Professional Documents
Culture Documents
Political Machines and The Curse of Public Resources in Subnational Democracies
Political Machines and The Curse of Public Resources in Subnational Democracies
Political Machines and The Curse of Public Resources in Subnational Democracies
https://www.emerald.com/insight/0144-3585.htm
Abstract
Purpose – This paper argues that decentralization reforms in Colombia, implemented since the 1980s,
have led to the decentralization of political clientelism rather than its demise. Clientelism is a system of
political and economic institutions that turns every local democracy into an extractive political
institution. The authors theoretically demonstrate that an increase in public resources will increase
corruption.
Design/methodology/approach – The authors develop and test a subnational public choice model, where
clientelism in elections and corruption in public administration constitute a stable long-term institutional
equilibrium. The model comprises two linked subgames: electoral tournament and corruption in public policy.
The model makes two predictions that currently oppose predominant approaches: (1) increasing the severity of
jail sentences to electoral crimes increases their price and the predominance of machine politics, instead of
improving the quality of electoral tournaments and (2) increasing local governments’ public finance increases
clientelism in elections and corruption in public administration.
Findings – The authors find evidence in favor of the theoretical model of curse of public resources, using
difference-in-differences estimation with a database 2016–17 of Colombia’s 1,034 municipalities. This country
is well-suited for our analysis because it has a long-term commitment to formal democratic processes (since
1958), while plagued by endemic corruption and clientelism problems.
Originality/value – (1) The theoretical approach is innovative and disruptive of current models on the
problem, (2) the model builds upon the Colombian situation, a country with prominent corruption and political
violence problems regardless of its relatively long-term commitment with free elections (since 1958) and (3) the
theoretical discussion is tested using a comprehensive set of difference-in-differences estimations.
Keywords Game theory, Clientelism, Corruption, Local public policy, Political machines, Curse of public
resources, Difference-in-differences
Paper type Research paper
1. Introduction
Although national democracies in Latin America have improved during the 1990 and 2000s,
not all of them are full democracies (the https://www.systemicpeace.org/polity/polity4.
htmPolity IV Project report that the only full democracies in the Americas are: Canada, United
States, Costa Rica, Chile and Uruguay.). Something similar occurs with subnational
democracies at the regional and municipal levels, of which the vast majority are not full
democracies. Moreover, in many cases, third-party movements that use clientelistic strategies
to guarantee their continuation in power have replaced traditional political parties. We
perform a journalistic and scholar literature review that characterizes the behavior of these
third-party political movements as political machines: “organizations, headed by a single Journal of Economic Studies
boss or a small autocratic group, which controls enough votes to sustain political and © Emerald Publishing Limited
0144-3585
administrative control of a city, county, or state” (Zimmer, 2015, p. 363) (See also: Stokes et al., DOI 10.1108/JES-03-2021-0148
JES 2013; and Scott, 1962). This is a context where, on the one hand, citizens do not electorally
punish political machines, and on the other, clientelism acquires a particular local expression
(Mansour et al., 2021; Robinson, 2016; Eaton and Chambers, 2014; Eaton, 2006; Eduardo and
Mu~ nloz, 2011).
Latin American countries are unitary states in which local governments have achieved
some degree of political, fiscal and administrative autonomy through decentralization
reforms. These reforms have been based on the liberal tradition, arguing that decentralization
helps to deepen and consolidate democracy by delegating power to local governments.
Consequently, decentralization reforms were made in the hope of a strengthening of the
quality of local public policies (See: Falleti, 2005; Diamond and Tsalik, 1999; Oates, 1972). This
paper argues that local governments have been strengthened but in a politically opposite
direction than expected. Decentralization reforms implemented since the 1980s have led to the
decentralization of political clientelism rather than its demise. Clientelism is a system of
political and economic institutions that turns every local democracy into an extractive
political institution (See: Eaton and Chambers, 2014, p. 90; Speck, 2003; Melendez, 2014).
This paper proposes a mechanism of clientelism and corruption, involving the transfer of
political and economic resources from the central to the local governments. In particular,
fiscal autonomy is a resource that the central government has transferred to the local
governments. Political machines know that gaining control of the local governments allows
them to gain control of central government transfers. Acemoglu and Robinson (2012) argue
that an important dimension of political institutions is the effectiveness and capacity of the
State. The authors point out that the Colombian State has been handing over state functions
to paramilitary groups and guerrillas. Such a process allows the configuration of extractive
political institutions of a criminal nature, destroying the effectiveness and capacity of the
State to promote economic development, the reduction of poverty and the supply of basic
public goods such as law and order, property rights and security. We agree with these
authors’ thesis. However, this paper proposes an even stronger thesis: each local democracy,
as an ecosystem of political institutions, is endemically extractive. The party system, the
electoral system and local governments are extractive institutions. Political parties are
political machines in charge of electoral clientelism and the local governments that corrupt
the local public administration.
We use Colombia as our case study and develop a game theory model that connects
clientelism in local elections with corruption in local public administration, based upon
narratives extracted from the academic and journalistic literature. Therefore, the
construction of the theoretical model has not arisen from a strictly deductive exercise. The
model demonstrates that an abundance of local government resources is a curse, encourages
clientelism and does not improve the provision of local public policy. Such a situation cannot
be fully understood when using neo-institutional theoretical frameworks (Acemoglu and
Robinson, 2012; Griebeler and Silva, 2020). Latin America offers counter-examples to such
logic, which in turn reinforce the logic of our model: harder sentences make more expensive
every vote, while increased public finance means a higher expected revenue and cost of
winning the elections. Both processes favor the predominance of machine politics, in
detriment of a healthy democracy (Sabioni and Harumi, 2013).
We argue that clientelism in local elections and corruption in local public administration in
the Latin American region constitutes a stable long-term institutional equilibrium. In this
sense, our curse of public resources coincides with Robinson et al. (2017) about the effect of
central government transfers. Our model is based on narratives about local governments
with a behavioral core. The analytical narratives methodology (Bates et al., 2000) uses archive
information on democratic performance at the local level and the use of network analysis to
explain clientelism. The model is based upon three well-established facts in the Colombian
case: (1) local democracies are not full democracies, (2) political parties are political machines
and (3) the median voter knows that these political machines perform as corrupt governments Political
in the municipality. The median voter is a low-income citizen living in a “Crisis of Democratic machines
Representation”, with a low marginal valuation of their vote, and who easily sell it to local
political brokers (Nieva, 2019; Robinson, 2016; Eaton and Chambers, 2014; Eaton, 2006;
Eduardo and Mu~ nloz, 2011; Acemoglu and Robinson, 2012). Our theoretical model is in this
sense similar to the regionalized presentation of Cendales et al. (2019), although ours has a
national scope and it is explicitly solved. In addition, we are performing the corresponding
empirical test.
The lack of state representation in peripheral regions determines a “Crisis of Democratic
Representation” in our model, which ultimately causes the low valuation of votes and the
predominance of machine politics in these peripheral municipalities. We use a national
classification of Colombian municipalities to perform difference-in-differences (DiD)
regressions that proves that in municipalities where third parties control the government,
the public policy performance worsens. Furthermore, this decrease is strongest in the
peripheral municipalities, evidence in favor of our model.
The paper is organized as follows: Section 2 presents the theoretical model, while Section 3
presents its empirical evaluation using DiD regression. We conclude in Section 4.
Figure 1.
Network of electoral
competition game with
clientelism
JES brokers (blue circles) and voters (green circles). This game takes place on the network
described in the right side of Figure 1. Let ci ∈ N be the social capital of a broker linked to
political machine i. That is, ci is the number of voters with whom each broker has social ties.
Assumption 1 (A.1). (1) The political machine i has private information about ci. (2) The
political machine 1 has links to brokers with higher social capital
than the ones linked to political machine 2, and hence, c1 > c2. (3)
Every broker has links with only one political machine. (4) Every
voter has links with only one broker.
The brokers are hired in a competitive process; however, once the links have been formed,
they are managed by the political machine through some degree of intimidation and
constraint. The competition for brokers is not fair and balanced, much less with symmetric
information. Consequently, the links tend to be stable and monopsonic (Ocampo, 2014). The
aforementioned is analogously fulfilled in the competition between brokers for links with
voters. The model assumes a network structure with the properties indicated, and it is not its
function to explain the process of networking.
The brokers have detailed local knowledge, which allows them to perform their activities
while effectively avoiding detection. Historiographic and fieldwork analyzes show the
importance of these brokers and their practices in clientelism (See: Brancaccio, 2011; Cutright
and Rossi, 1958; Sun et al., 2014; Acosta, 2017; El Heraldo, 2014).
Subgame of clientelism. The political machine i hires mi brokers with the “salary” si. The
expected salary of the brokers when linking to the political machine i is:
f$ðsi ðα$0 þ ð1 αÞ$SÞÞ þ ð1 fÞ$si (1)
We define the parameter f as the probability of being captured, α is the probability of not
being condemned by the judiciary and S is the monetary sanction if captured and condemned.
The brokers accept si (ρ 5 1) if the expected salary of engaging with the political machine i
is higher than Ii 5 u(w,ci), their income in legal activities using wealth w and social capital ci
such that 1 > ∇u(w, ci) > 0. The function u(w, ci) is concave in each variable separably
(Observe that if ci > ci then u(w, ci) > u(w, ci)), such that ci < u(w, ci). If the offer is not good
enough, they reject it (ρ 5 0), and the game ends. Our direct approach to the problem comes
from evidence in the narrative: 40 % of Colombians have been offered money for their vote,
47% consider themselves bribable and 78 percent say that reporting corruption exposes the
complainant to retaliation (Revista Semana, 2019). In terms of our model, a high value of α and
a low value of w. If the broker accepts si, they receive a budget equal to ci $θ from the political
machine to acquire votes, as documented by La Silla Vacia (2015). Each broker offers a bid
price β for every citizen’s vote. When si was accepted but the number of ci promised votes is
not delivered, the broker pays a sanction γ such that γ > >i þ ci $θ, where θ is a citizen’s vote
valuation.
Assumption 2 (A.2). We assume that θ 5 f(w, τ) is a common-knowledge reserve price
such that w is a voter’s wealth, and τ ∈ ½0; 1 is their degree of trust in
the democratic process.
A wealthier individual has a higher reserve price, and hence, fw(w, τ) > 0 such that f(0, τ) 5 0
(Stokes et al., 2013). The term τ is a proxy for quality of the democracy such that if τ 5 1, there
exists a full democracy, and if τ < 1, there exists an anocratic regime. If τ < 1 and decreasing,
there exists a crisis of democratic representation (in Colombia, the trust in elected officials is
low and decreasing. There is a crisis of democratic representation, which worsens because the
political system is unable to contain the clientelist and corrupt strategies of the different
political machines (Duque, 2019; and Jimenez, 2015). When τ 5 1, and there is a full
democracy, the reserve price θ tends to infinity and electoral fraud is impossible; hence, Political
fτ(w, τ) > 0 such that f(w, 0) 5 0 and f(w, 1) 5 ∞. In the fourth and last stage of the game, the machines
citizen will decide if accepting the bid price β for their vote (f 5 1), or if rejecting (f 5 0). The
bid price will be accepted if β ≥θ.
Electoral tournament. The political machines want to obtain the elected positions in order
to control the municipality’s revenues t > 0, which in the Colombian case mostly comprise
central government transfers. A total of 94% of Colombian municipalities are classified in the
highest two categories of dependence of central government transfers. These transfers, in
turn, must be spent according to Law 715 of 2001: health (24.5%), education (58.5%) and
general purpose (17%). The probability that the political machine i wins is a function of its
own expenses gi(ni) to obtain votes through electoral crimes and of the simultaneous gi(ni)
expenses made by its rival i, with ni representing the number of votes. The political
machine i will win the electoral tournament with probability payments.
gi ðni Þ
(2)
gi ðni Þ þ g−i ðn−i Þ
Let
ρ$ðsi þ ci $θ x$β μ$γÞ þ ð1 ρÞ$Ii (3)
be the payment function for a broker such that x is the number of votes that they obtain and
x ≤ ci. Suppose that the broker accepts si (ρ 5 1).
Case 1. If the number of votes delivered x is strictly inferior to ci, the broker pays the
sanction γ, and μ 5 1. In this case, their total payment is si þ ci $θ x $ β γ < 0.
Case 2. If x 5 ci, they will not pay the γ sanction (μ 5 0). In that case, total payment is
si þ ci $ (θ β) such that β ≥θ. Let
ð1 wÞ$θ þ w$β (4)
be the payment function for a citizen. If the citizen rejects the bid price β offered by the broker
(w 5 0), then their payment is θ. If they accept the price (w 5 1), then their payment is β. Let
gi ðni Þ
π i ðgi ðni Þ; g−i ðn−i ÞÞÞ ¼ t$ gi ðni Þ (5)
gi ðni Þ þ g−i ðn−i Þ
be the payment function for the political machine i. The payment π i includes information
about the brokers and their interactions with the citizens.
Lemma 1. Using backwards induction in the subgame of clientelism, the political
machine i anticipates that
s*
gi ðni Þ ¼ ni $ θ þ i (6)
ci
Every political machine i has private knowledge about ci (A.1), which means that the political
machine i can observe gi(ni), but not how many ni votes will be bought with that
expenditure. Every political machine chooses simultaneously the corresponding
expenditures, in a context of uncertainty about the result of the electoral tournament.
JES To find the equilibrium expenditure in the electoral tournament, we use the best response
method and lemma 2 is obtained.
Lemma 2. The Nash equilibrium of the electoral tournament is
t t
g *i ðni Þ; g *−i ðn−i Þ ¼ ; (8)
4 4
by each unit of public good that the contracted firms must produce, so that for each unit of
public good, whose marginal cost is equal to 1 (Assumption 1), the mayor pays a mark-up
equal to g * ðnÞ þ δ such that g * ðnÞ is the equilibrium expense established in (8).
Consequently, the public expenditure incurred by the mayor’s office is
ð1 þ g * ðnÞ þ δÞ$G ¼ G þ ðg * ðnÞ þ δÞ$G (10)
which corresponds to the total amount of money received by private firms to which public
procurement is assigned at the municipal level. We say that ðg * ðnÞ þ δÞ$G are the public
resources that the political machine diverts for clientelism and they are never executed in the
provision of public goods.
The political machine that wins the electoral tournament has preferences over the
quantities of private and public goods in the municipality. Let
U ðx; G; σ Þ ¼ x þ σ $νðGÞ þ a$ g * ðnÞ þ δ $G λ (11)
be its utility function such that x is the consumption of consumer goods, σ > 0 is the political
machine’s type that captures the intensity of the political machine from public goods, G is the
quantity of public good available to all citizens and ν($) is a continuous, concave and twice
differentiable function. The utility function of the political machine includes the net illegal
gains from corruption
a$ g * ðnÞ þ δ $G λ (12)
such that a ∈ ð0; 1Þ is the fraction of the total revenue g * ðnÞ þ δ $G used by the
political machine as private consumption, and ð1 − aÞ$ g * ðnÞ þ δ $G are resources paid to
intermediaries so that the contracts can be assigned to firms associated to the political
machine. Let λ be the expected value of the sanction for corruption, such that λ is fixed and
measured in units of the consumption good.
Transfers and fiscal revenue. The restriction of the public budget requires that transfers t
equal the public expenditure (10). These are the expenses that the central government must
cover with the transfers
t ¼ ð1 þ g * ðnÞ þ δÞ$G (13)
JES In the fiscal balance defined in (13), it is verified that the national government transfers are
greater than the clientelistic expenses if G > n. Likewise, the expression (13) shows that the
demands of local governments for a greater amount of transfers are distorted by a component
of clientelism. Moreover, given A.4, the national government does not know the magnitude of
such distortion.
The amount of resources spent on the provision of public goods comes from transfers from
the national government, and consequently, they are not assumed by the local government.
Therefore, G does not reduce the consumption of consumer goods of the political machine (See
expression (10)). If we replace the expression
x ¼ 1 g * ðnÞ þ δ $G (14)
The first set of rows of Table 1 shows information for all the municipalities, while the second
and third set of rows differentiating the municipalities controlled or not by third parties (panel
2 and 3 do not add up to the entire sample because there are municipalities that have no
majorities in the local government.). The row results show the value of the above-explained
DNP Public Policy Index. Total investment is the value of an index of total investment made
by the municipalities in different areas like public goods, education, etc. in the same period.
Investment per capita is the per capita investment of the municipality (in colombian pesos).
Finally, OG_T is an index of open government and transparency to account for access to
public information and good management of public resources.
Figure 2 is a map that portrays the peripheral location of the municipalities that depends
the most on central government transfers. The index in this map varies from 0 to 100, such
that 100 represents absolute dependence to government transfers. Based on our theoretical
model, public policy quality must be lower in the periphery, where there is a strong
dependence of central government transfers. The map shows that these heavily dependent
municipalities (identified with green on the map) are located in the Eastern, Caribbean and
Pacific coast peripheries. These regions are historically underdeveloped in relation to the
central (Andean) region, where most of the self-sufficient (identified with red and blue on the
map) municipalities are located. The peripheral municipalities are, therefore, ideal targets of
clientelist political machines, in the logic of our model. In Colombia, political reforms based
upon neo-institutional diagnostics have failed in controlling clientelism. They have just
promoted a transition from decentralized to atomized clientelism (Eaton and
Chambers, 2014).
Figure 2.
Dependence on
national government
transfers in category 5
and 6 municipalities
regressions on three different samples of municipalities, as categorized by DNP: (1) Political
municipalities that depend heavily on central government transfers (DNP categories 5 and 6, machines
around 94% of all the municipalities), (2) municipalities that can collect taxes and do not
specifically depend on government transfers (DNP categories 1 to 4) and (3) all the
municipalities. As explanatory variables, we use one period lags of total investment in the
municipality and investment per capita, which control for the resources that the municipality
owns. We also use the contemporaneous information of an index of open government and
transparency to account for access to public information and good management of public
resources. In our DiD, the treatment is to elect a third-party major. Our goal is to verify if third
parties, in municipalities of the periphery, underperform in the above-explained DNP
indicator of local public performance.
The general formulation is:
Resultsit ¼ α þ γ 1 Peripheryi þ γ 2 Thirdt þ γ 3 Peripheryi $Thirdt þ X it−1 β þ εit (16)
where: Resultsit is the indicator that measures the outcome of the municipality i in education,
health and access to public good at time t. Peripheryi, the first level, is an indicator variable
that takes the value of 1 if the municipality i belongs to the periphery and zero if it belongs to
one of the municipalities in the center. Thirdt, the second level is an indicator variable that
takes the value of 1 if in the municipality the political party in power belongs to the third
parties at time t. Xit1 is a set of control that includes total investment, investment per capita
and open government and transparency in the period t. Finally, «it is an econometric error
term that satisfies all the classical assumptions.
The coefficient of the interaction term between the two levels is our parameter of interest:
the difference in the outcome variable between the municipalities controlled or not by a third-
party, from the difference between the municipalities that belong to the periphery and the
center. That is (to simplify the notation we have excluded that in each expectation the
variable is conditional on X):
γ 3 ¼ f E ½Resultsit j i ¼ Periphery; t ¼ Third; X E ½Resultsit j i ¼ Center; t ¼ Third; X g
f E ½Resultsit j i ¼ Periphery; t ¼ NoThird; X E ½Resultsit j i ¼ Center; t ¼ NoThird; X g
This is a standard DiD set up that allows us to estimate the coefficient of interest by
controlling for different variables while easily calculating the standard errors. Our theory
suggests that γ 3 would reflect a negative relationship between the outcome variable and
being in the periphery, with third parties controlling the local government. By controlling for
covariates in Xit1, we are able to isolate the effect of the third parties on the outcome of the
municipalities at time t; assuming non-conflicting trends in the outcome variables by
municipalities, then our estimate will reflect the causal effect of being in the periphery, where
the local government is controlled by a third-party.
In causal inference studies with changes at the regional level (i.e. state, municipalities, etc.),
researchers typically conduct inference based on standard errors clustered at the regional
level to account for correlation in the policy variable (Jones et al., 2015). In this exercise, we use
robust cluster standard errors to conduct inference. In deciding the cluster, accounting for the
correlation in treatments status due to assignment is an important consideration in choosing
an appropriate level of clustering (Abadie et al., 2017).
Controlling for the clustering level is important because our theory identifies the treatment
status and determines four clusters (the treatment group are the municipalities in the
periphery with a third-party in control of the local government, whereas the control group are
the municipalities that do not have a third-party in control and can be located either in the
periphery or the center.): (1) municipalities in the periphery with third parties as majorities,
JES (2) with no third parties as majorities, (3) municipalities in the center with third parties as
majorities and (4) with no third parties as majorities. We would have problems in performing
inference if we do not control the problem of not having many clusters. Conventional methods
tend to fail when there are few clusters (Bertrand et al., 2004). Mackinnon and Webb (2017a, b)
showed that conventional cluster-adjusted standard errors cause rejection rates twice what
they should be. Thus, we use a second approach by conducting inferences that account for
clustering, using the wild bootstrap method and a six-point distribution (Webb, 2013). We
report p-values for both approaches, the conventional robust-clustering method and the wild
bootstrap in the Table 3 (the estimates report only our variable of interest, the interaction
term γ 3).
3.2 Results
Table 2 shows regression results for equation (16). Each column represents a different
specification. Columns (I) report the DiD equation for the poorest municipalities as above-
explained. Columns (II) include only the other municipality categories. Columns (III) use all
the municipalities. In addition, we verify that the results are consistent (signs and significance
do not change) when using an electoral year (2016) and its corresponding first year of
government (2017).
Note that, on average, municipalities in categories 1–4 present a higher value of the
outcome index when compared to the poorest. The left hand side of the table reports results
2016 2017
Variable (I) (II) (III) (I) (II) (III)
2016 2017
Variable (I) (II) (III) (I) (II) (III)
4. Conclusions
We have developed a model of clientelism and public policy, which we solve for the case of a
local government in a crisis of democratic representation. Our model is composed of two
subgames: a game of elections with clientelism, embedded in a game of public administration
with corruption.
In contrast to other contributions, our modeling strategy does not build upon neoclassical
behavioral assumptions, where incentives can change clientelist practices. Quite the opposite,
we build upon analytical narratives that include a review of cases from the academic and
journalistic literature and archival information. In our model, only fundamental determinants
of education and income level can change the game into one of fair and efficient elections.
In the aforementioned logic, we can notice that OG_T, the open government and
transparency variable, has a positive impact on public policy performance. Such a result goes
in line with our assertion that only fundamentals of education quantity and quality can break
the low-quality public policy equilibrium of contemporary local governments in Colombia. On
the other hand, the positive effect of long-run economic growth, which coincides with
observations made in the USA and the UK during the 19th and 20th centuries, pertains to a
JES policy option to a lesser degree. Long-run economic growth behaves as a national level,
possibly exogenous and source of social change.
There are two important neoclassical predictions that do not hold in our model: (1)
increasing municipalities’ resources does not decrease their degree of clientelism. The reason
is machine politics behavior, where once in power they must use the public resources to pay
their investors’ profit. In addition: (2) punitive measures do not diminish participation in
fraudulent electoral practices. The reason is that the expected value of the votes increases,
making machine politics the most feasible result in elections and eliminating legitimate
political parties.
We use a database of the 1,034 Colombian municipalities to test our theory using DiD
estimation during 2016–2017. The model assesses the impact of third-party local government
on municipal public policy performance. To perform the assessment, we use a national
classification of municipalities dependent upon their relative development level (reliance on
own public finance revenue). The test is built upon the idea that third-party movements have
emerged after decentralization efforts in the 1980 and 1990s, and following our analytical
narrative, their behavior is more akin to the political machines of our model. We find evidence
in favor of our theoretical model: local public administrations that suffer a crisis of democratic
representation (peripheral) worsen their performance when electing third-party mayors.
References
Abadie, A., Athey, S., Imbens, G.W. and Wooldridge, J. (2017), When Should You Adjust Standard
Errors for Clustering?, No. w24003, National Bureau of Economic Research.
Acemoglu, D. and Robinson, J.A. (2012), Why Nations Fail, Crown, Nueva York, NY.
Acosta, A. (2017), “Descentralizacion, poderes locales y clientelismo : un estudio sobre la creacion de
nuevos municipios en el departamento del Magdalena”, Tesis de Maestria en Estudios Politicos,
Universidad Javeriana, Bogota.
Bates, R.H., Greif, A., Levi, M., Rosenthal, J.-L. and Weingast, B.R. (2000), “The analytical narrative
project”, American Political Science Review, Vol. 94 No. 3, pp. 696-702.
Bertrand, M., Duflo, E. and Mullainathan, S. (2004), “How much should we trust differences-in-
differences estimates?”, Quarterly Journal of Economics, Vol. 119 No. 1, pp. 249-275.
Blume, L., Easley, D., Kleinberg, J. and Tardos, E. (2009), “Trading networks with price-setting
agents”, Games and Economic Behavior, Vol. 67, pp. 36-50, available at: https://www.
sciencedirect.com/science/article/abs/pii/S0899825608002182?via%3Dihub.
Brancaccio, L. (2011), “Among leaders and territories: the political networks of the district councillors
in Naples”, Quantity and Quality, Vol. 45, pp. 1127-1143.
Cendales, A., Garza, N. and Arroyo, S. (2019), “A model of public choice with clientelism and
corruption: introducing the analytical”, Analytical Narrative on Subnational Democracies in
Colombia, Springer, Cham, pp. 65-83.
Compra de votos en Colombia [Purchase of votes in Colombia] (2019), “Revista Semana”, Week
Magazine, available at: https://bit.ly/37CZKhzhttps://bit.ly/37CZKhz.
Cutright, P. and Rossi, P. (1958), “Grass roots politicians and the vote”, American Sociological Review,
Vol. 23 No. 2, pp. 171-179.
Diamond, L. and Tsalik, S. (1999), “Size and democracy. The case for decentralization”, in Diamond, L.
(Ed.), Developing Democracy. Toward Consolidation, The Johns Hopkins University Press,
Baltimore and London, pp. 117-160.
Duque, J. (2019), “Democracia electoral fraudulenta. La trashumancia electoral como estrategia para
ganar elecciones locales en Colombia” (Fraudulent electoral democracy. Electoral transhumance
as a strategy to win local elections in Colombia), Estudios Polıticos, Vol. 55, pp. 61-86.
Eaton, K. (2006), “The downside of decentralization: armed clientelism in Colombia”, Security Studies, Political
Vol. 15 No. 4, pp. 533-562.
machines
Eaton, K. and Chambers, C. (2014), “Teachers, mayors, and the transformation of clientelism in
Colombia”, in Brun, D.A. and Diamond, L. (Eds), Clientelism, Social Policy and the Quality od
Democracy, Johns Hopkins University Press, Baltimore.
nloz, P. (2011), “Democracy against parties? Party system deinstitutionalization in
Eduardo, D. and Mu~
Colombia”, Journal of Politics in Latin America, Vol. 3 No. 2, pp. 43-71.
El Heraldo (2014), “Anatomıa de la compra de votos en Barranquilla”, available at: https://www.
elheraldo.co/local/anatomia-de-la-compra-de-votos-en-barranquilla-136830.
Falleti, T. (2005), “A sequential theory of decentralization: Latin American cases in comparative
perspective”, American Political Science Review, Vol. 99 No. 3, pp. 327-346.
Garza, N. and Rodriguez, S. (2018), “Forced displacement and local development in colombia: spatial
econometrics analyses”, in Huang, B. (Ed.), Comprehensive Geographic Information Systems,
Elsevier, Vol. 3, pp. 42-58.
Griebeler, M.D.C. and Silva, A.R.D. (2020), “Signaling honesty: institutional strength and voters’
concern about corruption in a model of electoral competition”, Revista Brasileira de Economia,
Vol. 74 No. 3, pp. 277-304.
Jimenez, H. (2015), “Estudios sobre la distribucion polıtica de la inversion p
ublica: una revision desde
Colombia” (Studies on the political distribution of public investment: a review from Colombia),
Reflexion Polıtica, Vol. 17 No. 33, pp. 132-144.
Jones, L.E., Milligan, K.S. and Stabile, M. (2015), Child Cash Benefits and Family Expenditures:
Evidence from the National Child Benefit, No. w21101, National Bureau of Economic
Research.
La Silla Vacia (2015), “Profesion ‘Puya Ojos’”, available at: https://bit.ly/37wydOWhttps://bit.ly/
37wydOW.
La Silla Vacıa (2019), El dulce poder. Ası funciona la polıtica en Colombia (The sweet power This is how
politics works in Colombia), Editorial Aguilar, Colombia.
Mansour, S., Wallace, S., Sadiraj, V. and Hassan, M. (2021), “How do electoral and voice accountability
affect corruption? Experimental evidence from Egypt”, European Journal of Political Economy,
Vol. 68, 101994.
MacKinnon, J.G. and Webb, M.D. (2017a), “Wild bootstrap inference for wildly different cluster sizes”,
Journal of Applied Econometrics, Vol. 32 No. 2, pp. 233-254.
MacKinnon, J.G. and Webb, M.D. (2017b), “Pitfalls when estimating treatment effects using clustered
data (No. 1387)”, Queen’s Economics Department Working Pape Paper.
Melendez, C. (2014), Honduras Elecciones 2013: Compra de Votos y Democracia, National Democratic
Institute and Centro de Documentacion de Honduras, Tegucigalpa.
Nieva, R. (2019), “Corruption and paradoxes in alliances”, Economics of Governance, Vol. 20 No. 1,
pp. 41-71.
Oates, W. (1972), Fiscal Federalism, Harcourt Brace Jovanovich, New York.
Ocampo, G. (2014), Poderes regionales: etnografıas del poder y la polıtica en Cordoba (Regional powers:
ethnographies of power and politics in Cordoba), (Colombia), Cinep, Bogota.
Pani, M. (2011), “Hold your nose and vote: corruption and public decisions in a representative
democracy”, Public Choice, Vol. 148, pp. 163-196.
no, H. (2019), “Regiones para vivir: Indice sintetico de calidad de vida
Perez, C., Burbano, C. and Londo~
departamental”, Revista de Economıa del Caribe, Vol. 23, pp. 7-31.
Robinson, J. (2016), “La miseria en Colombia” (Misery in Colombia), Desarrollo y Sociedad,
Vol. 76, pp. 9-90.
JES Robinson, J., Torvik, R. and Verdier, T. (2017), “The political economy of public income volatility: with
an application to the resource curse”, Journal of Public Economics, Elsevier, Vol. 145 No. C,
pp. 243-252.
Sabioni, L. and Harumi, S. (2013), “Evid^encias do Impacto da Corrupç~ao Sobre a Efici^encia das
Polıticas de Sa
ude e Educaç~ao nos Estados Brasileiros”, Planejamento e Politicas Publicas,
Vol. 41, pp. 199-228.
Scott, J. (1962), “Corruption, machine politics and political change”, American Political Science Review,
Vol. 63 No. 4, pp. 1142-1158.
Speck, B. (2003), “A compra de votos: uma aproximaç~ao empırica”, Opini~
ao P
ublica, Vol. 9 No. 1,
pp. 148-169.
Stokes, S., Dunning, T., Nazareno, M.Y. and Brusco, V. (2013), Brokers, Voters, and Clientelism. The
Puzzle of Distributive Politics, Cambridge University Press, available at: https://www.cambridge.
org/core/books/brokers-voters-and-clientelism/2346382B38862E36C09042C779EA1510.
Sun, X., Zhu, J. and Wu, Y. (2014), “Organizational clientelism: an analysis of private entrepreneurs in
Chinese local legislatures”, Journal of East Asian Studies, Vol. 14 No. 1, pp. 1-30.
Webb, M.D. (2013), “Reworking wild bootstrap based inference for clustered errors (No. 1315)”,
Queen’s Economics Department Working Paper.
Zimmer, K. (2015), “The comparative failure of machine politics, administrative resources and fraud”,
Canadian Slavonic Papers, Vol. XLVII Nos 1-2, pp. 361-384.
Further reading
Acemoglu, D., Garcıa-Jimeno, C. and Robinson, J. (2015), “State capacity and economic develpment: a
network approach”, The American Economic Review, Vol. 105 No. 8, pp. 2364-2409.
Bonet, J. and Ayala-Garcıa, J. (2016), La brecha fiscal territorial en Colombia, Banco de la Rep
ublica-
Economıa Regional, available at: https://www.banrep.gov.co/sites/default/files/publicaciones/
archivos/dtser_235.pdf.
OECD (2019), Asymmetric Decentralization: Policy Implications in Colombia. Report Was Produced by
the OECD Centre for Entrepreneurship, SMEs, Regions and Cities (CFE) Led by Lamia Kamal-
Chaoui, Director.
Theoretical Appendix
Proof of Lemma 1.
Using backwards induction in the clientelism subgame, political machine i knows that each broker in the
second to last stage compares the payment to be obtained if accepting the salary ðρ ¼ 1Þ and delivering
her promised votes ðx ¼ ci ; μ ¼ 0Þ against the alternative when accepting but not delivering
ðx < ci ; μ ¼ 1Þ. Evaluating each case in expression (3), we have the following inequality
The political machine knows that if the punishment threat is credible, the best response is delivering the
promised ci votes. The political machine also knows that every vote will cost β such that β ≥ θ, and that
ultimately the price will be β 5 θ, with payment si to fund expenses ci $θ.
Consequently, political machine i obtains ni votes hiring mi ¼ ncii brokers, with expenses mi $ (si þ ci $ β),
and every broker will accept the salary offer si* if
f$ s*i ðα$0 þ ð1 αÞ$S Þ þ ð1 fÞ$s*i ¼ Ii (A2)
Solving, it holds that si* ¼ Ii þ f$ðα$0 þ ð1 − αÞÞ$SÞ and the political machine i hires mi brokers, with Political
salaries si* and a budget of ci $θ, while offering a salary of 0 (zero) to the other brokers in its network. The
political machine knows it will get ni votes through mi brokers, with a minimum expenditure defined in machines
the expression (8), where each broker obtains ci votes-
Proof of Lemma 2.
We have:
t gi ðni Þ
t$ 1¼0 (A4)
gi ðni Þ þ g−i ðn−i Þ ðgi ðni Þ þ g−i ðn−i ÞÞ2
for each i 5 1, 2. And the best response function for each i 5 1, 2 is:
1
BRi g *−i ðn−i Þ ¼ t$g *−i ðn−i Þ 2 g *−i ðn−i Þ (A5)
given that
vuðw; ci Þ uðw; ci Þ
<1< (A8)
vci ci
Hence, s1 > s2 and sc11 < sc22. In consequence,
s*2 s*
θþ >θþ 1 (A9)
c2 c1
We affirm that n1 > n2. The proof is by contradiction, assume that n2 ≥ n1. Hence,
s* s*
n2 $ θ þ 2 > n1 $ θ þ 1 (A10)
c2 c1
Corresponding author
Andres Arcila can be contacted at: aarcilav@uwaterloo.ca
For instructions on how to order reprints of this article, please visit our website:
www.emeraldgrouppublishing.com/licensing/reprints.htm
Or contact us for further details: permissions@emeraldinsight.com