World Development 162 (2023) 106128

Contents lists available at ScienceDirect

World Development
journal homepage: www.elsevier.com/locate/worlddev

Poverty, social networks, and clientelism

Nico Ravanilla a, Allen Hicken b,⇑
a
School of Global Policy & Strategy, University of California, San Diego, 9500 Gilman Drive, #0519, La Jolla, CA 92093-0519, United States
b
Department of Political Science, Uiversity of Michigan, 505 S. State Street, Ann Arbor, MI 48109-1045, United States

a r t i c l e i n f o a b s t r a c t

Article history: Why are the poor susceptible to clientelism, and what factors shield them from the influence of vote buy-
Accepted 20 September 2022 ing? We explore the role of both formal and informal social networks in shaping the likelihood of being
Available online 12 November 2022 targeted with private inducements. We argue that when the poor lack access to formal social networks,
they become increasingly reliant on vote buying channelled through informal networks. To test our the-
Keywords: ory, we build the informal, family-based network linkages between voters and local politicians spanning
Social networks a city in the Philippines. We then collect survey data on formal network connections, electoral handouts,
Poor
and voting behaviour of 900 voters randomly drawn from these family networks. We show first that cam-
Vote buying
Clientelism
paigns disproportionately target poorer voters. We then show that familial ties further influence target-
Voter behaviour ing among poor voters. Finally, we show that access to formal networks such as workers’ associations
Philippines mitigate voter fears of punishment for failing to reciprocate.
Ó 2022 Elsevier Ltd. All rights reserved.

This study analyzes the role that social networks play in shap- criminate between different sets of poor voters. Campaigns might
ing how poor voters are targeted with offers of electoral clien- prefer to target their core supporters in a bid to induce them to
telism—defined as the offer of material benefits to voters during turn out (Nichter, 2018; Aspinall, Edward, & Meredith Weiss,
election season in hopes of affecting their behavior (Hicken Forthcoming) or because targeting the poor, core supporters is less
2011). Most studies of electoral clientelism conclude that political costly (Stokes et al., 2013). Alternatively, campaigns may prioritize
machines disproportionately target the poor. Arguments for why voters who are weakly attached or undecided (Nichter, 2008;
this association exists vary, but many focus on the decreasing mar- Schady, 2000). Campaigns also make efforts to target those who
ginal utility of income. From the perspective of a candidate or party they think are most likely to be bound by norms of reciprocity
(the supply side), as incomes rise, so too does the value of material (Finan & Frederico, 2012; Ravanilla, Haim, & Hicken, 2021), and
resources politicians must offer in order to sway a potential voter. to reward those who show up to their political rallies (Kramon,
Faced with resource constraints politicians thus prefer to target 2017).
poorer, less expensive voters (Dixit & Avinash, 1996; Calvo & In this study we consider another factor that shapes the likeli-
Ernesto, 2004; Stokes, Dunning, Nazareno, & Brusco, 2013). From hood that a given poor voter is the target of clientelism: the nature
a voter’s perspective, as incomes rise preferences shift away from of their social networks. Our focus is on electoral clientelism, but
private consumption via clientelism and towards more program- like other recent work in the field, we examine electoral clien-
matic policies and the provision of public goods (Nathan, 2019; telism in the context of a broader array of clientelistic relationships
Kitschelt & Herbert, 2007; Bardhan & Mookherjee, 2012; Scott, (e.g. ongoing relational clientelism) (see Nichter, 2018 2018,
1977; Kitschelt, 2000; Stokes et al., 2013; Desposato, 2007).1 Szwarcberg, 2012 2012, and Aspinall et al., Forthcoming n.d.). We
However, while it may be true that poor voters are dispropor- argue that centrally located poor voters are more likely to receive
tionately targeted relative to richer voters, not all poor voters have handouts and we explore two competing mechanisms for why this
an equal likelihood of receiving a handout. A large literature might be the case. First, poor voters who occupy central positions
describes the various ways in which candidates and campaigns dis- in important social networks are better able to extract resources
from campaigns (a demand story). Second, social networks enable
⇑ Corresponding author. campaigns to better monitor and enforce the clientelist exchange,
E-mail addresses: [email protected] (N. Ravanilla), [email protected] (A. making poor voters who are central to those networks particularly
Hicken). appealing targets (a supply story).
1
Although whether changing demands actually translate into a decline in To test our theory, we rely on local naming conventions to build
clientelism is contingent on the broader sociopolitical context, as Nathan (2019)
the informal, family-based network linkages between voters and
and Weitz-Shapiro (2014) ably demonstrate.

https://doi.org/10.1016/j.worlddev.2022.106128
0305-750X/Ó 2022 Elsevier Ltd. All rights reserved.
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

local politicians spanning one entire city in the Philippines. (Family shocks (e.g. job loss, illness, or natural disaster) (Nichter (2018)). In
networks are the most salient networks in Philippines social and many contexts, it is the stated responsibility of governments to
political life). We then collect survey data on formal network con- ensure basic needs are met and to provide social insurance as a
nections, the distribution of electoral handouts, as well as the hedge against vulnerability and risk. Yet, governments often fail
beliefs and voting behavior of 900 voters randomly drawn from to effectively provide these needed goods and services. As a result
these family networks. Our resulting analysis makes three distinct people turn to other strategies for assistance. Accessing social net-
contributions to our understanding of how electoral clientelism works to obtain clientelist resources from social and/or political
operates among poor voters. First, while we find evidence that bro- patrons is one such strategy. Whether membership in such net-
kers do disproportionately target poorer voters, we also show that works is determined by fate (e.g. family or clan membership) or
network centrality has a positive effect on the probability that poor by choice (e.g. membership in a farmers cooperative) individuals
central voters will receive offers of electoral handouts. Second, we can use such networks to help meet their needs and reduce their
find no evidence that being central in these informal networks vulnerability. Obtaining resources is contingent on access to the
makes voters (poor or otherwise) more susceptible to monitoring given social networks and resources tend to flow through - and
or enforcement from politicians or their agents. Contrary to the thereby reinforce - these networks.
fears and expectations of many scholars and practitioners, electoral But access to social networks that can provide needed assis-
clientelism does not appear to trap poor voters in situations of tance is not evenly distributed across voters. Voters vary in the
‘‘perverse accountability” (Stokes, 2005). Finally, we show that degree to which they are central in their networks, as well as in
access to alternative networks such as workers’ associations, com- the number and types of networks to which they are connected.
munity organizations, or other formal organizations can mitigate We argue that voters who are more centrally located within their
voter fears of punishment for failing to reciprocate. networks should be better able to leverage their connections to
obtain access to patronage resources, including electoral clien-
telism. Why is this the case? First, being central conveys informa-
1. Theory: Voter Demands, Monitoring and Enforcement, and tion advantages—centrally-located voters have better information
Alternative Networks about how the network functions, who to ask for help, etc., and
political patrons are more likely to learn of the needs central vot-
To begin with, it is helpful to define what we mean by a social ers. Second, by virtue of their position, centrally-located voters
network in a political context. We follow Wasserman and Stanley’s are more likely to be connect to and able to make demands of polit-
(1994) general definition of a social network as a ‘‘finite sets of ical patrons. Third, campaigns may be more likely to respond to
actors and the relation or relations defined on them.” In political request from more-central voters in hopes of producing a ”social
contexts, these social networks would be made up of the [political] multiplier”— amplifying the effect of the electoral handout beyond
actors and the relationships that link them together. These nodes that recipient (Schaffer & Baker, 2015).2 Finally, it is often the case
and ties can be analyzed either by assessing the position of actors that electoral clientelism has its roots in ongoing clientelist relation-
in the overall network, the relative position of actors to each other, ships (Nichter, 2018; Szwarcberg, 2012). Central voters who are bet-
or features of the network structure as a whole. The existing liter- ter able to use their access to social networks to obtain clientelist
ature has found that social ties influence participation (Nickerson, resources from politicians in between elections, are also likely to
2008; McClurg, 2003), including voting behavior (Cruz, Labonne, & be at the top of the list when it comes to electoral handouts.
Querubin, 2017) and the ability of individuals to cooperate (Larson, Starting with the voter perspective may strike some as odd. We
2017). Social networks also transmit political information often approach the analysis of clientelism, particularly electoral
(Campos, N.d) and affect the capacity of politicians to engage in clientelism, from the perspective of giver–the politician, candidate
political exchange (Cruz, Labonne, & Querubin, 2018; Ravanilla or party providing the resource. We imagine the designing their
et al., 2021). In our theoretical discussion we will refer social net- mobilization strategies in a top-down fashion, deciding which vot-
works in general, intending the theory to be broadly applicable, ers to target, subject to a budget constraint. However, more recent
though our primary focus in this study are family networks, which work has redirected attention towards a more bottom-up dynamic,
are the foundation of political life in the Philippines. In this study where politicians fill the role of ”request fillers”, forced to respond
context, it would be the voters and politicians that are the nodes to demands from voters (Nichter, 2018). This certainly describes
of the social network, and the ties that connect the nodes are based the dynamic we have observed in our work in the Philippines,
on their kinship and marital ties. where voters readily approach politicians and candidates with
In what ways do social networks shape the distribution of elec- request for assistance or handouts, and where patrons feel the
toral clientelism, specifically, which poor voters are more or less need to respond favorably to those requests, particularly those that
likely to be targeted? Building on the rich literature in comparative come from voters with whom they have a personal connection or
clientelism we focus on two categories of explanations—one that from voters who occupy positions of potential influence (e.g. cen-
focuses on the role of social networks in helping certain individuals tral voters) (Ravanilla et al., 2021). Even so, candidates are not
make demands of politicians/campaigns, and another that focuses solely reactive. They also attempt to make strategic choices about
on the ways in which social networks can facilitate the monitoring which voters to target, and in the next section we discuss the fac-
and enforcement of the clientelist exchange. The expectation flow- tors that shape those choices.
ing from both sets of arguments is that poor voters who are cen- Monitoring and Enforcement. We have argued that poverty and
trally located in key social networks should be more likely to vulnerability drive voters to seek redress through clientelism,
receive electoral handouts. In this section we describe the logic and that they utilize the networks available to them to get access
behind why central voters are disproportionately targeted accord- to clientelistic resources. But what about the supply side? What
ing to each perspective, and the derive distinct observable implica- role do social networks play in shaping the incentives of campaigns
tions from each perspective. We then explore the extent to which to target electoral handouts to some poor voters over others? Here
access to multiple social networks can mitigate voter concerns we need to remember an ever-present concern among all candi-
about potential enforcement.
Voter Demands. By definition, the poor are more likely to lack
the resources to fully meet the basic needs for themselves and their 2
Note, Cruz (2015) finds no support for the social multiplier mechanism in her
family. Poverty also increases individuals’ vulnerability to external work on the Philippines.

2
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

dates distributing electoral clientelism—the possibility that voters From our argument we derive the following hypothesis, which
will take a handout and either shirk (not vote) or defect (vote for a we evaluate in the following sections.
different party/candidate). Concern over voter shirking/defection, H1: Poor voters are more likely to receive electoral handouts.
then, might help drive the choice of which voters to target.. H2: Among poor voters, those who are more centrally located in
Even as social networks facilitate access to clientelism for poor, their social networks are more likely to receive electoral handouts.
vulnerable voters, they can also help enable the monitoring and H3: Among poor voters, those who are more centrally located in
enforcement of clientelistic exchanges for candidates. Information their social networks are more likely to believe campaigns can
about individual political preference and behavior is easier to learn how they voted.
obtain when voters are part of identifiable social networks, partic- H4: Among poor voters, those who are more centrally located in
ularly so for voters who are central in those networks (Brusco, their social networks are more likely to fear punishment if they fail
Nazareno, & Stokes, 2004; Stokes, 2005; Kitschelt & Herbert, to reciprocate.
2007; Larreguy, 2012; Stokes et al., 2013; Cruz, 2019; Ravanilla H5: Access to alternative networks mitigates the extent to
et al., 2021). Social networks also create additional opportunities which voters fear enforcement.
for the punishment of non-reciprocating voters, including through
social sanctioning (Ravanilla et al., 2021).
2. Poverty and Clientelism in the Philippines
Both the voter demand model and the monitoring and enforce-
ment model predict that offers of electoral clientelism should
We evaluate our hypotheses in the context of the Philippines.
increase in the network centrality of voters. However, they have
For our purposes the Philippines is a good case in which to test
very different implications for voter beliefs about monitoring and
our hypotheses. The Philippines is a lower-middle income country
enforcement. If targeting is being driven by concerns over shirking
where the combination of poverty, poor or inconsistent gover-
or defection then we should observe that poor central voters are
nance, and exposure to natural disasters means that many citizens
more likely to believe that (a) politicians can learn about how they
face economic vulnerability. Both relational and electoral clien-
voted, and (b) they will punish the voter if they fail to vote as
telism are common in almost all parts of the Philippines. Nation-
expected.
ally, with 22 percent of respondents report receiving offers of
If both arguments hold, then access to social networks can be a
electoral handouts prior to the 2016 election (Pulse Asia
double-edged sword for poor voters. These networks help them
Research, 2019). Elections for national, provincial, and municipal
access on going clientelist benefits, for which they are especially
offices occur every three years, with electoral clientelism a ubiqui-
in need, including electoral clientelism, but at the potential cost
tous part of all races except for those elected from single national
of higher levels of monitoring and enforcement. This raises the
constituencies (i.e. president, vice president, and senators) (Hicken,
possibility that the poor can become effectively trapped in these
Aspinall, & Weiss, 2019). Philippines political parties are notori-
clientelist relationships. Given their reliance on clientelism voters
ously weak and candidates are political promiscuous, with most
may be unable/unwilling to hold incumbents accountable for poor
relying on personal, locally-based electoral machines to support
performance outside the clientelist sphere, undermining account-
their electoral bids (ibid.).
ability and freeing politicians from the need to address broader
Our data were collected in a city in Southern Luzon, which we
voter demands (Stokes, 2005; Weitz-Shapiro, 2014; Nathan,
will call Paros.4 In most respects Paros is a typical, medium-sized
2019). But again, we see variation across voters, even among those
city, with a mix of urban, semi-urban, and rural areas. It has a pop-
who are centrally located within social networks (Ravanilla et al.,
ulation of between 150,000 and 170,000, and a poverty incident rate
2021). Some voters report feeling bound by electoral handouts,
of between 25 and 30 percent—somewhat higher than the national
while others feel little pressure to reciprocate. How do we explain
average of 16.6 percent. Elections in the city are generally competi-
this?
tive, with no single political group able to dominate.
Alternative Networks. We argue that the extent to which voters
are caught in a clientelist trap is partially a function of the nature of
their social networks. Specifically, voters’ freedom to shirk or 3. Measuring Social Networks and Clientelism
defect grows in their access to multiple social networks. In our
empirical analysis we focus on voters who belong to formal orga- In order to evaluate our hypotheses we need measures of both
nizations (e.g. women’s associations) in addition to being part of voter positions in social networks and their exposure to electoral
the informal family networks that are the lifeblood of Philippine clientelism. To do so, we develop a map of informal family net-
politics, but the principle applies regardless of the nature of the works within each barangay (village) within Paros, and pair these
networks involved. Where voters belong to only a single primary data with data from original surveys of voters and brokers in Paros
network, the consequences of losing access to or running afoul of during the 2016 election. The two voter surveys were conducted
that network are dire. But where voters are part of multiple net- using a random sample of 659 voters, drawn from voters that
works that can provide access to patronage, the stakes involved appeared on the 2016 Certified Voter Lists. Partnering with a local
in non-reciprocity are lower. Of course, to the extent campaigns team of enumerators working with Innovations for Poverty Action
are aware of voters’ networks connections, we would expect them surveys were administered using iPads and an offline survey app
to view multi-network voters a riskier bets than single-network (iSurvey). We randomly selected target respondents from the Com-
voters.3 But, conditional on being targeted with electoral clientelism, mission on Elections’ (COMELEC) 2016 Certified Voters List (CVL).
we expect voters with access to more than one social network to be Stratifying randomization by barangay ensured that a larger sam-
less fearful of shirking or defecting, ceteris paribus. ple of respondents was drawn from larger barangays. The CVL lists
the complete name, birthday, gender, and barangay of residence of
3
The effect of access to alternative networks on the probability of being targeted is all registered voters in each municipality. Enumerators located pri-
actually not straightforward. On the one hand, to the extent access to alternative mary respondents at their residential addresses, invited them to
networks lowers the cost of defection, we would expect a negative relationship participate in the research study using a recruitment script, and
between alternative network access and targeting. On the other hand, the fact voters
are part of multiple networks could also raise the probability that they appear on
4
some broker’s voter list. In the end, this is empirical question. As we will discuss There are currently 135 cities and 1,496 municipalities in the Philippines. Cities
below the positive and negative effects appear to cancel each other out—voters with are distinguished from municipalities by their larger population size and annual
access to alternative networks are not any more or less likely to be targeted. revenue.

3
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

obtained their consent to participate in the study. When they could 3.2. Measuring access to formal networks
not interview a primary respondent due to their out-migration,
refusal, or death, the enumerator sought out a randomly selected To test whether access to alternative networks ameliorates the
alternate respondent, instead. The first survey was conducted effects of clientelism we also calculate whether voters are part of
about three weeks prior to the election, and a second survey, tar- formal social networks available in the barangay, municipality, or
geting the same sample of voters, was conducted about three province. We presented voters with a list of common groups and
weeks after the elections. We primarily rely on results from the organizations and asked voters to indicate to which groups they
post-election survey in this paper. belonged. The list of groups included: cooperatives, volunteer asso-
ciations (e.g. Rotary), religious organizations, professional organi-
zations (e.g. teachers associations), and various sectoral
3.1. Building kinship networks organizations (e.g. associations for women, youth, laborers, farm-
ers, etc.). These groups are ubiquitous in the Philippines, in both
Our analysis of voter social networks focuses on informal kin- urban and rural areas, and nearly every respondents would have
ship networks—for most Philippines voters the chief network to ample opportunity to join one or more of these groups. We code
which they have access. There are many kinds of networks that as 1 any respondent who claims membership in a least one of these
voters might utilize in order to get access to clientelist resources, formal organizations, and 0 for respondents who don’t belong to
including tribal or ethnic groups, religious networks, or profes- any of the organizations listed.
sional associations (Adida, 2015; Carlson, 2015; Aspinall,
Dettman, & Warburton, 2011; Stithorn, 2012; Carlin, Singer, & 3.3. Measuring electoral clientelism
Zechmeister, 2015). The salience of particular networks will neces-
sarily vary by political context, but there is a large body of litera- We measure whether an individual survey respondent was
ture on the Philippines that identifies informal kinship networks exposed to electoral clientelism in two ways. First, as part of the
as central to accessing patronage resources for Filipinos (see for voter survey respondents were asked whether, in the run up to
example Cruz et al., 2018, Hicken et al., 2019). Our primary mea- the 2016 election, any political campaign had offered them money,
sure of voters’ position in these social networks draws on methods food, or household items. There were two local slates that com-
developed by Cruz et al. (2017) and adapted for use at the individ- peted for seats during the 2016 election in Paros. The campaign
ual level by Ravanilla, Davidson, and Hicken (2022) and Ravanilla that agreed to be surveyed by our research team did not engage
et al., 2021. For each barangay in Paros we mapped the network in handing out private inducements of any kind to voters, but
ties between all pairs of individuals who share at least one com- the other campaign (which we refer to as Team A in this article)
mon surname. We create these barangay-level family networks did engage in vote-buying. We verified this information both in
for all 64 barangays in Paros. the brokers’ survey as well as in our independent survey of ran-
Two features of Filipino naming conventions allow us to map domly selected voters.
these local family networks with a high degree of confidence. First, One might understandably be concerned about the possibility
the Philippines adopted the Spanish naming convention of giving of biased responses due to social desirability. Indeed, in other con-
children two surnames, one from their father, and one from their texts social desirability bias leads to a severe under-estimate of the
mother (her maiden name). Second, colonial authorities assigned extent of electoral clientelism when respondents are asked directly
a unique set of Iberian surnames to each barangay. As a result, with (e.g. (Corstange, 2009)). However, in the context of the Philippines
the exception of large urban areas that experience lots of in- electoral clientelism carries very little social stigma. In our inter-
migration it is highly probable that individuals who share sur- views with candidates, brokers, and voters most spoke openly of
names in a given barangay are related. See Ravanilla et al. (2021) the practice (Hicken et al., 2019), and this is confirmed through
and Ravanilla et al. (2022) for more details about this method. the use of list experiments as an alternative way to measure of
the extent of electoral clientelism. There is very little difference
between the reported incidence of electoral handouts whether
measured directly, indirectly, or via a list experiments (Aspinall
3.1.1. Voter centrality et al., Forthcoming). This is certainly born out in our survey.
Our primary explanatory variable is voter position within the Around 60 percent of our surveyed voters reported receiving offers
family network based on the full roster of voters spanning one of a handout prior to the elections, substantially more than the
whole city in the Philippines.5 Specifically, for each individual voter national average of 22 percent in 2016. We measure both whether
in our survey we calculate their degree and betweenness centrality or not a voter received a offer of a handout, and the amount of
in the family network. Degree centrality is a count of the number money offered.
of direct family ties that connect to an individual and is essentially
a measure of the extent to which an individual is connected to the 3.4. Measuring voter perceptions of broker monitoring and
network. Betweenness centrality measures the number of shortest enforcement, and measuring poverty
paths between all pairs of individuals in the network on which the
relevant individual is present. This captures the degree to which Our voter survey also included questions that allow us to assess
an individual connects different parts of the network. In previous the extent to which voters feel obliged to vote for the candidate
work we found betweenness centrality to be the better predictor that offered them a handout. To get a sense for whether voters
of voter targeting by brokers (Aspinall et al., Forthcoming). We thus believe they could be monitored we asked respondents how likely
focus on betweenness centrality in our analysis, but because degree it was that different individuals, including campaign brokers, could
centrality can also play role, we control for it in all of our models.6 find out how they voted. We also asked respondents about what
would happen if they did not vote for a candidate whose broker
5
Building the full family network based on all the registered voters within the gave them money or goods, in order to estimate the extent to
study context is key to credibly identifying social network effects that avoids the
which voters are fearful of being punished for shirking or defecting.
measurement error arising from missing nodes in the network Advani and sArun,
2018.
Possible responses included no consequences, feelings of shame,
6
We note that our results are also robust to dropping degree centrality as a control. not receiving future electoral handouts, an inability to ask for help
See Appendix Table A.1. in the future, and being subject to verbal or physical abuse. An
4
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

Table 1 3.5. Control variables

Summary statistics.

Mean Standard Deviation In addition to our primary independent variables we control for
Vote buying (binary variable) a number of respondent characteristics, including, gender, age,
Offered by Team A 0.62 0.49 educational attainment, religion, employment, marital status, and
Vote buying (amount in USD) household size. We also control for a variety of respondent atti-
Offered by Team A 8.24 7.31
tudes and attributes, including expressed levels of support for
Voting attitudes & behavior
Voted in 2016 election 0.96 0.19 the vice-mayoral candidates prior to the election, levels of
Voted for Team A Mayor 0.68 0.47 reciprocity, levels of trust, levels of risk acceptance, and discount
Network centrality measures rates. We also use barangay fixed effects in all models. Table 1 lists
Degree centrality 54.78 58.98 all of the variables and along with the descriptive statistics.
Betweenness centrality 4.15 3.58
Demographics
Female 0.61 0.49 4. Empirical Results
Age 42.84 16.22
Educational attainment
Our goal in this paper is to explain why some poor voters are
Elementary and below 0.28 0.45
High school and below 0.41 0.49 more likely to be targeted by campaigns, and the extent to which
Above high school 0.31 0.46 such targeting undermines accountability. We are especially inter-
Catholic 0.94 0.24 ested in whether voters’ positions in family networks shapes their
Employed 0.55 0.50 probability of being targeted and their likelihood of reciprocating.
Never married 0.24 0.42
We also explore whether the power of electoral handouts to induce
Number of family members 5.50 2.55
Monthly household income (USD) 171.46 202.36 reciprocity is mitigated by ties with formal networks.
Other controls
Integrity 0.39 0.99 4.1. Are the poor more likely to be targets of clientelism?
Negative reciprocity 0.63 0.48
Risk preference 6.13 2.94
Time preference 0.71 0.46 Our first step is to test whether poverty is indeed a predictor of
Trust 0.47 0.50 being targeted with an electoral handout. To test H1 we run the
Number of observations 659 following ordinary least squares (OLS) regression. If our expecta-
Number of villages 61 tions hold, there should be a negative relationship between logged
Notes: For amount of vote buying and household income, the exchange rate used is: household income (Income_i) and being targeted with money, con-
USD:PHP $1/P45. trolling for observables (b1 < 0).
Targeted i ¼ b0 þ b1 Income i þ b2 Central Family i
þ b3 Formal Connection i þ C0Controls i þ  i
alternative approach to measuring enforcement would be to ask
voters directly whether they have suffered reprisals after failing As expected, poor voters are more likely to be targeted with hand-
to reciprocate. We prefer our approach to this alternative for three outs (Table 2), consistent with most existing work and with our
reasons. First, voters’ direct experience with punishment already hypothesis. This is true whether our measure of targeting is at the
informs their beliefs about the likelihood of potential punishment. intensive margin (the amount of money offered to voters) (Model
Second, we are interested in voter beliefs and perceptions. Even if a 1) or extensive margin (the probability that they the received a
voter has not been punished themselves in the past, a belief that handout). (Model 3).
they may be punished in the future could still shape their behavior.
And finally, asking only about past punishment presumes 4.2. What role do voter centrality in informal networks play in the
defection. The question excludes those who have always recipro- targeting of poor voters?
cated, and may have done so out of fear of being punished for
defecting. To what extent do informal social networks help shape the deci-
Our measure of poverty is the household monthly income (USD) sions of candidates about which poor voters to target. We hypoth-
as reported by respondents in the voter survey, logged. esized that poor voters who are central in family networks are

Table 2
Are the poor more likely to be targets of clientelism?

Dependent variable (DV): Vote-buying Amount Vote-bought = 1

(1) (2) (3) (4)
Log HH monthly income (HMI) 0.741* 0.266 0.058** 0.025
(0.419) (0.411) (0.025) (0.033)
Betweenness centrality (BC) 0.105 0.693* 0.005 0.046**
(0.090) (0.364) (0.003) (0.021)
Formal network ties = 1 0.729 0.780 0.023 0.025
(0.722) (0.591) (0.040) (0.035)
HMI x BC 0.122* 0.009**
(0.070) (0.004)
Full set of controls YES YES YES YES
Adjusted R2 0.060 0.061 0.046 0.048
Number of observations 629 629 659 659
Mean of DV 8.24 0.62

Notes: Unit of observation is voter. Robust standard errors clustered at the barangay level in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Barangay fixed effects, measure of
degree centrality, demographic controls, and ‘other controls’ included.

5
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

Fig. 1. What role do voter centrality in informal networks play in the targeting of poor voters?
Note: The figures display marginal effects from models 2 and 4 in Table 2. In poorer households, central voters are more likely to be targeted, and as household income
increases, voter centrality becomes less important.

more likely to be targeted (H2). To test this hypothesis we employ no longer retains its significance when we include the interaction
the following model specification, with the key variables of interest with centrality. However, our primary quantity of interest is the
being Income_i, Central_Family_i, and the interaction between the interaction term. As hypothesized, the coefficient on the interaction
two. Our expectation is that the coefficient on Income_i will be neg- term is negative. Fig. 1 displays the marginal effect of centrality on
ative, the coefficient on Central_Family_i will be positive, and the targeting across the range of income scores, using the results from
coefficient on the interaction term will be negative. Models 2 and 4, respectively. Among the poor, central voters are
more likely to be targeted.7 Further, network centrality matters
Targeted i ¼ b4 þ b5 Income i þ b6 Central Family i þ b7 Income i
much more for poor voters than for wealthier voters. As household
 Central Family i þ b8 Formal Connection i income increases, the positive effect of centrality attenuates. One
þ C0Controls i þ  i
7
Fig. A1 in the Supplemental Information (SI) illustrates the marginal effect of
Models 2 and 4 in Table 2 display the results. As expected, centrality income on targeting across the range of betweenness centrality, using the results
(specifically betweenness centrality) is positively associated with from Models 2 and 4, respectively. Among the central voters, the poor are more likely
being targeted. The coefficient on income remains negative, but to be targeted.

6
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

might wonder whether this relationship is driven by differences voters which, if any, of several types of actors could learn how they
between rich and poor voters in the extent to which they are central voted. For purposes of this analysis we analyze respondent views
in family networks. For example, perhaps wealthier voters are more about three different sets of actors: brokers (termed liders in the
likely to have immigrated to the city from other areas, and are less Philipines), family members, or other actors. We estimate the fol-
connected to local family networks. However, this does not seem lowing model. We expect a negative sign for Income_i, a positive
to be the case. With a correlation coefficient of 0.018 there appears sign for Central_Family_i, and a positive sign for the interaction
to be no significant association between income and network term.
centrality.
Monitored i ¼ b10 þ b11 Income i þ b12 Central Family i
4.3. Do social networks make it easier to monitor and enforce vote þ b13 Income i  Central Family i
buying among the poor? þ b14 Formal Connection i þ b15 Income i

We have thus far demonstrated that: (1) poor voters are more  Formal Connection i þ C0Controls i þ  i
likely to be targeted, and (2) that poor voters that are more
centrally-located in informal family networks are more likely to The results of the analysis reveal no clear support for the monitor-
be targeted. We now turn to the question of whether family net- ing hypothesis (Table 3). Income either has no significant effect or is
works are valuable to campaigns because they help facilitate mon- incorrectly signed in all of the models. Whether it be family mem-
itoring and enforcement. Are poor voters, particularly those who bers or non-family members, including brokers, central voters are
are centrally-located, easier to monitor and more vulnerable to no more or less likely to believe that others can learn how they
enforcement if they fail to reciprocate? We first consider the rela- voted than less central voters, and income has no effect on monitor-
tionship between poverty, network position, and a belief that ing beliefs, either on its own, or interacted with centrality, counter
others can learn how you cast your vote. As mentioned, we asked to our hypothesis.

Table 3
Do poor voters perceive brokers to be effective monitors?

Dependent variable (DV): Brokers knows Family knows Others knows

(1) (2) (3) (4) (5) (6)
Log HH monthly income (HMI) 0.009 0.009 0.030 0.066 0.007 0.016
(0.007) (0.009) (0.024) (0.048) (0.016) (0.020)
Betweenness centrality (BC) 0.000 0.003 0.007 0.049 0.001 0.007
(0.001) (0.004) (0.006) (0.039) (0.003) (0.016)
Formal network ties = 1 0.018 0.021 0.024 0.080 0.022 0.094
(0.016) (0.061) (0.047) (0.200) (0.019) (0.094)
HMI x BC 0.001 0.009 0.001
(0.001) (0.008) (0.003)
HMI x Formal network ties 0.008 0.011 0.015
(0.013) (0.041) (0.020)
Mean of DV among
voter respondents 0.01 0.37 0.05
Full set of controls YES YES YES YES YES YES
Adjusted R2 0.002 0.004 0.075 0.075 0.047 0.044
Number of observations 659 659 659 659 659 659

Notes: Unit of observation is voter. Robust standard errors clustered at the barangay level in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Barangay fixed effects, measure of
degree centrality, demographic controls, and ‘other controls’ included.

Table 4
Do poor voters perceive brokers to be effective enforcers?

Brokers can’t Will feel Will stop receiving Brokers

Dependent variable (DV): do anything ashamed future handouts will reprimand
(1) (2) (3) (4) (5) (6) (7) (8)
Log HH monthly income (HMI) 0.002 0.056 0.020 0.041 0.008 0.007 0.016 0.022
(0.023) (0.040) (0.017) (0.033) (0.014) (0.025) (0.015) (0.017)
Betweenness centrality (BC) 0.004 0.047* 0.004 0.010 0.002 0.011 0.003 0.012
(0.006) (0.027) (0.006) (0.019) (0.004) (0.017) (0.004) (0.018)
Formal network ties = 1 0.099* 0.493*** 0.014 0.147 0.031 0.168 0.008 0.447***
(0.054) (0.169) (0.036) (0.187) (0.024) (0.119) (0.021) (0.123)
HMI x BC 0.009* 0.003 0.002 0.003
(0.005) (0.004) (0.003) (0.004)
HMI x Formal network ties 0.082** 0.034 0.029 0.092***
(0.038) (0.040) (0.027) (0.025)
Mean of DV among
voter respondents 0.55 0.24 0.11 0.09
Full set of controls YES YES YES YES YES YES YES YES
Adjusted R2 0.039 0.044 0.118 0.116 0.106 0.104 0.101 0.111
Number of observations 659 659 659 659 659 659 659 659

Notes: Unit of observation is voter. Robust standard errors clustered at the barangay level in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Barangay fixed effects, measure of
degree centrality, demographic controls, and ‘other controls’ included.

7
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

Fig. 2. Do formal network ties weaken voter susceptibility to clientelism? (Voters believe brokers can’t do anything about defections.).
Note: The figure displays marginal effects from model 2 in Table 4. Poorer voters with access to formal networks are more likely to believe there will be no consequence for
not reciprocating. As income increases, the relationship between formal networks and punishment expectations attenuates.

What about enforcement? Are poor, central voters more likely mal networks are significantly more likely to believe that campaigns
to fear punishment (H4) compared to more wealthy, or less central can do nothing if they fail to reciprocate (Table 4, Model 1). In addi-
voters? Again, the answer appears to be no, counter to our expec- tion, there is evidence that access to alternative networks is particu-
tation, and to some findings elsewhere in the literature (Cruz, larly important for poorer voters. Model 2 in Table 4 includes the
2019).8 Table 4 repeats the analysis from Table 3, but with expecta- interaction of income and access to formal networks. The results
tions about different enforcement outcomes as the outcomes of indicate that poorer voters are less likely to believe they will not
interests. Specifically, we examine expectations about the following be punished for defecting compared to more wealthy voters, but
four enforcement outcomes: non action (brokers can’t do anything), membership in a formal organization attenuates that belief.10
feeling a sense of shame, losing access to future electoral handouts, Fig. 2 illustrates this with graphs of the marginal effects of formal
or other, more severe forms of punishment (namely, verbal repri- networks across the range of incomes. We can see that poor voters
mands, losing access to patronage, or violence and intimidation). with access to formal networks are more likely to believe there will
Across each of these outcomes neither centrality nor income, nor be no consequence for non-reciprocity. However, the relationship
their interaction, has any consistent effect on enforcement between formal networks and punishment expectations attenuates
expectations. as income increases.
The results from Model 8 in Table 4 tell a similar story. Access to
formal networks is associated with a reduced fear that campaigns
4.4. Do formal network ties weaken voter susceptibility to clientelism? might enact costly punishment as a result of a failure to reciprocate
where income is low. However, the negative impact of formal net-
Finally, we consider whether access to alternative networks works becomes smaller as incomes rise (Fig. 3).
affects voter beliefs about the likelihood of monitoring and
enforcement (H5). Recall that our measure of access to formal net-
works is operationalized as whether or not respondents belong to 5. Discussion
one or more formal organizations. First, we note that being part of
a formal network does not affect the likelihood that a respondent is So what have we learned? As we expected, and consistent with
targeted with an electoral handout in the first place—none of the the broader literature, we find that campaigns are more likely to
coefficients for the formal networks variable approach statistical target poor voters with electoral handouts (H1). We also find that
significance (Table 2). Access to formal networks, whether inde- all poor voters are not created equal in the eyes of candidates and
pendently, or interacted with centrality, also has no effect on voter brokers. Poor voters who are centrally located within informal
beliefs about monitoring, as we can see in Table 3.9 However, the family networks make especially appealing targets, as hypothe-
access to alternative networks does appear to matter when it comes sized (H2). (For wealthier voters, their location within family net-
to beliefs about enforcement. Those respondents who are part of for- works has no bearing on their likelihood of being targeted).
However, poor and centrally located voters are not more likely to
8
It is worth nothing, however, that Ravanilla et al. (2021) demonstrate that these believe that they will be monitored or punished if they defected.
effects are conditional on the density of the village network, which we do not This is contrary to our expectations and to some previous findings,
examine in this paper.
9
In Table A2 in the Supplemental Information, we use the count of formal network
10
membership instead of an indicator for being a member of any formal network, and In Table A3 in the Supplemental Information, we again use the count of formal
the null results hold. network membership and the results are even stronger.

8
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

Fig. 3. Do formal network ties weaken voter susceptibility to clientelism? (Voters believe brokers won’t be able to punish them physically or verbally.).
Note: The figure displays marginal effects from model 8 in Table 4. Poorer voters with access to formal networks are less likely to fear that campaigns might enact costly
punishment for failure to reciprocate. As income increases, the relationship between formal networks and punishment expectations attenuates.

but not inconsistent with the much of the recent empirical work in Finally, our results support the idea that access to alternative
contemporary electoral clientelism (Hicken & Nathan, 2020). networks can reduce the risk that clientelism generates perverse
But if centrality does not seem to aid monitoring and enforce- incentives that undermine the agency of voters (Stokes, 2005).
ment, why are poor, central voters more intensively targeted by We find that access to formal networks reduces the fear of punish-
campaigns? In the theory section we laid out for reasons why such ment for the most vulnerable voters, consistent with our
voters are more likely to receive electoral handouts, including an hypothesis.
information advantage, better connections to potential patrons,
their appeal as potential catalysts for a social multiplier effect,
6. Conclusion
and their embeddedness in ongoing clientelist relationships. In
future work we hope to collect data to better distinguish between
The goal of this paper has been to critically examine the link
these mechanisms, but let us say a bit more about central voters
between social networks, poverty, and electoral clientelism, using
and the relationship between ongoing and electoral clientelism.
unique data from the Philippines. We examined the role that
The targeting of central voters may be partially an artifact of the
social networks play in shaping which voters are targeted with
way that electoral machines are constructed in the Philippines.
handouts and the likelihood that voters then reciprocate. Like
Social networks help voters gain access to patronage, and not just
numerous studies before, we find that campaigns prefer to target
during election season. It is common for local politicians in the
poorer voters, but not all poor voters have an equal probability of
Philippines, for example, to have staff dedicated year-round to
receiving a handout. Being centrally located within informal fam-
receiving, reviewing, and selectively responding to requests from
ily networks is associated with more targeting for poorer voters,
voters. When electoral mobilization machines are created in the
while network centrality has no effect on more wealthy voters.
Philippines, the past recipients of patronage are generally the first
This is the first major take-away from the paper. What explains
in line to receive electoral handouts because campaigns believe
why being embedded in informal networks facilitates the target-
those voters are more likely to be loyal (Aspinall et al.,
ing of poorer voters? One possibility, often discussed in the exist-
Forthcoming). If centrally-located voters have better access to
ing literature, is that social networks lower the cost of monitoring
patronage, they should also be more likely to appear on brokers’
and enforcement, and hence, campaigns and their brokers prefer
lists for electoral handouts.
to target individuals who are central in those networks. However,
Cruz (2019) offers another possible explanation, arguing that
the results of our analysis are inconsistent with the monitoring
social networks allow candidates to engage in collective monitoring
and enforcement story. Brokers rely on social networks to target
and enforcement,11 encouraging voters to weigh the ‘‘group-level
poorer voters, but not always because these networks enable bet-
ramifications of falling out of favor with politicians” (p. 8). Thus, cen-
ter monitoring and enforcement. This is the second takeaway
tral voters are targeted because they are part of networks that are
from the paper.
easier to collectively monitor, and because they are more likely to
A third contribution of the paper is our exploration of the ways
feel a responsibility for that welfare of that network’s members.
in which access to alternative social networks impacts the dynam-
Future work could do more to disentangle the perceptions of
ics of electoral clientelism. We find that while access to formal net-
individual- vs. group-focused monitoring and enforcement.
works does not make it more likely that a voter will be targeted or
feel monitored, access to these alternative networks does appear to
11
For more on collective monitoring see Larreguy, Marshall, and Querubin (2016) lower the perceived risk of defecting for voters. This is especially
and Rueda (2017). true for poor voters. Poor voters who are part of formal networks
9
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

are more likely to believe they will not be punished if they fail to Calvo, Ernesto, & Murillo, Maria Victoria (2004). Who Delivers? Partisan Clients in
the Argentine Electoral Market. American Journal of Political Science, 48(4),
reciprocate compared to those who do not have access to formal
742–757.
alternatives to family networks. Campos, Jose Edgardo. N.d. Governance gone local: Does decentralization improve
Finally, our paper has interesting policy implications for those accountability?. Forthcoming.
concerned with mitigating the effects of electoral clientelism, par- Carlin, Ryan, Singer, Matthew, & Zechmeister, Elizabeth (2015). The Latin American
Voter: Pursuing Representation and Accountability in Challenging Contexts. Ann
ticularly when combined with other findings in the literature Arbor, Michigan: University of Michigan Press.
regarding the limits of anti-vote buying campaigns (Hicken, Carlson, Elizabeth (2015). Ethnic Voting and Accountability in Africa: A Choice
Leider, Ravanilla, & Yang, 2015). In addition to trying to shame vot- Experiment in Uganda. World Politics, 67(2), 353–385.
Corstange, Daniel (2009). Sensitive Questions, Truthful Answers? Modeling the List
ers into refusing money (or taking the money and voting their con- Experiment with LISTIT. Political Analysis, 17(1), 45–63.
science), or trying to crackdown on the use of electoral handouts Cruz, Cesi (2019). Social Networks and the Targeting of Vote Buying. Comparative
by campaigns, our findings suggest policymakers might wish to Political Studies, 52, 382–411.
Cruz, Cesi, Labonne, Julian, & Querubin, Pablo (2017). Politician Family Networks
consider two other sets of reforms. First, as others have pointed and Electoral Outcomes: Evidence from the Philippines. American Economic
out, policy reforms that reduce the vulnerability of voters, particu- Review, 107(10), 3006–3037.
larly the poorest, and take discretion for resource distribution out Cruz, Cesi, Labonne, Julien, & Querubin, Pablo (2018). Social Fragmentation,
Electoral Competition, and Public Goods Provision. American Political Science
of the hands of politicians and their agents, can help free voters Review.
from the need to appeal to politicians in the first place (see also Desposato, Scott W. 2007. How does vote buying shape the legislative arena.
Bobonis, Gertler, Gonzalez-Navarro, & Nichter, 2017). Second, our Elections for sale: The causes and consequences of vote buying pp. 101–22.
Dixit, Avinash, & Londregan, John (1996). The determinants of success of special
findings suggest that the creation of and investment in formal
interests in redistributive politics. Journal of Politics, 58(4), 1132–1155.
local-level civil society institutions could be an antidote to some Finan, Frederico, & Schechter, Laura (2012). Vote-Buying and Reciprocity.
of the more pernicious consequences of electoral clientelism. Econometrica, 80(2), 863–881.
Hicken, Allen, Aspinall, Edward, & Weiss, Meredith (2019). Electoral Dynamics in the
Philippines: Money Politics, Patronage and Clientelism and the Grassroots.
Singapore: National University of Singapore Press.
Declaration of Competing Interest
Hicken, Allen, & Nathan, Noah L. (2020). Clientelism’s Red Herrings: Dead Ends and
New Directions in the Study of Nonprogrammatic Politics. Annual Review of
The authors declare that they have no known competing finan- Political Science, 23(1), 277–294. https://doi.org/10.1146/annurev-polisci-
cial interests or personal relationships that could have appeared 050718-032657.
Hicken, Allen, Leider, Stephen, Ravanilla, Nico, & Yang, Dean (2015). Measuring
to influence the work reported in this paper. Vote-Selling: Field Evidence from the Philippines. American Economic Review,
105, 352–356.
Kitschelt, Herbert (2000). Citizens, politicians, and party cartellization: Political
Acknowledgements representation and state failure in post-industrial democracies. European
journal of political research, 37(2), 149–179.
Kitschelt, Herbert, & Wilkinson, Steven (2007). Citizen-Politician Linkages: An
We are grateful for Michael Davidson for excellent research Introduction. In Herbert Kitschelt & Steven Wilkinson (Eds.), Patrons, Clients and
assistance, and Grace Labalan for unparalleled field research man- Policies: Patterns of Democratic Accountability and Political Competition. New
agement. We thank those who provided advice and comments on York: Cambridge University Press.
Kramon, Eric (2017). Money for votes: The causes and consequences of electoral
this project, including Dotan Haim, Stephan Haggard, Ellen Lust, clientelism in Africa. New York: Cambridge University Press.
Krislert Samphantharak, and participants in UCSD-GPS Junior Fac- Larreguy, Horacio Alejandro. 2012. Monitoring political brokers: Evidence from
ulty Brown Bag and UNU-WIDER’s Clientelistic Politics and Eco- clientelistic networks in Mexico. In EPSA 2013 Annual General Conference
Paper. Vol. 655.
nomic Development Workshop. This paper was supported from Larreguy, Horacio, Marshall, John, & Querubin, Pablo (2016). Parties, Brokers, and
Discovery Project grant from the Australian Research Council Voter Mobilization: How Turnout Buying Depends Upon the Party’s Capacity to
(DP140103114), supplemented with grants from the Australian Monitor Brokers. American Political Science Review, 110(1), 160–179.
Larson, Jennifer M. (2017). Networks and interethnic cooperation. The Journal of
Department of Foreign Affairs and Trade and De La Salle University.
Politics, 79(2), 546–559.
This study received ethical approval from the UM-IRB McClurg, Scott D. (2003). Social networks and political participation: The role of
(HUM00111251). Declarations of interest: none. social interaction in explaining political participation. Political research
quarterly, 56(4), 449–464.
Nathan, Noah L. (2019). Electoral Politics and Africa’s Urban Transition: Class and
Ethnicity in Ghana. New York: Cambridge University Press.
Appendix A. Supplementary data Nichter, Simeon (2008). Vote Buying or Turnout Buying? Machine Politics and the
Secret Ballot. American Political Science Review, 102, 19–31.
Supplementary data associated with this article can be found, in Nichter, Simeon (2018). Votes for Survival: Relational Clientelism in Latin America.
New York: Cambridge University Press.
the online version, athttps://doi.org/10.1016/j.worlddev.2022. Nickerson, David W. (2008). Is voting contagious? Evidence from two field
106128. experiments. American political Science review, 102(1), 49–57.
Pulse Asia Research. 2019. Ulat ng Bayan June 2019 survey.
Ravanilla, Nico, Haim, Dotan, & Hicken, Allen (2021). Brokers, Social Networks,
References Reciprocity, and Clientelism. American Journal of Political Science.. URL: https://
onlinelibrary.wiley.com/doi/abs/10.1111/ajps.12604.
Ravanilla, Nico, Davidson, Michael, & Hicken, Allen (2022). Voting in Clientelistic
Adida, Claire (2015). Do African Voters Favor Coethnics? Evidence from a Survey
Social Networks: Evidence From the Philippines*. Comparative Political Studies,
Experiment in Benin. Journal of Experimental Political Science, 2(1), 1–11.
55(10), 1663–1697. https://doi.org/10.1177/00104140211060275.
Advani, Arun, & Malde, Bansi (2018). Credibly identifying social effects: Accounting
Rueda, Miguel R. (2017). Small Aggregates, Big Manipulation: Vote Buying
for network formation and measurement error. Journal of Economic Surveys. 32
Enforcement and Collective Monitoring. American Journal of Political Science,
(–):1016–1044.
61(1), 163–177.
Aspinall, Edward, Meredith Weiss, Allen Hicken and Paul Hutchcroft. Forthcoming.
Schady, Norbert R. (2000). The political economy of expenditures by the Peruvian
Mobilizing for Elections: Patronage and Political Machines in Southeast Asia.
Social Fund (FONCODES), 1991–95. American political Science review, 94(2),
Cambridge University Press.
289–304.
Aspinall, Edward, Dettman, Sebastian, & Warburton, Eve (2011). When Religion
Schaffer, Joby, & Baker, Andy (2015). Clientelism as Persuasion-Buying: Evidence
Trumps Ethnicity: A Regional Election Case Study from Indonesia. South East
From Latin America. Comparative Political Studies, 48(9), 1093–1126. https://doi.
Asian Research, 19(1), 27–58.
org/10.1177/0010414015574881.
Bardhan, Pranab and Dilip Mookherjee. 2012. Political clientelism and capture:
Scott, J. C. (1977). The Moral Economy of the Peasant: Rebellion and Subsistence in
Theory and evidence from West Bengal.
Southeast Asia. Yale University Press. URL:.
Bobonis, Gustavo J, Gertler, Paul, Gonzalez-Navarro, Marco, & Nichter, Simeon
Stithorn, Thananithichot (2012). Political engagement and participation of Thai
(2017). Vulnerability and clientelism. National Bureau of Economic Research.
citizens: the rural–urban disparity. Contemporary Politics, 18(1), 87–108.
Technical report.
Stokes, Susan (2005). Perverse accountability: A formal model of machine politics
Brusco, Valeria, Nazareno, Marcelo, & Stokes, Susan (2004). Vote Buying in
with evidence from Argentina. American Political Science Review, 99(3), 315–325.
Argentina. Latin American Research Review, 39, 66–88.

10
N. Ravanilla and A. Hicken World Development 162 (2023) 106128

Stokes, Susan, Dunning, Th. ad., Nazareno, Marcelo, & Brusco, Valeria (2013). Wasserman, Stanley, & Faust, Katherine (1994). Social network analysis: Methods and
Brokers, Voters, and Clientelism: The Puzzle of Distributive Politics. Cambridge, applications. Cambridge University Press.
United Kingdom: Cambridge University Press. Weitz-Shapiro, Rebecca (2014). Curbing clientelism in Argentina: Politics, poverty, and
Szwarcberg, Mariela (2012). Revisiting clientelism: A network analysis of problem- social policy. Cambridge University Press.
solving networks in Argentina. Social Networks, 34(2), 230–240.

11