Contents lists available at ScienceDirect
Electoral Studies
journal homepage: www.elsevier.com/locate/electstud
Natural disasters and clientelism: The case of ﬂoods and landslides in
Colombia
Jorge Gallego
Facultad de Economia, Universidad del Rosario, Colombia
A R T I C L E I N F O
Keywords:
Natural disasters
Incumbency advantage
Clientelism
A B S T R A C T
What are the eﬀects of natural disasters on electoral results? Some authors claim that catastrophes have a
negative eﬀect on the survival of leaders in a democracy because voters have a propensity to punish politicians
for not preventing or poorly handling a crisis. In contrast, this paper ﬁnds that these events might favor incumbents.
Disasters are linked to leader survival through clientelism: they generate an in-ﬂow of resources in the
form of aid, which increases money for buying votes. Analyzing the rainy season of 2010–2011 in Colombia,
considered its worst disaster in history, I use a diﬀerence-in-diﬀerences strategy to show that in the local election
incumbent parties beneﬁted from the disaster. The result is robust in regard to diﬀerent speciﬁcations and
alternative explanations. Moreover, I present evidence that goes against other common explanations found in the
literature and that goes in line with the clientelism mechanism.
1. Introduction
What are the eﬀects of natural disasters on incumbency advantage?
On the one hand, some authors purport that the occurrence of natural
disasters reduces incumbents' reelection prospects, since voters punish
their leaders even if the events are considered “acts of God” (Achen and
Bartels, 2016; Quiroz and Smith, 2013). On the other hand, in line with
other studies (Healy and Malhotra, 2009; Gasper and Reeves, 2011;
Bechtel and Hainmueller, 2011; Cole et al., 2012), this paper ﬁnds that
natural disasters might be beneﬁcial from an electoral perspective for
democratic leaders in developing countries. The mechanism proposed
in this paper, that has not been fully explored in the literature so far,
links disasters to incumbency advantage through clientelism: disasters
generate an in-ﬂow of resources in the form of donations and humanitarian
aid, which increases cash and resources available for buying
votes. Hence, disasters might increase subsequent vote shares and the
probability of reelection for those holding oﬃce, if the institutional
setting in which the disaster takes place favors patronage and vote-
buying.
Why is it that disasters might be beneﬁcial for incumbents? There
are at least two reasons why this may be true. First, if the politicians in
oﬃce are eﬃcient at handling the crisis, an adequate provision of humanitarian
assistance and relief might lead voters to reward this eﬃciency.
In other words, a catastrophe could serve as a good opportunity
for incumbents to signal their quality to voters (Ashworth, 2005;
Besley, 2007; Ashworth et al., 2018), and this eﬀect might be higher in
places where voters are more responsive and better informed (Besley
and Burgess, 2002; Lazarev et al., 2014). Second, incumbents might use
aid and relief to buy votes. After the disaster, victims are expected to
increase the propensity to sell their votes. An increased budget for
buying votes and a higher propensity to sell them will result in more
clientelism in favor of the incumbent. In developing countries, where
institutions overseeing public spending tend to be weaker, incumbents
face more favorable conditions to allocate strategically aid in order to
increase the odds of staying in power. In this paper, I present evidence
in favor of the clientelism mechanism and against alternative ex-
planations.
To support these claims, between 2010 and 2011 Colombia suﬀered
the worst rainy season of its history, while some weeks later—in
October of 2011—local elections were held. More than 3 million people
(about 8% of the Colombian population) were aﬀected by the disaster
and around 3.5 billion dollars had been allocated to ameliorate its
consequences. Naturally, some municipalities were more aﬀected than
others and some received more resources from the central government,
which created a source of variation in terms of the levels of victimization
and the distribution of disaster relief. Consequently, the empirical
strategy proposed in this paper utilizes a diﬀerence-in-diﬀerences
(DID) estimator to calculate the causal eﬀect of exposure to the
disaster on the probability of reelection of the incumbent party.
In the simplest speciﬁcation, municipalities' level of exposure to the
disaster is represented by the number of victims per capita caused by
ﬂoods and landslides. Other speciﬁcations include an exogenous
https://doi.org/10.1016/j.electstud.2018.08.001
Received 30 April 2018; Received in revised form 24 July 2018; Accepted 2 August 2018
E-mail address: jorge.gallego@urosario.edu.co.
Electoral Studies 55 (2018) 73–88
Available online 10 August 2018
0261-3794/ © 2018 Elsevier Ltd. All rights reserved.
T
measure of extreme rainfall shocks, quantiﬁed by the change in the
maximum level of precipitation in 24 h during the six months in which
the rainy season was more intense, for the 2006–2007 and 2010–2011
electoral periods. This measure of rainfall diﬀers from other precipitation
variables used in the development literature, which certainly tend
to be correlated to important economic and institutional features of
municipalities or countries. The strategy compares incumbents' probability
of reelection in the 2007 (before the disaster) and the 2011 (after
the disaster) mayoral races as a function of exposure to the disaster.
Under a common trend assumption—and compelling evidence is presented
to support it—the DID estimator provides a causal estimation of
the eﬀect of the disaster on incumbents' electoral performance.
In order to account for potential endogeneity concerns—given that
the number of victims or the amount of aid received by a municipality
are not completely exogenous to the political process or to municipality-level
unobservable variables—besides exploiting events of extreme
rainfall, this paper employs an instrumental variables approach
to identify the causal eﬀect of the disaster on electoral outcomes. Predisaster
surface runoﬀ at the municipal level, which is a measure of
water supply per capita (or the average volume of water contained by
major water bodies), is utilized as an instrument for levels of victimization.
The basic argument is that municipalities surrounded by voluminous
water bodies (major rivers, lakes, ponds, etc.) are more likely to
experience ﬂoods during days of heavy rainfall and consequently are
predicted to have a higher number of victims.
These results are robust in regard to several speciﬁcations and falsiﬁcation
tests. I conducted a placebo test that strengthened the causal
argument posed by this paper. A second rainy season took place after
the October 2011 election, which was less devastating than the ﬁrst
one, but still caused substantial damage and left in its wake a signiﬁcant
number of victims. Using the rate of victimization associated with this
post-election event, no relationship was found between this disaster and
electoral performance. The second rainy season was not helpful for
incumbents, because any aid associated with it was disbursed after the
election. This supports the argument that the ﬁrst rainy season—which
took place before the election—had an impact on electoral outcomes,
because incumbents strategically used aid to acquire votes.
If the lethality of disasters has a negative impact on leaders' survival
in democracies, and given that Colombia has a long tradition of democratic
rule, why is it then the case that exposure to ﬂoods and
landslides had a positive impact on the incumbent parties in the 2011
election? As previously stated, the basic mechanism proposed in this
paper links disasters to political survival through clientelism—the exchange
of private beneﬁts for political support usually through brokers.
Data on the Colombian disaster allows for the disaggregation of disaster
aid between relief and reconstruction, something that traditionally has
not been done within this literature (Besley and Burgess, 2002; Cole
et al., 2012; Bodet et al., 2016). Municipalities receiving higher
amounts of relief per capita—which is highly particularistic—exhibit a
higher increase in the probability of reelection for the incumbent party.
An equivalent association with reconstruction aid—which is more collective
in nature—is not found. Furthermore, once I control for vote
buying irregularities, the eﬀect of both exposure to the disaster and
humanitarian aid allocated to victims on reelection is absorbed by this
measure.
As previously mentioned, other explanations could account for the
fact that there is a positive association between exposure to the disaster
and electoral performance of the incumbent party. For instance, voters
could reward the incumbent—or at least not punish him—because he
eﬃciently handled the crisis (Cole et al., 2012; Lazarev et al., 2014;
Bodet et al., 2016). However, using diﬀerent measures of state capacity,
media penetration, literacy, and disaster response, no relationship is
found between incumbency diligence and subsequent electoral performance.
Moreover, the results also suggest that the positive eﬀect of the
disaster on incumbency advantage is not driven by changes in the dynamics
of civil conﬂict in Colombia.
An additional potential concern of the identiﬁcation strategy is that
the disaster might aﬀect agricultural production. Income shocks would
have an impact on political preferences, altering the electoral performance
of incumbents. To rule out this alternative channel, total planted
and harvested land per capita are controlled for at the municipality
level. In every speciﬁcation, the eﬀect of the disaster on the probability
of reelection remains positive and signiﬁcant and no heterogeneous
eﬀects are found among agricultural and nonagricultural places.
Therefore, even if the disaster aﬀects agricultural production, a positive
and alternative eﬀect on political survival still persists.
This paper is comprised of seven sections, including this introduction.
Section 2 reviews the literature on the political economy of natural
disasters, with particular emphasis on how these events aﬀect electoral
outcomes and, therefore, can be used strategically by political actors.
Background information pertaining to the 2010–2011 rainy season in
Colombia, as well as information on the 2011 local elections is provided
in section 3. Section 4 describes the empirical strategy employed within
this paper, founded on a diﬀerence-in-diﬀerences estimation complemented
by an instrumental variables approach. Section 5 describes
the data and its sources. Meanwhile, the main results of this paper are
reported in section 6, including the basic speciﬁcation, the instrumental
variables estimations, some tests of the mechanism, and the validity of
alternative explanations. Finally, section 7 provides the concluding
discussion for the paper, as a whole.
2. Disasters, preparedness, and clientelism
The political economy of natural disasters has focused on explaining
the variation in the lethality of these events (Anbarci et al., 2005; Cohen
and Werker, 2008; Kahn, 2005; Quiroz and Smith, 2013; Raschky and
Schwindt, 2016; Stromberg, 2007; Toya and Skidmore, 2007). In general,
these sources argue that the nature and quality of political institutions
greatly aﬀect the number of victims caused by a disaster.
While some authors focus on cross-national data that includes many
types of disasters (Kahn, 2005; Stromberg, 2007; Quiroz and Smith,
2013), others focus on speciﬁc events, such as earthquakes (Escaleras
et al., 2007; Brancati, 2007; Keefer et al., 2011), hurricanes (Abney and
Hill, 1966; Chen, 2013a; b), or ﬂoods (Mustafa, 2003; Congleton, 2006;
Eisensee and Stromberg, 2007; Fair et al., 2017).
This paper is situated within the literature on the political legacies
of natural disasters. Previous work suggests that voters punish incumbents
who do a poor job at implementing disaster relief policies. Achen
and Bartels (2016), for instance, argue that hard times tend to threaten
governments, “even in situations where objective observers can ﬁnd
little rational basis to suppose that those incumbents have had any part
in producing the voters' pain” (p. 9). However, recent studies (Gasper
and Reeves, 2011; Bechtel and Hainmueller, 2011; Cole et al., 2012;
Lazarev et al., 2014; Bodet et al., 2016; Ashworth et al., 2018) challenge
the “voter irrationality conclusion”, claiming that natural disasters—and
in general random shocks—can be informative in regards
to government preparedness. Consequently, if voters punish or reward
incumbents after a catastrophe, they do so not because they are irrational,
but because the event reveals information about the politician's
competency at handling a crisis.
In the Colombian case, as I show in this paper, not only do citizens
not punish incumbents in the aftermath of a disaster, they reward them.
A possible explanation for this phenomenon, as related in Besley and
Burgess (2002), is that governments are responsive because citizens are
informed and politically sophisticated. Using data from extreme climatic
events in India, the authors show that state governments are more
responsive to disasters in places where the mass media plays a more
important role and where electoral accountability is higher. Hence, this
explanation would argue that the disaster signals incumbents' competency
(Cole et al., 2012; Bodet et al., 2016), being the quality of the
signal a function of voter sophistication, which in turn depends on
factors such as education, political competition, and information
J. Gallego Electoral Studies 55 (2018) 73–88
74
(Besley and Burgess, 2002). However, using diﬀerent measures of voters'
levels of information and sophistication, the evidence from the
Colombian case provides little support for this explanation. The eﬀect
of relief allocation on political survival is not mediated by mass media
penetration, state capacity, literacy rates, or incumbent performance
after the disaster. On the contrary, it seems to be connected to the
strategic (i.e., clientelistic) allocation of relief among the constituency.
Moreover, an additional contribution of this paper with respect to
previous studies is that in this case it is possible to disaggregate the
electoral eﬀects of various forms of disaster aid, i.e., relief versus reconstruction
aid. In a similar fashion, Healy and Malhotra (2009) show
that there is a discrepancy between the electorate's reaction towards
relief spending and preparedness spending. Voters reward politicians'
eﬀorts to ameliorate the consequences of a disaster that already occurred,
but are relatively indiﬀerent towards eﬀorts to prepare ex ante
for these events. In both cases, the mechanism driving this result seems
to be the fact that relief spending takes the form of particularistic goods,
while preparedness spending and reconstruction aid, in each case, have
a collective nature. Consequently, this paper represents a contribution
with respect to Healy and Malhotra (2009) because it shows that even
in the case of post-disaster spending, the nature of aid matters in order
to understand how voters respond to politicians' actions.
Other studies, in the context of developed countries and especially
in the U.S., show how governments might strategically use relief
spending and humanitarian aid for electoral purposes. If retrospective
voters (Fiorina, 1981; Ferejohn, 1986) reciprocate in the polls towards
the plight of politicians who pursue actions that increase their welfare,
then aid becomes an important instrument for getting votes. In the
context of the 2004 hurricane season in Florida, Chen (2013a) and Chen
(2013b) show that awarding Federal Emergency Management Agency
(FEMA) aid to the challenger party's voters reduces their turnout, while
distributing aid to the incumbent party's supporters augments it. These
eﬀects are mediated by income and are greater a week before the
election. In a similar fashion, using a placebo test I show that preelection
aid has an impact on political outcomes, while post-election
relief has no eﬀects. Nonetheless, this paper goes even further because
it analyzes the eﬀects of diﬀerent types of aid and uses a clearer identiﬁcation
strategy. In any case, all of the evidence supports the claim
that aid may be strategically allocated by incumbents to increase the
odds of winning elections. The diﬀerence is that in developing countries,
like Colombia, patronage and clientelism may exacerbate these
incentives.
Other theories suggest that political institutions mediate the electoral
consequences of natural disasters. Quiroz and Smith (2013) frame
the analysis in terms of the selectorate theory (Bueno de Mesquita et al.,
2003). According to these authors, the occurrence and lethality of
disasters have divergent eﬀects on the political survival of autocratic
governments on the one hand, and democratic leaders on the other.
This is due to the divergent sizes of the coalitions that hold politicians
accountable. The authors ﬁnd that in large coalition systems such as
democracies, where mass support is required to retain oﬃce, the lethality
of a disaster, and not its occurrence per se, has a direct negative
eﬀect on political survival. In autocratic regimes, the situation is different,
as the occurrence of disasters serves as a coordination device for
protests that can overthrow the leader, while lethality of disasters has
no signiﬁcant eﬀects.
However, the situation quite diﬀerent in developing countries like
Colombia. Being a democracy, the winning coalition is certainly bigger
than in an autocratic regime. Nonetheless, and in contrast to developed
countries, the clientelist nature of political systems in most developing
countries makes the size of winning coalitions smaller, especially at the
regional and local levels. Clientelism erodes democratic institutions, as
it makes politicians less accountable (Stokes, 2005), undermines the
quality of public servants (Gallego et al., 2018), increases excessively
the size of the public sector (Robinson and Verdier, 2013), and results
in the under-provision of public goods (Keefer and Vlaciu, 2008).
Hence, in the context of a catastrophe, the delivery of particularistic
goods to strategic agents (e.g. the allocation of relief spending to inﬂuential
local intermediaries) attenuates the pervasive electoral eﬀects
of disaster lethality, even making them advantageous for democratic
leaders. In other words, when disasters hit broker-mediated democracies
(Stokes et al., 2013; Gallego and Wantchekon, 2018), relief may
represent a positive shock on the resources incumbents need to discipline
intermediaries in charge of delivering votes. In turn, brokers are
able of cultivating supporters for their candidates with these additional
resources. This behavior is feasible because anti-corruption and electoral
agencies, and institutions overseeing public spending, tend to be
weaker.
3. Background
3.1. Colombia's 2010–2011 rainy season
President Juan Manuel Santos described the 2010–2011 rainy
season as the worst natural disaster in the history of Colombia.1
Of
particular importance for understanding Colombia's current weather is
the ocean-atmosphere phenomenon known as La Niña, which forms
part of the broader El Niño Southern Oscillation (ENSO) pattern. La
Niña corresponds to a drop in the sea surface temperature across the
equatorial Eastern Central Paciﬁc Ocean along the coasts of Peru,
Ecuador, and Colombia. A signiﬁcant increase in rainfall takes place,
particularly in the Paciﬁc, Caribbean, and Andean regions of the
country, as a result of the combined decrease in temperatures and wind
patterns.
Even though this phenomenon occurs on a yearly basis, the event
that started in the second quarter of 2010 and persisted until at least
April of 2011 is considered the most intense in history. During this
period, rainfall registered at unprecedented levels with values that
presented 170 percent above average norms (Sanchez, 2011). Extreme
rainfall did not plague a particular region of the country, but struck
with varying intensity throughout 28 of the 32 departamentos and 93
percent of the municipalities.
An increase in the river levels was a direct consequence of the heavy
rain. Colombia's hydrology is one of the most abundant and complex
around the world, with almost every municipality having access to a
signiﬁcant body of water. As a result, ﬂoods were a common event,
causing damage and loss throughout the aﬀected population. In addition,
given the complex topography of the nation and the fact that an
important proportion of its habitants are located in the Andean region,
landslides caused by the accumulation of water and sediment in the
ground were the other major source of devastation.
According to oﬃcial records,2
the 2010–2011 La Niña rainy season
phenomenon aﬀected about 4 million citizens (more than 8 percent of
the population), with 490 human deaths, 595 people injured, 12,908
houses destroyed, and 441,579 aﬀected, as well as 1,080,000 ha of
productive land ﬂooded (Sanchez, 2011). The Colombian disaster was
not very lethal in terms of human loss, but it had an enormous social
and economic impact when measured by the number of people aﬀected
and the total property destroyed. Fig. 1 depicts the spatial distribution
of this Colombian disaster. In terms of the number of people aﬀected,
this event was particularly intense in the Paciﬁc, Caribbean, and Andean
regions of the country, which are also the most populous.
Compared to other natural disasters around the world, this type of
impact is common to events that result from heavy rain and ﬂoods. For
instance, using data from the Emergency Events Database (EM-DAT),
Stromberg (2007) documents that from 1980 to 2004, a total of 621
1
See http://www.semana.com/nacion/peor-tragedia-natural-historia-del-
pais/155398-3.aspx.
2
See http://www.regiones.gov.co/FenomenoNina/Fenomeno-Nina_110714.
pdf.
J. Gallego Electoral Studies 55 (2018) 73–88
75
earthquakes occurred worldwide, causing approximately 215 million
deaths and aﬀecting 78 million people. In contrast, 2102 ﬂoods took
place during the same period, causing 171 million deaths but aﬀecting
2490 million people. Hence, while other disasters are more lethal in
terms of fatalities, ﬂoods like the ones that took place in Colombia have
higher levels of aﬀectation.
Fig. 1. Spatial distribution of exposure to the disaster (No. Of victims).
J. Gallego Electoral Studies 55 (2018) 73–88
76
The signiﬁcant impact of La Niña led President Santos to create
“Colombia Humanitaria”, a National-Level Agency in charge of raising
aid and distributing it to the aﬀected communities. For this purpose, the
government implemented a three-stage strategy that included humanitarian
aid, reconstruction, and rehabilitation. More than U.S. $3.5
billion were raised and subsequently distributed from the central government
to local authorities at both the departmental and municipal
levels. To a great extent, mayors were in charge of allocating relief, and
in the case of reconstruction aid, 47% of aid resources was allocated to
mayors. Needless to say, mayors had a lot of discretion over these resources.
In fact, President Santos signed a decree that was later approved
by the Constitutional Court, that authorized mayors to leap-frog
certain steps in the execution of these contracts in order to fasten the
procurement process (Garbiras, 2012). Naturally, these simpliﬁcations
to legal requirements expedited the processes, but in some cases fostered
malfeasance and clientelism.
3.2. The 2011 local elections
This season of above-average rainfall and the associated number of
ﬂoods and landslides continued through the middle of 2011. Later that
year, on October 30th, local and regional elections took place with
more than 130,000 citizens running for governor, mayor, or local
councils. Elections for these oﬃces took place in the country's 32 departamentos
in over 1100 municipalities. Until 1988, mayors in
Colombia used to be centrally appointed, but now they are elected by
universal suﬀrage. Since then, local governments have been given increased
responsibilities and powers to raise local revenues. In this
sense, the mayor is a central ﬁgure in advancing the interests of the
municipality. Being a multiparty system, the immediate reelection of
mayors (the main focus of this paper) is prohibited in Colombia. For this
reason, reelection of the incumbent party is used as a measure of political
survival.
Even if mayors cannot be immediately reelected in Colombia, in
2007, and especially in 2011, they had strong incentives to support
their oﬃcial party. First, in 2003 a political reform was approved to
deter the fragmentation of candidacies into a plethora of parties and
political movements, as it used to be the case.3
Hence, compared to
previous years, in 2007 and 2011 the number of party options for
candidates reduced signiﬁcantly (Arevalo et al., 2013). More importantly,
before the 2011 local election, another reform was approved
to punish party indiscipline. Among other things, the reform punishes
members of one party—especially previously elected oﬃcials—that
support candidates from other parties. Hence, by law, former mayors
cannot publicly support candidates from other parties, unless they ofﬁcially
defect, a decision that is both rare and costly. These reforms
have been eﬀective: in 2002, 28% of the elected congressmen had
switched parties. This number increased to 51% in 2006 but decreased
to 10% in 2010 and 2% in 2014. While equivalent data for mayors is
hard to construct, given that these laws are valid for all corporations, it
is reasonable to believe that the trend is similar in those elections.
Hence, one would expect higher levels of party discipline during the
2011 post-disaster election.
Finally, qualitative and journalistic evidence supports the claim that
mayors tend to select and support a successor, who usually belongs
either to the same party or to the government (Lopez, 2010). As a
consequence, mayors tend to be aﬃliated to political and economic
groups that usually persist in power. UNDP (2011) calculated the 2007
political persistence index—a measure of the extent to which the same
group governed continuously in a given place—for a sample of Colombian
municipalities. The report found high levels of persistence in a
large number of municipalities. Hence, being part of these groups that
want to stay in power, it is common for mayors to inﬂuence the result of
the elections in favor of their parties.4
After all, non-consecutive reelection
of mayors is allowed in this country.
Election day was relatively calm, although illegal armed groups
(comprised by guerrillas and paramilitaries) tried to interfere and alter
the normal conduct of elections (Gallego, 2018). Given the overlap
between the political campaign associated with this election and the
humanitarian eﬀorts aimed at ameliorating the impact of the disaster,
several institutions monitoring the process warned about the possibility
of candidates using aid strategically for electoral purposes. On March of
2011, more than seven months before the election, the Electoral Observation
Mission—a traditional NGO that monitors democracy in Colombia—launched
an oﬃcial document in which it warned about the
risk of mayors using relief to buy votes.5
Similarly, the Oﬃce of the
Comptroller General, which is the institution in charge of ﬁscal control
in the country, on May of that same year warned about the same risks
and announced its strategy against these crimes.6
Despite these calls
and actions, mayors and diﬀerent party members were not completely
deterred from allocating resources strategically. After the election, in
more than 15% of Colombian municipalities at least one public servant—in
most cases the mayor—ended up being investigated by the
Oﬃce of the Inspector General, the institution that oversees the public
conduct of these oﬃcials, for “irregularities” associated to the allocation
of relief7
.
4. Empirical strategy
To determine the causal eﬀect of the natural disaster on party survival
in Colombia, this paper emphasizes the fact that the 2010–2011
rainy season was not concentrated on a particular region, but instead
exhibited spatial variation across the country. Therefore, it is possible
to claim that the intensity of rainfall and, to some extent, the occurrence
of disasters, such as ﬂoods and landslides, are exogenous to the
political process and to electoral results. Diﬀerent approaches will be
used in order to measure a municipality's level of exposure to the disaster.
In the two basic speciﬁcations used in this study, the changes in
the levels of abnormal rainfall during the rainy season and the rate of
victims per capita are used.
In this paper, a diﬀerence-in-diﬀerences estimator is utilized, in
order to compare party survival in municipalities heavily and weakly
aﬀected by the disaster, before and after it occurred. For this purpose,
this research controls for municipality and year ﬁxed eﬀects in a linear
regression. Municipality ﬁxed eﬀects control for any speciﬁc characteristics
that do not change over time and that might aﬀect the
probability of reelection or the vote share of an incumbent party, such
as historical conditions or geographic characteristics of the municipalities.
Time eﬀects control for speciﬁc events that occurred in a
particular year and that equally aﬀect every municipality, such as any
political reform that might have changed the rules of the game for a
particular election or certain economic characteristics that caused an
impact across the country. In this context, the basic speciﬁcations estimated
are:
3
The reform established that parties need to obtain at least a 2% vote share
in Congress elections in order to be able to run in future elections. Through
another reform, approved in 2009, this threshold increased to 3%.
4
Should mayors support candidates from other parties, measurement error
would ﬂaw our dependent variable. If this form of unloyalty of mayors is orthogonal
to the predictors of party reelection, estimates would not be biased. If
it is not, the identiﬁcation strategy (in particular the IV approach) should solve
any biasedness.
5
See http://www.semana.com/enfoque/emergencia-invernal-riesgo-
electoral/153227-3.
6
See https://www.dinero.com/economia/articulo/contraloria-advierte-
posible-desviacion-recursos-para-ola-invernal/119887.
7
See https://www.procuraduria.gov.co/portal/media/ﬁle/070911procesos.
pdf.
J. Gallego Electoral Studies 55 (2018) 73–88
77
= + + × + +Reelection α β Victims Post δ ϕ εX( 2010 )it i t i t it it (1)
= + + × + +Reelection α β Rainfall Post γ ϕ εX( 2010 )it i t i t it it (2)
where Reelectionit is a dummy variable8
indicating if the incumbent
party won the mayoral election in municipality i at election year t, for
=t 2007,2011. Furthermore, αi controls for municipality-level ﬁxed effects
while βt for time eﬀects. Xit is a vector representing time-varying
municipality level controls. In this case and to avoid potential endogeneity
caused by the inclusion of post-treatment covariates, I include
pre-disaster ﬁxed values of diﬀerent controls, and interact them
with a post-disaster time dummy (Post2010t). This vector includes demographic
controls (population and population density), a climatological
control (temperature), and a socio-economic control (poverty).
Finally, εit represents the error term for municipality i at election-year t.
The main variables of interest in (1) and (2) are the interaction
terms ×Victims Post2010i t and ×Rainfall Post2010i t. Here, Victimsi is the
number of victims per capita in municipality i, as a consequence of the
2010–2011 rainy season in Colombia. Post2010t is a dummy variable,
which indicates if the event (election) analyzed took place before or
after the disaster. Given that we only have two years, Post2010t also
represents the year eﬀects. Therefore, the coeﬃcient of interest in (1),
δ, measures the eﬀect of the disaster on incumbent party reelection in
2011, as compared to the 2007 contest. Note that the constituent terms
of the interaction, Victimsi and Post2010t, are absorbed by the ﬁxed and
time eﬀects respectively, and do not appear separately in (1). In all
speciﬁcations, standard errors are clustered at the municipality level, to
control for potential serial correlation at such level. Section A7 in the
Online Appendix shows that the main results of this paper are robust to
clustering at the departamento level.9
It can be claimed that Victimsi is not exogenous to municipality-level
observable or unobservable characteristics. For instance, it is likely that
in poor and unequal municipalities, more families will live in zones
prone to ﬂoods and landslides, such as riverbanks or hillsides.
Therefore, municipalities highly aﬀected by the disaster would diﬀer
from weakly aﬀected ones on other (possibly) non-observable characteristics,
making the identiﬁcation of causal eﬀects more diﬃcult.
This fact would be troublesome if such characteristics exert a diﬀerential
impact after 2010. An example of this would be if poorer places
voted diﬀerently in 2011. In order to account for this potential endogeneity
concerns, I employed a primary identiﬁcation strategy to
utilize a plausibly more exogenous measure of municipality exposure
from the disaster. Landslides and ﬂoods are more likely to occur when
extreme climatological events take place in short periods of time, such
as episodes of heavy rain.
For this reason, the second basic speciﬁcation is the one described
by (2). The diﬀerence between (1) and (2) is that in the second speciﬁcation,
the exposure to the disaster is captured by the variable
Rain falli, which quantiﬁes events of extreme rainfall at the municipal
level. This variable measures, originally in millimeters, the change in
the maximum level of rainfall in 24 h experienced by municipality i
during the six months between October of 2010 and March of 2011,
compared to the same period for years 2006 and 2007.10
The rationale
behind this measure, similar in nature to the one employed by Burgess
et al. (2014) in their analysis of the eﬀects of abnormally hot days in
India, is that in those places in which it rained more during a whole
day, the probability of having subsequent ﬂoods and landslides (and
hence more victims) is also higher. In other words, Rain falli measures,
for each municipality, how much more it rained the day that it rained
the most during the 2010–2011 rainy season, compared to the
2006–2007 period.
The claim in this case is that a variable like Rain falli is much more
exogenous than the rate of victims caused by the disaster and than traditional
measures of precipitation used in other studies (Benson and Clay,
1998; Miguel et al., 2004; Barrios et al., 2010; Bruckner and Ciccone,
2011). In essence, Rain falli captures outliers, as it ﬁlters for days in
which precipitation was extreme, causing catastrophes such as ﬂoods or
landslides. The literature on meteorology suggests that these types of
events are more likely when episodes of abnormal rainfall take place
during short periods of time rather than being the result of continued
days of heavy rainfall. Hence, it is less likely that this measure is correlated
to economic or social variables, which usually is the case for other
measures of rainfall, such as monthly deviations from historical levels.
Nonetheless, the rate of victims caused by this disaster remains to be
an endogenous variable. Therefore, this paper employs an instrumental
variables (IV) approach, using an exogenous source of variation correlated
with the number of victims per capita and uncorrelated with the
error term. For this purpose, pre-disaster water supply per capita at the
municipal level is used to predict the number of victims, as measured by
hydrologists for the 1998's National Water Study (IDEAM, 1998), more
than a decade before the catastrophe. Hence, our instrument is by no
means aﬀected by the 2010–2011 rainfall levels. This variable measures
surface runoﬀ, which is an estimation of the volume of water contained
in nonoceanic water bodies (rivers, lakes, ponds, etc.) and groundwater.
Naturally, municipalities with bigger rivers or lakes tend to have a
higher supply.
Water supply (interacted with the time dummy) is a valid instrument
for victims of ﬂoods and landslides (also interacted) if it is not
weak and if the exclusion restriction is satisﬁed. As shown below, it is
not weak being that an increase in supply predicts an increased likelihood
of ﬂoods and, hence, more victims. In terms of the correlation
with agricultural activities, it is not expected to be as high as in the case
of rainfall, because this variable simply measures the volume of water
contained in rivers. An interesting feature of Colombia's geography is
that almost every municipality has a river; therefore, this variable is not
an indicator of whether a population has access to a body of water as a
transportation or economic source. It simply measures the volume of
water surrounding a place, which in part is the result of topographic
characteristics, such as the size of the hydrographic basins that, perhaps,
are exogenous to covariates aﬀecting any political outcome. A full
discussion on the validity of the exclusion restriction, and a test based
on observable characteristics, are provided in section A9 of the Online
Appendix.
Deﬁning Wateri as the average level of water supply per capita in
municipality i before the disaster, the ﬁrst stage of the IV approach
estimates the following:
× = + + × + +Victims Post α β Water Post η ϕ ωX2010 ( 2010 )i i t i t it it (3)
While the second stage is given by:
= + + × + +Reelection α β Victims Post ρ ϕ εX( ˆ 2010 )it i t i t it it (4)
where ×Victims Postˆ 2010i t are the predicted values that result from
equation (3). Naturally, η is expected to be positive, as ﬂoods are more
likely to occur in places with more water surrounding. Additionally, in
order to test the mechanism—that disasters generate an inﬂow of resources
to local authorities that is used to buy votes and consequently
augments the probability of reelection—the following speciﬁcation is
estimated:
= + + × +
× + +
Reelection α β Relief Post δ
Reconstruction Post δ ϕ εX
( 2010 )
( 2010 )
it i t i t
i t it it
1
2 (5)
8
See section A1 in the Online Appendix for an explanation of the coding
scheme of this variable.
9
As posed by Dube and Vargas (2013), this test is quite stringent given that
the cross-sectional variation in the key explanatory variables is at the municipality
level, which are grouped into 32 departamentos.
10
These months, in both periods, where characterized by high levels of
precipitation. In fact, the 2006–2007 also caused victims, although the intensity
was much lower. In fact, the 2010–2011 event was so intense, that for the ﬁrst
time an agency like Colombia Humanitaria was created to handle the crisis.
J. Gallego Electoral Studies 55 (2018) 73–88
78
where Reliefi and Reconstructioni are measures of the amount of relief
(mainly food) and reconstruction aid per capita received by municipality
i after the 2010–2011 rainy season and before the election. In this
way, δ1 and δ2 test if these two versions of disaster aid have diﬀerential
eﬀects on the probability of reelecting the incumbent party. Subsequent
estimations and alternative speciﬁcations include measured irregularities
in the allocation of aid, media penetration, state capacity, literacy
rates, agricultural production, and civil conﬂict, yielding evidence in
favor of the clientelism mechanism and against alternative explanations,
such as government responsiveness, civil conﬂict dynamics, or
electoral consequences of income shocks. Additionally, a placebo test is
conducted, showing that disaster aid disbursed after the election has no
eﬀect whatsoever on the probability of reelecting the incumbent party.
Summing up, this section describes the basic identiﬁcation
strategy used in this paper, which employs a diﬀerence-in-diﬀerences
estimation of the eﬀect of the disaster on party survival. An instrumental
variables approach complements this strategy and diﬀerent
estimations are used to test the robustness of the results and the validity
of the mechanism.
5. Data
5.1. Data sources
Victim-related data used in this study comes from the National
Administrative Department of Statistics (DANE). A few weeks after the
disaster took place, and for the purpose of quantifying its eﬀects, DANE
developed the Unique Record of Victims. According to this record,
victims are “people that have suﬀered serious damage associated directly
to the event: partial or total loss of goods (estate, livestock or
crops) and/or disappearance, injury or death of family members”
(DANE, 2011). In order to control for municipality size, the total
number of victims is divided by the total population. Therefore, the
variable Victimsi used below corresponds to the number of victims per
capita in municipality i as a consequence of the 2010–2011 wet season
in Colombia.
Rainfall data comes from the Hydrology, Meteorology, and
Environmental Studies Institution (IDEAM), which is the oﬃcial
Colombian government environmental agency. This institution registers
rainfall on a daily basis, using more than 2500 pluviometric
stations located across the country. Given that in Colombia there are
about 1200 municipalities, and for some there is more than one station,
while for others there are none, this study geo-referenced both
stations and municipalities, calculated the distance between each,11
and determined the nearest one. As it was described above, the
variable Rainfalli is a measure of extreme climatological events and
corresponds to the change in the maximum level of rainfall in 24 h,
measured in millimeters, registered for municipality i from October
through March, for the periods corresponding to both elections.12
Water supply, which is an estimate of surface runoﬀ of major water
bodies and groundwater of Colombian municipalities in 1998, was
measured by IDEAM and is reported in the National Water Study
(IDEAM, 1998). Measured in millions of cubic meters, this variable
represents the average volume of water available for Colombian
municipalities during a regular year.13
Another disaster-related variable includes the total aid allocated to
aﬀected municipalities. This information comes from Colombia
Humanitaria, a national agency created by the central government after
the crisis for the purpose of facilitating assistance to victims.14
Two
types of humanitarian aid are analyzed in this paper: relief (measured
as the number of grocery and hygiene kits per capita allocated to affected
municipalities during or weeks after the disaster) and reconstruction
aid (measured as the amount of money per capita in Colombian
pesos allocated to municipalities to construct or rebuild
infrastructure assets destroyed by the disaster). Information on the
timing of reconstruction projects comes from Garbiras (2012). To
measure the eﬃcacy of local governments in facilitating these projects,
a dummy variable called Delays is employed, which is based on information
provided by the Oﬃce of the Inspector General of Co-
lombia.15
The report lists municipalities that were investigated in
September of 2011 (one month before the election), because of delays
and other irregularities in the execution of these projects. Additional
information on vote-buying allegations for the 2007 and 2011 local
elections was obtained from the Oﬃce of the Attorney General of Co-
lombia.16
Municipality-level covariates include pre-disaster values for demographic,
climatic, and socioeconomic controls. Demographic controls
include levels of population and population density in 2007, as measured
by DANE. The climatic covariate is the average temperature in
Celsius and is also provided by DANE. Poverty is used to control for
socioeconomic characteristics, measured as the proportion of citizens
with unsatisﬁed basic needs.17
This index was compiled by the National
Planning Department (DNP). Electoral variables for the 2007 and 2011
local elections are derived from the National Registry, the oﬃcial Colombian
government electoral agency. Information on radio stations
comes from the Ministry of Information Technologies and Communication,
newspaper data from Camacho and Conover (2011), and
measures of state capacity and literacy rates come from the DNP. Finally,
information on agricultural production comes from the Ministry
of Agriculture.
6. Results
6.1. Basic speciﬁcation: diﬀerence-in-diﬀerences approach
I begin by presenting a couple of graphs that summarize the basic
empirical strategy used in this paper. The main idea is to compare the
proportion of municipalities in which the incumbent party got reelected
in 2007 and 2011, in places more and less “aﬀected” by the 2010–2011
disaster. In Fig. 2, municipalities aﬀected by the disaster are those
whose number of victims per capita is above the 50th percentile.18
The
graph shows that in 2007, before the disaster occurred, a similar proportion
of municipalities in both groups again saw the incumbent party
win the mayoral election. However, the situation is quite diﬀerent in
the months after the disaster took place. In 2007, there was a sharp
increase in the proportion of municipalities heavily aﬀected by the
disaster where the incumbent party won, whereas this proportion remains
stable (decreases a little bit) for municipalities not aﬀected by the
catastrophe. Similar results were found when comparing places exposed
to higher changes in levels of extreme rainfall (above the 50th percentile)
with places less exposed to this situation, as shown in Fig. 3.
The rationale behind this analysis is that if the disaster had not
occurred, reelection patterns in municipalities aﬀected and not aﬀected
would have followed a common trend and, consequently, the diﬀerence
11
For this purpose, Stata's Vincenty package was used (Nichols, 2003).
12
This particular range is chosen because of it corresponds to the most intense
period. Results are robust to other rainfall speciﬁcations.
13
Regular years are deﬁned as those in which surface runoﬀ corresponds to
the average multi-year value of historical series of representative ﬂows (IDEAM,
1998).
14
See http://www.colombiahumanitaria.gov.co/Paginas/
QueesColombiaHumanitaria.aspx.
15
Procuraduría General de la Nación.
16
Fiscalía General de la Nación.
17
In Spanish the index capturing this number is called NBI, for Índice de
Necesidades Basicas Insatisfechas.
18
The shape of the ﬁgure and the intuition holds for lower and higher
thresholds.
J. Gallego Electoral Studies 55 (2018) 73–88
79
in the actual trajectories is the causal eﬀect of the disaster that wants to
be determined. Figs. 2 and 3 suggest that the disaster caused an increase
in the probability of reelection in aﬀected municipalities compared to
those not aﬀected in 2011, relative to 2007.
In fact, Table A1 in the Online Appendix yields evidence in support
for the common trend assumption. In this case, I use an extended period
of analysis, by including electoral years 2003, 2007, and 2011. In this
way, two pretreatment periods are included, making feasible the conventional
test, in which the treatment variables—the number of victims,
the change in extreme rainfall, and surface runoﬀ in the reduced
form—interact with the year dummies. Note that in these estimations
the dummy variable for the year 2003 is left as the benchmark.
Table A1 shows that before the disaster, in 2007, there was no
signiﬁcant change in the eﬀect of any of the three treatment variables
on the probability of reelecting the incumbent party. In contrast, the
coeﬃcient for the interaction between treatment and the 2011 year
dummy is positive and signiﬁcant in the majority of the models. In the
analysis that follows, the sample is restricted to the 2007 and 2011
timeframe, given that from 2003 backward, the political system in
Colombia was extremely fragmented, exhibiting a plethora of parties.
Because of this, tracking reelection is, at best, very noisy.19
Nonetheless,
robustness tests to prove that the main results of the paper hold
for the extended period of analysis are reported in Appendix section A8.
Table 1, which presents estimations for diﬀerent speciﬁcations
based on models (1) and (2), corroborates the intuition suggested by
Figs. 2 and 3. Every speciﬁcation includes municipality-level ﬁxed effects,
time eﬀects, and standard errors clustered at the municipal level.
Recall that the coeﬃcients of interest are δ and γ, the parameters
Fig. 2. Victims and reelection.
Fig. 3. Extreme rainfall and reelection.
19
For this same reason, the main outcome of interest is used in each period
as a level, rather than as a change.
J. Gallego Electoral Studies 55 (2018) 73–88
80
associated with ×Victims Post2010i and ×Rain fall Post2010i . As columns
1–8 reveal, the positive and signiﬁcant coeﬃcients of the different
speciﬁcations suggest that in 2011 the increase in the probability
of reelection is higher in municipalities more aﬀected by the disaster. In
other words, an increase in the rate of victims, or in the change in the
level of extreme rainfall, signiﬁcantly increases the probability of reelection
in 2011, relative to 2007. Columns 1 and 5 report the estimates
when municipality-level controls are not used; columns 2 and 6 show
that these results are robust to the inclusion of demographic controls,
such as population and population density; columns 3 and 7 include
temperature; and columns 4 and 8 incorporate poverty. In every case,
the result is the same: a positive and signiﬁcant coeﬃcient for the interaction
between the rate of victims per capita, or the measure of extreme
rainfall, and the post-disaster time dummy.20
The estimates imply substantial eﬀects. For instance, the coeﬃcient
in column 1 suggests that a one percentage point increase in the rate of
victims per capita increases the probability of reelection by approximately
0.23% in 2011. Thus, compared to a situation with no victims, a
municipality in which the whole population was aﬀected experiences
an increase in the probability of reelecting the incumbent party by 23
percentage points higher in 2011. Similarly, the coeﬃcient in column 5
suggests that an increase in extreme rainfall of one cubic meter translates
into an increase in the probability of reelection by 0.8 percentage
points higher.21
Endogeneity could bias the results presented above if our main
variable of interest, ×Victims Post2010, is correlated with error term εit
in (1). This situation may occur, for instance, if municipalities with
higher levels of income inequality have more inhabitants living in zones
prone to ﬂoods and landslides, such as riverbanks or hillsides. If inequality
also predicts whether an incumbent party is reelected, estimates
based on (1) would be biased. Another cause of endogeneity and
biasedness may be measurement error. For instance, it could be the case
that municipalities governed by corrupt politicians tend to over-report
the number of victims just for the sake of obtaining more aid. If there is
a correlation between this source of measurement error (more victims
registered than real victims in “corrupt municipalities”) and the likelihood
of reelection, the estimates based on (1) would be biased.
6.2. Instrumental variables approach: surface runoﬀ and victims
To account for potential endogeneity bias, in this section I present
an alternative approach: the use of a pre-disaster hydrotopographic
variable as an instrument for the levels of victimization associated with
the disaster. Colombia is a great power in terms of water supply.
Besides being the only South American country with access to both the
Atlantic and Paciﬁc oceans, its territory contains several rivers and
lakes that supply water to the population. In fact, it is diﬃcult to ﬁnd
municipalities that do not have access to water bodies. Naturally, accessibility
to this important natural resource is not only a function of
rainfall. It also depends on topographic characteristics, which determine
the size of river basins and soil accumulation, which in turn
determine the levels of water supply available for each municipality.
Given that most victims analyzed in this paper struggled with ﬂoods
that destroyed their homes, crops, land and other properties, and that
these events were not only a function of rainfall but also the existence
and abundance of surrounding water bodies, surface runoﬀ per capita
will be used as an instrument for victimization.
Naturally, for the 1998 level of surface runoﬀ to be a valid instrument,
the only channel through which the interaction of runoﬀ and the
post-disaster time dummy aﬀect reelection results is through their eﬀect
on the interaction of victims and its corresponding time dummy. It may
be claimed that runoﬀ predicts in which places people choose to live. It
is for this reason that I use per capita levels in this analysis; in fact, the
correlation between runoﬀ per capita and population density is weak
and statistically insigniﬁcant (results not reported). It can also be
claimed that surface runoﬀ aﬀects economic activities, as is the case
with rainfall. Nonetheless, given the nature of the instrument described
in (3), for this concern to be valid it would be necessary to explain why
is there a diﬀerential eﬀect of runoﬀ on these activities in 2011 as
Table 1
Victims, extreme rainfall, and reelection.
(1) (2) (3) (4) (5) (6) (7) (8)
Reelection Reelection Reelection Reelection Reelection Reelection Reelection Reelection
Victims × Post2010 0.225∗∗∗
0.212∗∗∗
0.163∗
0.151∗
(0.0767) (0.0770) (0.0840) (0.0879)
Rainfall × Post2010 0.806∗∗
0.833∗∗
0.783∗∗
0.744∗∗
(0.347) (0.348) (0.346) (0.345)
Constant 0.138∗∗∗
0.145∗∗∗
0.0608 0.0620 0.151∗∗∗
0.157∗∗∗
0.0785 0.0606
(0.0136) (0.0148) (0.0674) (0.0716) (0.00927) (0.0102) (0.0678) (0.0714)
Pop. and Density N Y Y Y N Y Y Y
Temperature N N Y Y N N Y Y
Poverty N N N Y N N N Y
Fixed Eﬀects Y Y Y Y Y Y Y Y
Time Eﬀects Y Y Y Y Y Y Y Y
N 2200 2196 2144 2134 2113 2111 2111 2102
Notes: Standard errors clustered at the municipality level are shown in parentheses. Election years in the sample are 2007 and 2011. Reelection is a dummy variable
that equals 1 if the incumbent party wins the election. Victims is the number of victims per capita. Rain fall is the change in the highest level of precipitation in 24 h,
measured in cubic meters, from 2006 to 2007 to 2010–2011. Post2010 equals 1 for 2011 observations and 0 otherwise. OLS Fixed Eﬀects estimation is used in every
speciﬁcation. * is signiﬁcant at the 10% level, ** is signiﬁcant at the 5% level, *** is signiﬁcant at the 1% level.
20
Note that all these models use OLS ﬁxed eﬀects estimations, despite having
a dichotomous dependent variable. This approach is preferred, instead of a nonlinear
Logit or Probit model, for at least three reasons. First, to ease the interpretation
of the results; second, because in the context of panel data, a linear
probability model seems better suited to incorporate diﬀerent methodological
tools, such as ﬁxed eﬀects estimators, instrumental variables, and/or clustered
standard errors (Greene, 2002; Arellano and Hahn, 2013); and third, because
Logit or Probit models lead to an important number of municipalities being
dropped because their outcome has no variation across time. However, section
A10 in the Online Appendix shows that these basic results are robust to estimations
based on Logit ﬁxed eﬀects models.
21
In the case of extreme rainfall, note that the treatment variable accounts
for the change in the maximum level of rainfall when going from the 2006–2007
period to the 2010–2011 one. These results, however, are robust to other
measures of rainfall, such as the cross-sectional level of rainfall during the
disaster or deviations from historical (average) levels of rainfall (results not
shown).
J. Gallego Electoral Studies 55 (2018) 73–88
81
compared to 2007, for a reason diﬀerent from the disaster itself.
Besides, as it was mentioned before, water supply is not an indicator of
whether a Colombian municipality has access to a river or lake for
agricultural or transportation purposes, because most municipalities
have access. Consequently, the basic assumption behind the use of
×Water Post2010 as an instrument for ×Victims Post2010 is that water
supply measures the size and presence of water bodies that may cause
ﬂooding after high levels of precipitation. A full discussion on the validity
of the exclusion restriction in the case, including a test on observable
characteristics, is provided in section A9 of the Online Ap-
pendix.
Table 2 reports the results of the instrumental variables estimation
based on models (3) and (4). Second stage results show that in 2011, as
compared to 2007, higher rates of victimization caused a higher increase
in the probability of reelecting the incumbent party. This ﬁnding
is robust to the inclusion of demographic controls (column 2), climate
(column 3), and poverty (column 4). The coeﬃcient in column 1 suggests
that a one percentage point increase in the rate of victims per
capita increases the probability of reelection by approximately 0.45%
more in 2011. Thus, compared to a situation with no victims, one in
which the whole population is aﬀected implies an increase in the probability
of reelecting the incumbent party 45 percentage points higher in
2011.
First stage results also conform to the expectations. After the disaster,
municipalities with higher levels of water supply in 1998 are
predicted to have higher levels of victims per capita in 2010–2011. In
every speciﬁcation, an F value higher than 40 on the Kleibergen-Paap
test provides evidence to reject the null hypothesis that the instrument
is weak. Overall, results reported in Table 2 suggest that in 2011,
municipalities surrounded by more voluminous rivers and lakes are
more likely to experience ﬂoods, and that places with a higher number
of victims exhibit a higher increase in the probability of reelecting the
incumbent party.
In this section, an instrumental variables approach has been used to
complement the diﬀerence-in-diﬀerences strategy proposed above and
to determine the causal eﬀect of exposure to a natural disaster on the
probability of reelection of the incumbent party. Using the 1998 level of
surface runoﬀ, interacted with the post-disaster time dummy as an instrument
for the number of victims, I found that in 2011, the disaster
caused an increase in the incumbent parties' survival in municipalities
heavily aﬀected by ﬂoods and landslides. In the next section, I
strengthen the argument through a placebo test and some robustness
checks.
6.3. A placebo test and two robustness checks
First, I present the results of a placebo test performed to corroborate
the hypothesis that a disaster increases the incumbent party's chance of
reelection. So far, the measure of exposure to the catastrophe includes
the number of people aﬀected by the disaster, which corresponds to the
victims of the so-called ﬁrst rainy season, which took place from the
second quarter of 2010 through April of 2011. However, after the
election took place in October 2011, a second rainy season occurred.
Although it was less intense and severe than the ﬁrst one, it also caused
damage and high levels of victimization. For the placebo test, the
number of victims per capita from November 2011 through June 2012
is utilized. Consequently, VictimsAfter is interacted with the Post2010
dummy variable to estimate models equivalent to those reported in
Table 1.
If the mechanism proposed in this paper is correct—that the disaster
implies an in-ﬂow of resources (aid) that can be used for buying votes—no
eﬀects whatsoever on the probability of reelection should be
found for these new speciﬁcations. The rate of post-electoral victims
should have no impact on the probability of reelecting the incumbent
party, because any relief allocated for these new victims cannot be used
to buy votes.
Table 3 corroborates the intuition. Columns 1–4 report the coeﬃcients
for the DID speciﬁcation, using the rate of post-election victims as
the treatment variable, while columns 5–8 report the results for the IV
approach. In any case, the results reveal that the level of exposure to the
post-election disaster has no eﬀect on the probability of reelecting the
incumbent party. In municipalities more aﬀected by the second rainy
Table 2
Disasters and reelection: Instrumental variables approach.
Second Stage
Reelection Reelection Reelection Reelection
Victims × Post2010 0.449** 0.441** 0.424* 0.457*
(0.208) (0.209) (0.241) (0.272)
Post2010 −0.0613* −0.0556 −0.0802 −0.0520
(0.0345) (0.0357) (0.0706) (0.0833)
First Stage
Vic × P2010 Vic × P2010 Vic × P2010 Vic × P2010
Water × Post2010 0.024*** 0.024*** 0.022*** 0.019***
(0.003) (0.003) (0.003) (0.003)
Post2010 0.132*** 0.136*** −0.018 −0.103***
(0.006) (0.006) (0.023) (0.024)
Population × Post2010 N Y Y Y
Popdensity × Post2010 N Y Y Y
Temperature × Post2010 N N Y Y
Poverty × Post2010 N N N Y
Fixed Eﬀects Y Y Y Y
Time Eﬀects Y Y Y Y
First Stage F-Stat 342.15 171.85 143.62 137.40
KP Wald F-Stat 70.28 69.32 51.01 40.40
N 2136 2134 2128 2122
Notes: Standard errors clustered at the municipality level are shown in parentheses. Election years in the sample are 2007 and 2011. Reelection is a dummy variable
that equals 1 if the incumbent party wins the election. Victims is the number of victims per capita. Post2010 equals 1 for 2011 observations and 0 otherwise. Water is
the 1998 level of surface runoﬀ per capita. IV Fixed Eﬀects estimation is used in every speciﬁcation. * is signiﬁcant at the 10% level, ** is signiﬁcant at the 5% level,
*** is signiﬁcant at the 1% level.
J. Gallego Electoral Studies 55 (2018) 73–88
82
season—from November 2011 through June 2012—the change in the
probability of reelection is not higher than in municipalities less affected.
Hence, this placebo test corroborates the idea that what matters
is the pre-electoral impact of the disaster. In future subsections, I will
claim that this is true because relief is used by incumbents to buy votes.
Two additional robustness checks of the main results are reported in
the Online Appendix. First, I explore more deeply the variation in the
reelection rates of parties in mayoral elections in Colombia. Given that
Colombian politics has been characterized in past decades by a multiplicity
of parties, especially at the local level, it cannot be taken for
granted that a party holding oﬃce will run for reelection in the next
period. For this reason, I construct a dummy variable called Continuityit,
which equals 1 whenever the incumbent party runs for reelection, and
zero otherwise. Not surprisingly, only 57% of incumbent parties ran for
reelection in 2007 and 2011.
Moreover, Table A2 in the Online appendix shows that the exposure
to the disaster—measured through the rate of victims (cols. 1–4), extreme
rainfall (cols. 5–8), or the IV approach (cols. 9–12)—does not
aﬀect the probability of running for reelection. In contrast, Appendix
Table A3 shows that exposure to the disaster has the same positive
eﬀect on the probability of reelection, even if we control for whether or
not the incumbent party runs for reelection. Similar results (not reported
and available upon request) are obtained if we code for whether
political parties disappear for at least one sample period.
For the second robustness check presented in this subsection, I extend
the period of analysis to include additional pre-disaster periods. As
claimed above, in the past decade the Colombian political system has
been highly fragmented and unstable, exhibiting a multiplicity of parties
that make it diﬃcult to track reelection rates. This situation is
particularly evident in 2000 and 2003, before the approval of political
reform that disciplined politicians and signiﬁcantly reduced the number
of contesting parties. Nonetheless, to prove the robustness of the estimations,
Table A10 reports the results of models based on (1), (2), and
(4), in which electoral years 2003, 2007, and 2011 are included in the
analysis. In this case, for every speciﬁcation, once more it is true that
higher exposure to the disaster implies a higher increase in the probability
of reelecting the incumbent party. In fact, the estimated coeﬃcients
are similar in magnitude to those reported in Tables 1 and 2.
Table 3
Placebo test (post election victims and reelection).
(1) (2) (3) (4) (5) (6) (7) (8)
Reelection Reelection Reelection Reelection Reelection Reelection Reelection Reelection
VictimsAfter × Post2010 0.0532 0.0505 −0.0326 −0.154∗
7.961 7.984 −30.50 −8.983
(0.0840) (0.0841) (0.112) (0.0866) (9.933) (10.21) (143.9) (13.39)
Pop. and Density N Y Y Y N Y Y Y
Temperature N N Y Y N N Y Y
Poverty N N N Y N N N Y
Fixed Eﬀects Y Y Y Y Y Y Y Y
Time Eﬀects Y Y Y Y Y Y Y Y
N 2200 2196 2144 2134 2136 2134 2128 2122
Notes: Standard errors clustered at the municipality level are shown in parentheses. Election years in the sample are 2007 and 2011. Reelection is a dummy variable
that equals 1 if the incumbent party wins the election. VictimsAfter is the number of victims per capita during the second rainy season. Post2010 equals 1 for 2011
observations and 0 otherwise. OLS Fixed Eﬀects estimation is used in columns 1–4. IV Fixed Eﬀects estimation is used in columns 1–4. * is signiﬁcant at the 10% level,
** is signiﬁcant at the 5% level, *** is signiﬁcant at the 1% level.
Table 4
Mechanism: Relief, reconstruction, vote-buying, and reelection.
(1) (2) (3) (4) (5) (6) (7) (8)
Reelection Reelection Reelection Reelection Reelection Reelection Reelection Reelection
Relief × Post2010 0.945∗∗
0.941∗∗
0.736∗
0.633 0.341 0.395 0.310 0.270
(0.370) (0.370) (0.415) (0.421) (0.585) (0.587) (0.617) (0.618)
Reconstruction × Post2010 −0.00338 −0.00334 −0.00183 −0.00163 −0.00349 −0.00369 −0.00228 −0.00205
(0.00529) (0.00531) (0.00523) (0.00522) (0.00529) (0.00533) (0.00526) (0.00525)
Victims × Post2010 0.184 0.167 0.137 0.126
(0.126) (0.126) (0.131) (0.137)
VoteBuying × Post2010 0.0187∗∗
0.0249∗∗
0.0248∗∗
0.0250∗∗
(0.00912) (0.0117) (0.0116) (0.0114)
Constant 0.160∗∗∗
0.161∗∗∗
0.0899 0.0768 0.137∗∗∗
0.141∗∗∗
0.0840 0.0794
(0.0118) (0.0140) (0.0708) (0.0753) (0.0162) (0.0171) (0.0705) (0.0749)
Pop. and Density N Y Y Y N Y Y Y
Temperature N N Y Y N N Y Y
Poverty N N N Y N N N Y
Fixed Eﬀects Y Y Y Y Y Y Y Y
Time Eﬀects Y Y Y Y Y Y Y Y
N 2114 2110 2058 2050 2114 2110 2058 2050
Notes: Standard errors are shown in parentheses. Election years in the sample are 2007 and 2011. Reelection is a dummy variable that equals 1 if the incumbent party
wins the election. Relief is the number grocery and hygiene kits distributed per capita. Reconstruction is the amount of Colombian thousand hundred pesos per capita
disbursed for reconstruction projects. Post2010 equals 1 for 2011 observations and 0 otherwise. Victims is the number of victims per capita. VoteBuying is the number
of vote buying allegations in the 2011 election, as reported by the Oﬃce of the Attorney General of Colombia. OLS Fixed Eﬀects estimation is used in every
speciﬁcation. * is signiﬁcant at the 10% level, ** is signiﬁcant at the 5% level, *** is signiﬁcant at the 1% level.
J. Gallego Electoral Studies 55 (2018) 73–88
83
6.4. Mechanism: relief versus infrastructure aid
If a disaster diminishes the well-being of those aﬀected, why would
it be beneﬁcial for a democratic leader? The mechanism proposed in
this paper suggests that after a disaster incumbents will have more
resources to buy votes. In order to test this hypothesis, this paper presents
two pieces of evidence, based on information provided by
Colombia Humanitaria, on the allocation of diﬀerent types of humanitarian
aid among aﬀected municipalities.
During the disaster and for weeks after it ended, the central government,
through Colombia Humanitaria, distributed relief to the affected
population in the form of food, hygiene kits, and other private
goods. But the disaster not only aﬀected citizens directly through the
loss of human lives, homes, or livestock. It also destroyed roads,
bridges, and other infrastructure assets, which naturally increased
economic costs to the aﬀected communities. For this reason, Colombia
Humanitaria allocated a fair amount of humanitarian aid for the purpose
of reconstructing local public goods destroyed by the disaster, and
for the construction of dikes and other structures that could prevent
future ﬂoods and landslides.
Is it necessarily true that high levels of reconstruction aid increased
the likelihood of reelection for incumbent parties in 2011? Facts and
intuition suggest that in this case, in contrast to the allocation of relief,
things are not so simple. On the one hand, mayors could allocate contracts
and jobs associated to reconstruction projects strategically. In
fact, orders of mayors concerning reconstruction projects represented
almost half of the total investment/spending in infrastructure in
Colombian municipalities. But on the other hand, although Colombia
Humanitaria allocated resources before the election took place, most of
these projects, for diﬀerent reasons, were not executed before October
30, 2011. Therefore, not all of the money was available during the
electoral campaign, as opposed to relief, that was distributed prior to
election day. Columns 1–4 in Table 4 report the estimates of a series of
OLS Fixed Eﬀects models showing the eﬀect of these diﬀerent forms of
aid on the probability of reelecting the incumbent party. These results
must be interpreted with caution, as these allocations cannot be expected
to be uncorrelated with unobservable characteristics aﬀecting
electoral outcomes. Consequently, they should be interpreted as tentative
evidence in favor of the mechanism proposed in this paper. The
results suggest that municipalities receiving higher levels of relief exhibit
higher increases in the probability of reelecting the incumbent
party.
In contrast, no signiﬁcant eﬀect on reelection is found for reconstruction
aid. In fact, to take into account that the timing of reconstruction
projects is diﬀerent and some of them were completed
after the election, Table A14 in the Online Appendix reports the results
of analogous models in which alternative and more contemporaneous
measures of reconstruction aid are used instead. The results are substantively
the same. Reconstruction aid has no eﬀect on the probability
of reelecting the incumbent party, while relief has a positive eﬀect.
Hence, these estimations reveal that particularistic beneﬁts such as
relief have a positive impact on party survival, while collective goods,
such as reconstruction aid, have no eﬀect. These results support the
mechanism proposed in this paper. As deﬁned and discussed in Gallego
(2015), political clientelism is a relationship in which a politician (the
patron) oﬀers private goods or services to the voter (the client) in exchange
for political support, which generally includes the vote. In this
case, voters receive private rewards in the form of relief. Incumbents, in
the meantime, are more likely to win in places more aﬀected by the
disaster and where more relief was allocated. Does it mean, then, that
relief is being allocated strategically to gain votes and win elections? In
the following subsections, I present more tentative evidence in support
of this claim.
6.5. Disaster aid and clientelism
So far I have shown that in places more aﬀected by the disaster—both
in terms of the number of victims or through the occurrence
of more severe episodes of extreme rainfall—there is a higher increase
in the probability of the incumbent party being reelected. Furthermore,
Table 4 suggests that reelection becomes more likely in places that
receive more relief and that the same does not hold for reconstruction
aid. The ﬁrst piece of suggestive evidence in favor of the clientelism
mechanism incorporates evidence from the Oﬃce of the Attorney
General of Colombia (Fiscalía General de la Nación). After every election,
this oﬃce reports the number of electoral crimes committed in each
municipality, using a typology comprised of eleven types of crimes. For
the purposes of this study, I use the fourth category, “voter corruption”,
which occurs when someone “promises, pays or handles money or gifts
to a citizen in exchange for his vote.” Accordingly, I construct the
variable VoteBuyingi, which corresponds to the count of vote buying
allegations in municipality i, as reported by the Oﬃce of the Attorney
General after the 2011 local elections.
This measure, which has been used elsewhere to capture clientelism
(Rueda, 2017), is far from perfect as it may suﬀer from report bias, but I
believe it provides a proxy of the intensity of vote buying at the local
level. Nonetheless, the results that follow should be interpreted with
caution. To test if clientelism plays any role at determining the eﬀect of
the disaster—and of the allocation of relief—on political survival, I
extend the previous estimation by including both the vote buying
measure and the rate of victims. Columns 5–8 in Table 4 report the
results of these estimations. In every speciﬁcation, once we control for
vote buying, the eﬀects of both the number of victims and relief allocation
diminish and become insigniﬁcant. In contrast, vote buying has a
positive and signiﬁcant eﬀect on the probability of reelecting the incumbent
party.
Therefore, these results tentatively suggest that the mechanism at
play is clientelism, as the explanatory power of disaster exposure and
relief allocation diminish once we control for vote buying allegations.
This goes in line with the abundant anecdotal and journalistic evidence
revealed during the election, which claimed that parties and political
groups allocated relief strategically to increase their chances of staying
in power.22
But this raises an additional question strongly connected to
the literature on clientelism: whether incumbents are using aid to buy
vote choices (Stokes, 2005) or turnout (Nichter, 2008). Instead of intending
to switch preferences, parties may try to mobilize voters that
otherwise would abstain. Following this theory, we should expect differential
eﬀects of the disaster on voter turnout. However, section A4 in
the Online Appendix shows that this is not the case. Correlational and
causal evidence reveals that there is no connection between exposure to
the disaster and turnout. This tentative evidence suggests that incumbents
bought vote choices, not participation.
6.6. Alternative mechanisms
So far this paper has shown that, in comparison to the 2007 mayoral
elections, incumbent parties in the 2011 election had a higher probability
of being reelected in municipalities heavily aﬀected by the disaster.
Moreover, a basic mechanism has been proposed. The disaster is
beneﬁcial for incumbent politicians because it generates an inﬂow of
aid and resources that can be used to buy votes. The patterns exhibited
by food aid allocation yield tentative evidence in support of this hypothesis.
Nonetheless, alternative explanations may justify these ﬁndings.
In particular, these results could be justiﬁed through theories of
electoral accountability and government responsiveness, the eﬀects of
agricultural income shocks, or the dynamics of civil conﬂict. In this
22
See, for instance, http://www.elespectador.com/noticias/politica/
proselitismo-subsidios-articulo-302748.
J. Gallego Electoral Studies 55 (2018) 73–88
84
subsection, I present evidence against these alternative explanations.
6.6.1. Electoral accountability and government responsiveness
A ﬁrst possibility would be that incumbent parties increase their
probability of reelection after the disaster because they are eﬃcient at
alleviating the pervasive consequences associated to the tragedy and,
consequently, voters reward this eﬃciency at the polls. This alternative
explanation is closely related to the retrospective voting theory, in
which voters punish incumbents that they consider diminished their
well-being through their actions. On the contrary, if they consider that
the incumbent's policies have increased in some way their welfare—and
this would be the case of citizens aﬀected by the disaster whose suffering
is ameliorated or whose losses are diminished thanks to the efﬁcient
strategies implemented by the authorities—a natural way to
reciprocate would be to support the party in oﬃce.
I present several tests to rule out this explanation. First, in
September 2011, about a month before the election, the Attorney's
Oﬃce published a list of municipalities that were being investigated
because relief and prevention projects presented delays in their execution.
Therefore, under the assumption that voters have knowledge of
this situation—either because they have access to the list or because
they are aware of the delays in their municipality—incumbent governments
investigated because of their ineﬃciency in alleviating the
crisis should have exhibited a lower probability of reelection. Table 5
reports evidence against this hypothesis. Incorporating the variable
Delaysi — an indicator of whether a municipality had any delayed
projects— into the basic speciﬁcation, columns 1–4 show that there is
no diﬀerential eﬀect of this variable on the probability of reelection.
Moreover, columns 5–8 show that the eﬀect of relief allocation remains
strong and positive even when we control for delays and that there are
no heterogeneous eﬀects at this level. The positive eﬀect of relief on
political survival is not lower (higher) in places where incumbents were
less (more) eﬃcient.
But perhaps citizens are not aware of these delays. It could be the
case that citizens in fact reward more responsive governments. Hence,
as stated in Besley and Burgess (2002), incumbents will have electoral
incentives to respond to the disaster, and responsiveness will be higher
in places where voters are more sophisticated. Consequently, according
to these arguments, the positive eﬀect of relief on reelection should be
higher in places with greater levels of state capacity, where voters have
more access to the media and higher levels of education. Tables A5-A7
in the Online Appendix show that in the Colombian case, these assertions
are not true. First, using the number of radio stations in each
municipality23
or the level of circulation of the main newspaper in
Colombia (Camacho and Conover, 2011),24
Table A5 shows that the
eﬀect of relief does not vary heterogeneously across places with higher
or lower levels of media penetration.
Similarly, using the Integral Performance Index, compiled by the
National Planning Department to measure state capacity, Table A6
shows that the eﬀect of relief is robust to the inclusion of this indicator
(columns 1–4), and that this impact is not higher in places with more
state capacity (columns 5–8). Finally, using the average years of
schooling as a measure of literacy, Table A7 again shows that the eﬀect
of relief is robust to the inclusion of this measure and that there are no
heterogeneous eﬀects at this level. In sum, all this evidence supports the
argument that voters are not rewarding more responsive governments,
since there is no association between eﬃciency (measured through the
occurrence of delayed projects) and reelection. The eﬀect of relief also
is not higher in places with stronger states, higher levels of media penetration,
or greater levels of literacy. Furthermore, the eﬀect of relief
survives to the inclusion of all these variables.
6.6.2. Agricultural income shocks
A third possibility is that the income shocks caused by the disaster,
especially among those families that heavily depend on agricultural
production, alter political preferences and hence exert some inﬂuence
on electoral results. Retrospective voting theory suggests that negative
income shocks should harm incumbents. Consequently, it would be
Table 5
Alternative mechanisms: Accountability and responsiveness.
(1) (2) (3) (4) (5) (6) (7) (8)
Reelection Reelection Reelection Reelection Reelection Reelection Reelection Reelection
Delays × Post2010 0.0409 0.0397 0.0347 0.0327 0.0181 0.0190 0.0161 0.0122
(0.0347) (0.0349) (0.0350) (0.0347) (0.0410) (0.0413) (0.0417) (0.0415)
Relief × Post2010 1.100∗∗
1.097∗∗
0.894∗
0.766
(0.432) (0.432) (0.474) (0.477)
Reconstruction × Post2010 −0.0138 −0.0138 −0.0114 −0.0112
(0.00839) (0.00845) (0.00851) (0.00850)
Delays × Relief × Post2010 −0.485 −0.492 −0.450 −0.370
(0.856) (0.857) (0.945) (0.944)
Delays × Recons × Post2010 0.0178∗
0.0177∗
0.0162 0.0162
(0.0107) (0.0107) (0.0107) (0.0108)
Constant 0.165∗∗∗
0.170∗∗∗
0.0456 0.0336 0.163∗∗∗
0.164∗∗∗
0.100 0.0852
(0.0109) (0.0119) (0.0677) (0.0711) (0.0157) (0.0177) (0.0719) (0.0767)
Pop. and Density N Y Y Y N Y Y Y
Temperature N N Y Y N N Y Y
Poverty N N N Y N N N Y
Fixed Eﬀects Y Y Y Y Y Y Y Y
Time Eﬀects Y Y Y Y Y Y Y Y
N 2200 2196 2144 2134 2114 2110 2058 2050
Notes: Standard errors are shown in parentheses. Election years in the sample are 2007 and 2011. Reelection is a dummy variable that equals 1 if the incumbent party
wins the election. Delays is a dummy variable that equals 1 if the municipality was investigated by the Oﬃce of the Inspector General of Colombia for delays in
infrastructure projects. Relief is the number grocery and hygiene kits distributed per capita. Reconstruction is the amount of Colombian thousand hundred pesos per
capita disbursed for reconstruction projects. Post2010 equals 1 for 2011 observations and 0 otherwise. OLS Fixed Eﬀects estimation is used in every speciﬁcation. * is
signiﬁcant at the 10% level, ** is signiﬁcant at the 5% level, *** is signiﬁcant at the 1% level.
23
In this case, Radioi measures the number of FM radio stations per capita in
2011. Similar results are obtained if AM stations are included as well.
24
In Colombia, Periódico El Tiempo is the main newspaper. Note that the
number of observations is lower in these speciﬁcations, as information is not
available for all municipalities.
J. Gallego Electoral Studies 55 (2018) 73–88
85
natural to assume that these shocks would downwardly bias the estimates
presented so far. To account for this potential confounder, I use
two measures of municipal agricultural dependence: the total extensions
of land that are planted and harvested, previous to the disaster, as
a proportion of the total area of the municipality.25
If the incomeshocks
hypothesis is true, diﬀerential impacts of the disaster should be
observed across municipalities that are more or less dependent on
agriculture.
Table 6 reports the results of DID models that include these two
measures. In this case, we are interested in the triple interactions between
these measures of agricultural dependence, exposure to the disaster,
and the post-disaster time dummy. Columns 1–4 show that there
are no heterogeneous eﬀects when we use the proportion of land
planted, while columns 5–8 report an equivalent result if we instead use
Table 6
Disaster impact and agricultural production.
(1) (2) (3) (4) (5) (6) (7) (8)
Reelection Reelection Reelection Reelection Reelection Reelection Reelection Reelection
Victims × Post2010 0.209∗∗
0.215∗∗
0.189∗
0.175 0.211∗∗
0.216∗∗
0.190∗
0.177∗
(0.103) (0.103) (0.106) (0.107) (0.0990) (0.0990) (0.102) (0.103)
Planted × Post2010 0.135 0.126 0.128 0.156
(0.113) (0.115) (0.114) (0.118)
Planted × Victims × Post2010 −0.351 −0.351 −0.318 −0.340
(0.548) (0.548) (0.551) (0.552)
Harvested × Post2010 0.156 0.145 0.147 0.176
(0.130) (0.132) (0.131) (0.135)
Harvested × Victims × Post2010 −0.437 −0.435 −0.394 −0.418
(0.604) (0.604) (0.607) (0.608)
Constant 0.119∗∗∗
0.116∗∗∗
0.0421 0.0209 0.119∗∗∗
0.116∗∗∗
0.0425 0.0220
(0.0211) (0.0219) (0.0714) (0.0775) (0.0212) (0.0220) (0.0718) (0.0782)
Pop. and Density N Y Y Y N Y Y Y
Temperature N N Y Y N N Y Y
Poverty N N N Y N N N Y
Fixed Eﬀects Y Y Y Y Y Y Y Y
Time Eﬀects Y Y Y Y Y Y Y Y
N 2050 2050 2050 2042 2050 2050 2050 2042
Notes: Standard errors clustered at the municipality level are shown in parentheses. Election years in the sample are 2007 and 2011. Reelection is a dummy variable
that equals 1 if the incumbent party wins the election. Victims is the number of victims per capita. Planted is the proportion of planted area in the municipality in
2005. Harvested is the proportion of harvested area in the municipality in 2005. Post2010 equals 1 for 2011 observations and 0 otherwise. Literacy is the pre-disaster
average number of years of education. OLS Fixed Eﬀects estimation is used in every speciﬁcation. * is signiﬁcant at the 10% level, ** is signiﬁcant at the 5% level, ***
is signiﬁcant at the 1% level.
Table 7
Disaster impact and civil conﬂict.
(1) (2) (3) (4) (5) (6) (7) (8)
Reelection Reelection Reelection Reelection Reelection Reelection Reelection Reelection
Victims × Post2010 0.227∗∗∗
0.214∗∗∗
0.164∗
0.152∗
0.239∗∗∗
0.226∗∗∗
0.175∗∗
0.162∗
(0.0766) (0.0769) (0.0838) (0.0880) (0.0788) (0.0791) (0.0867) (0.0911)
FARC × Post2010 47.04 30.15 44.51 57.69 182.7 147.8 59.33 52.27
(353.2) (353.2) (354.5) (370.2) (527.3) (528.2) (536.4) (548.3)
Paras × Post2010 −2436.4 −2405.3 −2478.4 −2497.6 −392.2 −357.9 −518.6 −555.6
(2375.9) (2380.3) (2350.9) (2385.4) (2714.9) (2724.9) (2729.5) (2822.6)
FARC × Victims × Post2010 −796.4 −694.8 −115.7 1.010
(2395.2) (2401.9) (2446.4) (2512.1)
Paras × Victims × Post2010 −10667.4 −10680.4 −10205.8 −9917.2
(8639.2) (8690.0) (8605.9) (8783.9)
Constant 0.139∗∗∗
0.146∗∗∗
0.0582 0.0592 0.137∗∗∗
0.144∗∗∗
0.0573 0.0587
(0.0141) (0.0153) (0.0675) (0.0719) (0.0144) (0.0156) (0.0678) (0.0722)
Pop. and Density N Y Y Y N Y Y Y
Temperature N N Y Y N N Y Y
Poverty N N N Y N N N Y
Fixed Eﬀects Y Y Y Y Y Y Y Y
Time Eﬀects Y Y Y Y Y Y Y Y
N 2114 2110 2058 2050 2114 2110 2058 2050
N 2114 2110 2058 2050 2114 2110 2058 2050
Notes: Standard errors clustered at the municipality level are shown in parentheses. Election years in the sample are 2007 and 2011. Reelection is a dummy variable
that equals 1 if the incumbent party wins the election. Victims is the number of victims per capita. FARC is the 2009 rate of FARC attacks. Paras is the 2009 rate of
paramilitary attacks. Post2010 equals 1 for 2011 observations and 0 otherwise. Literacy is the average number of years of education. OLS Fixed Eﬀects estimation is
used in every speciﬁcation. * is signiﬁcant at the 10% level, ** is signiﬁcant at the 5% level, *** is signiﬁcant at the 1% level.
25
Plantedi corresponds to the 2005 proportion of planted land. Similarly,
Harvestedi is the proportion of harvested land.
J. Gallego Electoral Studies 55 (2018) 73–88
86
the proportion of harvested land. In any case, places that depend more
on agriculture do not exhibit lower (or higher) eﬀects of the disaster on
reelection. Moreover, in general, the overall impact of this event remains
unchanged even when we control for these measures.
6.6.3. Civil conﬂict dynamics
The ﬁnal alternative mechanism that I explore in this study deals
with civil conﬂict dynamics, which has been proven to aﬀect electoral
outcomes in Colombia. Illegal armed organizations, whether right-wing
paramilitary groups or left-wing guerrilla movements, might respond
strategically to allocations of aid, as has occurred in other contexts.
Additionally, the disaster could represent a negative shock on state
capacity, as government forces might get distracted from their security
provision duties while they focus on the catastrophe. In any case, if
illegal armed groups respond accordingly and their actions aﬀect
electoral results, violence might explain the association between disaster
exposure, aid allocation, and reelection outcomes.
To test the role of conﬂict, Table 7 presents the results of models
that incorporate the 2009 rates of FARC and paramilitary forces at-
tacks,26
the main active illegal armed groups in Colombia during the
period of analysis for this study. Columns 1–4 show that the impacts of
the disaster on reelection remain unchanged once we control for these
attacks. Similar results are found if we use the other measures of exposure.
section A6 in the Online Appendix shows that this eﬀect also
remains unchanged once we control for these measures both in OLS FE
models with extreme rainfall used as the exposure measure, and in the
IV speciﬁcations that use surface runoﬀ as an instrument. Moreover,
columns 5–8 show that these eﬀects do not vary as a function of the
level of exposure to these armed groups. Hence, it is not true that in
places more aﬀected by conﬂict the eﬀect of the disaster on reelection is
higher (or lower). Similar results are found in models using relief allocation
(results available upon request): the eﬀect of relief on reelection
is robust to the inclusion of these conﬂict measures and no heterogeneous
eﬀects at this level are found.
7. Conclusion
I have shown that natural disasters might be beneﬁcial for political
survival. However, in contrast with other studies that ﬁnd similar results
(Healy and Malhotra, 2009; Cole et al., 2012), I focus on a mechanism
that has not been fully explored in the literature: the clientelistic
allocation of disaster aid. This study represents a contribution to
the literature for at least three reasons. First, my results are robust to
several speciﬁcations and tests, which show that the positive relationship
between disaster exposure and reelection is quite strong. Second,
tentative evidence suggests that places receiving more relief are more
likely to reelect their incumbent parties, while the same is not true for
reconstruction aid. This result represents a contribution to the literature
in itself, since it shows that the distinction between preparedness and
relief spending (Healy and Malhotra, 2009) is not the only one that
matters.
Third, I show that traditional theories, which diﬀer from the clientelism
story, are not satisfactory explanations in this case. Using ofﬁcial
evidence of the (in)eﬃcient allocation of aid by local governments,
measures of state capacity, media penetration, and education, I
show that it is not necessarily true that in Colombia incumbent parties
get reelected more in places more aﬀected by the disaster, because in
such places governments respond more and voters reward them accordingly.
Moreover, economic shocks, caused by agricultural losses
after the disaster, do not seem to play a mediating role in these eﬀects.
Finally, even though it is possible that illegal armed groups respond
strategically to disaster aid allocation, these responses are not confounding
with the eﬀect of the catastrophe on reelection outcomes. In
sum, I present evidence in favor of the clientelism mechanism and
against other plausible explanations.
Naturally, the conclusions of this paper do not imply that aid should
not be allocated after a disaster takes place because politicians will use
it strategically to buy votes. The ﬁnal message is that policies aimed to
diminish clientelism will also reduce the incentives to strategically
manipulate the utilization of disaster relief. In the meanwhile, alternative
mechanisms can prevent this behavior if aid is managed and
distributed by truly independent agencies and not by political organizations
with clear electoral goals. In the context of climate change and
global warming, in which extreme climatological events are expected to
become more frequent, it is important to understand what the political
consequences of natural disasters are and what can be done to ameliorate
the pervasive consequences of these events.
Acknowledgements
I would like to thank Adriana Camacho, Oeindrila Dube, Rebecca
Morton, Adam Przeworski, Alastair Smith, and Pablo Querubin for
helpful comments. Gabriel Angarita, Maria Paula Contreras, Diego
Eslava, Laura Moreno, and Elliot Motte provided superb research assistance.
The usual caveat applies.
Appendix A. Supplementary data
Supplementary data related to this article can be found at https://
doi.org/10.1016/j.electstud.2018.08.001.
References
Abney, G., Hill, L., 1966. Natural disasters as a political variable: the eﬀect of a hurricane
on an urban election. Am. Polit. Sci. Rev. 60 (4), 974–981.
Achen, C., Bartels, L., 2016. Democracy for Realists. Princeton University Press.
Anbarci, N., Escaleras, M., Register, C., 2005. Earthquakes fatalities: the interaction of
nature and political economy. J. Publ. Econ. 89 (9–10), 1907–1933.
Arellano, M., Hahn, J., 2013. Understanding Bias in Nonlinear Panel Models: Some
Recent Developments. Cambridge University Press, pp. 381–409 (chapter 12).
Arevalo, J., Angarita, G., Jimenez, W., 2013. Reformas electorales y coherencia ideologica
de los partidos politicos en Colombia, 1986-2013. Rev. Mex. Ciencias Polit. Soc. 58
(218), 233–270.
Ashworth, S., 2005. Reputational dynamics and political careers. J. Law Econ. Organ. 21
(2), 441–466.
Ashworth, S., de Mesquita, E.B., Friedenberg, A., 2018. Learning about voter rationality.
Am. J. Polit. Sci. 62 (1), 37–54.
Barrios, S., Bertinelli, L., Strobl, E., 2010. Trends in rainfall and economic growth in
africa: a neglected cause of the african growth tragedy. Rev. Econ. Stat. 92, 350–366.
Bechtel, M., Hainmueller, J., 2011. How lasting is voter gratitude? An analysis of the
short- and long-term electoral returns to beneﬁcial policy. Am. J. Polit. Sci. 55 (4),
852–868.
Benson, C., Clay, E., 1998. The Impact of Drought on Sub-saharan Economies. Techincal
Paper. World Bank.
Besley, T., 2007. Principled Agents: the Political Economy of Good Government. Oxford
University Press.
Besley, T., Burgess, R., 2002. The political economy of government responsiveness: theory
and evidence from India. Q. J. Econ. 117 (4), 1415–1451.
Bodet, M., Thomas, M., Tessier, C., 2016. Come hell or high water: an investigation of the
eﬀects of a natural disaster on a local election. Elect. Stud. 43, 85–94.
Brancati, D., 2007. Political aftershocks: the impacts of earthquakes on intrastate conﬂict.
J. Conﬂict Resolut. 51 (5), 715–743.
Bruckner, M., Ciccone, A., 2011. Rain and the democratic window. Econometrica 79 (3),
923–947.
Bueno de Mesquita, B., Smith, A., Siverson, R., Morrow, J., 2003. The Logic of Political
Survival. MIT Press.
Burgess, R., Deschenes, O., Donaldson, D., Greenstone, M., 2014. The Unequal Eﬀects of
Weather and Climate Change: Evidence from Mortality in Indi, WP.
Camacho, A., Conover, E., 2011. Manipulation of social program eligibility. Am. Econ. J.
Econ. Pol. 3 (2), 41–65.
Chen, J., 2013a. Voter Income and Mobilizing Eﬀect of Distributive Beneﬁts. Working
Paper. University of Michigan.
Chen, J., 2013b. Voter partisanship and the eﬀect of distributive spending on political
participation. Am. J. Polit. Sci. 57 (1), 200–217.
Cohen, C., Werker, E., 2008. The political economy of natural disasters. J. Conﬂict
Resolut. 52 (6), 795–819.
Cole, S., Healy, A., Werker, E., 2012. Do voters demand responsive governments?
Evidence from indian disaster relief. J. Dev. Econ. 97 (2), 167–181.
Congleton, R., 2006. The story of katrina: new orleans and the political economy of
26
Results are similar if we use attacks during any other pre-treatment year.
J. Gallego Electoral Studies 55 (2018) 73–88
87
catastrophe. Publ. Choice 127 (1–2), 5–30.
DANE, 2011. Informe ﬁnal registro unico de damniﬁcados por la emergencia invernal
2010-2011. Documento Oﬁcial.
Dube, O., Vargas, J., 2013. Commodity price shocks and civil conﬂict: evidence from
Colombia. Rev. Econ. Stud. 80 (4), 1384–1421.
Eisensee, T., Stromberg, D., 2007. News drought, news ﬂoods, and U.S. Disaster relief. Q.
J. Econ. 122 (2), 693–728.
Escaleras, M., Anbarci, N., Register, C., 2007. Public sector corruption and major earthquakes:
a potentially deadly interaction. Publ. Choice 132 (1–2), 209–230.
Fair, C., Kuhn, P., Malhotra, N., Shapiro, J., 2017. Natural disasters and political engagement:
evidence from the 2010-2011 pakistani ﬂoods. Quarterly Journal of
Political Science 12 (1), 99–141.
Ferejohn, J., 1986. Incumbent performance and electoral control. Publ. Choice 30, 5–25.
Fiorina, M., 1981. Retrospective Voting in American National Elections. Yale University
Press.
Gallego, J., 2015. Self-enforcing clientelism. J. Theor. Polit. 27 (3), 401–427.
Gallego, J., 2018. Civil Conﬂict and Voting Behavior: Evidence from Colombia. Conﬂict
Management and Peace Science.
Gallego, J., Li, C., Wantchekon, L., 2018. The Political Economy of Patrons, Brokes, and
Voters. Working Paper. Universidad del Rosario.
Gallego, J., Wantchekon, L., 2018. Clientelism: Concepts, Agents, and Solutions. Working
Paper, Universidad del Rosario.
Garbiras, N., 2012. Desastres Naturales y Uso de Recursos Publicos: Evidencia de la Ola
Invernal de 2010-2011 en Colombia. Master's thesis. Universidad de los Andes.
Gasper, J., Reeves, A., 2011. Make it rain? Retrospection and the attentive electorate in
the context of natural disasters. Am. J. Polit. Sci. 55 (2), 340–355.
Greene, W., 2002. The Bias of the Fixed Eﬀects Estimator in Nonlinear Models. Working
Paper, NYU.
Healy, A., Malhotra, N., 2009. Myopic voters and natural disaster policy. Am. Polit. Sci.
Rev. 103 (3), 387–406.
IDEAM, 1998. Estudio Nacional del Agua. Instituto de Hidrologia, Meteorologia y
Estudios Ambientales.
Kahn, M., 2005. The death tolls from natural disasters: the role of income, geography, and
institutions. Rev. Econ. Stat. 87 (2), 271–284.
Keefer, P., Neumayer, E., Plumper, T., 2011. Earthquake propensity and the politics of
mortality prevention. World Dev. 39 (9), 1530–1541.
Keefer, P., Vlaciu, R., 2008. Democracy, credibility, and clientelism. J. Law Econ. Organ.
24 (2), 371–406.
Lazarev, E., Sobolev, A., Soboleva, I., Sokolov, B., 2014. Trial by ﬁre: a natural Disaster's
impact on support for the authorities in rural Russia. World Polit. 66 (4), 641–668.
Lopez, C., 2010. Y Refundaron la Patria. De Como Maﬁosos y Politicos Reconﬁguraron el
Estado Colombiano, Debate.
Miguel, E., Satyanath, S., Sergenti, E., 2004. Economic shocks and civil conﬂict: an instrumental
variables approach. J. Polit. Econ. 112 (4), 725–753.
Mustafa, D., 2003. Reinforcing vulnerability? Disaster relief, recovery, and response to
the 2001 ﬂood in rawalpindi, Pakistan. Environ. Hazards 5 (3–4), 71–82.
Nichols, A., 2003. VINCENTY: Stata Module to Calculate Distances on the Earth's Surface.
Working Paper. Boston College.
Nichter, S., 2008. Vote buying or turnout buying? Machine politics and the secret ballot.
Am. Polit. Sci. Rev. 102 (1), 19–31.
Quiroz, A., Smith, A., 2013. Surviving disasters. Br. J. Polit. Sci. 43 (4), 821–843.
Raschky, P., Schwindt, M., 2016. Aid, natural disasters, and the Samaritan's dilemma.
Economica 83 (332), 624–645.
Robinson, J., Verdier, T., 2013. The political economy of clientelism. Scand. J. Econ. 115
(2), 260–291.
Rueda, M., 2017. Small aggregates, big manipulation: vote buying enforcement and
collective monitoring. Am. J. Polit. Sci. 61 (1), 163–177.
Sanchez, A., 2011. Despues de la inundacion, Documento de trabajo sobre economia
regional No. 150. Banco de la Republica.
Stokes, S., 2005. Perverse accountability: a formal model of machine politics with evidence
from Argentina. Am. Polit. Sci. Rev. 99 (3), 315–325.
Stokes, S., Dunning, T., Nazareno, M., Brusco, V., 2013. Brokers, Voters, and Clientelism.
The Puzzle of Distributive Politics Cambridge University Press.
Stromberg, D., 2007. Natural disasters, economic development, and humanitarian aid. J.
Econ. Perspect. 21 (3), 199–222.
Toya, H., Skidmore, M., 2007. Economic development and the impacts of natural disasters.
Econ. Lett. 94 (1), 20–25.
UNDP, 2011. Colombia Rural. Razones para la Esperanza, UNDP.
J. Gallego Electoral Studies 55 (2018) 73–88
88