Social Network Structures and the Politics of Public Goods Provision:

Evidence from the Philippines*

Cesi Cruz†
Julien Labonne‡
Pablo Querubin§

May 2019

Abstract

We study the relationship between social structure and political incentives for public goods
provision. We argue that when politicians–rather than communities–are responsible for the
provision of public goods, social fractionalization may decrease the risk of elite capture and lead
to increased public goods provision and electoral competition. We test this using large-scale data
on family networks from over 20 million individuals in 15,000 villages of the Philippines. We
take advantage of naming conventions to assess intermarriage links between families and use
community detection algorithms to identify the relevant clans in those villages. We show that
there is more public goods provision and political competition in villages with more fragmented
social networks, a result that is robust to controlling for a large number of village characteristics
and to alternative estimation techniques.

*A previous version of this paper circulated under the title Village Social Network Structures and Electoral Competition.
We thank Klaus Desmet, Nick Eubank, James Fowler, Yana Gorokhovskaia, Matias Iaryczower, Steph Haggard, Alex
Hughes, Franziska Keller, Stuti Khemani, Alexander Kustov, Anja Prummer, Sarah Shair-Rosenfeld, Jake Shapiro,
and Francesco Trebbi for helpful suggestions. We are also grateful to seminar and workshop participants at ASOG,
Barcelona Summer Forum, Barcelona’s Workshop on the Political Economy of Development and Conflict, Emory, LSE-
NYU Conference, Oslo, Princeton Political Economy Workshop, Ryerson University, University of Virginia, University
of British Columbia, University of California, San Diego, University of Pennsylvania, University of the Philippines
School of Economics, Warwick, and World Bank ABCDE for feedback. All remaining errors are ours.
†
University of British Columbia: [email protected]
‡
University of Oxford: [email protected]
§
New York University: [email protected]
 The campaign (...) is the culmination of a
contest to see who makes best use of the
social structure.

Mary Hollsteiner, The Dynamics of Power

1 Introduction

Political institutions are built on existing social structures, with their own hierarchies, rules, and

norms governing interactions and behavior. The existing literature often focuses on ethnic and

religious cleavages and associates social cohesion with a host of positive outcomes such as greater

public goods provision, arguing that fractionalization undermines communities’ collective action

and aggregation of preferences. However, in most developing countries, politicians (and not

communities) are responsible for the provision of public goods which are funded with transfers

(rather than local taxes). In these contexts, the implication of social fractionalization for collective

action and preference aggregation may be less relevant for understanding public goods provision.

Thus, it is essential to understand how social structures shape the incentives of local politicians to

allocate resources across communities.

We offer a new framework for theorizing about the link between social structure – the configu-

ration and relative position of social groups – and subsequent electoral strategies for redistribution.

We argue that in weakly institutionalized and clientelistic democracies, social cohesion can also

facilitate elite capture, by concentrating political influence in a handful of leaders who can help

provide votes in exchange for private transfers. Consequently, if society is divided into multiple

relevant social groups, the subsequent redistributive strategies adopted by politicians to attract

voters may shift towards greater–rather than lower–public goods provision.

We study these questions in the context of local politics in the Philippines. The provision of

public goods in the Philippines is partly the responsibility of elected municipal mayors whose

budgets depend mostly on transfers from the central government. Mayors must allocate their

budget across the different barangays (villages) that compose the municipality. A critical actor in

Filipino social and political life are clans or extended families: political alliances at the local level

often involve securing the support of clan leaders who can leverage strong norms of in-group

1
cooperation and reciprocity to deliver the votes of their family members and other members of the

community (Hollnsteiner, 1963; Timberman, 1991; Fegan, 2009).1

We demonstrate that mayor decisions to provide public goods across the different villages that

comprise their municipality vary as a function of the social structure of the village. To do this,

we use a unique dataset covering 20 million individuals in more than 15,000 villages across 709

municipalities of the Philippines. The dataset includes information on family names and we use

naming conventions in the Philippines to establish ties between families through intermarriages.

Following Padgett and McLean (2006) and Cruz, Labonne and Querubin (2017), we consider a

tie between two families to exist whenever we observe at least one marriage between members

of the two families. We graph the full family network in all villages in our sample. We then use

community detection algorithms (Girvan and Newman, 2002; Pons and Latapy, 2005) to identify

the configuration of clans in every village, based on the relative number of ties within and between

clusters of families in the network. To account for the relative influence of each clan, we create a

measure of social fractionalization with a Herfindahl index that captures how the village population

is distributed across the different clans. We hypothesize that in less fractionalized or more cohesive

villages, clan leaders are more effective in concentrating political influence and capturing politicians

for private, personal interests at the expense of public goods provision. Another implication of this

is that more fractionalized villages will also feature higher political competition since clan heads

will exert less persuasion and control over candidacy and the voting decisions of village residents.

Our results show that social fractionalization is positively correlated with village-level provision

of public goods such as schools, public marketplaces, water systems and health centers. These

correlations are sizable; for example, a one standard deviation increase in social fractionalization

increases the probability that a health center is located in the village by around 6 percentage points

(an increase in 10% relative to the sample mean). These correlations are robust to the inclusion

of municipality fixed effects and a broad set of village covariates such as village population, the

number of distinct families in the village, whether the village is classified as rural, as well as controls

for different occupations and educational levels, amongst others. To address concerns about reverse

causality (i.e. whether family ties, and thus our social fractionalization measures, are affected by

1
While there are important differences in ethnic composition and religious affiliation across regions and larger
administrative divisions, municipalities and villages in the Philippines are relatively homogeneous.

2
public goods provision) we restrict our network to ties between individuals 45 years old or older,

and use the social fractionalization measure in the restricted network - capturing marriages that

took place at least 20 to 25 years ago - as an instrument for the corresponding measure in the full

network. Our results are similar when we use this approach.

We also use two further approaches to establish that social fractionalization is correlated with

stronger political competition. First, we use data on local sources of political influence from an

original survey conducted in two provinces shortly after the 2013 election to show that, consistent

with our theory, less fractionalized villages are characterized by a greater concentration of political

influence. Second, we show that social fractionalization is associated with i) a larger number of

candidates running for office in village elections and ii) narrower vote margins for the winning

candidate in village and municipal elections.

Our results should be interpreted cautiously since naturally, social structure measures are not

randomly assigned across villages. While we control for several village-level characteristics that

may potentially confound our estimates, and follow an instrumental variables approach, we cannot

fully account for all variables that may have shaped intermarriage networks over many decades. We

show that fractionalization across clans is not correlated with preference heterogeneity or collective

action and thus rule out these variables as mediators in our context. While political competition

may partly mediate the effect of fractionalization on public goods (political competition and public

goods provision are indeed positively correlated in our context), it may also be an outcome and

there are other channels through which fractionalization may impact the redistributive strategies

of politicians. Our findings point to the importance of considering how social structure shapes the

concentration of power and the distributive strategies of politicians.

Much of the literature on public goods provision uses ethnic, linguistic, or religious fractional-

ization as a measure of social structure. This literature has shown mixed results, with many papers

showing a negative relationship between fractionalization and public goods provision across both

developed and developing countries (Easterly and Levine, 1997; Alesina, Baqir and Easterly, 1999;

Miguel and Gugerty, 2005), while others question these correlations on both methodological and

substantive grounds (Gao, 2016; Soifer, 2016; Wimmer, 2016; Kustov and Pardelli, 2018). We

contribute to this literature in several ways. First, we argue that when politicians, rather than com-

munities, are responsible for public goods provision, social cohesion may promote elite capture

3
and lead to the under-provision of public goods. Second, we focus on a different measure of frac-

tionalization based on a fundamental unit of social organization–the family–which may account for

social structure even in countries where ethnic and religious fractionalization are not as relevant.

We also add to the literature on local elite capture and the delivery of public goods (Bardhan,

2002). Closely related to our paper is the work by Acemoglu, Reed and Robinson (2014) who find

that places in Sierra Leone with more ruling families exhibit better development outcomes today.

In these places “chiefs constrained by greater competition will be less able to manipulate access to

land for their own benefit or will have to compete by offering and providing public goods” (p. 321).

We study how social structure can influence the distribution of power and the ability of key elite

actors (in our context, heads of large clans) to deliver and broker votes in exchange for targeted

transfers, at the expense of public goods.

Our paper is also related to a growing literature on the economic and political impact of families

and kinship ties (Todd, 1985; Padgett and McLean, 2006; Moscona, Nunn and Robinson, 2017). In

a closely related paper, Xu and Yao (2015) study the role of lineage groups on local governance

in China. However, to our knowledge, our paper is the first to use large-scale family networks

(based on the full sample of intermarriage ties for a large number of villages) to study how social

structures influence the incentives of politicians to provide public goods.

We also contribute to the empirical literature on the role of social networks on the distributive

strategies of politicians (Calvo and Murillo, 2009; Hughes, 2016; Larson and Lewis, 2017) and

other political outcomes such as turnout (Eubank et al., 2017). We build on Cruz, Labonne and

Querubin (2017) who show that a candidate’s centrality in family networks facilitates brokered

linkages with voters and contributes to higher vote shares during the elections. In this paper,

rather than exploiting the position of individual candidates in the network, we study how the

network structures of villages condition the distributive strategies of politicians.

Finally, we also contribute to the literature on social diversity and political competition. A

series of studies have documented that greater social diversity leads to a larger number of parties

and stronger political competition (Amorim-Neto and Cox, 1997; Potter, 2014; Lublin, 2017). This

is consistent with our finding that greater fractionalization is associated with a larger number of

candidates and less concentrated political influence.

4
2 Theory: Social Structure, Public Goods and Electoral Competition

Most existing theories on how social structures influence political and economic outcomes are based

on studies documenting a negative correlation between ethnic or religious fractionalization and

public goods provision (Easterly and Levine, 1997; Alesina, Baqir and Easterly, 1999). These theories

emphasize how fractionalization may undermine collective action and the aggregation of policy

preferences among citizens. These seem particularly relevant in contexts in which communities are

responsible for the provision of public goods, for example through the payment of taxes or other

contributions and must collectively agree on local priorities.

However, in most of the developing world, public goods are the responsibility of elected

politicians rather than of communities. These public goods are often funded with transfers from

the central government (not local taxes) and thus collective action and preference heterogeneity

may be less relevant. In these contexts it becomes essential to understand how social structure

shapes the distributive choices of politicians.2

We argue that in weakly institutionalized democracies, where politicians are responsible for

public service delivery, social cohesion can lead to an under provision of public goods. When

members in a society are concentrated in a relatively small number of politically relevant groups,

leaders of larger groups have high bargaining power and can demand private, targeted, excludable

transfers in exchange for the electoral support of its members. As a consequence, political influence-

– used here to refer to social persuasion and the ability to broker political exchange – becomes

more concentrated. This is especially important in clientelistic contexts where voters engage with

politicians through brokers and rely on information from friends, family, and neighbors in order

to make political decisions.3 In other words, social concentration increases the likelihood of elite

capture which can potentially undermine the incentives of politicians to provide public goods.

Social fractionalization can also increase the agency and transaction costs for politicians of

2
For example, Desmet, Ortuño-Ortín and Weber (2017) and Munshi and Rosenzweig (2018) study how diversity
affects elected politicians’ incentives to provide transfers vs. public goods.
3
The importance of local political influence is not limited to the developing world. For example, Katz and Lazarsfeld
(1955) show that individuals were most influenced by peers of similar social status. Similarly, research from surveys
in the United States identify discussion networks (Huckfeldt and Sprague, 1991) and social cohesion (Huckfeldt et al.,
1995) as important factors for the transmission of political information to citizens.

5
engaging in the clientelistic exchange of private transfers for votes. Consider a society in which

a large share of citizens belong to handful of groups; in this case politicians can secure a large

number of votes by brokering deals with a small number of influential leaders that they can more

easily monitor. As the number of groups increases and each group represents a smaller share of

the electorate, this electoral strategy becomes less attractive. Following Lizzeri and Persico (2004),

as society becomes more fragmented, the incentives for politicians to provide policies with diffuse

(as opposed to targeted and excludable) benefits increases, which can encourage the provision

of public goods. This argument is also related to Dahl’s theory on the benefits of pluralism and

diversity for the performance of democracy (Dahl, 1973).

Finally, another implication of our theory is that more fractionalized villages will also feature

higher political competition. First, vote buying may be a more appealing strategy in these villages

than providing public goods which will encourage clan leaders to mobilize voters in support of

one of the candidates leading to both high turnout and less competitive races). Second, even

absent vote-buying, the concentration of political influence implies that clan leaders can influence

and mobilize a large set of voters over whom they exert influence in support of their preferred

candidate. As a result we expect to observe a broader set of individuals running for public office

and tighter races. In the next section we illustrate some of these ideas in the Philippine context.

2.1 Clans and Elections in the Philippines

Local democracy in the Philippines is vibrant and highly relevant for studies of public service

delivery (Rogers, 2004; Abinales and Amoroso, 2017). The country is divided into roughly 1,600

cities and municipalities which are themselves divided into over 42,000 barangays (villages). Mu-

nicipalities are governed by a mayor, a vice-mayor and eight municipal councilors. All municipal

officials are elected in first-past-the-post elections organized, by law, at fixed intervals of three

years. Political parties tend to be weak and unstable, and there are typically large shifts in party

affiliations after each election (Hutchcroft and Rocamora, 2003; Mendoza, Cruz and Yap, 2014).

Every three years each barangay also elects a barangay captain (village head) and a barangay

council. These are responsible for the maintenance of public goods and assisting the mayor with

the implementation of municipal programs.

The 1991 Local Government Code devolved significant responsibilities for the delivery of a

6
number of social services to municipalities, including primary health care programs, repair and

maintenance of local infrastructure, and provision of agricultural, fishery, mines, and geoscience

services (Azfar et al., 2000). Municipalities are expected to finance these services through yearly

transfers from the central government, known as the Internal Revenue Allotment (IRA) (Llanto,

2012). While municipalities can also raise their own revenues through local taxes and fees, the IRA

provides 85 percent of their budgets on average (Troland, 2016).4

The mayor, as the chief executive of the municipal government, enjoys significant responsibili-

ties and discretionary powers. Even in sectors with national-level programs such as education and

health, the Local Government Code devolved responsibility for many client-facing services to the

municipalities, as well as significant fiscal and regulatory functions (Llanto, 2012; Capuno, 2012).

This is consistent with Rogers’s (2004) characterization that for the day-to-day life of Filipinos,

“government in every practical sense means local government.” Mayors play an important role

in deciding how to allocate the budget across various sectors and the different barangays that

compose the municipality (Hutchcroft, 2012; Capuno, 2012).

At the same time, service delivery and public goods provision remains a challenge. For example,

according to the 2010 Population Census, about 32 percent of villages do not have a health center

and close to 40 percent do not have modern water and sanitation systems. Not surprisingly,

health outcomes are also lacking. Under 5 infant mortality is about 28 per 1,000 live-births, higher

than neighboring countries with lower income levels like Vietnam. About 33 percent of children

under 5 are stunted, a rate similar to that of poorer countries in the region like Cambodia and

Myanmar. This suggests that under provision of public goods is not only due to low income or

lack of resources. In addition, the average municipality only spends 90 percent of its budget every

year.

These challenges are largely attributed to electoral incentives that center on clientelism rather

than public goods provision (Timberman, 1991; Azfar et al., 2000; Khemani, 2015). This includes

providing jobs (Lande, 1964; Fafchamps and Labonne, 2017), money (Khemani, 2015; Mendoza

et al., 2016; Cruz, 2019), and other private goods and services (Hutchcroft and Rocamora, 2003)

4
These funds are disbursed in tranches, without conditions; while technically there is a requirement that at least 20%
of funding should go towards development projects, the so-called Local Development Fund (LDF), but it is not enforced
in practice (Troland, 2016).

7
in exchange for political support. For example, surveys carried out in both urban and rural

communities after the 2010 and 2016 elections suggest that around 30 percent of voters were offered

money for their vote (Mendoza et al., 2016; Cruz, 2019). While clientelism can take different forms,

the logistical requirements are substantial: the identification of clients and the delivery of benefits

requires sophisticated networks to monitor actors and manage exchange relations (Kitschelt and

Wilkinson, 2007).5 Consequently, the practice of clientelism is predicated on social norms of

reciprocal exchanges and obligations, which in the Philippines is reinforced by family ties (Corpuz,

1965; Enriquez, 1986; Miralao, 1997; Covar, 1998).

Taken together, it becomes easier to understand why Philippine political culture revolves around

families and clans (Lande, 1964; Hutchcroft and Rocamora, 2003; McCoy, 2009; Abinales and

Amoroso, 2017). Sidel (1999) attributes the importance of families in politics to the overlaying of

democracy on the emerging socioeconomic landscape of hacienda-based clans and business elites.6

There is even a commonly used Tagalog phrase that highlights the linkages between families and

clientelism: kasal, binyag, libing, which literally means “weddings, baptisms, funerals." This well-

known term for clientelism refers to the fact that politicians are involved even in intimate family

events, serving as godfathers for baptisms, sponsors for weddings, and contributing towards

funeral costs.

Consequently, politicians competing in municipal and barangay elections must often seek the

support of extended families, or clans (Timberman, 1991; Abinales and Amoroso, 2017). Fegan

(2009) argues that families are key political actors because their reputation, loyalties, and alliances

are transferable from members who die or retire to the younger generations. One example is

the norm of utang na loob (literally, “inner debt”), which refers to a debt of gratitude that fosters

reciprocity and feelings of social obligation. As stated by Hollnsteiner (1963), “keeping with the

highly familistic orientation of Philippine society whereby an individual represents his family, utang

5
Even if vote buying is conducted as a one-off transaction, it still requires the ability to identify and contact voters,
and in the context of a secret ballot, the means to ensure with reasonable certainty that voters will keep their end of the
bargain once they are in the voting booth (Cruz, 2019).
6
While families are perhaps the most common example of groups that can coordinate votes, they are certainly not
the only ones: some religious and regional groups also use their ability to coordinate and enforce behavior to engage in
group-level political exchange with politicians. A notable example is the religious group Iglesia ni Cristo, which endorses
a slate of candidates that adherents are expected to vote for.

8
na loob is not limited to an individual-to-individual relationship but is rather seen as operative from

family to family” (p.79). While we do not want to overstate the cultural basis for clans as political

units, even conceptions of Filipino culture that would consider utang na loob as just a small part

of a broader value framework would still emphasize the relational basis of Filipino values and

the importance of the family unit for understanding politics (Enriquez, 1986; Aquino, 2004; Reyes,

2015). For example, Timberman (1991) cites the norms of utang na loob and pakikisama (ability

to engage with others) as contributing to a personalistic and patronage-based political culture.

Furthermore, while adherence to these traditional norms (as well as their social relevance) may

decline over time, a recent study of Filipino adolescents conducted by Clemente et al. (2008) show

that the relational values identified by Enriquez (1986) – utang na loob, pakikisama, and hiya (shame

as a result of unfulfilled social obligations) – were still considered to be important.

An implication of these features of Philippine society is that politicians can often secure a large

number of votes by brokering deals with the heads of clans who can commit to deliver votes

from their clan members in exchange for access to private transfers and services. These private

transfers often come at the expense of the provision of public goods that would benefit all village

residents equally. Barangay captains and councilors often operate as brokers between municipal

candidates and local clans and must monitor that clans vote as promised and that resources flow

to the families. While in some instances clan leaders can only imperfectly command the votes

of their clan, the shared expectations and norms of reciprocity among clan members give them

organizational advantages for engaging in political exchange.

3 Identifying Clans and Measuring Fractionalization

3.1 Clans and Communities

An empirical challenge in our context is characterizing the social structure in every village. Key to

our analysis is determining the number of politically relevant clans in each village. Theoretically,

for our purposes a clan is a set of families: (i) connected to each other by kinship or marriage;

and (ii) where relationships of exchange among members are governed by well-established norms

of cooperation and reciprocity. In other words, they are the set of individuals for whom existing

familial ties can facilitate coordination of votes for the politician that provides patronage or transfers

9
to the clan leader. While individuals in the Philippines can easily provide us with this information,

absent detailed village-level surveys it is very hard to collect this data at a large scale. We propose

to use social network analysis to address this issue, by identifying cohesive groups of families

in the inter-marriage network. Cohesive groups are those with many ties within the group and

relatively fewer ties to other groups.

Consider a social network in which a node is a family and edges between nodes imply that a

marriage has occurred between members of these families. An example is illustrated in panel a) of

Figure 1 that shows a network with 15 different families. This network features three components,

that is, groups within which nodes are path-connected, but disconnected from other sets of nodes

in the network (Jackson, 2010). One intuitive approach would be to identify each different clan with

the different components in the marriage network. This approach, while appealing, can be quite

restrictive in practice since family networks in real life (and in our Filipino context, in particular)

rarely feature neatly distinct components as those illustrated in panel a).

By contrast, the slightly modified network in panel b) of Figure 1 differs only from panel a) in

that we have added two additional edges (represented by dashed lines): one between families A

and K and another one between families I and M. The three distinct sets of nodes are still apparent,

but this modified network has only one component (the full network). Thus, an approach based

on components would identify only one clan in this network and all individuals would belong to

that clan. An alternative that considers both the distinct sets of nodes but also the additional links

among them is the concept of communities. In a social network, communities are groups of nodes

with dense connections internally (i.e. within the group) and sparser connections between groups

(Jackson, 2010). Intuitively, the social network in panel b) has three different communities even

though it has only one component.

Our approach is thus to use the different communities in the social networks as proxies for the

clans. At the same time, the community structure in a network is a latent feature that needs to be

uncovered; there are several potential ways to partition a network’s nodes into separate groups

that we describe in the following section.

10
 Panel a): Network with 3 components Panel b): Network with 1 component

Figure 1: Sample marriage networks. Nodes represent families and edges indicate a marriage between those
families

3.2 Community Detection

One approach to uncovering community structures in social networks is based on edge removal.

The intuition is as follows: if two groups of nodes are only loosely connected with each other, then

removing the edges between those two groups will generate components in the restricted network.

Communities correspond to those components in the restricted network. The networks in Figure 1

can be used to illustrate this approach. The two dashed edges in panel b) loosely connect groups of

nodes that are densely connected with each other. Removing those two edges will yield a restricted

network like the one illustrated in panel a) with three different components.

Approaches based on edge removal differ in terms of the selection rule regarding which edges

to remove. We follow an algorithm proposed by Girvan and Newman (2002) that consists in the

sequential removal of edges with high betweenness centrality. This centrality measure captures the

extent to which the edge serves as a link between different groups. It is calculated using the number

of shortest paths between nodes in the network that pass through that edge.7 For example, the

dashed edge between nodes J and E in Figure 2 has the highest betweenness centrality in that

network. Similarly, the dashed edges in panel b) of Figure 1 have high betweenness centrality.

7
For more information on calculating betweenness centrality, please refer to Appendix A.

11
 B
D

E A

F J

H
I

Figure 2: Edge with High Betweenness Centrality

The Girvan-Newman algorithm proceeds as follows:

1. Calculate betweenness for all edges in the network

2. Remove the edge with the highest betweenness

3. Recalculate betweenness for all edges affected by the removal

4. Repeat from step 2 until no edges remain

5. From resulting dendrogram, select the partition that maximizes network modularity

While in our baseline analysis we focus on communities identified by the Girvan-Newman

algorithm, for robustness we also implement the walktrap algorithm developed by Pons and

Latapy (2005). Intuitively, the algorithm relies on the idea that random walks on a graph tend to

get “trapped” into densely connected parts corresponding to communities. The algorithm thus

generates a large number of random walks and groups together nodes that are tied together through

those walks. See Pons and Latapy (2005) for more details.

3.3 Measuring fractionalization

The algorithm delivers a partition of C communities (indexed by c = 1, . . . , C), each containing

a share sc of distinct families. We then use this to compute our main independent variable, the

measure of social fractionalization (SF), using a standard Herfindahl-Hirschman index:

12
 C
X
SF = 1 − s2c
c=1

The measure can be interpreted as the probability that two randomly selected families are from

the different clans. We use this approach because we are interested in accounting for both the

overall configuration of clans in the village, as well as differences in relative size or strength among

clans.8

4 Data

In this section we present our various data sources and describe our main dependent and indepen-

dent variables.

4.1 Family networks

Our main independent variable is the extent to which a village’s population is fragmented into

several clans. To do this we identify network communities and use them as proxies for clans

in the family network of the village. To construct the family networks, we use data collected

for the National Household Targeting System for Poverty Reduction (NHTS-PR). This large-scale

household-survey, implemented between 2008 and 2010, reports several socio-economic charac-

teristics of the household as well as the gender, age, educational attainment and occupational

category of every household member. We focus on municipalities where full enumeration took

place.9 This leaves us with information on 20 million individuals in about 15,000 barangays in 709

municipalities.10 Importantly, we have access to the non-anonymized version of the dataset and

8
By contrast, other measures of social structure tend to focus on one or the other: for example, connectedness is
measured by density measures, while variance in connectedness among actors is measured using centralization measures
(essentially the distribution of centrality in the network).
9
Importantly, we do not use a sampled network to detect our communities and as such our analysis does not suffer
from the problems identified by Chandrasekhar and Lewis (2011). In the remaining municipalities, only households in
so-called pockets of poverty were interviewed.
10
As shown in Table A.1, villages in our sample are, on average, less populated and more rural than the national
average. There are no major differences in other socio-economic characteristics such as ethnic and religious fragmentation

13
have two family names (the middle and last name) for every individual.11

We are able to measure large scale family networks in the Philippines due to naming conventions

with two convenient features: (i) within a municipality, a shared family name implies family

connections; (ii) each individual carries two family names, which establishes that a marriage took

place between members of those two families.

More concretely, names in the Philippines have the following structure:

firstname midname lastname

where firstname corresponds to the individual’s given first name, midname corresponds to the

mother’s maiden name (for men and single women) or the father’s family name (for married

women) and lastname corresponds to the father’s family name (for men and single women) or the

husband’s family name (for married women).

The naming structure and distribution of family names in the Philippines can be traced back to

the 19th century. In 1849, concerned with the arbitrary way in which Filipinos chose their surnames

and the implications for tax collection, Governor Narciso Claveria y Zaldua created a catalog with

a list of 61,000 different surnames. Municipal officials throughout the country then assigned a

different name to each family. Since then, names have been transmitted through generations

according to well-established and enforced naming conventions. As a consequence very common

family names are not as prevalent in the Philippines as in other countries and thus, sharing a

family name is very strongly correlated with an actual family tie. This is especially the case within

municipalities and villages.

Given the full names of all individuals in an area, we are able to reconstruct all of the edges

in the family network by examining the joint occurrences of middle and last names.12 As noted

above, each individual maintains two family names: their father’s name and either their mother’s

maiden name or their husband’s name, in the case of married women. Thus each individual’s set

of family names indicates an intermarriage between the two families–either in their generation (in

the case of married women) or their parents’ generation (in the case of men and single women). As

or public goods provision. Our sample is representative of rural villages in the Philippines.
11
Fernandez (2012) describes the data in more detail.
12
A similar method is used by Cruz, Labonne and Querubin (2017), Haim (2018) and Rubin (2018).

14
a result, we are able to observe ties between families merely by the occurrence of the family names

within an individual.

For example, Figure 3 below depicts the family network that can be drawn from a list of

relatives of the previous Philippine President, Benigno Cojuangco Aquino. His middle name is his

mother’s maiden name, Cojuangco, and his last name is his father’s last name, Aquino, implying a

marriage tie between the Cojuangco and Aquino families. Similarly, we can show ties between the

Aquino, Abellada, and Aguirre families through the names of his sister Aurora Aquino Abellada

and cousin Bam Aguirre Aquino. On the Cojuangco side, we can show ties to the Sumulong

and Teodoro families through the names of his cousin Gilberto Cojuangco Teodoro and uncle Jose

Sumulong Cojuangco, as well as an indirect tie to the Prieto family through Gilberto’s wife Monica

Prieto Teodoro.

Aguirre
Sumulong

Cojuangco

Aquino

Abellada

Teodoro

Prieto

Figure 3: Family network for selected members of former President Aquino’s family.

Once the networks are constructed, we implement the Girvan-Newman and walktrap algo-

rithms and as discussed in Section 3, we then compute our main independent variable, the measure

of social fractionalization (SF).

Table A.2 presents some descriptive statistics on the number of communities and fractionaliza-

tion measures. The average (median) number of communities per village is 46 (34). However, the

two largest communities in the village contain, on average, close to 25% of the village population.

While more populated villages exhibit higher fractionalization, there is substantial variation in

fractionalization across villages in all terciles of the population distribution. Below we show that

our estimates are robust to controlling for village population and to dropping villages with extreme

fractionalization and population values.

15
4.2 Outcome Variables

4.2.1 Public Goods

For our main outcome, we use data from the 2010 population census that lists public goods available

in every barangay. We use this to code dummies for whether the barangay has an elementary school,

a high school, a public market, a health center or a community water system. Since our indicators

all capture the same concept, to address the multiple comparisons problem, we combine them into

a public goods index using the inverse covariance weighting approach proposed by Anderson

(2008).13 Table A.2 provides some descriptive statistics for our different outcome variables.

4.2.2 Preference Heterogeneity and Collective Action

We use data from from a survey conducted shortly after the 2013 local elections to examine the

extent to which social fractionalization is correlated with heterogeneity in preferences over public

goods and collective action.14 Previous studies argue that these variables play an important role in

understanding the effect of fractionalization on public goods provision.

More specifically respondents were asked about their preferred allocation of the municipality’s

Local Development Fund (LDF) across 10 different sectors.15 As a measure of heterogeneity in

preferences, we simply take the standard deviation in respondents preferred allocation for each

budget item.

To measure social capital and collective action we simply use dummy variables for whether

the household participated in any formal group (such as unions, farmer’s or other professional

associations, community development associations, micro finance groups and cooperatives) or in

communal voluntary work activities (known locally as bayanihan) and average those over all village

13
The computations are described in Appendix A.1.3.
14
The survey covers 3,408 households in 284 villages in 12 municipalities in the provinces of Ilocos Norte and Ilocos
Sur.
15
The 10 sectors on which we have data are: public health services, public education services, cash or in-kind transfers
(such as loans or job assistance), water and sanitation services, road construction and rehabilitation, construction of
community facilities (such as multipurpose halls or basketball courts), business loans and other private economic
development programs, agricultural assistance and irrigation systems, peace and security and community events and
festivals.

16
respondents.

4.2.3 Political Competition

To examine the correlation between social fractionalization and political competition we use elec-

toral outcomes from the 2010 municipal elections and the 2010 and 2013 barangay elections collected

from the Commission of Elections (COMELEC) website. For municipal elections, we have precinct-

level data on the number of registered voters, the number of individuals who voted and the number

of votes received by each mayoral candidate.16 For barangay-level elections, we have village-level

data on the votes obtained by every candidate for barangay head (punong barangay) and for the

barangay council (barangay kagawad).

Our main political competition variables are the win margin (vote share of the candidates that

received the most votes in that precinct minus vote share of the runner-up in that precinct) and the

number of candidates running in the race. We also use indices of effective number of candidates,

proposed by Laakso and Taagepera (1979) and Golosov (2010) and described in the Appendix. As

above, we combine these measures into a single political competition index following Anderson

(2008).

To capture the effects on elite capture–the concentration of political influence– we also use the

2013 survey. Respondents were asked to “name five individuals living in the barangay, but not

living in your household, whose opinions you respect the most when it comes to politics.” This

allows us to test whether social fractionalization affects the overall number of influential leaders

that villagers mention in their responses.

5 Empirical Analysis and Results

Our main analysis consists of village-level cross-sectional regressions between public goods and

political competition outcomes and our index of social fractionalization. More concretely, we

estimate OLS regressions of the form:

16
Importantly, in the Philippines there is at least one precinct per village, so we can perfectly merge our electoral,
public goods and social networks data.

17
 yvm = α + βSFvm + γXvm + δm + vm (1)

where yvm is the outcome variable in village v in municipality m, SFvm is our measure of social

fractionalization, Xvm corresponds to a full set of village covariates and δm is a full set of municipality

fixed-effects. Standard errors are clustered at the municipality level.

The inclusion of municipality fixed effects is important in this context since they allow us to

absorb all municipality-specific characteristics that may be correlated with both social fractionaliza-

tion and our different outcome variables. Moreover, we are interested in how mayors adjust their

electoral and distributive strategies across the different villages in their municipality as a function

of the level of social fractionalization. Thus, we are interested in exploiting within-municipality

variation. For ease of interpretation, in all regressions we include a standardized version (mean

zero, standard deviation one) of the social fractionalization index.

5.1 Public Goods

We begin by estimating the correlation between social fractionalization and the public goods

index, as well as dummies for the presence of each of the different public goods in the village.

Estimates of β are reported in Panel A of Table 1. The coefficients reveal a positive and statistically

significant correlation between social fractionalization and public goods provision. For example,

a one-standard deviation increase in social fractionalization is associated with an increase of 0.21

standard deviations in the public goods index. Looking at the individual public good dummies,

a one-standard deviation increase in social fractionalization is associated with an increase in 8

percentage points in the likelihood of having a secondary school in the village, and a 6 percentage

point increase in the likelihood of having a public market or a health center in the village. Some

of these estimates are sizable; relative to the mean they correspond to an increase of 40% for

high schools, 30% for public markets and 10% for health centers and waterworks. However,

these coefficients must be interpreted very cautiously since social fractionalization is not randomly

assigned across villages. While municipality fixed effects account for municipal-level confounders,

omitted variable bias remains a concern since other village characteristics may be correlated with

social fractionalization and public goods provision. For example, larger, heavily populated, urban

18
villages that feature higher social fractionalization may be more likely to have public goods. It may

also be the case that wealthier villages feature greater fractionalization and can use their resources

to secure more public goods from politicians. Finally, reverse causality may also be a concern: there

may be higher migration into villages with a larger supply of public goods, and new migrants may

generate more social fractionalization (i.e. more disperse marriage networks).

To deal with this concern, we follow two approaches. First, we control for a wide range of

village characteristics. To deal with potential demographic confounders we include average age,

gender ratio, village population, and the number of distinct families in the village. We also include

average length of stay in the village which may account for differential migration patterns. To

control for economic characteristics of the village we control for a dummy indicating whether the

village is classified as rural, as well as population in each of 17 educational and 11 occupational

categories, average per capita income and poverty incidence. The estimates of β once we include

this set of controls are reported in Panel B of Table 1. The point estimates become smaller but

remain statistically significant at conventional levels and substantively large for some outcomes

(for example, a one standard deviation increase in fractionalization leads to a 10% increase in the

likelihood of having a high school or market in the village).17

17
Please note that with our full set of controls the p-values for each of the 5 components of the public goods index
are below .01. Thus, even a very conservative Bonferroni correction would lead us to reject the null of no effects for all
the outcomes at the 5 percent level.

19
 Table 1: Network Fractionalization and Public Goods Provision

(1) (2) (3) (4) (5) (6)

Index Elem. School High School Market Health Center Waterworks
Panel A: No Controls
Fractionalization 0.12** 0.01* 0.08** 0.06** 0.06** 0.05**
(0.02) (0.01) (0.01) (0.01) (0.01) (0.01)

Observations 15,464 15,449 15,449 15,449 15,449 15,449

R-squared 0.040 0.001 0.027 0.020 0.014 0.010
Mean Dep. Var. 0 0.806 0.209 0.190 0.639 0.597
Std. Dev. Dep. Var. 0.571 0.395 0.407 0.392 0.480 0.490
Panel B: Full Controls
Fractionalization 0.05** 0.03** 0.02** 0.02** 0.03** 0.02**
(0.01) (0.01) (0.01) (0.00) (0.01) (0.01)

Observations 15,445 15,432 15,432 15,432 15,432 15,432

R-squared 0.169 0.075 0.172 0.139 0.049 0.037
Mean Dep. Var. 0 0.806 0.209 0.190 0.639 0.597
Std. Dev. Dep. Var. 0.571 0.395 0.407 0.392 0.480 0.490

Notes: Results from village-level regressions with municipal fixed-effects. The dependent variable is an
index (Column 1) capturing the availability of key public goods at the village-level (elementary schools,
high schools, markets, health centre and water systems), a dummy equal to one if there is an elementary
school in the village (Column 2), an high school in the village (Column 3), a market in the village (Column
4), a health center in the village (Column 5) and a waterworks system in the village (Column 6). In Panel B
regressions control for village-level average age, average length of stay in the village, gender ratio, village
population, the number of distinct families in the village, whether the village is classified as rural, as well
as education levels in the village, occupation in the village and average per capita income and poverty
incidence. Standard errors (in parentheses) are clustered by municipality. * p < 0.05, ** p < .01.

Our second approach to address concerns of endogeneity and reverse causality is to construct

networks based on individuals aged 45 or older. These networks would mostly reflect marriage

decisions made prior to when public goods are observed (i.e. a generation earlier) and thus the

social fractionalization measures based on these networks are less likely to reflect reverse causality.

All our subsequent robustness checks are reported only for the public goods index and include the

full set of village controls.18 In Column 1 of Table 2 we report the reduced form (OLS) estimate using

the social fractionalization index from the network restricted to those 45 and older while in Column

2 we instead use the social fractionalization index in the restricted network as an instrument for

social fractionalization in the full network and report the 2SLS estimate of β. Both point estimates

18
Results are very similar for each of the public goods dummies separately.

20
are positive and statistically significant. In Table 2 we also report the robustness of our estimates

to different ways of constructing our fractionalization measure such as weighting communities by

family size (Column 3) or number of voters (Column 4) or using the walktrap algorithm to identify

the set of communities (clans) in every village (Column 5). The point estimates remain essentially

unchanged which suggests our estimates do not depend on our particular choice of community

detection algorithm.

Table 2: Robustness to Alternative Fractionalization Measures

(1) (2) (3) (4) (5)

Over 45 Communities Weighted by: Walktrap
OLS IV Family Size # Voters Algorithm
Fractionalization 0.06** 0.11** 0.06** 0.06** 0.05**
(0.01) (0.02) (0.01) (0.01) (0.01)

Observations 15,441 15,441 15,445 15,445 15,445

R-squared 0.171 0.170 0.170 0.171

Notes: Results from village-level regressions with municipal fixed-effects (OLS in Columns 1, 3-5 and 2SLS
in Column 2). The dependent variable is an index capturing the availability of key public goods at the
village-level (elementary schools, high schools, markets, health centre and water systems). In Column 1 the
fractionalization measure is computed using communities obtained on the network restricted to individuals
over the age of 45. In Column 3, the fractionalization measure weights each community by total population in
the family. In Column 4, the fractionalization measure weights each community by the number of members
above the age of 18 in the family. In Column 5, the fractionalization measure is computed using communities
obtained with the walktrap algorithm. Regressions control for village-level average age, average length of
stay in the village, gender ratio, village population, the number of distinct families in the village, whether the
village is classified as rural, as well as education levels in the village, occupation in the village and average
per capita income and poverty incidence. Standard errors (in parentheses) are clustered by municipality. *
p < 0.05, ** p < .01.

We report additional robustness checks in TablesA.3-A.5. In Table A.3 we show that our results

are robust to dropping villages in the top and bottom 1% and 5% of the fractionalization (Columns

1-2) and population (Columns 3-4) distributions. This suggests that our findings are not driven

by villages with extremely high or low values of population or social fractionalization. In Table

A.4 we also show that our estimates are robust to: dropping urban areas (Column 1), dropping

the village where the largest number of relatives of the incumbent mayor reside (Column 2) and

dropping municipalities in the Autonomous Region of Muslim Mindanao (ARMM), a majority

muslim region and one of the poorest in the country (Column 3). Finally, we also show that our

estimates remain relatively unchanged when we control for characteristics of the incumbent and

21
the challengers’ families in the village (Columns 4 and 5).19 This addresses the concern that more

fragmented villages are more likely to house immediate relatives of politicians and this is what

drives the higher provision of public goods.

Most villages and municipalities in the Philippines are relatively ethnically and religiously

homogenous and thus our measure of social fragmentation is unlikely to capture fragmentation

across these two dimensions. In fact, in Column 6 of Table A.4 we show that our point estimate for

social fractionalization remains unchanged when we control for Herfindahl indices for ethnic and

religious fractionalization. One potential concern is that our measure of social fractionalization

is only relevant in an ethnically or religiously homogenous setting like the Philippines, a natural

concern in within-country studies. However, in Table A.5 we show that the correlation between

fractionalization and public goods provision is statistically significant and of similar magnitude in

villages above or below the median value of ethnic or religious fractionalization (Columns 1 and

2). Our estimate is also similar when we restrict the analysis to ARMM (Column 3).

In sum, our results provide evidence of a positive correlation between social fractionalization

and public goods provision. As previously highlighted, a key feature of our setting, of broader

theoretical and empirical relevance, is that public goods provision in the Philippines is not the

responsibility of local communities but rather of politicians with funds transferred by the central

government. In these settings, the risk of elite capture becomes particularly important and collective

action or heterogenous preferences may become less relevant. We explore this directly in Tables

3 and 4 where we show respectively that: i) more fractionalized villages do not exhibit more

heterogenous preferences over 10 different public goods categories, as measured by the standard

deviation of respondent’s average desired budget share on each item; ii) social fractionalization

is not robustly correlated with collective action as measured by participation in voluntary work

(bayanihan) or membership in groups and civil associations.20

19
These include the politician’s number of relatives, number of females relatives, number of relatives in each educa-
tion/occupation category and eigenvector centrality of the family.
20
The municipalities in which we conducted our surveys are not part of the NHTS-PR sample. Thus, to generate
social networks we rely on the lists (family names) of registered voters in every village. However, network statistics
based on NHTS-PR yield relatively similar results to those based on registered voter lists.

22
 Table 3: Network Fractionalization and Preferences over Public Goods

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Health Education Emergencies Water Road ComFaci EconProg Agriculture Peace Festivals
Panel A: No Controls
Fractionalization -0.05 -0.29 -0.21 -0.50* -0.00 0.15 0.19 -0.33 -0.14 -0.04
(0.53) (0.35) (0.30) (0.22) (0.25) (0.07) (0.32) (0.30) (0.13) (0.10)

Observations 283 283 283 283 283 283 283 283 283 283
R-squared 0.000 0.002 0.002 0.020 0.000 0.005 0.001 0.002 0.004 0.000
Mean Dep. Var. 11.19 11.19 8.285 7.425 6.836 5.526 7.798 15.14 5.855 4.064
Panel B: Full Controls
Fractionalization 0.05 -0.20 -0.09 -0.51* -0.03 0.12 0.28 -0.29 -0.12 -0.03
(0.51) (0.35) (0.30) (0.22) (0.27) (0.07) (0.29) (0.32) (0.11) (0.09)

Observations 283 283 283 283 283 283 283 283 283 283
R-squared 0.082 0.066 0.118 0.057 0.023 0.043 0.065 0.071 0.049 0.011
Mean Dep. Var. 11.19 11.19 8.285 7.425 6.836 5.526 7.798 15.14 5.855 4.064

Notes: Results from village-level regressions with municipal fixed-effects. The dependent variable is the
standard deviation in the budget share that voters in the village would like to spend on Health (Column 1),
Education (Column 2), Emergencies (Column 3), Water (Column 4), Roads (Column 5), Community Facilities
(Column 6), Business Loans (Column 7), Agriculture (Column 8), Security (Column 9) and Community
Events/Festivals (Column 10). In Panel B regressions control for village population, whether the village is
classified as rural, average education, age, household size, and length of residence, as well as the share of
population that is female, receives remittances from abroad, and benefits from a CCT program. Standard
errors (in parentheses) are clustered by municipality. * p < 0.05, ** p < .01.

Table 4: Network Fractionalization and Collective Action

(1) (2) (3) (4)

Bayanihan Group
Fractionalization 0.02 0.02 -0.01 -0.01
(0.01) (0.01) (0.01) (0.01)

Controls No Yes No Yes

Observations 283 283 283 283
R-squared 0.008 0.092 0.002 0.128
Mean Dep. Var. 0.751 0.751 0.658 0.658

Notes: Results from village-level regressions with municipal fixed-effects. The dependent variable is the
share of households that participates in voluntary work (Columns 1-2) and the share of households that is
a member of a formal group (Columns 3-4). In Columns 2 and 4, regressions control for village population,
whether the village is classified as rural, average education, age, household size, and length of residence, as
well as the share of population that is female, receives remittances from abroad, and benefits from a CCT
program. Standard errors (in parentheses) are clustered by municipality. * p < 0.05, ** p < .01.

5.2 Political Competition

We hypothesize that social fractionalization can limit the ability of clan heads to effectively deliver

a large number of votes in exchange for private transfers which in turn provides incentives for

23
mayors to provide more public goods in these areas. An additional implication of our theory is

that more fractionalized villages will also feature higher political competition since clan heads will

have a weaker influence over candidacy and voting decisions of village residents.

First, we provide evidence that higher social fractionalization undermines the ability of a

small set of elite members (for example, clan leaders) to exercise disproportionate influence on

the political choices of village residents. To study this, we consider the number of politically

influential individuals mentioned by village respondents in our 2013 survey.21 We consider both

the raw number of individuals nominated as well the effective number of nominees (where we

take into account the number of mentions). The estimates reported in Table 5 suggest that political

influence is less concentrated in highly fragmented villages: a one standard deviation increase

in social fractionalization is associated with approximately one additional politically influential

leader.
21
The surveys were pilot tested to ensure that the questions were capturing personalized political influence within
communities and not national media or politicians. Indeed, respondents were overwhelmingly reporting their local
village elected officials as politically influential, accounting for 47% of the responses. Consistent with the notion of social
persuasion, the rest are personal connections of the respondent: family members and friends/neighbors are 20% and
29% of responses, respectively.

24
 Table 5: Network Fractionalization and Politically Influential Individuals

(1) (2) (3)

# Influential Individuals
Raw Laakso Golosov
Panel A: No Controls
Fractionalization 0.74** 0.47* 0.45*
(0.24) (0.16) (0.15)

Observations 269 269 269

R-squared 0.017 0.014 0.014
Mean Dep. Var. 9.137 5.900 5.157
Panel B: Full Controls
Fractionalization 0.80* 0.54** 0.53**
(0.26) (0.16) (0.14)

Observations 269 269 269

R-squared 0.053 0.066 0.069
Mean Dep. Var. 9.137 5.900 5.157

Notes: Results from village-level regressions with municipal fixed-effects. The dependent variable is the
number of distinct individuals named as influential by survey respondents (Column 1), the effective number
of distinct individuals named as influential by survey respondents computed as Laakso (Column 2), the
effective number of distinct individuals named as influential by survey respondents computed as Golosov
(Column 3). In Panel B regressions control for village population, whether the village is classified as rural,
average education, age, household size, and length of residence, as well as the share of population that is
female, receives remittances from abroad, and benefits from a CCT program. Standard errors (in parentheses)
are clustered by municipality. * p < 0.05, ** p < .01.

Next we provide estimates of regression (1) but using different measures of political competition

as outcomes. In Table 6 we first report the correlation between social fractionalization and political

competition in barangay elections. In Column 1 we show that a one standard deviation increase

in social fractionalization is associated with an increase of 0.07 standard deviations in our index

of political competition for barangay elections. In Columns 2-6 we then show the estimates for

the individual outcomes that constitute the index.22 Social fractionalization is positively correlated

with the raw and effective number of candidates running for barangay captain (Columns 2-4)

and for the barangay council (Column 6). For example, a one standard deviation increase in

social fractionalization is associated with roughly an additional candidate in the barangay council

22
As above, the p-values for each of the 5 components of the political competition index are below .01. Thus, even a
very conservative Bonferroni correction would lead us to reject the null of no effects for all the outcomes at the 5 percent
level.

25
 Table 6: Network Fractionalization and Competition in Barangay Elections

(1) (2) (3) (4) (5) (6)

Index # Candidates Bgy. Cpt. Win # Candidates
Raw Laakso Golosov Margin Bgy. Councilors
Panel A: No Controls
Fractionalization 0.07** 0.07** 0.04** 0.03** -1.73** 0.72**
(0.01) (0.01) (0.00) (0.00) (0.25) (0.07)

Observations 31,344 31,344 30,986 31,344 30,597 31,344

R-squared 0.005 0.004 0.003 0.002 0.002 0.012
Mean Dep. Var. 0 2.176 1.875 1.667 36.89 16.83
Std. Dev. Dep. Var. 0.984 0.988 0.685 0.625 36.91 7.020
Panel B: Full Controls
Fractionalization 0.06** 0.05** 0.04** 0.03** -1.82** 0.61**
(0.01) (0.01) (0.01) (0.00) (0.30) (0.06)

Observations 31,306 31,306 30,948 31,306 30,559 31,306

R-squared 0.018 0.012 0.008 0.007 0.007 0.052
Mean Dep. Var. 0 2.176 1.875 1.667 36.89 16.83
Std. Dev. Dep. Var. 0.984 0.988 0.685 0.625 36.91 7.020

Notes: Results from village*election-level regressions with municipal*election fixed-effects. The dependent
variable is an index (Column 1) capturing the competitiveness of barangay elections (number of candidates
for barangay captains, win margin and number of candidates for barangay councilors), the number of
candidates for barangay captain (Column 2), the effective number of candidates for barangay captain
computed as Laakso (Column 3), the effective number of candidates for barangay captain computed as
Golosov (Column 4), the win margin in the barangay captain election (Column 5) and the number of
candidates for barangay councilor (Column 6). In Panel B regressions control for village-level average age,
average length of stay in the village, gender ratio, village population, the number of distinct families in the
village, whether the village is classified as rural, as well as education levels in the village, occupation in the
village and average per capita income and poverty incidence. Standard errors (in parentheses) are clustered
by municipality. * p < 0.05, ** p < .01.

elections. Also, social fractionalization is positively correlated with more competitive races as

measured by the win margin between the winner and runner-up in barangay captain elections

(Column 5). A one-standard deviation increase in social fractionalization is associated with a

decrease in the win margin of almost 2 percentage points, an effect of almost 5 percent relative

to the sample mean. The point estimates are remarkably stable to controlling for the same set of

village covariates included in the public goods regressions (Panel B, Table 6) or to reduced form or

instrumental variables regressions based on the network of individuals older than 45 (Table A.6).

In Tables A.5-A.8 we report, for our political competition index, the same set of robustness checks

conducted on the public goods estimates.

26
 Finally, we study how social fractionalization correlates with political competition in mayoral

elections. In this case, we keep the number of candidates constant, allowing us to explore whether

races in more fragmented villages are more competitive. The estimates reported in Columns 1-2

of Table 7 are consistent with previous findings and suggest that mayoral races are more tightly

contested in highly fragmented villages.

Identifying all the potential mediators between social fractionalization and public goods provi-

sion in our context is challenging. The evidence presented above suggests that collective action and

preference heterogeneity play a limited role in our context. Previous research provides evidence

of a positive correlation between electoral competition and public goods provision (Besley and

Burgess, 2002; Crost and Kambhampati, 2010; Rosenzweig, 2015), which suggests that political

competition may be an important mediator between fractionalization and public goods provision.

Indeed, in Table A.9 we show that political competition and public goods provision are positively

correlated in our sample. However, political competition may simply be a separate outcome

(not a mediator) and there are other channels through which fractionalization may impact the

redistributive strategies of politicians.23

Our theory suggests that socially fragmented villages receive more public goods at the expense

of private or clientelistic transfers to clan leaders. In fact, in the Philippines, Khemani (2015) reports

a strong negative correlation between the extent of clientelism (proxied by vote buying) and public

goods provision at the local level. Unfortunately we do not have access to systematic measures of

vote buying or private transfers across a wide number of villages. However, a very strong correlate

of vote-buying or clientelism in the Philippines is turnout.24 In villages with a high prevalence of

clientelism, clan leaders deliver the votes of their clan members and this maps into particularly

high turnout rates. Consistent with this, in Columns 3-4 of Table 7 we show that a one standard

deviation increase in social fractionalization is associated with a 1 percentage point decrease in

turnout. While indirect, this evidence is consistent with lower clientelism in highly fragmented

23
For example, Tajima, Samphantharak and Ostwald (2018) argue that segregation can reduce or neutralize the
negative effects of diversity on public goods provision. Their effect operates through the possible exclusion of ethnic
groups living in different villages in their constituency. In our context, exclusion of clans within villages is unlikely
given the small geographical area and population of barangays in our sample.
24
This is consistent with the theory of turnout buying put forward by Nichter (2008).

27
villages.

Table 7: Network Fractionalization, Turnout and Win Margin in Municipal elections

(1) (2) (3) (4)

Win Margin Turnout
Fractionalization -1.94** -0.63 -0.80** -0.68***
(0.35) (0.33) (0.14) (0.15)

Controls No Yes No Yes

Observations 17,023 17,021 18,352 18,350
R-squared 0.006 0.021 0.004 0.016
Mean Dep. Var. 33.60 33.60 72.64 72.64

Notes: Results from precinct-level regressions with municipal fixed-effects. The dependent variable is win
margin in the 2010 municipal elections (Columns 1-2) and turnout in the 2010 municipal elections (Columns
3-4). In Columns 2 and 4, regressions control for village-level average age, average length of stay in the
village, gender ratio, village population, the number of distinct families in the village, whether the village is
classified as rural, as well as education levels in the village, occupation in the village and average per capita
income and poverty incidence. Standard errors (in parentheses) are clustered by municipality. * p < 0.05, **
p < .01.

6 Conclusion

We find compelling evidence that network structures—specifically, the fractionalization of social

networks—are associated with greater incentives for politicians to provide public goods and with

higher levels of electoral competition. Our paper is among the first to provide such evidence using

large-scale social networks data. We use a number of approaches to account for potential con-

founders of social fractionalization and establish robustness of our findings to alternative samples

and estimation strategies.

We highlight alternative ways in which social fractionalization may impact local governance

depending on the institutional context that shapes the incentives for politicians and citizens to exert

effort towards the provision of public goods. While fractionalization may indeed make it difficult

for the community to act collectively and/or to aggregate heterogenous preferences, these may be

less relevant in contexts where politicians (and not communities) are responsible for the provision

of public goods which are funded with transfers (rather than local taxes). In these contexts, we

argue that fractionalization may also be associated with lower concentration of political influence,

making it more difficult for politicians to mobilize voters through elite capture.

28
 Our work highlights the challenges of local public goods provision in decentralized contexts.

Previous literature has highlighted the potential for elite capture to undermine the potential benefits

of decentralization on local governance (Bardhan, 2002; Ostwald, Tajima and Samphantharak,

2016). Our study shows how social structures can shape the degree of elite capture, electoral

competition, and the incentives of politicians to provide public goods. Thus, we contribute to

a growing literature showing that the effects of policy interventions depend on the cultural and

social context (Rao and Walton, 2004; Ashraf et al., 2016).

29
References

Abinales, Patricio N and Donna J Amoroso. 2017. State and Society in the Philippines. Rowman &

Littlefield.

Acemoglu, Daron, Tristan Reed and James Robinson. 2014. “Chiefs: Economic Development and

Elite Control of Civil Society in Sierra Leone.” Journal of Political Economy 122(2):319–368.

Alesina, Alberto, Reza Baqir and William Easterly. 1999. “Public Goods and Ethnic Divisions.” The

Quarterly Journal of Economics 114(4):1243–1284.

Amorim-Neto, Octavio and Gary W. Cox. 1997. “Electoral Institutions, Cleavage Structures, and

the Number of Parties.” American Journal of Political Science 41(1):149–174.

Anderson, Michael L. 2008. “Multiple Inference and Gender Differences in the Effects of Early

Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects.”

Journal of the American Statistical Association 103(484):1481–1495.

Aquino, Clemen C. 2004. “Mula sa Kinaroroonan: Kapwa, Kapatiran and "Bayan in Philippine

Social Science.” Asian Journal of Social Science 32(1):105–139.

Ashraf, Nava, Natalie Bau, Nathan Nunn and Alessandra Voena. 2016. “Bride price and female

education.” National Bureau of Economic Research .

Azfar, Omar, Tugrul Gurgur, Satu Kähkönen, Anthony Lanyi and Patrick Meagher. 2000. “Decen-

tralization and governance: An empirical investigation of public service delivery in the Philip-

pines.” Washington, DC: University of Maryland Center for Institutional Reform and the Informal Sector

and the World Bank .

Bardhan, Pranab. 2002. “Decentralization and Governance in Development.” Journal of Economic

Perspectives 16(4).

Besley, Timothy and Robin Burgess. 2002. “The Political Economy of Government Responsiveness:

Theory and Evidence from India*.” The Quarterly Journal of Economics 117(4):1415–1451.

Calvo, Ernesto and Maria Victoria Murillo. 2009. “Selecting Clients: Partisan Networks and

Electoral Benefits of Targeted Distribution.”.

30
Capuno, Joseph. 2012. “The PIPER Forum on 20 Years of Fiscal Decentralization: A Synthesis.”

Philippine Review of Economics 49(1):191–202.

Chandrasekhar, Arun and Randall Lewis. 2011. “Econometrics of Sampled Networks.” mimeo .

Clemente, Jose Antonio, Delia Belleza, Angela Yu, Effie Vinia Catibog, Goyena Solis and Jason La-

guerta. 2008. “Revisiting the Kapwa Theory: Applying Alternative Methodologies and Gaining

New Insights.” Philippine Journal of Psychology 41:1–32.

Corpuz, Onofre D. 1965. The Philippines. Englewood Cliffs, New Jersey: Prentice-Hall.

Covar, Prospero. 1998. Larangan: seminal essays on Philippine culture. National Commission for

Culture and the Arts.

Crost, Benjamin and Uma S. Kambhampati. 2010. “Political Market Characteristics and the Provi-

sion of Educational Infrastructure in North India.” World Development 38(2):195 – 204.

Cruz, Cesi. 2019. “Social Networks and the Targeting of Vote Buying.” Comparative Political Studies

52(3):382–411.

Cruz, Cesi, Julien Labonne and Pablo Querubin. 2017. “Politician Family Networks and Electoral

Outcomes: Evidence from the Philippines.” American Economic Review 107(10):3006–3037.

Dahl, Robert A. 1973. Polyarchy: Participation and Opposition. Yale University Press.

Desmet, Klaus, Ignacio Ortuño-Ortín and Shlomo Weber. 2017. “Peripheral diversity: transfers

versus public goods.” Social Choice and Welfare 49(3):787–823.

Easterly, William and Ross Levine. 1997. “Africa’s Growth Tragedy: Policies and Ethnic Divisions*.”

The Quarterly Journal of Economics 112(4):1203–1250.

Enriquez, Virgilio G. 1986. “Kapwa: A core concept in Filipino social psychology.” Philippine world

view pp. 6–19.

Eubank, Nicholas, Guy Grossman, Melina Platas and Jonathan Rodden. 2017. “Social Networks,

Social Context, and Political Participation: Evidence from Uganda.”.

31
Fafchamps, Marcel and Julien Labonne. 2017. “Do Politicians’ Relatives Get Better Jobs? Evidence

from Municipal Elections in the Philippines.” Journal of Law, Economics & Organization 33(2):268–

300.

Fegan, Brian. 2009. Entrepreneurs in Votes and Violence: Three Generations of a Peasant Political

Family. In An Anarchy of Families: State & Family in the Philippines, ed. Alfred McCoy. Madison,

WI: University of Wisconsin Press pp. 33–108.

Fernandez, Luisa. 2012. “Design and Implementation Features of the National Household Targeting

System in the Philippines.” World Bank - Philippines Social Protection Note No 5 .

Freeman, Linton C. 1977. “A Set of Measures of Centrality Based on Betweenness.” Sociometry

40:35–41.

Gao, Eleanor. 2016. “Tribal Mobilization, Fragmented Groups, and Public Goods Provision in

Jordan.” Comparative Political Studies 49(10):1372–1403.

Girvan, M. and M. E. J. Newman. 2002. “Community structure in social and biological networks.”

Proceedings of the National Academy of Sciences 99(12):7821–7826.

Golosov, Grigorii V. 2010. “The Effective Number of Parties: A New Approach.” Party Politics

16:171–192.

Haim, Dotan. 2018. “Civilian Social Networks and Credible Counterinsurgency.” UCSD, mimeo .

Hollnsteiner, Mary. 1963. The Dynamics of Power in a Philippine Municipality. University of the

Philippines.

Huckfeldt, Robert and John Sprague. 1991. “Discussant Effects on Vote Choice: Intimacy, Structure,

and Interdependence.” The Journal of Politics 53:122–158.

Huckfeldt, Robert, Paul Allen Beck, Russell J. Dalton and Jeffrey Levine. 1995. “Political Environ-

ments, Cohesive Social Groups, and the Communication of Public Opinion.” American Journal of

Political Science 39(4):1025–1054.

Hughes, D. Alex. 2016. Political Actors Use Social Networks to Coordinate Behavior PhD thesis

University of California, San Diego.

32
Hutchcroft, Paul. 2012. “Re-slicing the pie of patronage: the politics of internal revenue allotment

in the Philippines, 1991-2010.” Philippine Review of Economics 49(1):109–134.

Hutchcroft, Paul and Joel Rocamora. 2003. “Strong Demands and Weak Institutions: The Origins

and Evolution of the Democratic Deficit in the Philippines.” Journal of East Asian Studies 3:259–292.

Jackson, Matthew O. 2010. Social and Economic Networks. Princeton University Press Princeton

University Press.

Katz, Elihu and Paul F. Lazarsfeld. 1955. Personal Influence, the Part Played by People in the Flow of

Mass Communications. New York: Free Press.

Khemani, Stuti. 2015. “Buying votes versus supplying public services: Political incentives to

under-invest in pro-poor policies.” Journal of Development Economics 117:84–93.

Kitschelt, Herbert and Steven Wilkinson. 2007. Patrons, Clients, and Policies. Cambridge: Cambridge

University Press.

Kustov, Alexander and Giuliana Pardelli. 2018. “Ethnoracial Homogeneity and Public Outcomes:

The (Non)effects of Diversity.” American Political Science Review pp. 1–8.

Laakso, Markku and Rein Taagepera. 1979. “Effective Number of Parties: A Measure with Appli-

cation to West Europe.” Comparative Political Studies 12:3–27.

Lande, Carl H. 1964. Leaders, Factions, and Parties: the Structure of Philippine Politics. New Haven:

Yale University Press.

Larson, Jennifer M and Janet I Lewis. 2017. “Ethnic Networks.” American Journal of Political Science

61(2):350–364.

Lizzeri, Alessandro and Nicola Persico. 2004. “Why did the Elites Extend the Suffrage? Democracy

and the Scope of Government, with an Application to Britain’s ‘Age of Reform’.” The Quarterly

Journal of Economics 119(2):707–765.

Llanto, Gilberto M. 2012. “The Assignment of Functions and Intergovernmental Fiscal Relations

in the Philippines 20 Years after Decentralization.” Philippine Review of Economics 49(1):37–80.

33
Lublin, David. 2017. “Electoral Systems, Ethnic Heterogeneity and Party System Fragmentation.”

British Journal of Political Science 47(2):373–389.

McCoy, Alfred W. 2009. An anarchy of families: State and family in the Philippines. Univ of Wisconsin

Press.

Mendoza, Ronald U., Edsel L. Beja Jr., Victor S. Venida and David B. Yap. 2016. “Political dynasties

and poverty: measurement and evidence of linkages in the Philippines.” Oxford Development

Studies 44(2):189–201.

Mendoza, Ronald U., Jan Cruz and David B. Yap. 2014. “Political Party Switching: It’s More Fun

in the Philippines.” Asian Institute of Management (AIM) Working Paper No. 14-019 .

Miguel, Edward and Mary Kay Gugerty. 2005. “Ethnic diversity, social sanctions, and public goods

in Kenya.” Journal of Public Economics 89(11):2325 – 2368.

Miralao, Virginia A. 1997. “The Family, Traditional Values and the Sodocultural Transformation of

Philippine Society.” Philippine Sociological Review 45(1/4):189–215.

Moscona, Jacob, Nathan Nunn and James A. Robinson. 2017. “Keeping It in the Family: Lin-

eage Organization and the Scope of Trust in Sub-Saharan Africa.” American Economic Review

107(5):565–71.

Munshi, Kaivan and Mark Rosenzweig. 2018. “Ethnic Diversity and the Under-Supply of Public

Goods.” NBER Working Paper 19197 .

Nichter, Simeon. 2008. “Vote Buying or Turnout Buying? Machine Politics and the Secret Ballot.”

American Political Science Review 102(1):19–31.

Ostwald, Kai, Yuhki Tajima and Krislert Samphantharak. 2016. “Indonesia’s Decentralization

Experiment: Motivations, Successes, and Unintended Consequences.” Journal of Southeast Asian

Economies 33(2):139–156.

Padgett, John F. and Paul D. McLean. 2006. “Organizational Invention and Elite Transformation:

The Birth of the Partnership Systems in Renaissance Florence.” American Journal of Sociology

111(5):1463–1568.

34
Pons, Pascal and Matthieu Latapy. 2005. Computing Communities in Large Networks Using

Random Walks. In Computer and Information Sciences - ISCIS 2005, ed. pInar Yolum, Tunga Güngör,

Fikret Gürgen and Can Özturan. Berlin, Heidelberg: Springer Berlin Heidelberg pp. 284–293.

Potter, Joshua D. 2014. “Demographic Diversity and District-Level Party Systems.” Comparative

Political Studies 47(13):1801–1829.

Rao, Vijayendra and Michael Walton. 2004. Culture and Public Action: A Cross-Disciplinary Dialogue

on Development Policy. Stanford University Press.

Reyes, Jeremiah. 2015. “Loób and Kapwa: An Introduction to a Filipino Virtue Ethics.” Asian

Philosophy 25(2):148–171.

Rogers, Steven. 2004. “Philippine Politics and the Rule of Law.” Journal of Democracy 15(4):116.

Rosenzweig, Steven C. 2015. “Does electoral competition affect public goods provision in dominant-

party regimes? Evidence from Tanzania.” Electoral Studies 39:72 – 84.

Rubin, Michael. 2018. “Rebel Territorial Control and Civilian Collective Action in Civil War.

Evidence from the Philippines.” UCSAE, mimeo .

Sidel, John T. 1999. Capital, Coercion, and Crime: Bossism in the Philippines. Contemporary issues in

Asia and the Pacific Stanford University Press.

Soifer, Hillel David. 2016. “Regionalism, Ethnic Diversity, and Variation in Public Good Provision

by National States.” Comparative Political Studies 49(10):1341–1371.

Tajima, Yuhki, Krislert Samphantharak and Kai Ostwald. 2018. “Ethnic Segregation and Public

Goods: Evidence from Indonesia.” American Political Science Review 112(3):637–653.

Timberman, David G. 1991. A Changeless Land: Continuity and Change in Philippine Politics. M.E.

Sharpe.

Todd, Emmanuel. 1985. The Explanation of Ideology: Family Structures and Social Systems. B. Blackwell.

Troland, Erin. 2016. “Do Fiscal Transfers Increase Local Revenue Collection? Evidence From The

Philippines”. Working Paper. Available at SSRN: https://ssrn.com/abstract=3242240.

35
Wimmer, Andreas. 2016. “Is Diversity Detrimental? Ethnic Fractionalization, Public Goods Provi-

sion, and the Historical Legacies of Stateness.” Comparative Political Studies 49(11):1407–1445.

Xu, Yiqing and Yang Yao. 2015. “Informal Institutions, Collective Action, and Public Investment in

Rural China.” American Political Science Review 109(2):371–391.

36
 Social Network Structures and the Politics of Public Goods Provision:

Evidence from the Philippines

Online Appendix

A.1 Variables Definition

A.1.1 Network Measures

Edge Betweenness Centrality

Edge betweenness centrality is the extent to which an edge (marriage between 2 families) serves

as a link between different groups of families. It assesses centrality by looking at whether the edge

is an important hub in the paths traversing the network and is calculated using the number of

shortest paths in the network that necessarily pass through the edge. This measure is analogous to

betweenness centrality, which considers the relative importance of a node in the paths traversing

the network (Freeman, 1977).

In the family network f , let Pe (k j) indicate the number of shortest paths between family k and

family j that necessarily pass through edge e, while P(k j) is the total number of shortest paths

between k and j.

The ratio Pe (k j)/P(k j) approximates the importance of edge e in connecting k and j. If Pe (k j) =

P(k j), yielding a ratio of 1, then edge e lies on all of the shortest paths connecting families k and

j. Conversely, if Pe (k j) = 0, then the intermarriage represented by edge e is not important for

connecting families k and j.

Edge betweenness centrality is calculated by averaging this ratio for the entire network.

X Pe (k j)
Betweennesse ( f ) = (2)
P(k j)

A.1
A.1.2 Effective Number of Candidates Indices
Pn
The Laakso and Taagepera (1979) index is given by N = 1/ 2
i=1 si , where n is the number of

candidates and si is the share of votes of candidate i.

The index by Golosov (2010) is defined as N = ni=1 si /(si + s21 − s2i ), where s1 is the vote share of
P

the candidate with the largest number of votes.

A.1.3 Computing Indices, following Anderson (2008)

Here we explain how we compute our public goods and political competition indices. The public

goods index aggregates dummies for whether the barangay has an elementary school, a high

school, a public market, a health center or a community water system. The political competition

index aggregates the win margin (vote share of the candidate that received the most votes in that

precinct minus vote share of the runner-up in that precinct) and the number of candidates running

in the race. We also use the indices of effective number of candidates, proposed by Laakso and

Taagepera (1979) and Golosov (2010).

As carefully explained by Anderson (2008), we compute the public goods index and the electoral

competition indices as follows:

1. For all outcomes (yik ) in each of the two categories, switch signs where necessary so that the

positive direction always indicates a better outcome.

yik − y¯k y
2. Compute zik = σk
y , where y¯k is the sample average and σk the standard deviation of yik .

3. Compute si = (10 Σ̂−1 1)−1 (10 Σ̂−1 zi ), where 1 is a column vector of 1’s. Σ̂−1 is the inverted

covariance matrix, and zi is vector of all outcomes for individual i.

A.2
A.2 Descriptive Statistics

Table A.1: Describing our Sample

(1) (2)
Sample National Average
Population 1,412.28 2,196.38
(1535.37) (4629.95)
Urban 0.11 0.24
(0.31) (0.43)
Religious diversity 0.20 0.21
(0.22) (0.20)
Ethnic diversity 0.18 0.21
(0.23) (0.24)
Elementary school 0.81 0.77
(0.40) (0.42)
High school 0.21 0.23
(0.41) (0.42)
Market 0.19 0.18
(0.39) (0.38)
Health centre 0.64 0.68
(0.48) (0.47)
Waterworks 0.60 0.62
(0.49) (0.49)

Source: 2010 Census. Column 1 reports the mean and standard deviation (in parenthesis) of the relevant
variables for villages in our sample. Column 2 reports the mean and standard deviation (in parenthesis) of
the relevant variables for villages in the country.

A.3
 Table A.2: Descriptive Statistics

Mean Std. Dev. Median Min Max

Number of communities 45.89 (41.04) 34.00 1.00 614.00
Share largest community 0.13 (0.06) 0.12 0.02 1.00
Share 2nd largest community 0.10 (0.04) 0.10 0.02 0.50
Fractionalization (edge) 0.00 (1.00) 0.20 -21.98 1.46
Fractionalization (walktrap) 0.00 (1.00) 0.23 -14.69 1.30
Fractionalization (edge) 1st population tercile -0.60 (1.25) -0.35 -21.98 1.30
Fractionalization (edge) 2nd population tercile 0.09 (0.62) 0.19 -6.54 1.34
Fractionalization (edge) 3rd population tercile 0.51 (0.66) 0.60 -21.98 1.46
Elementary School 0.81 (0.40) 1.00 0.00 1.00
High School 0.21 (0.41) 0.00 0.00 1.00
Market 0.19 (0.39) 0.00 0.00 1.00
Health centre 0.64 (0.48) 1.00 0.00 1.00
Waterworks 0.60 (0.49) 1.00 0.00 1.00
Number of candidates barangay captain 2.17 (0.99) 2.00 1.00 21.00
Effective no. of candidates barangay captain (Laakso) 1.87 (0.69) 1.95 1.00 8.17
Effective no. of candidates barangay captain (Golosov) 1.67 (0.62) 1.70 0.00 7.36
Win margin (barangay captain elections) 36.96 (36.94) 19.85 0.07 100
Number of candidates barangay council 16.82 (7.01) 15.00 0.00 96.00

A.3 Additional Results

Table A.3: Fractionalization and Public Goods: Robustness to Excluding Outliers

(1) (2) (3) (4)

Fractionalization: Population:
1% 5% 1% 5%
Fractionalization 0.07** 0.06** 0.03** 0.02**
(0.01) (0.01) (0.01) (0.01)

Observations 15,147 13,908 15,133 13,888

R-squared 0.182 0.175 0.191 0.171

Notes: Results from village-level regressions with municipal fixed-effects. The dependent variable is an
index capturing the availability of key public goods at the village-level (elementary schools, high schools,
markets, health centre and water systems). In Column 1, we remove all villages in the top and bottom 1%
of the distribution of fractionalization. In Column 2, we remove all villages in the top and bottom 5% of
the distribution of fractionalization. In Column 3, we remove all villages in the top and bottom 1% of the
distribution of population. In Column 4, we remove all villages in the top and bottom 5% of the distribution
of population. Regressions control for village-level average age, average length of stay in the village, gender
ratio, village population, the number of distinct families in the village, whether the village is classified as
rural, as well as education levels in the village, occupation in the village and average per capita income and
poverty incidence. Standard errors (in parentheses) are clustered by municipality. * p < 0.05, ** p < .01.

A.4
 Table A.4: Fractionalization and Public Goods: Robustness to Alternative Samples and Controls

(1) (2) (3) (4) (5) (6)

Exclude: Control for:
Incumbent Incu. & Chall. Ethnic & Relig.
Urban “Hometown” ARMM Family Family Fractionalization
Fractionalization 0.05** 0.05** 0.07** 0.06** 0.05** 0.05**
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)

Observations 13,740 10,921 13,147 9,704 8,746 15,445

R-squared 0.141 0.157 0.174 0.186 0.193 0.173

Notes: Results from village-level regressions with municipal fixed-effects. The dependent variable is an
index capturing the availability of key public goods at the village-level (elementary schools, high schools,
markets, health centre and water systems). In Column 1, we drop all villages classified as urban. In Column
2, we drop villages where the incumbent has the most number of relatives. In Column 3, we drop all
villages in ARMM from our sample. In Column 4, we further control for characteristics of the incumbent’s
family. In Column 5, we further control for characteristics of both the challenger and the incumbent’s
family. In Column 6, we further control for ethnic and religious fractionalization. Regressions control for
village-level average age, average length of stay in the village, gender ratio, village population, the number
of distinct families in the village, whether the village is classified as rural, as well as education levels in the
village, occupation in the village and average per capita income and poverty incidence. Standard errors (in
parentheses) are clustered by municipality. * p < 0.05, ** p < .01

A.5
 Table A.5: Ethnic and Religious Fragmentation, Public Goods and Elections

(1) (2) (3) (4) (5) (6)

Public Goods Electoral Competition
Fractionalization*Below 0.05** 0.04** 0.06** 0.06**
(0.01) (0.01) (0.01) (0.01)
Fractionalization*Above 0.05** 0.07** 0.05** 0.06**
(0.02) (0.01) (0.01) (0.01)
Fractionalization 0.05** 0.01
(0.01) (0.02)

Sample Full Full ARMM Full Full ARMM

Interaction with: Ethnic Religious Ethnic Religious

Observations 15,445 15,445 2,298 31,306 31,306 4,039

R-squared 0.170 0.170 0.217 0.018 0.018 0.019

Notes: Results from village-level regressions with municipal*election fixed-effects. In Columns 1 and 4
we interact the fractionalization variable with dummies capturing whether a village is below or above the
median in the distribution of ethnic fractionalization. In Columns 2 and 5 we interact the fractionalization
variable with dummies capturing whether a village is below or above the median in the distribution of
religious fractionalization. In Columns 1-3, the dependent variable is an index capturing the availability of
key public goods at the village-level (elementary schools, high schools, markets, health centre and water
systems). In Columns 4-6, the dependent variable is an index capturing the competitiveness of barangay
elections (number of candidates for barangay captains, win margin and number of candidates for barangay
councilors). Regressions control for village-level average age, average length of stay in the village, gender
ratio, village population, the number of distinct families in the village, whether the village is classified as
rural, as well as education levels in the village, occupation in the village and average per capita income and
poverty incidence. Standard errors (in parentheses) are clustered by municipality. * p < 0.05, ** p < .01.

A.6
Table A.6: Fractionalization and Political Competition: Robustness to Alternative Fractionalization Measures

(1) (2) (3) (4) (5)

Over 45 Communities Weighted by: Walktrap
OLS IV Family Size # Voters Algorithm
Fractionalization 0.07** 0.12** 0.05** 0.05** 0.05**
(0.01) (0.01) (0.01) (0.01) (0.01)

Observations 31,298 31,298 31,306 31,306 31,306

R-squared 0.019 0.017 0.017 0.017

Notes: Results from village*election-level regressions with municipal*election fixed-effects (OLS in Columns
1, 3-5 and 2SLS in Column 2). The dependent variable is an index capturing the competitiveness of barangay
elections (number of candidates for barangay captains, win margin and number of candidates for barangay
councilors). In Column 1 the fractionalization measure is computed using communities obtained on the
network restricted to individuals over the age of 45. In Column 3, the fractionalization measure weights
each community by total population in the family. In Column 4, the fractionalization measure weights each
community by the number of members above the age of 18 in the family. In Column 5, the fractionalization
measure is computed using communities obtained with the walktrap algorithm. Regressions control for
village-level average age, average length of stay in the village, gender ratio, village population, the number
of distinct families in the village, whether the village is classified as rural, as well as education levels in the
village, occupation in the village and average per capita income and poverty incidence. Standard errors (in
parentheses) are clustered by municipality. * p < 0.05, ** p < .01.

Table A.7: Fractionalization and Political Competition: Robustness to Excluding Outliers

(1) (2) (3) (4)

Fractionalization: Population:
1% 5% 1% 5%
Fractionalization 0.08** 0.09** 0.05** 0.03**
(0.01) (0.01) (0.01) (0.01)

Observations 30,690 28,191 30,676 28,170

R-squared 0.019 0.022 0.019 0.019

Notes: Results from village*election-level regressions with municipal*election fixed-effects. The dependent
variable is an index capturing the competitiveness of barangay elections (number of candidates for barangay
captains, win margin and number of candidates for barangay councilors). In Column 1, we remove all
villages in the top and bottom 1% of the distribution of fractionalization. In Column 2, we remove all
villages in the top and bottom 5% of the distribution of fractionalization. In Column 3, we remove all
villages in the top and bottom 1% of the distribution of population. In Column 4, we remove all villages in
the top and bottom 5% of the distribution of population. Regressions control for village-level average age,
average length of stay in the village, gender ratio, village population, the number of distinct families in the
village, whether the village is classified as rural, as well as education levels in the village, occupation in the
village and average per capita income and poverty incidence. Standard errors (in parentheses) are clustered
by municipality. * p < 0.05, ** p < .01.

A.7
 Table A.8: Fractionalization and Political Competition: Robustness to Alternative Samples and Controls

(1) (2) (3) (4) (5) (6)

Exclude: Control for:
Incumbent Incu. & Chall. Ethnic & Relig.
Urban “Hometown” ARMM Family Family Fractionalization
Fractionalization 0.05** 0.05** 0.07** 0.05** 0.05** 0.06**
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)

Observations 27,249 21,704 27,267 19,703 17,777 31,306

R-squared 0.024 0.021 0.023 0.021 0.027 0.018

Notes: Results from village-level regressions with municipal fixed-effects. The dependent variable is an
index capturing the competitiveness of barangay elections (number of candidates for barangay captains,
win margin and number of candidates for barangay councilors). In Column 1, we drop all villages classified
as urban. In Column 2, we drop villages where the incumbent has the most number of relatives. In Column
3, we drop all villages in ARMM from our sample. In Column 4, we further control for characteristics of
the incumbent’s family. In Column 5, we further control for characteristics of both the challenger and the
incumbent’s family. In Column 6, we further control for ethnic and religious fractionalization. Regressions
control for village-level average age, average length of stay in the village, gender ratio, village population,
the number of distinct families in the village, whether the village is classified as rural, as well as education
levels in the village, occupation in the village and average per capita income and poverty incidence. Standard
errors (in parentheses) are clustered by municipality. * p < 0.05, ** p < .01

Table A.9: Electoral Competition and Public Goods Provision

(1) (2) (3) (4) (5)

Electoral Competition 0.03** 0.01** 0.01** 0.01* 0.01*
(0.00) (0.00) (0.00) (0.00) (0.00)

Observations 31,344 31,306 31,306 31,306 31,306

R-squared 0.002 0.151 0.165 0.169 0.171

Notes: Results from village*election-level regressions with municipal*election fixed-effects. The dependent
variable is an index capturing the availability of key public goods at the village-level (elementary schools,
high schools, markets, health centre and water systems). The main independent variable is an index
capturing the competitiveness of barangay elections (number of candidates for barangay captains, win
margin and number of candidates for barangay councilors). In Columns 2-5, we control for village-level
average age, average length of stay in the village, gender ratio, village population, the number of distinct
families in the village , whether the village is classified as rural. In Columns 3-5, regressions also control
for education levels in the village. In Column 4-5, regressions also control for occupation in the village. In
Column 5, regressions also control for average per capita income and poverty incidence. Standard errors (in
parentheses) are clustered by municipality. * p < 0.05, ** p < .01.

A.8