in certain instances with a unique database of potentially
wasteful political spending (e.g. Bandiera et al, 2009; Ferraz and Finan, 2011; Kaufman
and Vicente, 2011) this can indeed be achieved. This paper opens up scope to an entirely new
research in this direction aimed at linking corruption and misuse of public oﬃce to long-lasting
mandates in some levels of local, and perhaps even national, government.
The paper anticipates that if agents are inﬁnitely patient they can stay in oﬃce for inﬁnite
amounts of time, provided that they face a favorable economic shock each period. Even though
this may sound implausible, the attractiveness of holding power, particularly on a local level,
actually does yield results where certain politicians and political parties retain oﬃce for as long
as they like, or at least until some exogenous shock disturbs their position. From a multitude of
examples and anecdotal evidence in the developing world, the most striking one actually comes
from the United States and the former major of a small town Bell, California, Robert Rizzo,
who managed to stay in power for 17 years and pay himself a salary close to $800,000 per year,
even though the majority of Bell’s citizens are relatively poor (Bueno De Mesquita and Smith,
2011). Rizzo made sure they remain poor by levying high taxes to pay for the cronies that
were keeping him in power. Even though this example testiﬁes of a complex environment which
is more likely to resemble state capture than pure rent-extraction, the implications are obvious:
it is indeed possible to successfully overcome the trade-oﬀ between rent-extraction and holding
Empirically the paper conﬁrms the possibility of seizing the opportunity of higher rent-
extraction once in oﬃce by ﬁnding the cut-oﬀ level of wasteful spending the politicians need
to respect in order to maintain power. It also ﬁnds that parties react diﬀerently to the term
limit constraint than individual politicians. The main ﬁnding is that in times of economic down-
turns politicians use public good spending to increase their electoral chances; however this only
works up until a certain point, where further spending on public goods is likely to be perceived
as wasteful spending by the voters, who will then punish the incumbents. These ﬁndings open
up scope to a new research direction aimed at uncovering the actual reasons behind long-lasting
mandates characterized by rampant corruption and rent-extraction.
Proof of Lemma 1. Solving a non-linear optimization problem is most easily done using the La-
grange multiplier method. Formulating the maximization problem as:
s.t. f (σv
t) = c
The ﬁrst order conditions are:
L(σ, ψ, λ) =
t) + λ(c−f(σv
∂σ = 0
∂ψ = 0
∂λ =c−fσv ,∗
According to the implicit function theorem the extreme points at σv,∗
tof the voter
utility function u, for any Lagrange multiplier λ∈Rsatisfy:
The two equations representing partial derivatives over σand ψare satisﬁed at the point where
the extreme values σv,∗
toccur. The ﬁrst order conditions characterize the maximum
diﬀerence between the objective function and the constraint. The optimal level of ψv,∗
tis the maximum diﬀerence, and therefore presents the constrained optimum
Proof of proposition 2. Let Gbe a ﬁnite stage game between voters and politicians, where the
strategy of the voters is an action proﬁle (ar, a−r)∈A, while the cooperative strategy of an
incumbent iis si= (si1, . . . , sin), for every si∈S. A cooperative strategy infers respecting
the voter re-election threshold ψ≤ψ, implying an expected utility of ui(si). Let the deviation
strategy of an incumbent be denoted as s−i, with an expected utility of ui(s−i).
In a one period game, politicians maximize their immediate payoﬀs by choosing a defection
strategy s−isince E[U0
I(br|bg, λ)] > E[U0
I(r|g, λ)] which is true for br > r and bg > g ∀r, g. The
best response of the voters is to apply a punishment strategy, a−r. A one period game ends
up with a non-cooperative Nash equilibrium regardless of the shock β0since both players are
aware that no future periods exist. Deﬁne (xe1, . . . , xen)∈D(s−i, a−r) as the one period Nash
equilibrium of Gfor which the payoﬀs are (e1, . . . , en), and (xp1, . . . , xpn )∈C(si, ar) as the set
of cooperative actions of both players for which the optimal payoﬀs are (p1, . . . , pn).
In an inﬁnitely repeated stage game G(∞, δ) the players apply a trigger strategy where they
both play xpi ∈C(si, ar) in the ﬁrst stage, while at the tth stage if the outcome of all preceding
periods has been (p1, . . . , pn), they play xpi; otherwise they play xei ∈D(s−i, a−r). If both
players adopt this strategy than the outcomes of every period are (xp1, . . . , xpn), with expected
payoﬀs of (p1, . . . , pn). The expected utility of an incumbent following a cooperative strategy in
a repeated game is a weighted average of payoﬀs in each stage, weighted by the common discount
factor and an introduced economic shock, β0, as speciﬁed in Eq 19.
According to Friedman’s (1971) Theorem if the repeated game satisﬁes all the above
properties, if pi≥ei, and if the discount factor is suﬃciently close to one (which is by assumption
of using political parties always true), then there exists a subgame-perfect Nash equilibrium of
the inﬁnitely repeated game G(∞, δ) that results in (p1, . . . , pn) as the average payoﬀ.
For the Friedman Theorem to hold in this case, it must be shown that pi≥ei, or ui(si)≥
ui(s−i) for any incumbent i. The incumbent plays a cooperative strategy if and only if the
payoﬀ from a cooperative strategy is higher than the payoﬀ from a defection strategy, as stated
in Eq 20:
I(r|g, λ)] + (1 + β0)
I(r|g, λ)] ≥E[U0
I(br|bg, λ)] + E[Ut
Solving the upper equation for β0yields the optimal strategy for the incumbent, as speciﬁed in
Proposition 1. An incumbent cannot get a better payoﬀ by deviating for the given conditions
of β0, meaning that the cooperative strategy solved for β0≥β∗yields a Nash equilibrium of
the tit-for-tat game for the incumbent. The game G(∞, δ) is a repeated stage game, repeated
in every single period. A subgame-perfect equilibrium of a repeated game includes a stage game
Nash equilibrium in every sub game. Since the stage game Nash equilibrium is played every
period, or in every sub game, it is by deﬁnition a subgame-perfect Nash equilibrium.
Proof of proposition 3. Any level of public goods g < g implies two eﬀects; a non-optimal amount
of rents (r= 0) and no re-election (as the voter re-election threshold Ω ∈ψ,ψisn’t satisﬁed).
Any level of public goods g=gimplies re-election since the voter threshold is respected but the
level of rents is still r= 0 by assumption of Eq 4 where g= (1 −λ)Pm
j=1 Gj. The incumbent
utility maximization function is according to Eq 16 depended on rent-extraction (any r > r),
thus disabling the incumbent from choosing any g=gand therefore obtaining no rents. Since
it isn’t plausible for the incumbent to choose any g≤g, the chosen level of public goods always
has to be g > g.
Proof of proposition 4. From the assumption implied by the model that the level of rents in-
creases with public good expenditures in Eq 4 it is obvious that the higher level of gchosen from
the set P ∈ [g0, . . . , gi, . . . , gn],∀i∈Nincreases the utility an incumbent gets. The set Pcontains
increasing elements for every level of expenditures chosen, meaning that g0< g1< g2< . .. < gn.
According to the deﬁnition of ψfrom Eqs 5 and 6, the choice of ψis also determined within a
set containing increasing elements; O ∈ [ψ0, . . . , ψn] where ψ0< ψ1< ψ2< . . . < ψn, and where
ndenotes the decision on the size of spending and taxes; ψ0is the lowest level chosen implying
no taxes and no spending, while ψnis the highest level chosen implying maximum taxes and
If an incumbent is playing a cooperative strategy as implied in Proposition 3 (β0≥β∗) it
chooses any level of ψwithin the set Ω ∈ψ, ψ, where Ω ⊆ O (a subset of O). By assumption
ψ0< ψ and ψ < ψn, meaning that the highest level of ψ∈ O is higher than ψand that the lowest
level of ψ∈ O is lower than ψ. If Ω and Oare both sets containing increasing elements and if
ψ0< ψ and ψ < ψn, then by choosing the highest ψwithin the re-election threshold set Ω in
order to maximize its utility from rents and still stay in power, the incumbent will always choose
the level ψ∗=ψ. The decision of optimal g∗=gfollows the same intuitive conclusion.
 Acemoglu, D., and Verdier, T., 2000. “The Choice between Market Failures and Corruption.”
American Economic Review, 90(1): 194-211.
 Alt, J., de Mesquita, E.B., and Rose, S., 2011. “Disentangling Accountability and Compe-
tence in Elections: Evidence from US Term Limits.” Journal of Politics, 73(1): 171-186.
 Austen-Smith, D., and Banks, J., 1989. “Electoral Accountability and Incumbency.” In P.C.
Ordeshook, (ed.), Models of Strategic Choice in Politics. Ann Arbor: University of Michigan
Press, pp. 121-148.
 Bandiera, O., Prat, A., and Valletti, T. (2009). “Active and Passive Waste in Government
Spending: Evidence from a Policy Experiment.” American Economic Review, 99(4): 1278-
 Banks, J., and Sundaram, R., 1993. “Adverse Selection and Moral Hazard in a Repeated
Elections Model.” in W. Barnett, M. Hinich, and N. Schoﬁeld (eds.), Political Economy: In-
stitutions, Information, Competition and Representation. New York: Cambridge University
Press, pp. 295-311.
 Barro, R., 1973. “The Control of Politicians: An Economic Model.” Public Choice 14: 19-42.
 Besley, T., 2004. “Paying Politicians: Theory and Evidence.” Journal of the European Eco-
nomic Association 2(2-3): 193-215.
 Besley, T., 2006. Principled Agents? The Political Economy of Good Government. New
York: Oxford University Press.
 Besley, T., and Case, A., 1995a. “Incumbent Behaviour: Vote-Seeking, Tax-Setting and
Yardstick Competition.” American Economic Review, 85(1): 25-45.
 Besley, T., and Case, A., 1995b. “Does Political Accountability Aﬀect Economic Policy
Choices? Evidence from Gubernatorial Term Limits.” Quarterly Journal of Economics,
 Besley, T., and Case, A., 2003. “Political Institutions and Policy Choices: Evidence from
the United States.” Journal of Economic Literature, 41(1): 7-73.
 Besley, T., and Smart, M., 2007. “Fiscal Restraint and Voter Welfare.” Journal of Public
Economics, 91: 755-773.
 Brender, A., and Drazen, A., 2008. “How Do Budget Deﬁcits and Economic Growth Aﬀect
Reelection Prospects? Evidence from a Large Panel of Countries.” American Economic
Review, 98(5): 2203-2220.
 Brennan, G., and Buchanan, J., 1980. The Power to Tax: Analytical Foundations of a Fiscal
Constitution. Cambridge: Cambridge University Press.
 Buchanan, J., and Tullock, G., 1962. The Calculus of Consent. Logical Foundations of
Constitutional Democracy. Ann Arbor: University of Michigan Press.
 Buchanan, J., 1975. The Limits of Liberty. Between Anarchy and Leviathan. Chicago: Uni-
versity of Chicago Press.
 Bueno de Mesquita, B., Smith, A., Siverson, R., and Morrow J., 2005. The Logic of Political
Survival. Cambridge: MIT Press.
 Bueno de Mesquita, B., and Smith, A., 2011. The Dictator’s Handbook. Why Bad Behavior
is Almost Always Good Politics. New York: PublicAﬀairs.
 Caselli, F., and Morelli, M., 2004. “Bad Politicians.” Journal of Public Economics, 88:
 Coate, S., and Morris, S., 1995. “On the Form of Transfers to Special Interests.” Journal of
Political Economy, 103(6): 1210-1235.
 Dzhumashev, R., 2014. “The two-way relationship between government spending and cor-
ruption and its eﬀects on economic growth.” Contemporary Economic Policy 32(2), 403-419.
 Ferejohn, J., 1986. “Incumbent Performance and Electoral Control.” Public Choice 50: 5-25.
 Ferraz, C., and Finan, F., 2011. “Electoral Accountability and Corruption: Evidence from
the Audits of Local Governments.” American Economic Review, 101(4): 1274-1311.
 Friedman, J., 1971. “A Non-cooperative Equilibrium for Supergames.” Review of Economic
Studies, 38, 1-12.
 Goel, R., and Nelson, M., 1998. “Corruption and government size: A disaggregated analy-
sis.” Public Choice 97: 107-120.
 Higgs, R., 1987. Crisis and Leviathan. Critical Episodes in the Growth of American Gov-
ernment. New York: Oxford University Press.
 Kaufmann, D. and Vicente, P., 2011. “Legal Corruption.” Economics and Politics, 33(2):
 Kramer, G., 1971. “Short-term ﬂuctuations in US voting behavior, 1896-1964.” American
Political Science Review, 65: 131-43.
 Krueger, A., 1974. “The Political Economy of the Rent Seeking Society.” American Eco-
nomic Review, 64: 291-303.
 Maddison, A., 2001. The World Economy - A Millennial Perspective. Paris: OECD Devel-
 Mauro, P., 1998. “Corruption and the Composition of Government Spending.” Journal of
Public Economics 69(2): 263-279.
 Peltzman, S., 1992. “Voters as ﬁscal conservatives.” Quarterly Journal of Economics, 107(2),
 Persson, T., Roland, G. and Tabellini, G., 1997. “Separation of Powers and Political Ac-
countability.” Quarterly Journal of Economics, 112(4): 1163-1202.
 Persson, T., and Tabellini, G., 2000. Political Economics. Explaining Economic Policy.
Cambridge: MIT Press.
 Rogoﬀ, K., 1990. “Equilibrium Political Budget Cycles.” American Economic Review, 80(1):
 Smart, M., and Sturm, D., 2013. “Term Limits and Electoral Accountability.” Journal of
Public Economics, 107, 93-102.
 Tanzi, V., and Schuknecht, L., 2000. Public Spending in the 20th Century: a global perspec-
tive. Cambridge: Cambridge University Press
 The Tax Foundation, 2011. State and Local Tax Burdens: 1977 2009. [online] Available at:
http://www.taxfoundation.org/news/show/335.html [Accessed 5th August 2011].
 Tullock, G., 1967. “The Welfare Costs of Tariﬀs, Monopolies and Theft.” Western Economic
Journal, 5(3): 224-232.
 United States Census Bureau, 2011. “Statistical Abstract of the United States.” [online]
Washington DC, years 1992 2010. Available at: http://www.census.gov/statab/ [Ac-
cessed 27th June 2011].
 United States Department of Commerce, Bureau of Economic Analysis, 2011. Interactive
Data, GDP, years 1990 2008. [online] Available at: http://www.bea.gov/iTable/index_
nipa.cfm [Accessed 30th July 2011].
Table 1: Election summary data
Elections/Parties Governor State Senate(Upper) State House(Lower)
Total Democrats 96 218 242
Total Republicans 115 205 181
Total Independent 3 - -
Total elections 214 423 423
All 48 states included over the period from 1992 until 2008. Total Democrats and total Republicans includes
every time when a Democrat or Republican governor or party would either win oﬃce or hold oﬃce. Source and
description of data: Election data on both gubernatorial and state legislature election (upper and lower house)
was taken from the Statistical Abstract of the United States from the years 1992 - 2008 published by the Census
Bureau (2011). Notes on electoral results: Nebraska state legislature is unicameral and non-partisan, so only
gubernatorial changes are observed in this state (every four years). In California in 2003 gubernatorial recall
elections are taken into account instead of the 2002 elections. The democrat governor in power at the time,
Gary Davis, instead of ensuring his second term was recalled a year later. On the new elections the Republican
candidate Arnold Schwarzenegger won. The dummy value given for 2002 is 0, since it is accounted as an
incumbent defeat. Gubernatorial and state legislature elections are all being held in even years except for
Kentucky, Louisiana, Mississippi, New Jersey and Virginia which are held in odd years. The growth eﬀects are
all taken into account for these 5 states.
Table 2: Summary statistics
Variable Observations Mean Std.Dev. Min Max
Re-election 432 0.6041 0.4895 0 1
Capital outlays p/c 432 0.7535 0.2844 0.2644 2.713
Capital outlays p/c sq 432 0.6484 0.5776 0.0698 7.359
Total expenditures p/c 432 6.263 1.8094 3.053 14.108
Current expenditures p/c 432 4.649 1.3712 2.326 10.247
Total expenditures to GDP 432 0.1874 0.0261 0.1233 0.2683
Total expenditures 432 3.82 ×1075.03 ×1072.45 ×1064.15 ×108
Current expenditures 432 2.8×1073.62 ×1071.76 ×1063.01 ×108
Capital outlays 432 4.56 ×1065.98 ×1061.87 ×1054.67 ×107
Capital outlaysto current spending 432 0.1630 0.0399 0.0745 0.3138
Current spendingto total spending 432 0.7415 0.0372 0.574 0.83
Term limit 432 0.2176 0.4131 0 1
GDP 432 2.04 ×1072.5×1081.25 ×1071.91 ×109
Real GDP growth 432 0.0362 0.0378 -0.0483 0.3597
Lag real GDP growth 432 0.0554 0.0241 -0.0536 0.1399
Two year average growth 432 0.0442 0.0426 -0.0299 0.2135
Expenditures growth 384 0.0748 0.0482 -0.0207 0.3016
Revenue growth 389 0.0667 0.1385 -0.3817 0.5898
Unemployment rate 432 0.0505 0.0135 0.022 0.112
Unemployment change 384 -0.0039 0.2421 -0.4384 1.027
Deﬁcit to GDP 432 0.0102 0.0199 -0.0412 0.1928
Deﬁcit to GDP change 384 -0.0343 9.585 -95.61 115.73
Income tax 432 0.0942 0.0118 0.062 0.127
Change in income tax 384 -0.0015 0.0219 -0.098 0.106
Personal income 432 30779.44 8780.61 15606.07 63889.87
Personal income growth 384 0.0957 0.0434 -0.033 0.2809
Population change 389 0.0133 0.0139 -0.007 0.1045
Share of under 17 384 0.2519 0.0197 0.207 0.352
Share over 65 384 0.1314 0.0682 0.085 1.425
Sources and description of data: Data on public good spending, budget revenues and expenditures decomposed
into the data on capital outlays and current expenditures was taken from the US Census Bureau (2011) for
the entire period observed. Capital outlays are deﬁned as: Direct expenditure for contract or force account
construction of buildings, grounds, and other improvements, and purchase of equipment, land, and existing
structures. Includes amounts for additions, replacements, and major alterations to ﬁxed works and structures.
However, expenditure for repairs to such works and structures is classiﬁed as current operation expenditure. (US
Census Bureau, 2011). Current expenditure include direct expenditure for compensation of own oﬃcers and
employees and for supplies, materials, and contractual services except amounts for capital outlay, assistance and
subsidies, interest on debt, and insurance beneﬁts and payments. (US Census Bureau, 2011). Data on GDP
and unemployment is taken from the US Bureau of Economic analysis (2011). Income taxes and personal
income data was taken from the Tax Foundation (2011). Population data was taken from the Statistical
Abstract of the United States published by the Census Bureau (2011). The dummy variables on re-election
were assigned as speciﬁed under equation (26), and according to the data from Table 1.
Table 3: Public spending and economic shocks
Lag real GDP growth
(one year before elec-
Two year average
Term limit -0.033
Revenue growth -0.097
Expenditure growth 0.556
Unemployment rate -3.61
Deﬁcit to GDP -1.276
Income tax change 1.229
Population growth -0.593
Share under 17 -10.33
Share over 65 -0.107
Observations 384 384 384 384 384 384
F test 51.3 56.35 66.78 60.48 66.58 62
R squared 0.5125 0.5276 0.6298 0.6472 0.6306 0.6484
Hausman Chi value 159.63 156.22 74.36 60.44 94.90 87.99
p-value 0 0 0 0 0 0
Notes: See notes to Table 2 for information on sample variables. For years 2001 and 2003 there was no data available
for state revenues and expenditures, making the panel unbalanced. All regressions are panel data OLS ﬁxed eﬀects
regressions that include a constant and real GDP growth as the main explanatory variable (as according to equation
26). For the Hausman test a p-value of 0 implies a rejection of the null hypothesis and a suggestion to use ﬁxed eﬀects.
Standard erros are shown in parentheses and are robust to heteroskedasticity and clustered by state. *** denotes
signiﬁcance at 1%, ** at 5% and * at 10%.
Table 4: Re-election and wasteful spending
(1) Probit (2) Probit
(3) Logit (4) Logit
per capita squared
Term limit -0.722
Revenue growth -1.927
Expenditure growth -1.302
Deﬁcit to GDP 11.94
Deﬁcit change 0.0015
Personal income -0.00001
Income change 0.235
Population change -11.01
Under 17 1.592
Over 65 -2.012
Observations 384 384 384 384 384
62.33% 62.33% 69.67% 69.67% 62.34%
Pseudo R-squared 0.2038 0.2038 0.2031 0.2031 0.1358
Log likelihood -230.784 -230.784 -230.992 -230.992
Notes: See notes to Table 2 for information on sample variables. Regressions in columns (1) and (2) are calcualted
using a random eﬀects probit, in columns (3) and (4) a random eﬀects logit, while in column (5) a standard linear
probability model. Columns (2) and (4) present the average marginal eﬀects of the probit and logit estimates. The
re-election dummy variable is the dependent variable. For the linear probability model results reported in column (5),
the regular R-squared is calculated instead of the pseudo R-square. The pseudo R-square used is the McFadden pseudo
R-square. Standard errors are shown in parentheses and are robust to heteroskedasticity and clustered by state. ***
denotes signiﬁcance at 1%, ** at 5% and * at 10%.
Table 5: Robustness checks
(1) (2) (3) (4)
per capita squared
per capita squared
Capital outlays to
Capital outlays to
Current spending to
Current spending to
total spending squared
Term limit -0.695
Controls YES YES YES YES
Observations 384 384 384 384
62.28% 62.15% 61.98% 61.99%
Pseudo R-squared 0.1877 0.1890 0.1932 0.1849
Log likelihood -235.443 -235.0678 -233.837 -236.252
Notes: A random eﬀects probit regression has been used in each case. The pseudo R-square used is the McFadden
pseudo R-square. Standard errors are shown in parentheses and are robust to heteroskedasticity and clustered by state.
*** denotes signiﬁcance at 1%, ** at 5% and * at 10%.
Table 6: Decomposing term limits and governor and legislature elections
per capita squared
Controls YES YES YES YES
Observations 272 112 200 184
Pseudo R-squared 0.1431 0.1679 0.2909 0.1206
Log likelihood -170.7 -62.13 -114.8 -109.38
Notes: States that don’t have any type of a term limit are: Conneticut, Idaho, Illinois, Iowa, Massachusetts,
Minnesota, New Hampshire, New York, North Dakota, Texas, Utah, Vermont, Washington and Wisconsin. All other
apply at least some form of a term limit rule. A random eﬀects probit regression has been used in each case, and the
probit coeﬃcients are reported instead of the usual marginal eﬀects. The pseudo R-square used is the McFadden
pseudo R-square. Standard errors are shown in parentheses and are robust to heteroskedasticity and clustered by state.
*** denotes signiﬁcance at 1%, ** at 5% and * at 10%.