multiplexity as a conduit and sieve of information …...multiplexity as a conduit and sieve of...

55
Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate. Valentina A. Assenova Yale University, New Haven, CT, 06520, USA Abstract This study uses a quasi-experiment in the introduction of information about the benefits of microfinance among 16,984 people in 43 villages in India to understand how the presence of multiple kinds of social ties among people – multiplexity – aected information transmission and program enrollment over time. The arti- cle posits that multiplexity deterred microfinance enrollment by aecting infor- mants’ motives to share word-of-mouth information objectively and impartially. Results from social interaction models demonstrate that social connections to informants with conflicting interests deterred enrollment among eligible program participants. The article illuminates the consequences of multiplex information structures for social diusion through word-of-mouth. Keywords: Information Structure, Social Diusion, Multiplexity Word count: 100 words (abstract), 9,378 words (text), 9 figures. Correspondence: [email protected]. I thank Brandy Aven, Delia Baldassari, Jim Baron, Daylian Cain, Rodrigo Canales, Jared Curhan, Julia DiBenigno, Emily Erikson, Marissa King, Bruce Kogut, Bal´ azs Kov´ acs, Ko Kuwabara, Ray Reagans, Olav Sorenson, Dan Wang, Amy Wrzesniewski, Ezra Zuckerman Sivan, and seminar participants at Yale, MIT, Columbia, University of British Columbia, and the 2016 Academy of Management for helpful comments. I also thank Abhijit Banerjee, Arun Chandrasekhar, Esther Duflo, and Matthew Jackson for collecting and sharing the data, and the staof the Center for Science and Social Science Information, the Lillian Goldman Law Library, and Sterling Memorial Library at Yale University for materials and space to conduct this research. All errors are my own.

Upload: others

Post on 19-Apr-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity as a Conduit and Sieve of

Information in Embedded Markets

Preliminary and Incomplete. Please do not cite or circulate.

Valentina A. Assenova⇤

Yale University, New Haven, CT, 06520, USA

Abstract

This study uses a quasi-experiment in the introduction of information about thebenefits of microfinance among 16,984 people in 43 villages in India to understandhow the presence of multiple kinds of social ties among people – multiplexity –a↵ected information transmission and program enrollment over time. The arti-cle posits that multiplexity deterred microfinance enrollment by a↵ecting infor-mants’ motives to share word-of-mouth information objectively and impartially.Results from social interaction models demonstrate that social connections toinformants with conflicting interests deterred enrollment among eligible programparticipants. The article illuminates the consequences of multiplex informationstructures for social di↵usion through word-of-mouth.

Keywords: Information Structure, Social Di↵usion, MultiplexityWord count: 100 words (abstract), 9,378 words (text), 9 figures.

⇤Correspondence: [email protected]. I thank Brandy Aven, Delia Baldassari, Jim Baron,Daylian Cain, Rodrigo Canales, Jared Curhan, Julia DiBenigno, Emily Erikson, Marissa King, Bruce Kogut,Balazs Kovacs, Ko Kuwabara, Ray Reagans, Olav Sorenson, Dan Wang, AmyWrzesniewski, Ezra ZuckermanSivan, and seminar participants at Yale, MIT, Columbia, University of British Columbia, and the 2016Academy of Management for helpful comments. I also thank Abhijit Banerjee, Arun Chandrasekhar, EstherDuflo, and Matthew Jackson for collecting and sharing the data, and the sta↵ of the Center for Scienceand Social Science Information, the Lillian Goldman Law Library, and Sterling Memorial Library at YaleUniversity for materials and space to conduct this research. All errors are my own.

Page 2: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

1 Introduction

Socially connected people are often assumed to be the best individuals to spread word-of-

mouth information about new practices, technologies, and ideas. The di↵usion literature

(for reviews, see Strang and Soule, 1998; Wejnert, 2002) posits that having many social

connections to others promotes innovation adoption by increasing information flows at the

stage of awareness and strengthening social influence at the stage of evaluation (e.g. Coleman

et al., 1966; Rogers, 1983). Yet, having many connections often entails having di↵erent kinds

of connections, a property called multiplexity (Smith and Papachristos, 2016). Although

having multiple kinds of connections can increase the number of social pathways that a

person has to inform and influence others, competing motives across these ties can also

create conflicts of interest to share information (Merton, 1957; Snoek, 1966; Coser, 1991).

Prior research has not examined how these conflicting interests a↵ect di↵usion through

word-of-mouth. This question is important, however, because individuals who are connected

to others through multiple networks are positioned to communicate opportunistically (cf.

Padgett and Ansell, 1993). The duality of enjoying others’ trust while having private motives

and conflicts of interest enable actors with multiple social ties to be strategic about the

information they share (Moore and Loewenstein, 2004; Moore et al., 2010; Loewenstein

et al., 2011). Skewing, biasing, and withholding information can a↵ect di↵usion outcomes

by limiting awareness about an innovation and by creating social influence to not adopt

an innovation. Understanding how di↵erent kinds of social ties a↵ect di↵usion through

word-of-mouth can inform a larger question of when and why novel and beneficial practices,

ideas, and technologies fail to gain acceptance in communities with dense network topologies

(Brummitt et al., 2012; Centola, 2015).

This article advances new theory and evidence about the consequences of multiplex in-

1

Page 3: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

formation structures to explain when and why having di↵erent kinds of social pathways for

obtaining information deters di↵usion through word-of-mouth. By connecting prior research

on di↵erent types of social roles and relationships (e.g. Hecht, 2001; Clark and Mills, 1993;

Coser, 1991) to theories about how people disseminate information through their social net-

works (Centola, 2015; Aven, 2015), I posit that multiplexity a↵ects social di↵usion through

two distinct mechanisms. First, multiplexity increases the number of social pathways to

obtaining information and thus serves as a conduit of word-of-mouth among people. Second,

multiplexity in informants’ social a�liations creates competing interests when it involves

incompatible relational logics and thus interferes with informants’ ability to be objective

and impartial. When multiplex ties activate conflicts of interest, the presence of more so-

cial pathways to information can deter – rather than facilitate – di↵usion by decreasing the

benefits of word-of-mouth.

1.1 Explaining Variance in Di↵usion

The theory I propose o↵ers an explanation of an empirical puzzle that is unexplained by

existing theories of di↵usion: limited di↵usion despite dense social connectivity. Figure 1

shows the fraction of eligible villagers who enrolled in microfinance over 30 months following

the program’s launch and the social pathways through which people could have obtained

information from their contacts. The village with the highest enrollment (village 21) and

the village with the lowest enrollment (village 36) had similar concentrations of information

flows, yet village 21 had more than six times the uptake of village 36. The two villages

were otherwise similar in network features relevant for di↵usion outcomes, such as network

size and homophily (Centola, 2015; Aral et al., 2009), percent initially informed (Rogers,

1983), and graph centralization (Freeman, 1979). What factors produced the di↵erences in

2

Page 4: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

enrollment?

To explain the reasons for limited di↵usion, I develop a relational model of information

transmission that takes into account di↵erences in informants’ relational logics for sharing

word-of-mouth information. I posit that di↵erent types of social ties, which involve di↵erent

relational norms, a↵ect how people share information with others and impede di↵usion when

they activate conflicts of interest for informants. Drawing on data from an informational

intervention in the social networks of 16,984 people in 43 Indian villages between 2007

and 2010 (Banerjee et al., 2013), I explain how conflict in informants’ interests to share

information impeded the di↵usion of microfinance.1

Although the intervention was intended to increase the di↵usion of microfinance through

word-of-mouth and social influence, only a fraction of eligible participants (20 percent on

average) enrolled in the program.2 I show that social ties to informants with conflicts of inter-

est in the villages decreased microfinance enrollment among eligible participants. Moreover,

these e↵ects were stronger the more multiplex that the relationships between informants and

their contacts were. This evidence runs counter to prevailing theories of di↵usion through

social contagion, which contend that both tie strength and the number of social ties that

people have to others promote di↵usion (De Domenico et al., 2016; Centola, 2015). Prior the-

ories derive their predictions from assumptions that all social ties serve similar functions and

exert similar influence on people’s norms for communicating information (cf. Centola, 2015;

Brummitt et al., 2012). Extant theories also do not explicitly consider how the presence of

multiple kinds of social ties a↵ects conflicts of interests for individuals to share information

1Microfinance is the practice of providing loans to poor individuals, usually women, in developing coun-tries (Armendariz and Morduch, 2005). Microfinance provides an alternative to borrowing money fromfamily, friends, and neighbors.

2Among eligible participants (women from the most socially and economically disadvantaged groups)the microfinance program had many potential economic benefits, including larger loan amounts and lowerinterest rates than loans from peer lenders in the villages (Banerjee and Duflo, 2011; Banerjee et al., 2013).

3

Page 5: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Figure 1: Microfinance Enrollment in 43 Villages, 2007-2010

MoneyGoodsSocialMedicalVisitAdviceKinshipReligious

Homophily = 89%Centralization = 0.02Initially informed = 13%N=1041

(a) Village 21

MoneyGoodsSocialMedicalVisitAdviceKinshipReligious

Homophily = 89%Centralization = 0.02Initially informed = 13%N=1203

(b) Village 36

Note: Enrollment (scaled by data density) in microfinance across the 43 villages over 30 months (top) andsocial pathways for obtaining information in two of the villages (bottom). Homophily measured asproportion of all social interactions that occured within people’s in-groups (e.g. Caste, language, religion).Graph centralization captures the concentration of information flows (Freeman, 1979).

4

Page 6: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

and influence others to adopt new practices, ideas, or technologies.

My findings contribute to our understanding of why informational interventions can be

rendered less e↵ective at promoting di↵usion by providing new evidence that word-of-mouth

can deter di↵usion when informants have conflicting interests to share information. Fur-

ther, I show that multiplex pathways for obtaining information can create inertia in existing

practices by creating conflicts of interest to change these practices. Finally, I show that the

benefits of selecting socially “central” informants to di↵use innovations depend on whether

informants have competing private interests to endorse these innovations. When informants

have competing interests and multiplex social ties, they are positioned to behave opportunis-

tically by virtue of (1) being trusted, and (2) having competing interests to share information

impartially and objectively. These factors enable opportunism in how central actors spread

word-of-mouth and influence others, and thus a↵ect whether central informants promote

di↵usion.

2 Multiplexity and Information

Despite the fact that people often share multiple kinds of ties in markets (cf. Cohen et al.,

2010), how these ties a↵ect individual behavior remains an open question (Smith and Pa-

pachristos, 2016). The recent availability of high-dimensional data and novel methods in data

science have enabled analysis of multiplex structures, yet how multiplexity a↵ects di↵usion

dynamics in empirical markets remains unclear. And although we know from relational

perspectives on social interaction (e.g. Clark and Mills, 1993) that communal and exchange

ties a↵ect behavior di↵erently, it is unclear how the presence of multiple kinds of ties a↵ects

di↵usion through word-of-mouth mechanisms. Similarly, although we know that complex-

ity in social roles influences individual behavior and autonomy (Coser, 1975, 1991), these

5

Page 7: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

insights have remained largely disconnected from recent theories of di↵usion through social

contagion (e.g. Centola, 2015; Rossman, 2014; Aral et al., 2009). Thus, gaps remain in our

understanding of how the presence of multiple kinds of ties (multiplexity) in the networks

through which people obtain information a↵ect the di↵usion of innovations.

2.1 What is Multiplexity?

Multiplexity, defined as “the overlap of roles, exchanges, or a�liations in a social relation-

ship” (Verbrugge, 1979:1286) arises when people interact in more than one ways, for example

as colleagues and friends (Merton, 1957; Coser, 1991). Multiplex social ties, such as between

one’s work and social domains, present multiple social pathways for obtaining information

and for enabling di↵usion through word-of-mouth (cf. Brummitt et al., 2012; De Domenico

et al., 2016). Gluckman (1962) coined the term to refer to the superimposition of multiple

sets of norms within the same relationship. For example, two people that are at once friends

and colleagues are said to have a multiplex relationship. Multiplexity refers to the number

of ways in which people are connected, as contrasted with simplexity, which refers to con-

nections that involve only one kind of tie (e.g. friendship). Multiplexity also encompasses

overlap in di↵erent domains for interaction. Overlaps in people’s domains for interaction,

such as social clubs and organizations, creates multiple kinds of social ties and motives for

action (e.g. as a school friend and professional colleague).

Prior research has shown that multiplexity increases over the course of life as people

develop more interests and find new domains for interaction (Verbrugge, 1979). Multiplex

social ties emerge from a combination of a person’s (i) opportunities for contact with others,

(ii) preferences for special similarities (common activities, similar social groups), and (iii)

desires for deep and meaningful relationships with others (Verbrugge, 1979). For example,

6

Page 8: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

individuals who have few opportunities for outside interaction (e.g. in a neighborhood) are

more likely to develop multiplex ties with others. The stronger that is the division of labor,

the more that people’s social, kinship, and occupational domains for interaction represent

separate and non-overlapping spheres (Durkheim, 1984). Individuals develop multiplex re-

lationships because each type of relationship (e.g. economic, familial) provides di↵erent

resources (e.g. information, social support) and meets di↵erent needs (e.g. to belong, to

solicit the help of others) (Verbrugge, 1979; Baumeister and Leary, 1995).

The presence of multiple social norms that guide behavior across people’s social ties has

been shown to require individuals to “segment their activities” and to “behave di↵erently

at di↵erent places and at di↵erent times” (Coser, 1975:237). This conceptualization views

actions and behaviors as dependent upon the particular social roles that people enact at

a particular point in time, conditional on the social situation and the setting in which the

action unfolds (Go↵man, 1961, 1959).

Invariably, in their daily course of life, people interact in multiple ways with others.

Figure 2 illustrates, for example, the di↵erent types of social interactions among people in a

village community in India. Interactions in this setting are highly multiplex (overlapping in

di↵erent domains) as people enact multiple kinds of social roles (e.g. as advice givers, friends,

kin, goods traders, and money lenders). This diagram is organized around multiplexity in

people’s social roles. The largest “nodes” represent people who had more social roles in the

village. The most “central” people in this village based on the total number of social ties in

fact shared multiple kinds of relationships with others and enacted many di↵erent roles.

I posit that multiplexity in people’s patterns of social interaction a↵ects di↵usion dy-

namics through two distinct mechanisms. First, multiplexity increases the number of social

pathways through which people can obtain information from their contacts and learn about

7

Page 9: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Figure 2: Multiplex Pathways for Obtaining Information in an Indian Village

Note: Multiplexity in an Indian village (main) and its components (inset). Gray shading based on type ofinteraction. Larger nodes denote greater role multiplicity. Money denotes the structure of money lending;Goods denotes the structure of goods exchange (e.g. kerosene and rice); Social denotes the structure ofsocial relationships (e.g. attending each others’ marriages and festivals); Medical denotes the structure ofadvice giving and receiving about medical emergencies; Visit denotes the structure of home visits; Advicedenotes the structure of general advice giving and receiving; Kinship denotes the structure of kinship(relations by blood or marriage); and Religious denotes the structure of praying together (e.g. at temple,church, or mosque) in the village.

8

Page 10: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

new practices, technologies, and ideas. Sharing more kinds of social relationships increases

the valence of word-of-mouth and social influence (Brummitt et al., 2012; Cardillo et al.,

2013; Kim and Goh, 2013; Simpson, 2015), but this valence has the potential to amplify

positive as well as negative influence. Second, multiplexity in people’s social ties can create

conflicts of interest to share information when relationships involve incompatible sets of re-

lational norms. These conflicts can deter di↵usion by motivating people to skew and bias

information (cf. Moore and Loewenstein, 2004). In what follows, I propose that multiplexity

a↵ects social di↵usion by increasing the number of social pathways for sharing and obtain-

ing information (i.e. serving as a conduit of information) and by enabling actors to spread

information opportunistically (i.e. serving as a sieve of information).

2.2 Multiplexity as a Conduit of Information

Multiplexity can act as a powerful conduit of information by increasing contact, commitment,

and trust among people (Friedkin and Johnsen, 2011). Relationships that involve interaction

across multiple domains are generally associated with greater frequency of interaction and

more relational resources (e.g. trust, rapport) (Verbrugge, 1979; Ferriani et al., 2013; Uzzi,

1997). The norm of reciprocity enables relational stability in multi-role structures (Evan,

1962; Gemmetto et al., 2014) as it creates inter-dependence in actions and interests across

di↵erent domains.

Although embeddedness and network expansiveness a↵ect information flows (Baker, 1984;

Uzzi, 1997), we do not fully understand how multiplexity is implicated in these processes.

Extant research suggests that multiplexity should enable di↵usion cascades (Brummitt et al.,

2012; Centola, 2015). Multiplex ties can provide benefits such as fine-grain information trans-

fer, increased trust, and reciprocity that amplify the valence of word-of-mouth in di↵usion

9

Page 11: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

processes.

People who are connected in multiple ways have more pathways to being influenced

by and obtaining information from others. For example, neighbors who are also friends

are more likely to influence each other than people who live in adjacent houses but never

interact (McPherson et al., 2001). Multiplex relationships and exposure to similar social

settings can also strengthen the correspondence between individual and collective beliefs,

expectations, and behavior that increase the probability that two people come to behave

and think similarly (Friedkin and Johnsen, 2011; Brummitt et al., 2012; Goldberg, 2011).

People who interact across many dimensions of life are far more likely to observe each other’s

behaviors and choices over time and to hold shared understandings of the world, compared

to people who rarely see each other and are exposed to di↵erent social worlds (Baldassarri

and Goldberg, 2014; Goldberg and Stein, 2016).

Prior research in social psychology (Maitner et al., 2010) demonstrates that the social

context within which information is shared a↵ects the meaning of information beyond its face

content. At a basic level, a person’s relationships with others provide a social context in which

information can be shared, interpreted, and understood. Having more social relationships

and more frequent contact with others helps an actor to form expectations about others’

behavior and creates cognitive boundaries between people seen as trusted insiders and those

dismissed as outsiders (Merton, 1972; Burt, 2012). Information is more credible if it comes

from trusted insiders who share multiple types of relationships with another person than

from illegitimate outsiders who are socially disconnected (Merton, 1972). All things being

equal, the more multiplex that a relationship between two people is, the more opportunities

that these people will have to share information and influence each other.

Proposition 1. Multiplexity increases the valence of word-of-mouth information in conta-

10

Page 12: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

gion processes by creating more social pathways for sharing and obtaining information.

In addition to being conducive to social influence and information, multiplexity can

also pose conflicts for individuals that a↵ect how people communicate and influence others

through their relationships. The activation of conflicts in people’s roles and interests can

influence di↵usion dynamics.

2.3 Multiplexity as a Sieve of Information

Diversity in people’s interests and social commitments produces heterogeneity in people’s so-

cial roles and relationships (Snoek, 1966:364). When people’s roles and relationships involve

incompatible sets of norms, they can create competing interests for people to share informa-

tion. Consider two people, Tom and Jerry, who are close friends and co-workers. Suppose

that Tom learns about a great job opportunity at their organization that he is interested in,

but for which his friend Jerry is better qualified. By the norms of friendship, Tom should

tell Jerry about this great job opportunity. Yet, the competing interest created by Tom’s

economic relationship with Jerry and Tom’s own aspirations for this job could prevent Tom

from sharing job-related information with Jerry.3

Multiplexity can pose conflicts in people’s interests when it involves di↵ering logics for

sharing information. Relational logics define the norms and expectations that guide peo-

ple’s interactions with others. In the prior example, the two logics present in Tom and

Jerry’s relationship were communal (friendship) and exchange (employment relationship).

Prior literature in social psychology draws a distinction between the functions and con-

tent of communal and exchange relationships in social interactions and patterns of resource

3The counter-factual scenario is that if Tom and Jerry were only friends and had no other relationshipsamong them that complicated matters, Tom would share information with Jerry about job opportunitiesthat were well suited to Jerry’s qualifications.

11

Page 13: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

sharing (Clark and Mills, 1993). Communal logics pertain to norms and expectations to

share resources with another person in response to a general concern for this person’s needs.

These logics arise in relationships such as between a parent and a child, and between personal

friends, and guide prosocial behavior such as charitable donations and selfless acts of courage

(e.g. rescue workers). By contrast, exchange logics pertain to norms and expectations of

sharing resources with the condition of receiving reciprocal benefits (e.g. lending money

with the expectation of receiving back principal and interest). Exchange logics arise in rela-

tionships such as between two trade partners, and between people who exchange advice and

information (with the expectation of receiving similarly beneficial advice and information).

Social ties involving communal logics in which the norm is to provide resources in response

to concern for the other person’s needs should promote the spread of information even when

information has no personal advantages to informants. By contrast, social ties involving

exchange logics (e.g. advice exchange, money exchange) in which the norm is to provide

resources with the expectation of receiving comparable benefits, should deter the spread

of information unless informants derive personal advantages from sharing information (e.g.

they are financially incentivized to spread information).4 These di↵ering logics for spreading

word-of-mouth should a↵ect whether social ties promote or impede di↵usion.

Multiple role-sets can a↵ord a person greater autonomy (Coser, 1991), but also enable op-

portunism (Padgett and Ansell, 1993) by creating conflicts of interest. A conflict of interest,

also termed a “conflict of roles” arises when a person is (1) in a position to influence others’

behavior to his personal advantage and (2) in a position where he has competing interests

by virtue of having multiple roles (Davis, 2012; Moore et al., 2010; Moore and Loewenstein,

2004). A conflict of interest defines a “a situation in which some interest of a person has a

4For a study that examines the e↵ects of financial incentives on di↵usion by word-of-mouth and socialinfluence see BenYishay and Mobarak (2016).

12

Page 14: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

tendency to interfere with the proper exercise of his or her judgment on another’s behalf”

(Davis, 2012:571) and is related to but distinct from role conflict, defined as “a situation

in which incompatible demands are placed upon an actor (either an individual or a group)

because of his role relationships with two or more groups” (Gullahorn, 1956:299). Conflicts

in people’s motives and interests can arise when people occupy multiple roles and therefore

face opposing interests that guide their judgments and actions (Simmons, 1968:482).

Prior research in social psychology has shown that conflicts a↵ect whether people dis-

close information and whether the information they disclose is biased (cf. Cain et al., 2005;

Loewenstein et al., 2011). Research in sociology has also shown that role multiplicity in-

creases a person’s agency to deploy multivocality, a style of communication in which an

actor conveys di↵erent messages to di↵erent sets of people (Padgett and Ansell, 1993; Coser,

1991). As Coser (1991) notes, when people face conflicts in their roles and interests, “they

...must decide whether to abide strictly by the rules or to reinterpret or even defy them, and

they must weigh each decision in relation to their own purposes of action and the purposes

of others” (Coser, 1991:66-67). Actors in multiplex positions are thus able to act as change

agents and as sphinxes by using multivocality and robust action to communicate strategi-

cally (Padgett and Ansell, 1993; Go↵man, 1959). Although brokers often inhabit multiple

roles that enable them to span structural holes (Burt, 1992) and behave opportunistically

(Padgett and Ansell, 1993; Baker, 1984), most theories of di↵usion assume that the messages

conveyed through social ties are uniform in their e↵ects on the di↵usion of social practices,

ideas, and technologies. In many contexts, however, it is simplistic to assume that people

share the same information with everyone they know, or that they influence all of their

social contacts in similar ways. Rather, a more realistic scenario, which takes information

and influence as emergent properties of social ties, is that people share information oppor-

13

Page 15: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

tunistically and strategically with their social contacts, the greater the multiplexity of these

contacts.

Brokers, spies, sphinxes, and crooks, for example, gain personal advantages by carefully

managing information across di↵erent audiences (Padgett and Ansell, 1993; Aven, 2015;

Smith and Papachristos, 2016). Cosimo de’ Medici’s multiple connections in the marriage,

economic, and patronage elite structures of Renaissance Florence, for example, enabled him

to wield unusual influence over international trade and state a↵airs. Al Capone’s multiple

involvement in the structure of organized crime in Prohibition Era Chicago with his brothers,

bootleggers, and criminals similarly contributed to the resilience of his Syndicate (Smith and

Papachristos, 2016). When they have conflicting interests, multiplex actors can influence

di↵usion processes by biasing information toward their best personal interest.

If the most central individuals in a communication network hold multiple roles and if these

roles create competing interests, then multiplexity should a↵ect how central informants share

information with others and influence di↵usion dynamics. An informant who has conflicts

of interest may be motivated to share information with others more selectively than an

informant who does not have these conflicts.5 Therefore, a second proposition is needed to

place a limit on the informational conductivity of multiplex ties in the presence of conflicts

of interest:

Proposition 2. Multiplexity deters the benefits of word-of-mouth information for social

di↵usion when it impedes informants’ ability to provide objective and unbiased information.

Proposition 2 places a limit on the informational conductivity of multistranded ties

through the activation of conflicts that constrain and bias communication. This limit inheres

5As a testament to the importance of conflicts for individual behavior, most academic journals requireauthors to disclose conflicts of interest resulting from role conflicts (e.g. as impartial researchers and ben-eficiaries of funding from private interest groups) when publishing information that has the potential toinfluence behavior.

14

Page 16: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

from informants’ agency to skew or withhold information that is not in their personal inter-

est. The combination of having (1) multiple types of relationships and (2) private conflicts

of interest are necessary conditions to dampen di↵usion through word-of-mouth.

Based on these propositions, informants in the villages who had competing private inter-

ests to spread word-of-mouth about the benefits of microfinance should have constrained the

di↵usion of microfinance. Further, the more multiplex that these informants’ relationships

were with eligible program participants, the more that they should have influenced their

contacts’ choices and probability of enrollment. I present evidence that supports each of

these explanations for di↵erent patterns of di↵usion.

3 Methods

There are many methodological challenges to studying the e↵ects of multiplexity on contagion

by word-of-mouth. The first challenge is that the endogeneity of social interaction (and the

formation of multiplex relationships) renders the causal e↵ects of multiplexity on behavior

di�cult to isolate from the e↵ects of individual attributes (cf. Aral et al., 2009). The second

challenge is bounding the sampling space for a set of social interactions and being able to

isolate information flows and social interaction so that they are purely endogenous over time

in di↵erent communities. The third challenge is obtaining sociometric and demographic data

about individuals in order to isolate the e↵ects of individual attributes on the propensity to

adopt in relation to the e↵ects of social processes.

The present study overcomes the first challenge by using quasi-random variation in the

seeding of information. By examining the e↵ects of selectively injected information in di↵er-

ent communities, this study identifies how exposure to information through di↵erent social

pathways a↵ected individual behavior. The study also overcomes the challenge of bounding

15

Page 17: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

the sampling space for social interactions by using a sample of geographically remote villages

that had limited contact with people outside the villages during the observation period. Se-

lecting social networks in this way bounds potential social interactions to individuals within

the same community and thus limits the likelihood that people were a↵ected by outside

sources of information. This study overcomes the final challenge by using a detailed dataset

that includes people’s characteristics (e.g. Caste, gender, occupation) as well as their pat-

terns of social interaction, thus enabling analysis of how these characteristics shaped patterns

of interaction and individuals’ baseline probability of adoption.

3.1 Setting

The 43 villages selected for the informational intervention about the benefits of microfinance

were located in Karnataka, a state in southwestern India. Karnataka is the seventh largest

Indian state by area and the eight largest by population. The state spans an area of about 192

thousand square kilometers and is inhabited by about 61 million people. This area includes

the Deccan plateau, the Western Ghats, and the coastal region of Karavali. Agriculture and

horticulture dominate economic life (Roy, 2000). The region is irrigated by two major river

systems – the Krishna in the north and the Kaveri in the south – that flow eastward into the

Bay of Bengal and support a variety of agricultural crops (e.g. rice, finger millet, mulberries).

Kannada, the o�cial language of the state, is one among many spoken, including Tamil,

Urdu, and Hindi. Islam was introduced into the region under the Bahamani and Bijapur

sultanates that ruled parts of Karnataka in the fifteenth and sixteenth centuries (Eaton,

2005). Christianity also arrived to the region with the Portuguese in the sixteenth century

(Keay, 2000). Today, the majority of the population is Hindu, but many villages have

prominent Muslim and Christian minorities.

16

Page 18: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

3.1.1 Microfinance Model of Lending

The type of microfinance program that was introduced in each village used a joint-liability

group-based method of lending small sums of money to poor women, similar to group lend-

ing methods used widely among other microfinance organizations (Armendariz and Morduch,

2005). This method allows interested and eligible participants in a given village to enroll in

microfinance once the program becomes available in the village. Eligible microfinance par-

ticipants in the villages were women aged between 18 and 57 who were permanent residents

of the village, who were able to work but also received government food subsidies, and whose

husbands or guardians allowed them to enroll in microfinance. Only one woman from each

family could join a lending group and receive loans.

Eligible participants could borrow micro-loans from the microfinance organization (“bank”)

by joining a lending group. An eligible borrower who decided to enroll in microfinance was

placed in a group with four other borrowers, who become jointly liable for repaying loans

from the bank. To be placed into a lending group, prospective members were asked to con-

firm that they knew each other and that they were willing to take joint responsibility for the

loans of the other members in the lending group. Lending group members, however, were

not allowed to be blood relatives or kin and were also barred from being members of more

than one lending group, to prevent exploitation of the joint-liability model.

From a purely economic perspective, microfinance lending provided lower interest rates

and more capital than informal money lending in the villages.6 The maximal initial loan

that each woman was eligible to receive from the microfinance organization was 10,000 rupees

(about $200). This loan was repayable over 50 weeks in small weekly installments and carried

an interest rate of about 28 percent per year. The loans were uncollateralized and without

6Peer lending is the model of obtaining loans directly from other villagers rather than from a microfinanceorganization.

17

Page 19: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

any form of security deposit. By comparison, money lenders in the villages o↵ered loans that

were typically much smaller, around 50 rupees, and carried varying interest rates ranging

from 40 percent to 200 percent per year (Banerjee and Duflo, 2011:160). For these purely

economic reasons, in theory every woman eligible to enroll in microfinance who had borrowed

money from other villagers should have benefited from microfinance.

3.2 Seeding Information in the Villages

This study uses a quasi-experiment in the introduction of information about the benefits

of microfinance in each village to understand how multiplexity a↵ected microfinance en-

rollment.7 In June 2006, an independent non-profit microfinance organization, Bharatha

Swamukti Samsthe (BSS), identified people from 43 villages in rural southern Karnataka,

India to inform about the benefits of microfinance. BSS’s vision is “to see an India where

every child, woman and man can be the best that God meant for him or her to be, without

the burden of poverty,” and its mission is “to do large scale poverty alleviation by provid-

ing micro-finance services to poor women, and through them to their families, facilitating

increased earnings, better money management, and life quality improvement.”8 Given this

vision and mission, BSS chose people in each village to act as informants about microfinance

that it believed would be good communicators and advocates for microfinance: teachers,

shopkeepers, and leaders of self-help groups.

Figure 3 shows the time-line for the informational intervention in the villages between

January 2006 and September 2010. The intervention was designed to enable di↵usion through

a two-stage process intended to first raise awareness about the benefits of microfinance (June

7The intervention in this setting was a “quasi” experiment because people connected to and unconnectedto informants in the villages were not independent groups.

8BSS Microfinance Private Limited, accessible at: http://www.bssmicrofinance.co.in. AccessedNovember 11, 2016.

18

Page 20: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Figure 3: Time-line

2006-January 2007) and leverage social influence mechanisms within the villages to promote

program enrollment (February 2007- September 2010).

The informational intervention in each village was a private meeting between BSS credit

o�cers and selected informants from the village. The selection of informants was made blind

to the structures of peer lending relationships that existed in each village prior to the in-

troduction of microfinance (Fig. 4 illustrates the structural positions of informants in select

villages). The selection of individuals without prior knowledge of their social roles (e.g. as

money lenders) produced a natural experiment by virtue of introducing exogenous varia-

tion in whether informants had conflicts of interest and in how multiplex their relationships

were with eligible microfinance participants. At the informational meeting with selected

informants, BSS explained how microfinance worked, described its benefits, and asked in-

formants to tell their contacts who were eligible for the program. In February 2007, BSS

19

Page 21: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Figure 4: Informant Positions in Select Villages

MoneyGoodsSocialMedicalVisitAdviceKinshipReligious

(a) village 1

MoneyGoodsSocialMedicalVisitAdviceKinshipReligious

(b) village 10

MoneyGoodsSocialMedicalVisitAdviceKinshipReligious

(c) village 25

MoneyGoodsSocialMedicalVisitAdviceKinshipReligious

(d) village 43

Note: Node size increasing with multiplexity. Informants shown in red.

20

Page 22: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

introduced microfinance in the 43 villages, and eligible individuals were able to enroll. Dy-

namic microfinance enrollment data within villages were collected at the end of three-month

periods between February 2007 and September 2010 in each village.

3.3 Data

3.3.1 Household Surveys

In June 2006, prior to BSS’s entry into each village, the research team that collected the data

administered village questionnaires and conducted a complete census of all households. The

questionnaires used to collect data asked individuals about village leadership, the presence of

non-governmental organizations in the village, the presence of savings self-help groups, and

various geographical features of the area (rivers, mountains, and roads). The census collected

demographic information about individuals in each household, the locations of their homes,

and data about the types of amenities to which they had access (e.g. latrine, electric power).

These surveys were used to select villages for the intervention.

3.3.2 Individual Surveys

Upon completion of each village survey and household census, the research team administered

detailed individual demographic and sociometric surveys to a random sample of villagers,

stratified by religion and geographic location. More than half of the BSS-eligible partici-

pants and their spouses completed the individual questionnaire before the introduction of

microfinance in February 2007. Villagers were asked to name people with whom they (1)

borrowed and lent money, (2) exchanged goods such as kerosene and rice, (3) visited each

others’ homes, (4) shared advice about di�cult personal decisions, (5) were related by blood

or marriage (kinship), (6) helped with medical advice and emergencies, (7) attended social

21

Page 23: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

events in the village (e.g. marriages and festivals), and (8) prayed together (e.g. at temple,

church, or mosque). The relations were directed.

The sociometric data from these surveys were compiled into adjacency matrices for each

village. Individuals, households, and villages were identified with unique codes that enabled

data matching to demographic and enrollment data. The demographic data included a per-

son’s gender, age, marital status, Caste and sub-Caste group, religion, mother tongue, lan-

guages spoken, occupation, highest education level, ration card color (socioeconomic status),

and an indicator for whether a person had savings. The household level information included

the type of house the family lived in, the person IDs of other members of the household,

and an indicator for whether anyone in the household enrolled in microfinance. Since micro-

finance eligibility was restricted to one woman per household, household participation data

captured whether an eligible woman in a given household enrolled in microfinance. Data on

microfinance enrollment were collected in three-month intervals following the introduction of

microfinance until September 2010, or until there were no changes in the month-over-month

rate of enrollment, whichever happened sooner.

3.4 Measures

The individual outcome of interest in this study is whether an eligible microfinance par-

ticipant in a village enrolled in microfinance during the observation period, between 2007

and 2010. The aggregate outcomes of interest are the rates of microfinance enrollment in

each village over time. I examine how these outcomes varied with the multiplexity between

eligible microfinance participants and their informants, depending on whether informants

had private conflicts to endorse microfinance.

Conflict defines whether an informant had competing motives to disclose information

22

Page 24: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

about or endorse microfinance. Peer lenders to eligible microfinance participants in the vil-

lages who were selected as informants plausibly had such conflicts to endorse microfinance

for several reasons. First, the loan amounts that BSS o↵ered – 10,000 rupees – were equiva-

lent to borrowing from 200 peer lenders. Keeping in mind that the average eligible villager

borrowed money from about three peer lenders, and the most lenders from whom anyone in

the villages borrowed money was 39 people (about five times fewer than needed to match

the loan amounts o↵ered by BSS), microfinance meant that a woman would no longer need

to borrow money from money lenders in her village. Money lenders who were inadvertently

selected as informants could plausibly have been disinclined to share information about the

benefits of a competing model. By contrast, villagers selected as informants who had not lent

money to eligible participants plausibly had no competing interests to endorse microfinance.

Relational multiplexity refers to how multistranded a relationship was between an eligible

microfinance participant and each of her contacts prior to the introduction of microfinance.

This variable ranges from zero to seven, with higher values denoting more overlapping roles

and relationships (e.g. as lender, friend, and kin). The variable takes on a value of zero if

two people shared only a peer lending relationship, and a value of seven if two people shared

other social relationships across every dimension of interaction measured in their village.

In order to assess whether the di↵erent types of roles and relationships in the villages

involved di↵erent types of relational logics (e.g. communal and exchange), I examined the

extent of structural overlap between di↵erent relational structures in the village (e.g. kin-

ship, home visits). I measured overlap using the structural distance between the adjacency

matrices of di↵erent sets of role-relations (Butts and Carley, 2001), where the structural

distance was operationalized as the Hamming distance (cf. Hamming, 1950). The Hamming

distance between a set of social interactions (e.g. among three people A, B, and C) can be

23

Page 25: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

represented graphically as a cube in Euclidean space, as shown in Figure 5.

Figure 5: Graphical Representation of the Hamming Distance

A

B

C

Consider the path 010!101, for example. This path is denoted by dashed lines in the

cube. We can see from the diagram that the minimal number of permutations (coordinate

changes in Euclidean space) needed to transform the binary string 010 into the binary string

101 is three. That is, the structural di↵erence between the network represented by the

presence of a tie between actors B and C, and the network represented by the presence of

ties between A and B and B and C, is three permutations. Similarly, the path 011!101,

which is denoted by bold lines in the cube, has a Hamming distance of two. More generally,

the Hamming distance between the elements of two graphs g1

and g

2

with adjacency matrices

A(1) and A(2) can be computed as

dH(g1, g2) =NX

i 6=j

hA

(1)

ij 6= A

(2)

ij

i

The formula above gives the number of permutations needed to transform the set of edges

of g1

into that of g2

. The Hamming distance represents the relative similarity between two

24

Page 26: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

or more patterns of interaction among the same set of actors.9

To illustrate how this measure works, suppose that Apurva shares di↵erent connections

to two other people in a village – Bharat and Chanda – who exchange money, pay each

other home visits, and share advice. Suppose that Apurva borrows money exclusively from

Chanda, but visits both Bharat and Chanda at home. Apurva’s interactions with Bharat

and Chanda can be represented as the binary string 010 (money) and the binary string 101

(visits). The structural distance between the patterns of money exchange and home visits

among these actors can be computed by measuring the length of the path 010 ! 101 in

Euclidean space, which is exactly three. This measure allows comparison of the similarity in

interaction patterns within a community with a fixed number of actors.

The analyses also control for social heterogeneity in the villages. Heterogeneity defines

the dispersion of people in each village across di↵erent social groups (Blau, 1977:9). This

measure is computed as 1�PN

i=1

p

2

i and ranges from zero to one.10 If all people are in one

group, the measure will take on a value of zero, indicating that there is no heterogeneity.

Conversely, if people are about evenly distributed across many groups, then the measure will

approach one, indicating perfect heterogeneity. Prior research has shown that heterogeneity

a↵ects the extent to which people adopt innovations by learning from the judgments of others

about the value of these innovations (Assenova, 2016).

3.5 Characteristics of Eligible Participants

Table A1 provides a glimpse into the characteristics of microfinance-eligible participants

(the risk set of people who could have enrolled in microfinance). Among the 16,984 residents

9The graph product-moment correlation (Butts and Carley, 2001:28-29) between the structures of moneyexchange and social visits were used as a robustness check to compare the relative overlap in these structures.The graph correlations yielded similar results to the Hamming distance.

10This measure is the same as Simpson’s index of ecological diversity among species.

25

Page 27: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

surveyed in the villages, women comprised 55 percent, and 93 percent of them were eligible

for microfinance. Microfinance eligible participants were 36 years old on average at the time

of the intervention. The majority had no household savings and no education at the primary

school level or beyond. Many (39 percent) received food rations from the government as part

of famine relief programs. Most were occupied in agricultural work (growing and harvesting

finger millet, rice, and mulberries), but also in animal husbandry (e.g. cow herding and

poultry farming), sericultural work (e.g. silkworm breeding and silk production), and crafts

(e.g. tailoring and weaving).

Eligible participants were located in villages with about 238 households and about 1,118

residents on average. The average household size in these villages was about five people.

The villages were relatively religiously homogeneous (mostly Hindu), but had substantial

linguistic heterogeneity (e.g. Kannada, Urdu, Tamil, Hindi), and Caste heterogeneity (e.g.

Brahmin, Srivaishnava, Vishnava).11

The majority of eligible villagers (58 percent) came from Other Backward Class (OBC)

Castes, a designation used by the Indian government to denote the most socially and educa-

tionally disadvantaged groups of people. About four percent of microfinance-eligible villagers

were religious minorities (Christians and Muslims). The majority (73 percent) were native

Kannada speakers.13

Microfinance-eligible villagers borrowed money from about three other people in their

villages prior to the introduction of microfinance. At the time of the informational inter-

11The Caste groups in each village include Adi Dravida12, Adi Karnataka, Bhovi, Brahmin, Nayak, SriVishnav, Srivaishnava, Vishnava, and Vokkaliga. Traditionally, Brahmins occupied the highest status socialpositions while Adi Dravida (Dalit) occupied the lowest status social positions (Ambedkar, 1917; Vaid, 2014).The other classifications are Scheduled Castes (the higher-ranking Castes) and Scheduled Tribes (membersof indigenous tribes).

13Many other languages were spoken in the villages, including Urdu, Telugu, Tamil, Marathi, Tulu, Hindi,Konkani, Malayalam, and Kodava Takk, reflecting the cultural diversity of southwestern India.

26

Page 28: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

vention, about 17 percent of these villagers were informed by BSS and about 17 percent

had a peer lender who was informed by BSS. By the end of the observation period, about

20 percent of eligible villagers who had previously obtained loans from other villagers had

enrolled in microfinance.

3.6 Characteristics of Interactions in the Villages

Cluster analysis of the similarity across the patterns of social interaction in the 43 villages

supports the presence of di↵erent relational logics. Figure 6 plots a dendrogram of the

Hamming distance among structures of interaction across the 344 networks. This diagram

shows two structurally distinct clusters: a communal cluster (bottom left) and an exchange

cluster (top right). Examining the individual networks in each cluster further supports

the presence of di↵erent motives and expectations to share resources. Patterns of goods

exchange, for example, appear most structurally similar to patterns of money lending in

the villages, meaning that money and goods transactions involved similar logics of tit-for-tat

exchange (Table A3 gives the precise distance). The graph correlations between peer lending

and advice exchange across the 43 villages were 0.64 on average (Hamming distance = 1,156).

Meanwhile, the communal patters of interaction involved religious worship, attendance of

marriages and festivals in the villages, and the provision of medical help to others. These

structures appear distinct from the exchange structures, as evidenced by their low graph

correlation of 0.38 with money lending and goods exchange (Hamming distance = 4,751).

Patterns of interaction and money lending in the villages were also highly multiplex.

Among villagers who borrowed money from each other, 78 percent also shared advice, 85

percent also exchanged goods, 84 percent also paid home visits, 77 percent also shared medi-

cal advice, and 63 percent also attended temple together. The observed levels of multiplexity

27

Page 29: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Figure 6: Communal and Exchange Relationships

Note: Diagonal elements denote perfect structural similarity. Lighter shading denotes greater structural(Hamming) distance.

are unsurprising given the small size of the villages and the tight-knit fabric of social life in

these relatively isolated communities.

Most of the peer lending relations that existed in the villages prior to the introduction

of microfinance were among people of the same Caste, language, and religious groups. For

example, 85 percent of all peer loans occurred within people’s Caste in-groups, more than

97 percent occurred within people’s religious in-groups, and 85 percent occurred within

people’s language in-groups. These high proportions of in-group exchange demonstrate that

peer lending depended on trust, and that in-group members were more trusted than out-

group members (Abascal and Baldassarri, 2015; Tajfel and Turner, 1986). Common in-group

identity appears to have a↵ected the functioning of the peer lending structures.14

14Social assortment based on Caste, language, and religious in-groups in this context disadvantaged peoplefrom low socio-economic status groups, because they were restricted to borrowing money from people with

28

Page 30: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

3.7 Models

I use linear in-means models to understand what e↵ect exposure to multiplex informants had

on an eligible participant’s probability of enrolling in microfinance. Linear in-means models

have gained traction in econometrics as a method for identifying the e↵ects of exposure to

social referents on individual choices and outcomes (Graham and Hahn, 2005; Brock and

Durlauf, 2007; Bramoulle et al., 2009; Blume et al., 2015). The goal of these models is to

estimate an individual outcome of interest as a function of exposure to social contacts. For

example, one can use these models to estimate the probability of an individual engaging in

a particular kind of behavior (e.g. adoption) conditional on the behavior of her contacts. In

special cases when exposure is attributable to a controlled or natural experiment, the e↵ects

of exposure on behavior can be interpreted as causal (Fafchamps, 2015).

Econometric identification in this study relies on di↵erential exposure to two types of

information pathways in the villages: pathways in which informants had private conflicts of

interest to share information and pathways in which informants did not have these conflicts.

The in-means models estimate the probability that an eligible participant (Ego) in village

k enrolled in microfinance as a function of having a contact (Alter) who was informed by

BSS about the benefits of microfinance. Because information was as-if randomly seeded with

respect to the level of social multiplexity with contacts in each village, eligible participants

did not choose whether they were exposed to informants with conflicting interests. The

identification condition is that the level of multiplexity and the presence of conflict of interest

in the information pathway between Ego and Alter before the introduction of information

plausibly a↵ected Ego’s decision to enroll in microfinance, but was orthogonal to whether

similarly low economic means, producing a cycle of poverty. One advantage of microfinance was thereforethat it allowed people from low status Castes and religious minorities to gain greater access to finance.

29

Page 31: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Alter was informed by BSS.15

3.8 Baseline Linear In-Means Models

The baseline in-means models parse the e↵ects of endogenous (word-of-mouth) and exogenous

(direct) information pathways and social influence on individual behavior. Word-of-mouth

comes from social contacts who were selected as informants. By contrast, direct information

comes from the microfinance organization. Social influence means that a person’s social

contact enrolled in the program. The functional form of these models is:

y

ki = ↵ + �Ego

ki| {z }

direct information

+ ⇢

"gij/

X

j

gij

#Alter

kj

| {z }word-of-mouth

+ �

"gij/

X

j

gij

#Enroll

kj

| {z }social influence

+Z0�+V0 + "

ki

where y

ki is an indicator if a person i in village k enrolled in microfinance during the obser-

vation period. Ego

ki is an indicator if an eligible microfinance participant in village k was

informed about microfinance by BSS. The variable gij is an indicator that takes on a value

of one if Ego and Alter shared a relationship (gij = 1) in the adjacency matrix G (gij 2 G).

In the conflict condition, G includes the intersection between a person’s peer lending rela-

tionships and her other social relationships in the village (e.g. money \ kinship \ advice,

etc.). In the no conflict condition, G is the union between a person’s social relationships

across all social domains except peer lending (e.g. home visits [ social [ advice, etc.). The

termhgij/

Pj gij

irepresents the weight of the influence of Alter j on Ego in relation to all

other contacts. Alter

kj is an indicator if alter was a BSS-selected informant. Enroll

kj is an

15Indeed, if BSS had known about informants’ competing interests, it might have not selected particularindividuals to act as informants in the villages.

30

Page 32: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

indicator if Alter enrolled in microfinance.16

� is the exogenous e↵ect of receiving information directly from BSS. By contrast, ⇢ is

the endogenous e↵ect of receiving information from contacts within one’s village. Similarly,

� is the endogenous e↵ect of the behavior of contacts (social influence) on Ego’s behavior.

The models include a vector of individual controls, Z, and village intercepts, V. Individual

controls are a person’s gender, age, Caste group, occupation, religion, highest educational

level, language, ration card color (socioeconomic status), and an indicator for whether a

person had any savings. The inclusion of these controls is important for isolating the e↵ects

of individual heterogeneity and distinguishing them from the e↵ects of information. Standard

errors are clustered to account for the multiple membership of the same individuals across

dyads (cf. Aronow et al., 2015). I estimate separate models using structures of information

transmission in which informants had no conflicts and structures in which informants had

conflicts of interest to understand how these conflict a↵ected the di↵usion of microfinance

among eligible participants.17

3.9 Linear In-Means Models with Multiplexity

The baseline model can be expanded to account for multiplexity by adding a variable that

measures the multiplexity of social contacts. This modification enables analysis of how

multiplexity a↵ected the probability of enrollment jointly with endogenous information and

16Note that alters who were eligible to enroll were fellow women who met the eligibility criteria (low SESstatus, not related by kinship ties, working age) to enroll in microfinance.

17The structures with the conflicts were composed of all the multiplex relationships among eligible partici-pants and their social contacts in which informants were also peer lenders to eligible microfinance participants.The structures without conflicts were composed of multiplex relationships in which informants were not peerlenders to eligible participants.

31

Page 33: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

social influence. The functional form of the in-means model with multiplexity is:

y

ki = ↵ + �Ego

ki| {z }

direct information

+ ⇢

"gij/

X

j

gij

#Alter

kj

| {z }word-of-mouth

+ �

"gij/

X

j

gij

#Enroll

kj

| {z }social influence

+ ⇣Mij|{z}multiplexity

+ ⇢

"gij/

X

j

gij

#(Mij ⇥ Alter

kj )

| {z }word-of-mouth from multiplex contact

+ �

"gij/

X

j

gij

#(Mij ⇥ Enroll

kj )

| {z }influence from multiplex contact

+Z0�+V0 + "

ki

The model above is similar to the baseline linear in-means model but includes the relational

multiplexity between i and j as an independent variable, Mij. The coe�cient ⇣ captures

the main e↵ect of multiplexity on a person’s probability of enrolling in microfinance. In the

conflict condition, multiplexity with contact means that a person had multiple social ties to

her peer lender. In the no conflict condition, multiplexity with a contact means that a person

had multiple social ties with a person who was not a peer lender. This model also includes

interactions between multiplexity and word-of-mouth information and multiplexity and social

influence. The first interaction captures the e↵ects of word-of-mouth from multiplex contacts

and the second interaction captures the e↵ects of social influence from multiplex contacts.

4 Results

Coe�cient estimates of the in-means models, along with their 95 percent confidence intervals,

are shown in Figure 7 (baseline models) and Figures 8- 9 (models with multiplexity). These

estimates show the e↵ects of word of mouth (endogenous information) on the probability of

enrollment in microfinance among eligible participants. The gray scale is darker for models

with more controls for individual and village-level heterogeneity. Black dots and whiskers

indicate estimates from the most saturated models.

32

Page 34: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Figure 7: Word-of-Mouth and the Probability of Microfinance Enrollment

Probability of Enrollment-0.4 -0.2 0.0 0.2 0.4

Word ofMouth

(a) No conflict

Probability of Enrollment-0.4 -0.2 0.0 0.2 0.4

Word ofMouth

(b) Conflict

Note: Coe�cient estimates (dots) with 95% confidence intervals (lines). Darker shading denotes estimatesfrom more saturated models inclusive of individual and village-level controls. Results are also shown inTable A4 (no conflict) and Table A5 (conflict).

Beginning with Figure 7b, we can see from the coe�cient estimates of word of mouth that

eligible participants connected to informants with competing interests were about 30 percent

less likely to enroll in microfinance than participants connected to informants without these

conflicts of interest (cf. Figure 7a). Recall that informants with conflicts in the villages were

peer lenders who were selected to spread the word about the benefits of microfinance. By

contrast, informants without conflicts were people who were selected as informants but who

had not previously lent money to eligible program participants.

Figure 8 shows that eligible participants who had (audience-side) conflicts because they

were in multiplex relationships with their peer lenders were less likely to enroll in micro-

finance. Specifically, every additional relationship that two people shared was associated

with a 3 percent decrease in the probability of enrollment when an eligible microfinance

33

Page 35: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Figure 8: Multiplexity and the Probability of Microfinance Enrollment

Probability of Enrollment-0.03 -0.02 -0.01 0.00 0.01 0.02 0.03

Multiplexity

(a) No conflict

Probability of Enrollment-0.03 -0.02 -0.01 0.00 0.01 0.02 0.03

Multiplexity

(b) Conflict

Note: Coe�cient estimates (dots) with 95% confidence intervals (lines). Darker shading denotes estimatesfrom more saturated models inclusive of individual and village-level controls. Results are also shown inTable A6 (no conflict) and Table A7 (conflict).

participant was multiplex with a peer lender (conflict) versus when she was multiplex with

a contact who was not a peer lender (no conflict). This e↵ect translates to a decrease in

the probability of enrollment of about 15 percent, given the high level of multiplexity, be-

cause the average person in these villages shared about five relationships with every contact.

These results demonstrate that conflicts on both sides of a relationship contributed to lower

enrollment.

Figure 9 plots the coe�cient point estimates and 95 percent confidence intervals for the

joint e↵ects of word of mouth and multiplexity on the probability of enrollment for eligible

participants. Figure 9b shows that word of mouth from multiplex informants had a negative

e↵ect on the probability of enrollment as compared to word of mouth from simplex infor-

mants, but that these di↵erences are contingent on the presence of conflict for informants.

34

Page 36: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Figure 9: Word-of-Mouth from Multiplex Informants

Probability of Enrollment-0.4 -0.2 0.0 0.2 0.4

Word of Mouthfrom SimplexInformants

Word of Mouthfrom MultiplexInformants

(a) No conflict

Probability of Enrollment-0.4 -0.2 0.0 0.2 0.4

Word of Mouthfrom SimplexInformants

Word of Mouthfrom MultiplexInformants

(b) Conflict

Note: Coe�cient estimates (dots) with 95% confidence intervals (lines). Darker shading denotes estimatesfrom more saturated models inclusive of individual and village-level controls. Results are also shown inTable A6 (no conflict) and Table A7 (conflict).

5 Discussion

Multiplexity within villages was associated with dampened contagion about an economically

beneficial lending program (microfinance) despite enabling more pathways for people to ac-

cess information about the benefits of this program. These results appear attributable to

two mechanisms. First, multiplexity appears to have activated conflicts of interest for infor-

mants to share objective and impartial information about microfinance. Among informants,

these conflicts appear to have resulted in negative e↵ects of word-of-mouth on enrollment

in microfinance. These results appear consistent with explanations that informants misin-

formed their program eligible contacts or withheld information from these contacts about the

benefits of microfinance. Among eligible program participants, people who were exposed to

information but had multiple relationships with money lenders in their village were also less

35

Page 37: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

likely to enroll in microfinance, perhaps because they faced social repercussions for severing

ties with peer lenders who were also kin, neighbors, and friends.

Second, multiplexity appears to have further deterred microfinance enrollment by in-

creasing commitment to money lenders. Eligible microfinance participants were less likely

to enroll if their informants were multiplex money lenders than if they were simplex money

lenders. This evidence shows that multiplexity in the village money lending structures af-

fected individual decisions to stay in their existing lending relationships in the presence of

economically lower-cost alternatives such as microfinance.

These findings run counter to prevailing arguments that network density increases di↵u-

sion through social contagion (e.g. Centola, 2015; Brummitt et al., 2012). Extant theories

of di↵usion predict that informing socially central people in the villages (such as teachers,

shop keepers, and leaders of self-help groups) should have increased the rate of di↵usion by

influencing more people to enroll in microfinance. The key assumption behind these argu-

ments is that information spreads like diseases, whereby the most connected people are most

infectious. But this conceptualization of social contagion misses an important consideration:

People who are very central in social structure are often insiders into multiple social groups.

This multiplicity of roles makes very connected people some of the most conflicted about

which identities and information to deploy across di↵erent group boundaries. Occupancy of

multiplex social positions both gives people greater autonomy and agency in how they de-

ploy identities and enables them to be opportunistic about how they share information. The

potential for conflicts of interest that accompany multiplexity explains why greater density

in relationships does not necessarily promote di↵usion.

In the setting examined in this study, multiplexity a↵ected informants and eligible pro-

gram participants in ways that undermined the di↵usion of microfinance. Informants plausi-

36

Page 38: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

bly had conflicting motives to endorse microfinance, and eligible program participants plau-

sibly were concerned about severing their multiplex relationships with peer lenders and thus

reluctant to enroll in microfinance. Vested peer lending relationships with kin and friends

may have created social repercussions in the form of hurting the feelings of loved ones if an

eligible participant enrolled in microfinance. Multiplex peer lending relationships may also

have been more di�cult to re-organize than simplex relationships into smaller lending groups

with joint-liability to the microfinance organization, because peer lending relied on expan-

sive networks of multiplex relationships, whereas microfinance operated on an organizational

model of borrowing money from a bank.

The evidence supports the idea that the introduction of information activated conflicts

of interest for peer lenders who were selected as informants because informants occupied

multiplex positions in the villages. The evidence further supports the idea that multiplex-

ity amplified these conflicts and undermined the benefits of word-of-mouth information for

di↵usion. Conflicts appear to have exerted negative e↵ects on di↵usion through two distinct

social mechanisms. First, conflicted informants appear to have exerted negative influence

through word-of-mouth on eligible participants. Conflicted informants were in a position to

behave opportunistically (e.g. Baker, 1984) by either misinforming eligible participants or by

withholding information from them. Second, participants who shared both communal and

exchange ties with money lenders appear to have been more likely to stay in their lending

relationships despite having access to microfinance as an alternative way to obtain loans.

These findings demonstrate that multiplexity a↵ected the di↵usion of microfinance through

both informant-side and audience-side e↵ects.

The findings and theory contribute to research on conflict and contagion by explaining

how conflicts of interest a↵ect the di↵usion of new practices. The findings bear implica-

37

Page 39: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

tions for the impediments to di↵usion in embedded markets and for the social deterrents of

change in complex organizations. People tasked with implementing change in organizations

and markets are often socially central, but they also frequently face competing interests by

virtue of being multiple insiders into di↵erent groups. Central structural positions both ex-

pose individuals to more contacts that they can influence and enable these individuals to

communicate opportunistically. Although central informants are well positioned to di↵use

new practices, technologies, and ideas, they can also bias information and thereby deter

social di↵usion among their contacts. For example, if a manager who sits on the board

of another organization learns about a new process that can radically improve her organi-

zation’s performance, but jeopardizes her own job, she may never propose this innovation

despite its benefits. Who di↵uses information influences whether a new idea, practice, or

technology gains widespread social acceptance.18 The present research brings agency into

network-analytic approaches by showing that conflicts of interest among informants a↵ected

the benefits of word-of-mouth for di↵usion.

Important to explaining the puzzle of why we observe limited di↵usion in multiplex struc-

tures, despite high density of social relationships, is elaborating theoretically how communi-

cation di↵ers across di↵erent kinds of social ties and relationships, and in particular across

conflicted relationships where both communal and exchange relational logics are present.

Communication involves strategic choices about how, when, why, and with whom to share

information (Aven, 2015). Conflicts in people’s motivations for sharing information can

decrease communication below the level expected purely on the basis of social density.

Although density in social relationships can contribute to structural resilience and the

enforcement of social norms (Centola, 2015; Karlan, 2007), it can also create conflicting in-

18Interestingly, paying people or otherwise providing financial incentives to share information may over-come impediments to di↵usion, but may also undermine the credibility of informants.

38

Page 40: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

terests and norms of behavior (Merton, 1957; Coser, 1975; Padgett and Ansell, 1993). In

Indian villages, this conflict appears to have deterred participation in a program intended to

help women living in poverty. Inertia in social practices may be contingent on the multiplex-

ity of social structure, increasing with high levels of multiplexity (through role conflict) and

trumping the benefits of fine-grain information transfer (through socially close relationships).

The tradeo↵ between the benefits and detriments of multiplexity may depend on how fast

the density of social ties promotes information cascades (Brummitt et al., 2012; Centola,

2015) in relation to how much conflict it creates for individuals (Merton, 1957; Padgett and

Ansell, 1993; Burt, 2012).

The findings and theory presented in this article also bear implications for social theories

of market development (Banerjee and Duflo, 2011). These theories posit that greater ratio-

nalization can contribute to better outcomes, such as economic growth and reduced poverty.

Yet, many of the economic prescriptions for how to reduce economic inequality (e.g. by

providing information to enable better economic decisions) are based upon an assumption

that people can choose what is economically optimal without considering what is socially

optimal. In developing countries, where economic exchanges occur across di↵erent levels of

social structure (e.g. kinship, friendship), programmatic solutions to poverty through infor-

mational interventions should consider their social ramifications. Economic logics of rational

action, comparative advantage, and marginal utility (cf. Mas-Colell et al., 1995) are often

superseded by social logics of tradition, reciprocity, and cooperation in these societies. These

logics can conflict in ways that a↵ect the likelihood of successful market development and

change. Without accounting for the ways in which people’s social and economic lives are

inter-dependent, it is di�cult to recommend interventions that can reduce poverty without

eroding community structure.

39

Page 41: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Finally, the findings demonstrate some of the downsides of social cohesion for development

and change. Although theories of social cohesion (e.g. Kuwabara, 2011) posit that cohesion

is conducive to a variety of benefits for individuals, cohesion can also create insularity and

impede change. This article shows that cohesion arising from multiplexity can enable the

persistence of economically sub-optimal behavior. Multiplex social structures can promote

cohesion but can also impede the di↵usion of novel and potentially beneficial practices and

ideas. Microfinance failed to gain widespread acceptance because peer lending was multiplex

with other patterns of social interaction in the villages. Peer lending remained, even though

microfinance provided greater access to credit and reduced the economic costs of borrowing

money. The co-mingling of social and economic relationships created social disincentives for

severing existing lending relationships in favor of enrolling in microfinance.

There are many limitations to this study. First, verbal and non-verbal communication

is di�cult to observe and measure directly. This study infers the e↵ects of multiplexity on

individual behavior through a quasi-experiment in information seeding that inadvertently

created conflicts of interest for some informants. It is unclear, however, if informants were

cognizant of how their conflicts a↵ected their communications with microfinance-eligible par-

ticipants.19 On average, word-of-mouth from informants with conflicts of interest appears to

have exerted negative e↵ects on enrollment, but I am unable to distinguish between whether

an informant withheld information or misinformed contacts who were eligible participants.

Another challenge is that the e↵ects of hearsay and information from secondary sources (e.g.

friends of friends) are di�cult to isolate. Measuring the e↵ects of hearsay and further distin-

guishing between ways in which informants and their social contacts bu↵er conflicting roles

19The e↵ects of exposure to information from informants who were themselves not eligible to participate inmicrofinance show that enrollment was even lower among eligible participants exposed to informants who (1)plausibly had competing interests to endorse microfinance, and (2) were themselves ineligible to participatein microfinance.

40

Page 42: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

and motives for action are important avenues for future research.

6 Conclusion

This article has sought to explain why multiplex structures can deter contagion about new

practices despite having the property of dense connectivity among people. The article pro-

posed two mechanisms that a↵ect individual behavior in multiplex structures. The first

mechanism, increased social pathways to obtaining information from multiplex ties, was

linked to a higher valence of word-of-mouth and social influence. The second mechanism,

conflict in informants’ social and economic interests, was linked to informants’ ability to

be objective and impartial. Multiplexity appears to have deterred di↵usion by impeding

the unbiased transmission of word-of-mouth about an organizational model of lending that

jeopardized the integrity of existing money lending relationships in the villages.

Future research can explore the boundary conditions on these e↵ects and moderators

of the magnitude of information loss and distortion in the presence of conflicting interests

among informants. Finer grain data about the content and patterns of communication across

multiplex informants can further enable identification of audience-side e↵ects (commitment

to existing practices, ideas, technologies) and informer-side e↵ects (information skewing,

opportunistic communication). The key finding in this article, that multiplexity dampens

di↵usion when it activates conflicts of interest for informants, has ramifications for other

types of socially influenced behavior (e.g. voting, insurgency) and explains why potentially

useful innovations often fail to di↵use. Practically, this research places important scope

conditions on prior findings about the network topologies that promote di↵usion and on the

benefits of targeting central actors (cf. Ryan and Gross, 1943; Coleman et al., 1966; Aral

et al., 2009; Centola, 2015; Rossman, 2015), by noting that centrality often involves having

41

Page 43: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

multiple kinds of ties that can create conflicts of interest and undermine di↵usion. Thus, the

best actors to target about spreading the word may not be the most connected, but rather

the least conflicted.

42

Page 44: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

References

Abascal, Maria and Delia Baldassarri. 2015. “Love Thy Neighbor? Ethnoracial Diversityand Trust Reexamined.” American Journal of Sociology 121:722–782.

Ambedkar, Babasaheb. 1917. “Castes in India: Their Mechanism, Genesis and Develop-ment.” Indian Antiquary XLI.

Aral, Sinan, Lev Muchnik, and Arun Sundararajan. 2009. “Distinguishing Influence-BasedContagion from Homophily-Driven Di↵usion in Dynamic Networks.” Proceedings of theNational Academy of Sciences of the United States of America 106:21544–21549.

Armendariz, Beatriz and Jonathan Morduch. 2005. The Economics of Microfinance. Cam-bridge, MA: MIT Press.

Aronow, Peter, Cyrus Samii, and Valentina A. Assenova. 2015. “Robust-Cluster VarianceEstimation for Dyadic Data.” Political Analysis 23:564–577.

Assenova, Valentina A. 2016. “Social Learning in the Presence of Inequality and Hetero-geneity.”

Aven, Brandy L. 2015. “The Paradox of Corrupt Networks: An Analysis of OrganizationalCrime at Enron.” Organization Science 26:980–996.

Baker, Wayne E. 1984. “The Social Structure of a National Securities Market.” AmericanJournal of Sociology 89:775–811.

Baldassarri, Delia and Amir Goldberg. 2014. “Neither Ideologues nor Agnostics: AlternativeVoters’ Belief System in an Age of Partisan Politics.” American Journal of sociology120:45–95.

Banerjee, Abhijit, Arun G. Chandrasekhar, Esther Duflo, and Matthew O. Jackson. 2013.“The Di↵usion of Microfinance.” Science 341:1236498.

Banerjee, Abhijit and Esther Duflo. 2011. Poor Economics: A Radical Rethinking of theWay to Fight Global Poverty . New York: Perseus Books Group.

Baumeister, Roy F. and Mark R. Leary. 1995. “The Need to Belong: Desire for InterpersonalAttachments as a Fundamental Human Motivation.” Psychological Bulletin 117:497–529.

BenYishay, Ariel and Ahmet Mushfiq Mobarak. 2016. “Social Learning and Incentives forExperimentation and Communication.” Review of Economic Studies .

Blau, Peter M. 1977. “A Macrosociological Theory of Social Structure.” American Journalof Sociology 83:26–54.

43

Page 45: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Blume, Lawrence E., William A. Brock, Steven N. Durlauf, and Rajshri Jayaraman. 2015.“Linear Social Interactions Models.” Journal of Political Economy 123:444–496.

Bramoulle, Yann, Habiba Djebbari, and Bernard Fortin. 2009. “Identification of peer e↵ectsthrough social networks.” Journal of Econometrics 150:41–55.

Brock, William A. and Steven N. Durlauf. 2007. “Identification of Binary Choice Modelswith Social Interactions.” Journal of Econometrics 140:52–75.

Brummitt, Charles, Kyu-Min Lee, and Kwang-Il Goh. 2012. “Multiplexity-Facilitated Cas-cades in Networks.” Physical Review E 85:045102(R).

Burt, Ronald S. 1992. Structural Holes: The Social Structure of Competition. Cambridge,MA: Harvard University Press.

Burt, Ronald S. 2012. “Network-Related Personality and the Agency Question: MultiroleEvidence from a Virtual World.” American Journal of Sociology 118:543–591.

Butts, Carter T. and Kathleen M. Carley. 2001. “Multivariate Methods for InterstructuralAnalysis.” Unpublished manuscript. pp. 16–29.

Cain, Daylian M., George Loewenstein, and Don A. Moore. 2005. “The Dirt on ComingClean: Perverse E↵ects of Disclosing Conflicts of Interest.” The Journal of Legal Studies34:1–25.

Cardillo, Alessio, Jesus Gomez-Gardenes, Massimiliano Zanin, Miguel Romance, DavidPapo, Francisco del Pozo, and Stefano Boccaletti. 2013. “Emergence of Network Fea-tures from Multiplexity.” Scientific Reports 3:1344.

Centola, Damon. 2015. “The Social Origins of Networks and Di↵usion.” American Journalof Sociology 120:1295–1338.

Clark, Margaret S. and Judson Mills. 1993. “The Di↵erence between Communal and Ex-change Relationships: What it is and is Not.” Personality and Social Psychology Bulletin19:684–691.

Cohen, Lauren, Andrea Frazzini, and Christopher Malloy. 2010. “Sell-Side School Ties.”Journal of Finance 65:1409–1437.

Coleman, James S., Elihu Katz, and Herbert Menzel. 1966. Medical Innovation: A Di↵usionStudy , volume 9. Indianapolis: Bobbs-Merrill Co.

Coser, Rose Laub. 1975. “The Complexity of Roles as a Seedbed of Individual Autonomy.”In The Idea of Social Structure : Papers in Honor of Robert K. Merton, edited by Lewis A.Coser, pp. 237–263. New York, NY: Harcourt-Brace.

44

Page 46: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Coser, Rose Laub. 1991. “Anomie and Restricted Role-Set.” In In Defense of Modernity:Role Complexity and Individual Autonomy , pp. 48–67. Stanford, CA: Stanford UniversityPress.

Davis, Michael. 2012. “Conflict of Interest.” In Encyclopedia of Applied Ethics , pp. 571–577.

De Domenico, Manlio, Clara Granell, Mason A. Porter, and Alex Arenas. 2016. “The Physicsof Multilayer Networks.” arXiv pp. 1–22.

Durkheim, Emile. 1984. The Division of Labor in Society . New York: The Free Press.

Eaton, Richard M. 2005. “A Social History of the Deccan, 1300-1761.” In The New Cam-bridge History of India, Volume I-8 , edited by Gordon (University of Cambridge) Johnson,pp. 78–104. Cambridge, England: Cambridge University Press.

Evan, William M. 1962. “Role Strain and the Norm of Reciprocity in Research Organiza-tions.” American Journal of Sociology 68:346–354.

Fafchamps, Marcel. 2015. “Causal E↵ects in Social Networks.” Revue Economique 66:657–686.

Ferriani, Simone, Fabio Fonti, and Ra↵aele Corrado. 2013. “The Social and EconomicBases of Network Multiplexity: Exploring the Emergence of Multiplex Ties.” StrategicOrganization 11:7–34.

Freeman, Linton C. 1979. “Centrality in social networks conceptual clarification.” SocialNetworks 1:215–239.

Friedkin, Noah E. and Eugene C. Johnsen. 2011. Social Influence Network Theory: ASociological Examination of Small Group Dynamics . Cambridge, England: CambridgeUniversity Press.

Gemmetto, Valerio, Tiziano Squartini, Francesco Picciolo, Franco Ruzzenenti, and DiegoGarlaschelli. 2014. “Multiplexity and Multireciprocity in Directed Multiplexes.”

Gluckman, M. 1962. “Les Rites de Passage.” In Essays on the Ritual of Social Relations ,edited by M. Gluckman. Manchester, England: Manchester University Press.

Go↵man, Erving. 1959. “Discrepant Roles.” In The Presentation of Self in Everyday Life,pp. 141–167. New York: Doubleday and Co.

Go↵man, Erving. 1961. “Role Distance.” In Encounters , pp. 85–152. Indianapolis: Bobbs-Merrill Co.

45

Page 47: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Goldberg, Amir. 2011. “Mapping Shared Understandings Using Relational Class Analy-sis: The Case of the Cultural Omnivore Reexamined.” American Journal of Sociology116:1397–1436.

Goldberg, Amir and Sarah K. Stein. 2016. “Beyond ”Social Contagion”: AssociationalDi↵usion and the Emergence of Cultural Variation.”

Graham, Bryan S. and Jinyong Hahn. 2005. “Identification and Estimation of the Linear-In-Means Model of Social Interactions.” Economics Letters 88:1–6.

Gullahorn, John T. 1956. “Measuring Role Conflict.” American Journal of Sociology 61:299–303.

Hamming, Richard W. 1950. “Error Detecting and Error Correcting Codes.” Bell SystemTechnical Journal 29:147–169.

Hecht, Laura M. 2001. “Role Conflict and Role Overload: Di↵erent Concepts, Di↵erentConsequences.” Sociological Inquiry 71:111–121.

Karlan, Dean S. 2007. “Social Connections and Group Banking.” Economic Journal117:F52–F84.

Keay, John. 2000. “Other Indias: 1320-1525.” In India, A History , pp. 262–288. London:Harper Collins.

Kim, Jung Yeol and Kwang-Il Goh. 2013. “Coevolution and Correlated Multiplexity inMultiplex Networks.” Physical Review Letters 111:058702.

Kuwabara, Ko. 2011. “Cohesion, Cooperation, and the Value of Doing Things Together: HowEconomic Exchange Creates Relational Bonds.” American Sociological Review 76:560–580.

Loewenstein, George, Daylian M. Cain, and Sunita Sah. 2011. “The Limits of Transparency:Pitfalls and Potential of Disclosing Conflicts of Interest.” American Economic Review101:423–428.

Maitner, Angela T., Diane M. Mackie, Heather M. Claypool, and Richard J. Crisp. 2010.“Identity Salience Moderates Processing of Group-Relevant Information.” Journal of Ex-perimental Social Psychology 46:441–444.

Mas-Colell, Androu, Michael D. Whinston, and Jerry R. Green. 1995. Microeconomic The-ory . Oxford: Oxford University Press.

McPherson, Miller, Lynn Smith-Lovin, and James M. Cook. 2001. “Birds of a Feather:Homophily in Social Networks.” Annual Review of Sociology 27:415–444.

46

Page 48: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Merton, Robert K. 1957. “The Role-Set: Problems in Sociological Theory.” British Journalof Sociology 8:106–120.

Merton, Robert K. 1972. “Insiders and Outsiders: A Chapter in the Sociology of Knowledge.”American Journal of Sociology 78:9–47.

Moore, Don A. and George Loewenstein. 2004. “Self-interest, Automaticity, and the Psy-chology of Conflict of Interest.”

Moore, Don A, Lloyd Tanlu, and Max H Bazerman. 2010. “Conflict of Interest and theIntrusion of Bias.” Judgment and Decision Making 5:37–53.

Padgett, John K. and Christopher K. Ansell. 1993. “Robust Action and the Rise of theMedici.” American Journal of Sociology 98:1259–1319.

Rogers, Everett M. 1983. Di↵usion of Innovations . New York: The Free Press, 3rd ed.edition.

Rossman, Gabriel. 2014. “The Di↵usion of the Legitimate and the Di↵usion of Legitimacy.”Sociological Science 1:49–69.

Rossman, Gabriel. 2015. Climbing the Charts: What Radio Airplay Tells Us about theDi↵usion of Innovation. Princeton, N.J.: Princeton University Press.

Roy, Tirthankar. 2000. The Economic History of India, 1857-1947 . New Delhi: OxfordUniversity Press.

Ryan, Bryce and Neal C Gross. 1943. “The Di↵usion of Hybrid Seed Corn In Two IowaCommunities.” Rural Sociology 8:15–24.

Simmons, Roberta G. 1968. “The Role Conflict of the First-Line Supervisor: An Experi-mental Study.” American Journal of Sociology 73:482–495.

Simpson, Cohen R. 2015. “Multiplexity and Strategic Alliances: The Relational Embedded-ness of Coalitions in Social Movement Organisational Fields.” Social Networks 42:42–59.

Smith, Chris M. and Andrew V. Papachristos. 2016. “Trust Thy Crooked Neighbor: Multi-plexity in Chicago Organized Crime Networks.” American Sociological Review 81:644–667.

Snoek, J. Diedrick. 1966. “Role Strain in Diversified Role Sets.” American Journal ofSociology 71:363–372.

Strang, David and Sarah A. Soule. 1998. “Di↵usion in Organizations and Social Movements:From Hybrid Corn to Poison Pills.” Annual Review of Sociology 24:265–290.

47

Page 49: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Multiplexity and Information

Tajfel, Henri and John C. Turner. 1986. “The Social Identity Theory of Intergroup Behavior.”In Psychology of Intergroup Relations , edited by S. Worchel and W. G. Austin, volume2nd ed. of Key Readings in Social Psychology , chapter 1, pp. 7–24. Nelson-Hall.

Uzzi, Brian. 1997. “Social Structure and Competition in Interfirm Networks: The Paradoxof Embeddedness.” Administrative Science Quarterly 42:35–67.

Vaid, Divya. 2014. “Caste in Contemporary India: Flexibility and Persistence.” AnnualReview of Sociology 40:391–410.

Verbrugge, Lois M. 1979. “Multiplexity in Adult Friendships.” Social Forces 57:1286–1309.

Wejnert, Barbara. 2002. “Integrating Models of Di↵usion of Innovations: A ConceptualFramework.” Annual Review of Sociology 28:297–326.

48

Page 50: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Table A1: Descriptive Statistics for the Peer Lending Relationships in the Villages

Statistic N Mean St. Dev. Min Max

Share Advice 31,335 0.781 0.413 0 1Exchange Goods 31,335 0.852 0.356 0 1Pay Social Visits 31,335 0.843 0.363 0 1Share Medical Advice 31,335 0.775 0.417 0 1Interact Socially 31,335 0.762 0.426 0 1Part of Same Kinship 31,335 0.657 0.475 0 1Pray Together 31,335 0.638 0.481 0 1Multiplexity 31,335 5.309 2.358 0 7Age 31,335 36.581 11.882 10 85Indegree 31,335 3.397 2.483 0 39Outdegree 31,335 6.451 3.753 1 31Total number of rooms in house 31,335 2.821 1.817 0 19Bedrooms in house 31,335 1.221 2.132 0 24Married 31,335 0.036 0.186 0 1Directly Informed by BSS 31,335 0.165 0.372 0 1Money Exchange Partner Informed 31,335 0.172 0.378 0 1Enrolled in Microfinance 31,335 0.197 0.397 0 1No savings 31,290 0.575 0.494 0 1No education 31,290 0.462 0.499 0 1Government food subsidies 31,290 0.389 0.487 0 1OBC Caste (socially disadvantaged) 31,290 0.584 0.493 0 1Kannada native speaker 31,290 0.732 0.443 0 1Agricultural laborer 31,290 0.133 0.340 0 1Religious minority (Christian or Muslim) 31,290 0.040 0.195 0 1

Page 51: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Table A2: Descriptive Statistics for Village-Periods

Statistic N Mean St. Dev. Min Max

Enrollment Rate 325 16.194 10.241 0.000 44.390Number of Households in Village 325 221 53 114 356Fraction Villagers Informed by BSS 325 0.120 0.031 0.064 0.193Village size 325 1,032 257 446 1,771Household size 325 4.669 0.374 3.912 5.554Caste Heterogeneity 325 0.507 0.154 0.066 0.747Language Heterogeneity 325 0.311 0.221 0.000 0.694Religious Heterogeneity 325 0.077 0.135 0.000 0.487Caste In-group 325 0.852 0.128 0.000 0.987Religious In-group 325 0.972 0.123 0.000 1.000Language In-group 325 0.855 0.146 0.000 1.000Hamming distance to Goods 325 1,639 894 355 4,039Hamming distance to Visits 325 1,710 1,219 236 4,565Hamming distance to Advice 325 1,217 790 218 3,094Hamming distance to Kinship 325 4,866 1,540 1,791 8,764Hamming distance to Medical 325 4,318 1,519 1,669 8,466Hamming distance to Social 325 4,661 1,704 1,385 9,574Hamming distance to Religious 325 4,524 1,704 1,543 8,692Graph correlation with Goods 325 0.606 0.308 0.000 0.918Graph correlation with Visits 325 0.643 0.332 0.000 0.915Graph correlation with Advice 325 0.660 0.334 0.000 0.948Graph correlation with Kinship 325 0.391 0.201 0.000 0.618Graph correlation with Medical 325 0.449 0.231 0.000 0.668Graph correlation with Social 325 0.448 0.229 0.000 0.639Graph correlation with Religious 325 0.394 0.211 0.000 0.652

Page 52: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Table A3: Graph Correlations and Hamming Distance

1 2 3 4 5 6 7 8

1 Money dH 0⇢ 1.00

2 Goods dH 1,602 0⇢ 0.59 1.00

3 Visits dH 1,717 1,562 0⇢ 0.62 0.62 1.00

4 Advice dH 1,156 1,377 1,840 0⇢ 0.64 0.63 0.60 1.00

5 Kinship dH 4,751 4,685 5,005 4,166 0⇢ 0.38 0.40 0.37 0.40 1.00

6 Medical dH 4,240 4,410 4,650 4,166 2,877 0⇢ 0.44 0.43 0.41 0.40 0.58 1.00

7 Social dH 4,644 4,703 4,634 4,166 4,105 3,390 0⇢ 0.42 0.42 0.44 0.40 0.52 0.57 1.00

8 Religious dH 4,443 4,508 4,647 4,166 3,480 3,123 3,687 0⇢ 0.38 0.39 0.38 0.40 0.56 0.61 0.54 1.00

Note: Average Hamming distance (dH) and graph correlations (⇢) between exchangestructures are shown on the top and bottom rows, respectively.

Table A4: In-means Probability Model of Microfinance Enrollment, No Conflict

Dependent Variable:Probability that an Eligible Individual Enrolled

Model 1a Model 2a Model 3a Model 4a

Word-of-mouth �0.05 �0.05 0.04 0.05(0.06) (0.06) (0.06) (0.06)

Social Influence 0.68⇤⇤⇤ 0.68⇤⇤⇤ 0.53⇤⇤⇤ 0.33⇤⇤⇤

(0.07) (0.07) (0.07) (0.07)Direct Information 0.01 0.02 0.03

(0.02) (0.02) (0.02)(Intercept) 0.18⇤⇤⇤ 0.18⇤⇤⇤ �0.28 �0.47⇤

(0.01) (0.01) (0.18) (0.20)Individual Controls X XVillage Controls XR

2 0.01 0.01 0.16 0.21N 22,955 22,955 22,955 22,955⇤⇤⇤p < 0.001, ⇤⇤p < 0.01, ⇤p < 0.05 (two-tailed tests). Cluster robust standard errors in parentheses.

Page 53: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Table A5: In-means Probability Model of Microfinance Enrollment, Conflict

Dependent Variable:Probability that an Eligible Individual Enrolled

Model 1b Model 2b Model 3b Model 4b

Word-of-mouth �0.15⇤⇤⇤ �0.37⇤⇤⇤ �0.33⇤⇤⇤ �0.31⇤⇤⇤

(0.04) (0.08) (0.07) (0.07)Social Influence 4.68⇤⇤⇤ 4.69⇤⇤⇤ 4.37⇤⇤⇤ 4.14⇤⇤⇤

(0.03) (0.09) (0.09) (0.08)Direct Information 0.05⇤⇤⇤ 0.07⇤⇤⇤ 0.06⇤⇤⇤

(0.01) (0.02) (0.01)(Intercept) 0.09⇤⇤⇤ 0.09⇤⇤⇤ �0.26 �0.21

(0.00) (0.00) (0.29) (0.28)Individual Controls X XVillage Controls XR

2 0.37 0.37 0.44 0.46N 31,290 31,290 31,290 31,290⇤⇤⇤p < 0.001, ⇤⇤p < 0.01, ⇤p < 0.05 (two-tailed tests). Cluster robust standard errors in parentheses.

Page 54: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Table A6: Probability of Microfinance Enrollment with Multiplexity, No Conflict

Dependent Variable:Probability that Eligible Individual Enrolled

Model 5a Model 6a Model 7a Model 8a

Word-of-mouth 0.04 0.04 0.10 0.05(0.12) (0.12) (0.11) (0.11)

Social Influence 0.57⇤⇤⇤ 0.57⇤⇤⇤ 0.41⇤⇤⇤ 0.16(0.12) (0.12) (0.12) (0.12)

Direct Information 0.01 0.02 0.03(0.02) (0.02) (0.02)

Multiplexity (M) 0.00 0.00 0.00 �0.00(0.00) (0.00) (0.00) (0.00)

Word-of-mouth ⇥ M �0.06 �0.06 �0.04 0.00(0.06) (0.06) (0.06) (0.06)

Social Influence ⇥ M 0.06 0.06 0.07 0.10(0.06) (0.06) (0.06) (0.05)

(Intercept) 0.18⇤⇤⇤ 0.18⇤⇤⇤ �0.28 �0.47⇤

(0.01) (0.01) (0.19) (0.20)Individual Controls X XVillage Controls XR

2 0.01 0.01 0.16 0.21N 22,955 22,955 22,955 22,955⇤⇤⇤p < 0.001, ⇤⇤p < 0.01, ⇤p < 0.05 (two-tailed tests). Cluster robust standard errors in parentheses.

Page 55: Multiplexity as a Conduit and Sieve of Information …...Multiplexity as a Conduit and Sieve of Information in Embedded Markets Preliminary and Incomplete. Please do not cite or circulate

Table A7: Probability of Microfinance Enrollment with Multiplexity, Conflict

Dependent Variable:Probability that Eligible Individual Enrolled

Model 5b Model 6b Model 7b Model 8b

Word-of-mouth �0.06 0.10 0.13 0.11(0.07) (0.08) (0.07) (0.07)

Social Influence �1.87⇤⇤⇤ �1.86⇤⇤⇤ �1.76⇤⇤⇤ �1.89⇤⇤⇤

(0.07) (0.07) (0.07) (0.07)Direct Information 0.05⇤⇤⇤ 0.07⇤⇤⇤ 0.06⇤⇤⇤

(0.01) (0.01) (0.01)Multiplexity (M) �0.02⇤⇤⇤ �0.02⇤⇤⇤ �0.02⇤⇤⇤ �0.02⇤⇤⇤

(0.00) (0.00) (0.00) (0.00)Word-of-mouth ⇥ M �0.00 �0.08⇤⇤⇤ �0.08⇤⇤⇤ �0.07⇤⇤⇤

(0.01) (0.02) (0.02) (0.02)Social Influence ⇥ M 1.25⇤⇤⇤ 1.25⇤⇤⇤ 1.18⇤⇤⇤ 1.17⇤⇤⇤

(0.01) (0.01) (0.01) (0.01)(Intercept) 0.22⇤⇤⇤ 0.20⇤⇤⇤ �0.38⇤ �0.31

(0.00) (0.00) (0.19) (0.19)Individual Controls X XVillage Controls XR

2 0.51 0.51 0.56 0.58N 31,290 31,290 31,290 31,290⇤⇤⇤p < 0.001, ⇤⇤p < 0.01, ⇤p < 0.05 (two-tailed tests). Cluster robust standard errors in parentheses.