Nationalism and Interethnic Trust: Evidence from an...

Nationalism and Interethnic Trust:Evidence from an African Border Region∗

Amanda Lea Robinson†

Prepared for:Coevolution of Behaviors and Institutions Conference

January 13-15, 2014Santa Fe Institute

Abstract

Low levels of interethnic trust have been identified as a key link betweenethnic diversity and economic underdevelopment. I argue that increased identi-fication with a territorially-defined nation, common to all ethnic groups, shouldreduce the degree to which trust is ethnically bounded. To assess the relation-ship between national identification and interethnic trust, I report the results ofan original “lab-in-the-field” experiment, conducted near the Malawian-Zambianborder, which measured pre-existing strength of national identification, experi-mentally manipulated the salience of national identity, and relied on behavioralmeasures of trust. I find that common nationality is a robust predictor of inter-personal trust, and equal in magnitude to the impact of shared ethnicity. Further,increasing the salience of a common national identity reduces the magnitude ofthe co-ethnic trust advantage. These results offer micro-level evidence that astrong and salient national identity can indeed facilitate interethnic trust in di-verse societies.

∗Supplemental materials are available at: www.stanford.edu/~alrobins/Amanda_Lea_Robinson/Research.†Assistant Professor of Political Science, The Ohio State University. Email:

[email protected]. I am grateful to James Fearon, David Laitin, Michael Tomz, andJeremy Weinstein for their support of this project in all its stages. I also thank Claire Adida, JessicaGottlieb, Justin Grimmer, Eric Kramon, Alexander Lee, Avital Livny, and Kenneth Schultz fortheir feedback on previous versions of this paper. Finally, this project would not have been possiblewithout the excellent research assistance of Augustine Harawa, Hector Honde, Innocent Mwale,Paul Mwera, and Charles Sisya, and the institutional support of Invest in Knowledge Initiative inZomba Malawi. This research was supported financially by a Russell Sage Foundation Small Grant inBehavioral Economics, the O’bie Shultz Dissertation Travel Grant from the Freeman Spogli Instituteof International Studies, and a Graduate Research Opportunity Grant from the School of Humanitiesand Social Sciences at Stanford University.

www.stanford.edu/~alrobins/Amanda_Lea_Robinson/Research


mailto:[email protected]

www.investinknowledge.org

Nationalism and Interethnic Trust 1

Interpersonal trust is necessary for promoting economic exchange, especially in states

that lack strong institutions for facilitating and monitoring such transactions (??).

Thus, there is a robust positive relationship between aggregate levels of generalized trust

in societies, and the rate at which their economies grow (????). One of the strongest

predictors of interpersonal trust is the degree of ethnic diversity in society (???), since

individuals tend to trust coethnics more than non-coethnics (????). Because trust is

necessary for growth, and diversity reduces trust, diverse societies tend to experience

slower economic growth than more homogeneous societies (???). Does this then mean

that ethnically diverse states are doomed to underdevelopment?

I argue that increased identification with the overarching national identity within eth-

nically diverse African states reduces the negative effects of ethnic diversity on inter-

personal trust. This argument builds on a large literature in social psychology showing

that increasing identification with an overarching common identity improves intergroup

relations. When applied to the case of ethnically diverse African states, the argument

is developed in two parts. First, I argue that territorially-defined national identities in

these settings have the power to facilitate interpersonal trust, an endeavor that runs

counter to the widely held view that state-based nationalism in Africa is generally too

weak to influence behavior. Second, I argue that variation in identification with that

nation, both across individuals and across contexts, explains the degree to which trust is

concentrated within ethnic groups. In particular, an increase in national identification

should reduce the degree to which individuals base their trust on shared ethnicity.

To evaluate these expectations, I carried out a “lab-in-the-field” experiment in an ethni-

cally diverse region of Malawi near the international border with Zambia. By situating

the study at the intersection of an ethnic and a national border, where coethnicity

and conationality are orthogonal, I am able to empirically evaluate the impact of one

shared identity while controlling for the other. To assess whether national identifica-

tion impacts ethnic trust biases, the research design combines a novel psychological

measure of each individual’s strength of national identification with an experimental

manipulation of the situational salience of the national identity, and evaluates whether

these two forms of variation in national identification explain the degree to which trust

is conditioned on shared ethnicity. Trust is measured within-subjects, using standard

behavioral economic trust games implemented in rural markets near the field site.

Contrary to the popular image of African societies as purely tribal, with no loyalty tied


to the territorial state, I find evidence that, on average, shared nationality is just as

important as shared ethnicity for decisions about whom to trust. This suggests that

the national identity in Malawi, and perhaps in other ethnically diverse post-colonial

African states, has the potential to complement weak formal institutions by facilitating

the kind of interpersonal trust necessary for efficient economic transactions.

In terms of the influence of territorial nationalism on interethnic trust, I find that

the “coethnic trust premium” decreases as preexisting identification with the Malaw-

ian nation increases. In fact, among the strongest national identifiers, coethnic and

non-coethnic Malawians are trusted at the same rate. However, this trend is only

apparent when the national identity is not experimentally primed: once the common

national identity is made contextually salient, the coethnic trust premium is reduced to

zero, regardless of pre-existing levels of national identification. Thus, while increasing

the salience of the common national identity reduces, on average, the degree to which

individuals discriminate their trust along ethnic lines, this effect is driven by individ-

uals with weak underlying degrees of national identification. As a result, the national

prime causes weak national identifiers to trust coethnic and non-coethnic Malawians

at equal rates, making them behaviorally indistinguishably from strong national iden-

tifiers. Contrary to expectations, though, this saliency effect results from a reduction

in trust of coethnics, rather than increased trust in non-coethnic members of the na-

tion. This means that while exposure to national symbols reduces the “coethnic trust

premium”, it may do so at the expense of overall trust.

In sum, these results suggest that interethnic trust may be increased in diverse set-

tings either by increasing the degree to which individuals consistently identify with the

common national identity or by promoting contextual nationalism through the ubiqui-

tous use of national symbology such as the national flag, national anthem, or national

currency. However, the latter approach is much less costly than attempts to increase

individuals’ levels of identification with the nation through education and socialization,

and appears to be just as effective. Thus, making the common national identity salient

in everyday contexts, which has been termed “banal nationalism” by ?, may work to

reduce the impact of ethnic difference on interpersonal trust in commonplace economic

transactions. The ultimate value of national identity salience for economic development,

however, rests on how important aggregate levels of trust are for economic performance

relative to the detriments of ethnically-bound trust.


Theory

A growing literature in economics and political science has identified a robustly nega-

tive correlation between ethnic diversity and economic growth around the world (see ?,

for a review). One of the mechanisms suggested to account for this relationship works

through ethnic-based interpersonal trust. First suggested by ?, the argument asserts

that low levels of generalized interpersonal trust inhibit economic growth (????), and

that generalized trust at the state level is typically lower in ethnically diverse societies

(???). State-level generalized trust is lower in diverse states because individuals tend

to trust coethnics more than non-coethnics (?????), and the proportion of citizens

sharing one’s ethnicity decreases as diversity increases. The existing scholarship has

thus concluded, as ? succintctly puts it, that “some societies, particularly those deeply

divided by ethnic or racial divisions, may have strong ties and high levels of ‘thick’ trust

within particular communities, but this does not generalize to society as a whole.” Such

ethnic based trust translates into poorer economic performance by increasing transac-

tion costs, and thus hampering trade, across ethnic lines (?). Under such conditions,

national markets are fragmented into local ethnic markets, and potential gains to trade

are not realized.1 This characterization of the negative impact of diversity on economic

development via reduced trust has been particularly applied to African states, which

are among the most ethnically diverse in the world (?) and exhibit the lowest levels of

generalized trust across regions (?).

So, when faced with the reality of a multi-cultural society, how can trust be generalized

to society as a whole? I argue that increased identification with a common, overarching

national identity group can form the basis of a trust community, even amid ethnic

diversity. However, this conjecture is at odds with the general perception that national

identities in Africa are too weak to impact behavior among average citizens. Africanists

have long been skeptical of the power of the territorially-defined nation2 as an “imagined

community” in Africa, mostly because of the colonial origins of African states (?).

The borders of modern African states were determined by colonial partition of the

continent without regard for existing patterns of group identification (??), resulting

1This argument is developed further and empirically evaluated across Malawian markets as part ofa larger project on the impact of ethnic diversity and interpersonal trust on economic development inAfrica.

2By nation I mean the territorially-defined, state-based identity group. Thus, for the purposes ofthis study, the national ingroup is defined by citizenship.


in the amalgamation of many cultural groups into a single state and the partition of

other groups into multiple states (??). As a result, most African states at independence

lacked a common indigenous language, shared historical memories, or similar cultural

traditions – the basic building blocks of a coherent national identity (??). Further, the

processes of “modernization” that allowed European states to overcome sub-national

attachments and engender national identification (??) are the same forces that are

blamed for the supposed failure of African nations and their fragmentation along ethnic

lines (????).

Thus, the theoretical literature on national identification in Africa paints a pessimistic

picture of the impact national identities are likely to have on behavior. However, his-

torical accounts of the rise of widespread national identification in Europe document

that many of the problems purported to block territorial nationalism in Africa also ex-

isted in pre-national Europe, including partitioned cultural groups (???) and culturally

diverse states (??). It is not clear why these hurdles were overcome in Europe, but are

assumed to be insurmountable in post-colonial Africa. Consistent with this skepticism

of African exceptionalism, the (very small body of) existing empirical evidence sug-

gests that national identities are in fact relevant to regular Africans. ? looked at the

relative importance of multiple identities among the Hausa of Niger and Nigeria, and

found that for this partitioned ethnic group, national identity was more important than

ethnic identity. Similarly, in other work, I have found that processes of modernization

across African states result in increased national identification relative to ethnic identi-

fication, resulting in a net increase of national unity with modernization (?). However,

both these findings are based on self-reported identification (attitudinal) and not on

whether national identities are relevant for real behavior. In order to make the case

that national identification can promote interpersonal trust within a nation, it must

first be shown that shared nationality is relevant for behavioral trust. This produces

the first observable implication:

Conationality, along with coethnicity, will be a relevant determinant of interper-

sonal trust.

My central argument, however, goes beyond the claim that territorially-defined national

identities are relevant in ethnically diverse African societies. I argue that increased

identification with that nation reduces the degree to which trust is conditioned on shared

ethnicity, thereby reducing the negative impact of ethnic diversity on trust. While not


explicitly focused on trust, ? similarly posits that stronger nationalism in Tanzania

may help explain higher levels of interethnic cooperation there compared to neighboring

Kenya. The expectation is also consistent with ?’s (?) insistence that diverse societies

must “create new forms of social solidarity and dampen the negative effects of diversity

by constructing new, more encompassing identities. Thus, the central challenge . . . is to

create a new, broader sense of ‘we”’ (p.139). In the context of ethnically diverse African

states, I argue that the territorially-defined nation can offer just such a pan-ethnic sense

of “we.”

This argument builds theoretically on two key findings in the social psychological study

of intergroup relations. First, individuals tend to perceive members of their own ingroup

to be more trustworthy than members of outgroups, now matter how those groups are

defined, and thus to trust ingroup members at a higher rate (????). This ingroup

favoritism results from the psychological desire to see groups to which one belongs as

distinct from and favorable to other groups, a central tenet of social identity theory

(???). However, this trust bias due to categorization of others into ingroups and out-

groups generally results from positive ingroup bias rather than negative outgroup bias

(?). Thus, in the context of ethnically divided societies, we expect that observed trust

differences between coethnics and non-coethnics result from a “trust premium” for co-

ethnics rather than reduced trust in non-coethnics (relative to some baseline context in

which no groups are relevant). In contrast to strategic explanations of the coethnic trust

premium (e.g., ?), this psychological mechanism does not require that coethnics are ac-

tually more trustworthy, only that they are perceived to be when ethnic differences are

salient.

Second, this positive ingroup bias can be extended to outgroup members when individ-

uals cognitively recategorize themselves primarily as members of an overarching, super-

ordinate identity group. This expectation is laid out in a theory known as the Common

Ingroup Identity Model (CIIM) (?). The CIIM contends that creating or emphasizing

a more inclusive ingroup should result in the positive benefits of ingroup membership,

such as greater perceived trustworthiness, being extended to former outgroup members

(??). In the real-world context of ethnically diverse states, the territorially-defined na-

tion provides a more inclusive ingroup composed of individuals of different ethnicities.3

3While not the focus of this paper, in theory the overarching identity could be based on any groupthat transcends ethnic lines, such as common religion or class. It is an open question as to whethersome superordinate identities are better able to facilitate intergroup trust than others, and, if so, why.


Thus, as individuals identify more with the common ingroup (territorial nation), they

should trust members of (ethnic) outgroups more, up to the point at which they are

trusted at rates equal to (ethnic) ingroup members.

Existing evidence for the CIIM comes from both minimal group laboratory experiments

and real world contexts, such as an ethnically diverse high-school (?), executives after

a corporate merger (?), children in blended families after remarriage (?), and racially

diverse football fans (?). While fewer studies have used the CIIM to try to understand

the effect of national identification on intergroup relations, there are a few notable

exceptions. One study evaluates the effect of increasing the salience of a common

American identity on the relations between college-aged Democrats and Republicans,

finding that priming the common identity (American) was effective at reducing inter-

group bias based on political affiliation (?). In another study, ? finds that increasing

the saliency of a shared national (American) identity increases support among whites

for policies favoring a minority ethnic group in the US. To my knowledge, no existing

scholarship uses the CIIM to evaluate the impact of a common national identity on

interethnic relations in an African context, despite the tremendous amount of work on

understanding ethnic relations on the continent.

The theoretical literature on the impact of common ingroup identification is not just

about categorization or nominal membership in a common identity, but instead about

the degree to which individuals identify with that group. But what is meant by “identi-

fication”? Across the social sciences, there is a classic debate between primordialist and

circumstantialist approaches to the study of identity. The primordial approach, most

often linked to ?, argues that ethnic and national identities are perceived to be fixed,

deep-rooted, and naturally given, and thus tend to be fairly stable within individuals

across time.4 In contrast, the circumstantialist approach emphasizes group identities

as malleable and flexible in response to different circumstances. While this debate has

revolved around the degree of stability in group identities themselves, the same logic is

easily extended to the study of individual identification with a given group identity. In

this case, we can think of the primordialist form of group identification as the degree

or strength of identification with the group, which is fairly stable within an individual

4Both ? and ? make the cogent point that while primordialists only argue that individuals perceivetheir ethnic and national identities to be naturally given, their approach is often portrayed as pur-porting that the identities are naturally given. For this reason, I use the label of “circumstantialist”rather than “constructivist” to refer to the opposing view, as most primordialists agree that identityis socially constructed.


across contexts.5 The circumstantialist form of group identification, on the other hand,

is the relative salience of a particular identity, which is likely to vary across different

contexts, resulting in stronger or weaker identification in any particular situation.

While these two approaches are typically viewed as mutually exclusive, from a psycho-

logical perspective they are not incompatible (??). Research shows that while strength

of identification with a social category is generally stable across individuals, it is also

sensitive to context. In their study of racial identity among African-Americans, for

example, ? find that while degree of racial identification across individuals is fairly

stable, situations still influence which identity is particularly salient at a given time. ?

characterizes these two forms of group identification in cognitive terms as “chronically

accessible” versus “situationally accessible” identities, while ? refer to “predisposing

factors” and “situational triggers” (?). This project incorporates both approaches to

the study of group identification.

Existing political science literature, which has tended to focus on the effects of ethnic,

racial, or religious identities rather than national identities, has dealt almost entirely

with nominal group identity rather than degree of group identification. For example,

while most theories of ethnic politics are built on ideas of the relative salience or im-

portance of ethnic identities, the typical measures used are simply demographic. The

ethnolinguistic fractionalization measure (ELF), for example, captures the likelihood

that two randomly selected individuals from a population are nominal members of dif-

ferent groups, and can be constructed for any list of groups within a country and their

population shares. Various lists of the ethnic make-up of a population have been of-

fered in order to calculate indices of fractionalization (???), but all only offer lists of

the relevant groups and assume that strength of identification with those categories is

constant across individuals. Similarly, in micro-level studies of coethnicity, individu-

als are typically coded as coethnics or not, and, thus, variation in identification with

the ethnic group are not typically considered. Thus, one contribution of this project

is to make identification, rather than just identity, a variable under consideration in

exploring when and how identities shape behavior.6

5While strength of group identification may be stable within an individual over a long period of time(even a lifetime), this does not mean that such identification is not socially or politically constructed.For example, ? argues that Malawian Chewas and Tumbukas exhibit, on average, stronger ethnicidentification than Zambians of the same ethnic groups. Even though this difference is the product ofdifferent political mobilization in the two countries, the politically constructed difference is presumablyfairly stable.

6This approach follows ?’s (?) concern with variation in the “thickness” of identity categories,


In doing so, I consider both forms of variation in national identification: interpersonal

differences in psychological strength of identification with the nation and the situational

salience of the national identity in a particular context. I define strength of identification

as comprising three different ways in which individuals identify with social groups (?):

affectively (?), behaviorally (?) and cognitively (?). I assume that these three forms

of group identification are fairly stable within individuals across time, resulting from

long-term processes such as education, socialization, and life experience. We expect

that higher levels of stable identification with the common group identity should be

associated with less sub-group discrimination. This generates the second observable

implication of the theory:

The more strongly one identifies with the nation, the less inclined he will be to trust

his coethnics over non-coethnics.

Situational identification with the nation, on the other hand, is defined as the salience

of the national identity relative to other social identities in a particular context. When

the national identity is made salient or relevant in a given context, individuals will

identify more strongly with their national ingroup in that moment, and will thus be

more inclined to use that identity in making decisions about whom to trust. Salience is

crucial to understanding how a particular identity influences behavior given the circum-

stantialist insight that individuals possess multiple identities (?). ? has argued that

the most consequential form of national identification is when the national identity is

made salient in a subtle way across many everyday contexts. He refers to this form

of national identification as “banal nationalism,” which includes the ubiquitous use of

national symbols, support for national sports teams, routinized national practices, and

the use of first person plural pronouns that imply togetherness in the national media.

The more cognizant someone is about a common shared identity, the less she should

base her trust on differences along other identity dimensions. This suggests a third

observable implication:

If the national identity is made contextually salient, individuals will be less likely

to condition their trust in other members of the nation on coethnicity.

In addition to understanding the independent impacts of the strength of national iden-

tification and salience of national identity on interpersonal trust, it is also important to

understand how these two forms of increased national identification interact. Assuming

across both individuals and contexts.


there is an impact of increased saliency at all, there are three possible ways in which

that effect may interact with underlying individual differences in strength of identifi-

cation. First, increasing the salience of the national identity may affect both strong

and weak national identifiers equally. As a result, the increased salience would reduce

the size of the coethnic trust premium on average, but individuals with weak national

identification would still exhibit a larger coethnic trust premium than strong national

identifiers. Studies that evaluate the impact of identity primes without accounting for

individual differences in strength of identification (e.g., ?) implicitly assume either no

variation in strength of identification or a constant effect of increased salience across

such variation. Second, increased salience may impact strong national identifiers more

than weak national identifiers. Work in cognitive psychology suggests just such an

additive effect – the more strongly one identifies with the group, the more cognitively

accessible that identity is, and thus the more sensitive an individual should be to the

contextual salience of that identity (??). Consistent with this expectation, ? found that

experimentally increasing the salience of the (American) national identity had a strong

negative impact on hostility toward minorities among highly nationalistic participants,

but no effect on weak national identifiers. Third, increased salience may impact weak

national identifiers more than strong national identifiers. This could occur if strong

national identifiers are already so cognizant of their national identity that there is no

added effect of increased salience, while that same increased salience reminds weak

national identifiers of their national identity, and mobilizes them to “catch up” with

strong national identifiers. Consistent with this possibility, ? find that the experimen-

tal priming of the (Dutch) national identity has a larger effect on concern for national

cultural preservation among citizens with weak national identification.

What accounts for these inconsistent findings and what should we expect in the con-

text of ethnically diverse African societies? To some degree, the differential impact of

identity saliency across different degrees of identification depends on the baseline rela-

tionship between the outcome of interest and underlying strength of identification. For

example, if the coethnic trust premium is similar across levels of national identification,

then we may very well expect a bigger impact of increased salience on strong identifiers

due to cognitive accessibility. However, if the coethnic trust premium is very small or

non-existent among strong national identifiers, as expected, then the marginal impact

of increasing the salience of that national identity will necessarily be smaller than the

impact on weak national identifiers. Thus, a priori it is not clear how the interaction


between strength and salience of national identification will impact interethnic trust in

the Malawian context, but the research design allows me to evaluate this relationship

empirically.

Data and Methodology

To assess the impacts of conationality and coethnicity on interpersonal trust, as well

as the impact of national identification on the coethnic trust premium, I carried out a

multi-method research project in an ethnically diverse border region of Malawi between

July and December of 2012.

Malawi is a small, densely populated, landlocked country in south-central Africa, bor-

dered to the west by Zambia, the south and east by Mozambique, and to the north

by Tanzania. Malawi is home to at least ten ethnic groups, though processes under

colonialism reinforced three main ethno-regional identities – the Tumbuka in the North,

the Chewa in the Center, and the Yao in the South (?) – and these ethnic divisions

have been relevant for political behavior at least since the introduction of multiparty

elections in 1994 (??).

Further, and most relevant to the topic at hand, survey data suggest that trust in

Malawi is particularly ethnically-defined and that territorial nationalism is weak relative

to ethnic identification. The third round of the Afrobarometer public opinion surveys

(?) asked questions in several African states about trust in coethnics, trust in non-

coethnics, and degree of identification with the nation relative to one’s ethnic group.

Across the sixteen states in the sample, Malawi ranks 14th in terms of the percentage

of respondents that identified with the nation more than the ethnic group, and 15th in

terms of the rate at which non-coethnics were trusted relative to coethnics. In short,

Malawi appears to be particularly characteristic of the popular image of African states,

with its relatively weak territorial nationalism and trust concentrated within ethnic

groups.7 Therefore, any observed effects of conationality on interethnic trust within

Malawi are likely to generalize to a large set of other states.

7The 446 Malawians in my study sample have even weaker national identification and lower gener-alized trust than a nationally representative sample from across Malawi (?) using the same questionwording. Study participants appear less nationalistic and less trusting even compared only to demo-graphically similar Malawians from the Afrobarometer sample.


Field Site

An ideal sample would allow me to isolate the independent effects of conationality and

coethnicity on trust. Locating the study at the intersection of a national and an ethnic

boundary allows me to approximate this ideal sample, by creating a setting in which

nationality and ethnicity are orthogonal at the local level. Thus, data were collected

among individuals living in Traditional Authority Chulu in the northwestern corner of

Kasungu District, Malawi, focused around the Chisinga trading center. This region was

selected because it is located at the approximate intersection of the ethnic boundary

between the Chewa and the Tumbuka and the national boundary between Malawi and

Zambia (see Figure 1).

[Figure 1 about here.]

This simple fact lends great power to the research design by allowing me to vary common

ethnic group membership and common national group membership independently. This

in turn allows me to measure within-subject differences in trust based on coethnicty

and conationality. For example, for Malawian Chewa respondents living in this border

region, I was able to measure trust in four different types of individuals: a Chewa from

Malawi (same nationality, same ethnicity), a Tumbuka from Malawi (same nationality,

different ethnicity), a Chewa from Zambia (different nationality, same ethnicity), and

a Tumbuka from Zambia (different nationality, different ethnicity). I then evaluate the

degree to which trust is conditioned on shared membership in each type of identity

group, both on average and as a function of interpersonal differences and experimental

manipulation. This design is part of a long tradition of taking advantage of ethnic

groups partitioned by national boundaries in Africa (???). Further, the research builds

on previous work in the same border region in which ? shows that the Chewa-Tumbuka

divide is particularly politically relevant to individuals living on the Malawian side of the

border, because these identities map onto different political coalitions at the national-

level. Thus, the field site offers a particularly pertinent setting in which to evaluate the

impact of national identification on interethnic relations.

Individuals from across 16 villages (8 Chewa villages and 8 Tumbuka villages) near

the Zambian border were interviewed in the first stage of the study. These 16 villages

were selected based on the following criteria: they were officially registered as a village

with the National Statistics Office, they were predominantly ethnically homogeneous


(either Chewa or Tumbuka), their residents could walk to the nearest trading center in a

reasonable amount of time, and their residents regularly attended the weekly market at

that trading center. These last two criteria were necessary logistically, as participation

in the second stage of the study required participants to be willing and able to meet

the research team in the weekly market held at the local trading center.

Protocol

The protocol included three instruments, used over two time periods:

1. Identity Survey: includes questions on individual demographics, strength of

national group identification, policy attitudes, interpersonal trust, and social net-

works.

2. Market Survey: short survey, includes questions on market participation and

embedded exposure to the national prime (increased salience of the national iden-

tity) for those randomly assigned to the treatment group.

3. Trust Games: within-subjects behavioral measure of the degree to which indi-

viduals condition their trust on conationality and/or coethnicity

Over eight weeks, the research team visited two villages per week for the Identity

Surveys (one Chewa and one Tumbuka) and held two game sessions per week. The

study was conducted over two time periods for each participant.

Stage 1: In the Village

In the first time period, 32 residents in each of the 16 villages were randomly selected to

participate. Because household lists were unavailable or unreliable at the village level,

households were selected using point sampling.8 Within a selected household, surveyors

listed all members of the household and the household head would then blindly select

from a set of cards, each representing a household member. If the selected household

8The research team always started from the geographic center of a village. A glass soda bottlewas spun to determine the direction in which each surveyor would proceed from that center point.Based on the village chief’s estimate of the number of households, the surveyors would select everynth household such that the number of households divided by n would equal the total required samplesize (32).


member agreed to participate, he or she was privately interviewed by a research assis-

tant. The “Identity Survey” included questions on individual demographics, strength

of group identification, policy attitudes, interpersonal trust, and social networks. This

survey took, on average, around 38 minutes.

At the end of the Identity Survey, the participant was given a ticket with his or her

respondent ID number and information about a second study to be held in a local

market the following week. The participant was invited to attend this second session,

and told that he or she would be given 500 MWK ($3.50) as a show-up payment and

have the opportunity to earn additional money through behavioral activities. As there

were two market sessions per week, half of the respondents within a given village would

be invited to the first market session while the other half would be invited to the second

market session.

Stage 2: In the Market

The second time period includes the Market Survey and the Trust Games. This session

was held in a building near to the public market, during the weekly market day. Every

Monday the game session was held in the weekly Chisinga market, and every Tuesday

the game session was held in the weekly Chimaliro market.9

15 individuals from four different villages were invited to attend each game session.10

Two of the four villages were those that the team had visited in the previous week to

conduct Identity Surveys (one Malawian Chewa village and one Malawian Tumbuka

village). The other participants were invited from a Chewa village and a Tumbuka

village just across the border in Zambia. Thus, in each game session, there were a

maximum of 15 Malawian Chewa, 15 Malawian Tumbuka, 15 Zambian Chewa, and 15

Zambian Tumbuka.11

9Because Chimaliro was too far for many participants to walk, we asked participants invited to theChimaliro market session to meet at Chisinga trading center by 8 am. We then provided transport tothe Chimaliro market (and back) by lorry.

10The research design called for 15 residents from each of four villages, but 16 residents were invitedfrom each village. We chose to over-invite in order to allow for some no-shows. In cases were all 16invited participants showed up, the 16th individual was thanked, paid the show-up fee, and dismissed.

11Because of no-shows, especially among Zambians, in some game sessions there would be less than15 participants of one type. Because random pairings for each trust game were determined ahead oftime, some individuals were paired in a trust game with someone who failed to show up. In thesecases, the research assistants “played” for that no-show in a way that was most generous to thepartner (trusting all of the endowment, and returning the entire amount entrusted). While making


At the beginning of each session, the Trust Game was explained to the entire group of

participants in both local languages (Chichewa and Chitumbuka) and they were told

as a group which four villages had sent participants that day. It was noted, publicly,

that the group included both Zambian and Malawian individuals, and both Chewa and

Tumbuka individuals from each country. Thus, which villages contained individuals

of which type (Malawian Chewa, Malawian Tumbuka, Zambian Chewa, and Zambian

Tumbuka) was made public knowledge, although anecdotal evidence suggests this was

already common knowledge among participants. Appendix A provides the scripts used

to explain the trust game.

Respondents were called one at a time into private rooms with a research assistant where

the rules of the Trust Game were explained again and informed consent was obtained.

Then, the market survey, with an embedded experimental prime, was carried out. For a

randomly assigned half of the participants, the market survey had two extra questions,

which served as a national prime. In July 2010, the Malawian national flag was officially

changed (see Figure 2).12 Because there was an ongoing debate about which flag should

be used at the time data was collected, it was not odd to ask respondents their opinion

on the two flags.13 Research assistants had large images of each flag in front of the

participant, and explained the symbolism of each flag (they are very similar in meaning)

and then asked the respondent which flag they thought best represented the Malawian

nation. Their actual preference is not of interest: simply asking respondents to consider

the historical symbolism of the flags should have served to increase the salience of their

Malawian national identity. The flag images were left on the table for the duration

of that individual’s four decisions about how much to trust their partner in the Trust

Game. The use of the Malawian national flag(s) as a national identity prime builds on

work in both political science (?) and social psychology (????).

game decisions, a participant paired with a no-show partner would not be informed that their partnerwas not there and would not be playing, but in the final payouts at the end of the game session thisinformation would be disclosed. This allows me to use game decisions for all Malawian participants,even when an insufficient number of other players showed up.

12In 2012, after the death of President Bingu wa Mutharika and the installation of the new president,Joyce Banda, the national flag was changed back to its original design.

13Using questions about the flag change, rather than simply exposing participants to the imageof the national flag, reduced the likelihood that they were aware of our intention to prime nationalidentity. A potential drawback of this choice of prime is that it might have inadvertently signalednational divisions. However, this potential weakness of the national prime biases against finding anypositive impact on interethnic trust; thus any identified effect is likely to underestimate the potentialimpact of national symbols, in general.



After completing the Market Survey, each respondent played the Trust Game with

four anonymous partners. The trust game is a two-player game in which Player A

is given a sum of money and asked to decide how much money to send to Player B.

Any money transferred from Player A to Player B is tripled by the experimenter, and

Player B then decides how much of the tripled money to return to Player A. The

amount of money transferred from Player A to Player B is interpreted as the degree of

trust Player A holds in Player B, while the proportion of the tripled amount returned to

Player A by Player B is interpreted as Player B’s degree of trustworthiness in relation to

Player A. Each respondent played as Player A (Trustor) four times, with four different

partners. Then, he or she played as Player B with four different anonymous partners.

Each participant played each role with four different “types” of partners: a conational

coethnic, a conational non-coethnic, a non-conational coethnic, and a non-conational

non-coethnic.14 There were four different orders in which game partners were assigned,

and use of these four order sets were balanced across subjects.

Originally designed for the lab (?), the trust game has been increasingly used in the

field as a measure of trust (????). However, the game has been criticized for conflating

trust and risk acceptance (??). By comparing an individual’s behavior across different

types of partners, this research design allows me to take advantage of within-subject

variation in trust due to shared nationality and shared ethnicity, while controlling for

individual differences in general levels of trust and risk acceptance, which vary widely

across individuals (?).

In addition to the standard trust game instructions, the protocol also included an ex-

plicit frame for understanding the game. In particular, during the general instructions

to the entire group of participants, we framed the game as analogous to the decision

about whether to sell one’s surplus maize locally versus sending the maize with a vir-

tual stranger to be sold in the capital for a much higher price (see Appendix A for the

exact language used). The frame was included for three reasons. First, because of its

abstract nature, the behavioral economic Trust Game can be difficult to understand.

Focus group discussions during the piloting phase of the project suggested that the

maize selling frame most closely resembled the logic of the trust game and made the

14For example, a Malawian Chewa was partnered with a Malawian Chewa, a Malawian Tumbuka, aZambian Chewa, and a Zambian Tumbuka.


game instructions much easier for participants to understand. Second, the theoretical

motivation for using the trust game to measure interpersonal trust at the micro-level

is to help understand the impact of ethnic difference on economic transactions under

different levels of national identification. By framing the trust game as an economic

transaction that most participants had engaged in, the trust game became both more

familiar and more connected to the theoretical motivation of the project. Third, existing

scholarship has shown that the way in which a game is understood vis-a-vis different

cultural or economic frames affects the way in which individuals behave within that

game (?????). Thus, explicitly providing a frame with which to understand the game

reduces the likelihood that different individuals used different frames in making behav-

ioral decisions.

For each game played as the Trustor, the participant was told which village their partner

was from, thus indicating both the nationality and ethnicity of the partner.15 The

participant was then given 60 MWK in the form of three 20 MWK bills. Thus, there

were four options for how much the participant could send to each of his or her four

partners: 0 MWK, 20 MWK, 40 MWK, or 60 MWK. While the research assistant was

in the room with the participant as this decision was made, participants made their

decisions privately by turning away from the research assistant or moving to the corner

of the room for each decision.

Sample Demographics

The Identity Survey data was collected for 508 Malawians (31–32 per village), and the

Market Survey and Trust Game behavior for 421 of those 508 Malawians (210 Chewa

and 211 Tumbuka).16 While not a main focus of the project, we also collected (as a

15Recall that this information was also conveyed explicitly during the introductory instructions givento the entire group of participants. Thus, the relationship between village and nationality/ethnicitywas common knowledge among participants.

16465 (92%) of the 508 Malawians who were invited actually showed up for their assigned gamesession. Of those 465, data were not collected on 14 because the other 15 participants had alreadyshown up from their village, and 30 people were excluded from the dataset because their demographicinformation (age, gender, marital status, level of education) did not match across the two surveys,suggesting that the market study participant was not the same person as the individual originallyinterviewed and invited. The 421 participants do not differ significantly from the 87 participantsexcluded as mismatches, no shows, and late shows in terms of strength of national identification,gender, or tribe. However, excluded potential participants were, on average, two years younger thanparticipants (t = 1.77, df = 506, p = 0.08) and had completed around half of a year less of education(t = 1.73, df = 506, p = 0.08).


by-product of the research design) a Market Survey and Trust Game behavior for 341

Zambians (168 Chewa and 172 Tumbuka), although the Zambian participants were not

selected randomly and a few Zambian villages were invited to send participants more

than once.

The 421 Malawian participants from whom we have all three sources of data ranged

in age from 18 to 50, which was our target age range, and were on average 31 years

old (SD = 9.5 years). A little over half of the participants (56%) were male, and 90%

were married. By design, 50% of participants identified their tribe as Chewa, and 50%

as Tumbuka. The median number of years of education was 6 years, with 30% having

completed primary school (8 years) and only 2% having completed secondary school

(12 years). Due to the rural nature of the field site, all participants were subsistence

farmers with only 3 participants reporting any form of wage income.

Empirical Measures

For the empirical analyses, I focus on three key variables: strength of national identi-

fication, salience of national identity through the experimentally assigned flag prime,

and interpersonal trust behavior. I discuss the measurement of each of these in turn.

Summary statistics can be found in Appendix B, Table B.1.

Measuring Strength of National Identification

The Identity Survey measured strength of national identification (NatID) using agree-

ment with 6 first-person statements expressing affective, behavioral, and cognitive iden-

tification with the national group, adapted from ?. Affective identification is associ-

ated with the literature on social cohesion, and emphasizes emotional attachment to

the group and its other members. Behavioral identification is central to the common

fate literature, and focuses on the interdependence of members as a source of group

identification. Thus, behavioral identification with the nation should increase the more

that an individual depends on other citizens or perceives her fate to be dependent on

the nation’s fate as a whole. Cognitive identification comes from the social identity

literature, and theorizes that individuals categorize themselves as a member of a group

based on shared attributes and homogeneity of characteristics. Thus, the more homo-

geneous an individual perceives the national group to be, and the more he sees himself


as a typical member of that group, the stronger he should identify with the nation. The

components of the measure of national group identification are listed in Table 1 along

with their classification as affective, behavioral, or cognitive.

To construct a composite measure of strength of national identification, I averaged over

the six items on the scale, and standardized the resulting measure across the whole

sample. The scale is negatively skewed, ranging from -3.5 to 1.3.

[Table 1 about here.]

Experimental Manipulation of National Identity Salience

The dichotomous variable Flag indicates whether or not a participant was assigned to

the treatment condition and thus exposed to the national identity prime. In the treat-

ment condition, the common national (Malawian) identity was made salient through

the discussion of the symbolism of two different versions of the Malawian national flag.

51% of participants were assigned to the national identity prime.

Because assignment to treatment was randomized, treatment status should be orthogo-

nal to all participant characteristics, in expectation. In practice, treatment and control

groups were balanced in terms of age, gender, level of education, tribe, and strength of

national identification (Table 2).


Unfortunately, treatment was not balanced across the four different game-partner or-

ders. Because treatment and control participants were alternated, and order of partner

type was serially assigned across subjects, treatment status and the order in which

participants were paired with the four different partners are not orthogonal.17 Because

order of play, in general, is a strong predictor of both whether someone sends money

and how much money they send18 – participants tend to trust more in the first game

17Within the treatment group 38% of participants played with the four partner types in the order ofsame nationality and same ethnicity (SNSE), same nationality and different ethnicity (SNDE), differentnationality and same ethnicity (DNSE), and different nationality and different ethnicity (DNDE); 12%played in the order of SNDE, DNSE, DNDE, SNSE; 37% played in the order of DNSE, DNDE, SNSE,SNDE; and 13% played in the order of DNDE, SNSE, SNDE, and DNSE. In contrast, within thecontrol group 15%, 33%, 14%, and 38% played in those four orders, respectively.

18This effect is likely driven by social desirability concerns. In the first game, participants mayhave been hesitant to make a socially undesirable decision. However, after they observed that theenvelope containing their first transfer was not opened or discussed, they felt more free to make


played than in the subsequent three games – all analyses will employ game round fixed

effects. As a result, I analyze the impact of the national identity prime on conditional

trust within a game round (e.g., 1st, 2nd, 3rd, or 4th game in the series). Results are

also presented with game sequence fixed-effects in Table C.3 of Appendix C.

Measuring Trust

Interpersonal trust in anonymous individuals of four “types” was measured for each

respondent in the trust game: trust in individuals of the same nationality and same

ethnicity, same nationality and different ethnicity, different nationality and same eth-

nicity, and different nationality and different ethnicity. I construct the measure of trust

from an individual’s decision in each of the trust games he or she played.

The measure of trust (Trust) maps the amount of money sent in the first stage of

the trust game onto three values of the dependent variables, corresponding to no trust

(Trust = 0), moderate trust (Trust = 1), and high trust (Trust = 2). Individuals that

sent no money to their partner in the trust game are coded as having no trust in their

partner. If an individual sent either 20 MWK or 40 MWK, their trust level is coded as

moderate. Finally, if the participant sent all 60 MWK, trust in their partner is coded

as high. Overall, participants had no trust in their partner 12% of the time, exhibited

moderate trust 69% of the time, and trusted fully 19% of the time. Compared to

other trust games carried out across Africa (e.g., ????????), on average, trust among

Malawians in the sample is quite strong (see Table B.2 in Appendix B).

A continuous measure of how much money is sent in the first stage of the trust game is

a more common operationalization of trust. However, in the context of this experiment,

it is likely to be a noisier measure than ordinally ranked decisions to trust none, some,

or all of the endowment for two principal reasons. First, anecdotal evidence suggests

that participants felt that sending some amount of money was a socially desirable way

to the play the game, while evaluations of how much money constituted an acceptable

amount varied across individuals.19 As a result, the strongest signal in the data comes

socially undesirable decisions in subsequent games. There is no observable order effect after the firsttrust game.

19While we assured participants that their decisions would be made in private, it seems that theydid not fully trust this assurance until after making their first decision. Order of the games is a verystrong predictor of amount sent in the first stage of the trust game, with participants sending the mostmoney in the first game, before observing that the decision was actually private. For example, less


from the decisions to send nothing or to send everything, while the decision of whether

to send 20 MWK or 40 MWK is less informative. Second, because participants only

had four options of how much money to send in the first stage of the trust game, if they

start from varying baselines of the minimum acceptable amount to send, there is very

little observable variation around that minimum.20 Other potential operationalizations

of trust include a dichotomous indicator of trusting anything or a dichotomous indicator

of trusting everything. The main analyses are repeated using these three alternative

dependent variables, and those results are reported and discussed in Appendix C.

Empirical Models and Results

To model the impacts of coethnicity, conationality, national identification, and national

identity salience on conditional trust, I construct a dataset that includes four trust

games per individual – thus, the unit of analysis is the individual-trust game. While

the measure of trust is a three-level ordinal variable, ordinary least squares regression

estimates are presented for ease of interpretation. The results are robust to use of

ordered probit models, and those results are reported in Appendix C.

Shared Identity and Conditional Trust

Figure 3 shows the average degree of trust in each of the four types of partners, pooling

across individuals. The error bars represent Statistical Significance Bars (?), with

which the lack of an overlap represents statistical significance at the 0.05 level. The

graph makes apparent that individuals trust conational coethnics the most, conational

non-coethnics and non-conational coethnics at similar rates, and non-conational non-

coethnics the least. These averages suggest that conationality and coethnicity are given

roughly equal weight in decisions about whom to trust.


than 5% of individuals sent nothing in their first trust game.20A combination of budget concerns and currency denominations in Malawi limited me to giving

participants four choices. The MWK 20 bill is the smallest bill, and MWK 5 and MWK 10 coins aremuch more difficult to acquire. Because my budget was constrained to offering a maximum of MWK60 per game, participants were offered only four options instead of the 5-11 options common in othergames (see Table B.2 in Appendix B).


These averages, however, pool decisions over individuals and do not account for the

within-subject component of the research design. In order to identify the within-subject

impact of shared identity on trust, I estimate the degree to which trust is conditioned

on conationality and coethnicity using the following regression model with participant

fixed-effects:

Trustij = β1CoNationalij + β2CoEthnicij + µi + νk + εij

where Trustij is the three-level ordinal measure of trust in game j by respondent

i, CoNationalij is a dummy for whether the partner is a conational in that game,

CoEthnicij is a dummy for whether the partner is a coethnic in that game, νk denotes

a vector of fixed effects for the round in which a particular game was played (1st, 2nd,

3rd, or 4th,) and µi and εij represent the individual fixed-effect and the game-specific

error term, respectively. In a second model, I include an interaction of coethnicity and

conationality in order to determine whether trust in conational coethnics is driven by

the additive effect of being both a conational and a coethnic, or if there is some added

trust benefit from sharing both identities. The model estimated is:

Trustij = β1CoNationalij + β2CoEthnicij + β3CoNatij × CoEthij + µi + νk + εij

The results of these two estimations are reported as Model 1 and Model 2 of Table 3.


The coefficients of interest are on the CoNational and CoEthnic variables in both

models, and their interaction in Model 2. Model 1 of Table 3 shows that participants

on average exhibit more trust in conationals than non-conationals, and more trust

in coethnics than non-coethnics. For both forms of shared identity, the effect size

is 0.06, which translates into a 0.15 standard deviation increase in trust. When the

interaction of conationality and coethnicity is included (Model 2), the coefficient on

the interaction is positive and of a similar magnitude as the independent effects of

either coethnicity or conationality, but the estimate is estimated precisely enough to

be statistically distinguishable from zero. In other words, while conational coethnics

are trusted more than any other type of partner, I cannot conclude that the impact of

sharing both identities is any larger than the added independent effects of conationality

and coethnicity.


Models 1 and 2 employ participant fixed-effects and thus estimate the within-subject

effects of coethnicity and conationality on trust. Including participant random-effects

instead of fixed effects would produce a more efficient estimate of the effect sizes by

using a weighted average of within- and between-subject effects. Further, participant

random-effects in the form of random intercepts allow me to account for unobserved in-

dividual differences that impact trust, while also allowing me to include individual level

covariates of interest in subsequent analyses. However, including participant random-

effects could produce inconsistent estimates if independence between covariates and

the random-effects is violated. The results of estimating the same models with µi as

participant random-effects rather than participant fixed-effects are presented as Models

3 and 4 in Table 3. The fact that the random-effects models produce the exact same

coefficients as the fixed-effects models (Hausman test, χ2(5) = 0.52, p = 0.99) suggests

that the random-effects model can be confidently employed.

In sum, individuals are conditioning their trust in roughly equal magnitudes on both

coethnicity and conationality. While participants trust partners who share both iden-

tities the most, this effect is not statistically different from the additive effect of being

both a conational and a coethnic.21 That nationality would be just as important as

ethnicity in decisions about whom to trust in a rural region of sub-Saharan Africa goes

against conventional expectations. The result is even more surprising given the context

in which it appears. First, all participants are members of an ethnic group partitioned

by colonial, and subsequently state, borders – a condition that is expected to make na-

tional identification less likely due to the perception that the states resulting from such

partitioning are illegitimate (????). In addition, the border between partitioned coeth-

nics in this region is quite porous, with over 70% of participants having close friends or

family across the international border and around a third of them having crossed the

border in the month preceding the study. The fact that this particular population is

considering the nationality of their partner with equal weight as ethnicity, then, is an

important indication that the identity group defined by the territorial nation is more

important to interpersonal relations in rural Africa than previously appreciated.

21As a result, in subsequent models the interaction of conationality and coethnicity is not included.


National Identification and the Coethnic Trust Premium

The results above provide evidence that the national identity in Malawi is not, as is so

often assumed, irrelevant for social decisions, such as whom to trust. However, these

results only show that Malawians trust other Malawians more than they trust Zambians,

controlling for coethnicity. This is an important corrective to the common dismissal of

the importance of territorial nationalism in post-colonial Africa; but it does not tell us

anything about the impact of territorial nationalism on interethnic relations within the

diverse national community. Does increased identification with the national identity

reduce the degree to which interpersonal trust is ethnically-based, as predicted?

To answer this question, I restrict the sample to trust games played between Malawians,

which allows me to hold conationality constant. Thus, in this subset of the data, there

are two observations per participant, one when deciding how much to trust a conational

coethnic and one when deciding how much to trust a conational non-coethnic. For now,

I restrict the sample to individuals in the control condition, who did not see the flags

prior to making trust decisions. Doing so allows for an estimate of the impact of

underlying strength of national identification on interethnic trust, without the national

identity being made salient. Among this subsample, I estimate the following model:

Trustij = β1CoEthnicij + β2NatIDi + β3CoEthnicij ×NatIDi + µi + εij

where Trustij is the measure of trust in game j by respondent i, CoEthnicij is a dummy

for whether the partner is a coethnic in that game, andNatIDi is the standardized

measure of strength of national identification. As in the previous models, νk denotes

a vector of fixed effects for the round in which a particular game was played (1st, 2nd,

3rd, or 4th,) and µi and εij represent the individual random-effect and the game-specific

error term, respectively. The results of this estimation are reported in Model 1 of Table

4.


The coethnic trust premium is large and statistically significant, with coethnic cona-

tionals being trusted 14% more than non-coethnic conationals at average levels of na-

tional identification (NatID = 0). The estimated impact of an individual’s strength

of national identification, NatID, on trust in non-coethnics is positive and statistically


significant, suggesting that increased nationalism is related to increased trust in cona-

tionals from a different ethnic group. Finally, the interaction of coethnicity and strength

of national identification is negative. This means that, on average, participants with

stronger national identification are exhibiting a decreasing coethnic trust premium, as

the theory predicted. Figure 4 shows the estimated interaction effect graphically. The

graph shows that among weak nationalists – those who do not identify very strongly

with their Malawian identity – coethnics are trusted at a higher rate than non-coethnics.

However, among the strongest nationalists in the sample, ethnicity is irrelevant for trust,

with coethnics and non-coethnics trusted at equal rates.22 Thus, consistent with ex-

pectations, results suggest that nationalism at the individual level does indeed reduce

ethnic-based trust.


National Identity Salience and the Coethnic Trust Premium

The analysis above evaluated the impact of interpersonal differences in strength of

national identification on the size of the coethnic trust premium. Next, I evaluate the

impact of experimentally increasing the salience of the national identity on the size of the

coethnic trust premium, ignoring interpersonal differences in strength of identification

for the moment. The effect of national identity salience on trust is evaluated by looking

at the relationship between randomly assigned exposure to the Malawian flag (a national

identity prime) and the degree to which trust is conditioned on coethnicity. Thus, I

estimate the following model:

Trustij = β1CoEthnicij + β2Flagi + β3CoEthij × Flagi + νk + µi + εij

22Who are these strong national identifiers? In other work, I have found that at the individuallevel gender (male), urbanization, education, and formal employment are all positively related togreater national identification (?). In the present study, I again find that males are significantly morenationalist than women. However, there is very little variation in terms of education and no variation inurbanization and formal employment among the participant population. Instead I find that, consistentwith ?’s (?) argument about the role of mass media in “imagining the nation,” the frequency of radiolistenership is a robust predictor of national identification. In addition, greater political knowledgeand having non-coethnic friends are positively related to national identification, while the frequencyof crossing the national border is negatively related to national identification, emphasizing the role oflife experience in shaping individual identification with the nation.


where Trustij is the measure of trust in game j by respondent i, CoEthnicij is a dummy

for whether the partner is a coethnic in that game, and Flagi is a dichotomous indicator

for whether an individual was exposed to the national flag prime or not. νk denotes a

vector of fixed effects for the round in which a particular game was played (1st, 2nd, 3rd,

or 4th, with the first play as the omitted category), µi represent the individual random-

effects, and and εij captures the game-specific error term. The results of estimating

this model are presented in Model 2 of Table 4.

In this model, the coefficient on CoEthnic estimates the impact of coethnicity on trust

only when the national identity is not made explicitly salient (the control group). The

average impact of the national flag prime is zero across both partners, meaning that

increasing the salience of the common national identity does not increase trust in cona-

tionals in general. However, the interaction between coethnicity and the national iden-

tity prime is negative and statistically significant at the 0.10 level, meaning that the size

of the coethnic trust premium is reduced when the national identity is made contextu-

ally salient. Importantly, it is not the case that exposure to the national flag makes one

less cognizant of coethnicity in general, as the size of the coethnic trust premium when

playing with Zambians (non-conationals) is unaffected by exposure to the Malawian

national flag (see Table C.4 in Appendix C.3).

These result suggest that, on average, priming the common national identity among

Malawians reduces the degree to which they discriminate their trust along ethnic lines,

but only among conationals. This means that even ignoring underlying levels of national

identification, we can detect an impact of the contextual salience of the national identity

on the degree to which ethnic difference is relevant for trust.

Strength of National Identification, National Identity Salience, and the Co-

ethnic Trust Premium

Thus far, the results have shown that when the national identity is not salient, the

coethnic trust premium disappears among very strong national identifiers and (ignor-

ing strength of national identification) increasing the salience of the common national

identity reduces the size of the coethnic trust premium. Next, I consider the interaction

between strength of identification and salience of the common identity to determine if

underlying strength of national identification has a differential effect depending on the

contextual salience of the common national identity. Figure 6 shows the average size of


the coethnic trust premium, by treatment status, over three categories of strength of

national identification.23 While none of these differences are statistically significant at

conventional levels when pooled across individuals, the trend in the data is that increas-

ing the salience of the national identity reduces the size of the coethnic trust premium

among weak and moderate national identifiers, and that this reduction amounts to their

coethnic trust premium being similar to that of strong national identifiers.


To estimate the impact of national identity salience on the coethnic trust premium

by strength of national identification at the individual level, I estimate the following

model, which includes a triple interaction term:

Trustij =β1CoEthnicij + β2NatIDi + β3Flagi + β4CoEthnicij ×NatIDi+

β5CoEthnicij × Flagi + β6CoEthnicij ×NatIDi ∗ Flagi + νk + µi + εij

where Trustij is the measure of trust in game j by respondent i, CoEthnicij is a

dummy for whether the partner is a coethnic, NatIDi is the standardized strength of

national identification, and Flagi is a dichotomous indicator for whether an individual

was exposed to the national flag prime or not. The model also includes their pair-wise

interactions, as well as the triple interaction. Finally, νk denotes a vector of fixed effects

for the round in which a particular game was played (1st, 2nd, 3rd, or 4th, with the first

play as the omitted category), while µi are participant random effects and εij is random

error. The results of this estimation are reported in Model 3 of Table 4.

The coefficient on CoEthnic is 0.14: this means that for individuals with an average

level of national identification (NatID = 0), when the national identity is not made

salient (Flag = 0), coethnics are trusted a third of a standard deviation more than non-

coethnic conationals, an effect which is highly statistically significant. Because of the

difficulty in interpreting triple interaction terms and their statistical significance (?), I

present the marginal effect of coethnicity on trust graphically as a function of strength of

national identification and the experimental manipulation of national identity salience

in Figure 6. The marginal effect is plotted over strength of national identification

ranging from -2 to 2, a range which includes 98% of the sample, although NatID = 1.35

23Low, medium, and high national identification categories were assigned by splitting the sampleinto three equal quantiles based on strength of national identification.


is the highest national identification observed in the data.


Figure 6 shows that when the national identity is not made contextually salient (ex-

perimental control condition), the size of the coethnic trust premium is decreasing in

strength of national identification. This is the same data presented above in Figure 4.

The coethnic trust premium is indistinguishable from zero for those in the top quartile

of national identification (NatID >= 0.8). In contrast, when the national identity is

made contextually salient (experimental treatment condition) there is a slight positive

effect of national identification, but the coethnic trust premium is indistinguishable

from zero at all levels of national identification. Another way to think about this in-

teraction is as a heterogeneous treatment effect: increasing the salience of the national

identity has a larger causal impact on the coethnic trust premium among weaker na-

tional identifiers. In fact, as 6 shows, we observe a statistically significant treatment

effect of national identity salience on coethnic trust only for participants with below

average strength of national identification (NatID < 0).

Together, these results suggest two important conclusions. First, under conditions in

which the national identity is not made contextually salient, an individual’s pre-existing

strength of national identification is negatively related to the coethnic trust premium

– the more strongly one identifies with the nation personally, the less she discrimi-

nates between other members of the nation based on sub-national ethnic differences.

As a result, there is a subset of people who identify very strongly with the Malawian

nation and who trust coethnics and non-coethnics equally. Second, national identity

salience causes everyone, regardless of their pre-existing degree of national identifica-

tion, to behave like the strongest national identifiers, eliminating ethnic-based trust

discrimination.

This reduction in the size of the coethnic trust premium among strong national iden-

tifiers, and under conditions of high national identity salience, could result from in-

creasing trust in non-coethnics, reducing trust in coethnics, or both. The nature of this

reduction in the trust differential between coethnics and non-coethnics has implications

for aggregate levels of trust. Figure 7 shows average levels of trust in coethnics and

non-coethnics by strength of national identification and salience treatment condition.



The patterns suggest that, on average, strength of national identification is reducing

the coethnic trust premium by increasing trust in non-coethnics, as expected. How-

ever, in contrast to expectations, national identity salience is reducing the coethnic trust

premium by reducing trust in coethnics. At all three levels of national identification

strength, exposure to the flag reduces trust in coethnics, while trust in non-coethnics

remains constant for weak and strong identifiers and decreases for moderate identifiers.

These results suggest that increased national identity salience leads to ethnic decate-

gorization rather than national recategorization. As a result, aggregate levels of trust

are actually lower under the national salience condition. Thus, the ultimate value of

increased contextual nationalism on trust depends on the importance of absolute levels

of trust versus the degree to which coethnicity determines trust.

Discussion

This paper has presented a study of national identification and interpersonal trust in

one ethnically diverse region of Africa. The theoretical motivation of the study is inter-

disciplinary, applying well-founded theories from social psychological research on inter-

group relations and the historical and sociological literature on nationalism in Europe,

to the study of a central question in comparative political science – how to facilitate

interethnic cooperation in diverse African states. The research design takes advantage

of the natural experiment afforded by the intersection of an ethnic and a national bor-

der to identify the impacts of conationality and coethnicity, independently from each

other. The combination of an experimental manipulation with observational data at

the individual-level allows for the assessment of both interpersonal and contextual ef-

fects of national identification on interpersonal trust. Finally, the “lab-in-the-field” use

of multiple trust games allows for a within-subject behavioral measure of trust. Using

a within-subject measure of trust alleviates concerns about individual-level differences

in risk acceptance and general trust, and also provides behavioral observations that are

less prone to social desirability than attitudinal measures of trust. Three principal sets

of findings offer important insight into how we understand the power of nationalism to

transform interpersonal relations in ethnically diverse states of sub-Saharan Africa.

The first major finding is that, on average, conationality is just as important as coeth-

nicity in decisions about whom to trust. This finding runs counter to the general image


of African states as being almost exclusively organized around tribal loyalty, with little

credence given to the power of territorial nationalism (????). Instead, the evidence is

consistent with an alternative view in which territorially-defined national identities in

Africa are meaningful to the average citizen. For those that have argued – based on

anecdotal evidence – that national forms of group identification should be considered

alongside ethnic and tribal identification (e.g., ?), this finding provides more rigorous

empirical support, at least for one locale. Further, the previous finding that rural, sub-

sistence farmers in Africa valued their national identities at least as much, if not more,

than their ethnic identities (?) is bolstered by my finding that this pattern extends to

real behavioral decisions.

The second important result is that individual differences in strength of identification

with the territorial nation are related to the degree to which conationals of different

ethnicities are trusted equally. More specifically, when the national identity is not made

contextually salient, individual-level variation in pre-existing strength of national iden-

tification is negatively related to the degree to which coethnics are trusted more than

non-coethnics. For individuals with low to moderate strength of national identification

(below the 75th percentile), coethnics are trusted at a higher rate than non-coethnics.

For strong national identifiers (the top quarter of the sample), however, this coethnic

trust premium disappears. This is significant because it means that even in a setting

where we expect relatively weak national identification – rural border villages compris-

ing members of partitioned ethnic groups – there exists a subset of people for whom

strong identification with the territorially defined nation leads them to ignore ethnicity

in decisions about trust in fellow citizens.

The third major finding emerges from the experimental manipulation of national iden-

tity salience. When the national identity is made contextually salient, the coethnic trust

premium is reduced to zero for strong and weak national identifiers alike. This means

that making the national identity contextually salient, through the use of a national

symbol, induced the entire population to behave as if they were strong national iden-

tifiers. This finding is analogous to what ? call a “mobilizing effect”: the presence of

a national flag “mobilized” Malawian citizens who would otherwise be inclined to base

their trust on coethnicity to ignore ethnic differences. However, questions remain as

to how long such a saliency effect will persist,24 as well as the relative effectiveness of

24One study finds that a single exposure to the American national flag impacts political attitudesup to eight months later (?)


different national symbols.

Taken together, these results suggest that a focus on the common national identity

reduces ethnic discrimination in trust decisions. For strong national identifiers, the

national identity is always consequential, while for weak to moderate national identifiers

a common nationality only leads to less ethnic discrimination in the presence of a

national symbol. These findings lend empirical support to the theoretical shift away

from thinking about social identity categories as fixed groups in the world. As ? has

strongly argued, strength of identification with a category – what he calls “groupness”

– varies across groups, across individuals within groups, and across different contexts

for a single individual. Thus, studies that fail to take such variation into account at

best miss an important source of consequential variation, and at worse risk reifying

irrelevant social identities or dismissing identities that are indeed relevant to a subset

of the population or in particular contexts.

These findings may also have important policy implications for nation-building in eth-

nically diverse African states. If we think in terms of ideal types, existing literature

suggests that there are two main ways to build an effective nation in which all citizens,

regardless of ethnicity, are part of the same trust community. The first is to foster

wide-spread identification with the national group such that average citizens’ primary

allegiance is to the nation, above and beyond their loyalty toward other groups to which

they belong. This form of nation-building has been most often tied to strong, central-

ized, and universally free education systems (??). Other ways to foster this kind of

strong national identification are to institute the use of a common national language,

fund print media accessible to a majority of citizens, and post bureaucrats outside their

home region. This process is slow and resource intensive. The second way to build

a unified nation is to activate the national identity in everyday contexts through the

ubiquitous presence of national flags, mundane exposure to national symbols on cur-

rency, and creation and promotion of national sports teams, among other things. ? has

argued that such “banal nationalism” is an effective form of nation-building in that it

serves as a subtle but constant reminder of the common national identity.

The results of this study suggest that either form of nation-building can be effective

at reducing the degree to which ethnic diversity impacts generalized trust within the

nation. In fact, the coethnic trust premium is nonexistent both for individuals with

strong national identification and for individuals experimentally exposed to a national


symbol regardless of their strength of national identification. However, making the

national identity salient is less resource intensive, which is important in developing

states such as Malawi. Given their equal efficacy, the results of this study suggest that

nation-building policies aimed at reducing the impact of ethnic difference on generalized

trust should focus their efforts on increasing the salience of the territorially-defined

national community.

There are several reasons, however, to be cautious in our optimism. First, this study

focuses only on ethnic-based trust discrimination: policy makers must also weigh the

potential costs of increased nationalism on inter-state relations and conflict (???) and

the possible loss of multiculturalism in the long run (??). Second, in contrast to theoret-

ical expectations, increasing the salience of the national identity reduced ethnic-based

trust discrimination by reducing the coethnic trust premium, rather than extending

that premium to non-coethnic members of the nation (as increased strength of national

identification did). If this pattern generalizes beyond the context of the experiment,

then increased national salience may actually reduce aggregate levels of generalized trust

by reducing the particularized trust directed toward coethnics. Thus, the ultimate eco-

nomic value of increased national identity saliences rests on how important aggregate

levels of trust are for economic performance relative to the detriment of ethnically-

bound trust. Third, at least within the context of this experiment, trust was rarely

reciprocated with trustworthiness. In fact, on average, trust was exploited, as the av-

erage rate of return was only eighty-percent of the original investment.25 This suggests

that policies aimed at promoting greater trust in society will be counter-productive in

the absence of assurances that such trust will be reciprocated.

A final concern is that these results are based on a single field site in rural Malawi.

It is thus important to consider the scope of their generalizability. For example, the

size of the coethnic trust premium identified within the study population is likely to be

smaller than would be expected in less ethnically diverse areas (?). As a consequence

of the research design, participants were recruited from a very localized setting – about

240 square kilometers right on the border of two ethnic communities, across which

there is significant “mixing.” For example, around one-third of the sample have a

least one grandparent of a different ethnicity than the one they claim for themselves,

25While this rate was slightly higher for conationals, there was no benefit of coethnicity in termsof trustworthiness. Analyses of trustworthiness as a function of trust, shared identities, and groupidentification are explored further in other work.


a majority speak the other ethnic groups’ language fluently, and around half list a

non-coethnic as one of their four closest friends. While we do not have comparable

statistics for the Malawian population at large, or for other countries in the region, it is

likely that interethnic mixing would be lower in the more homogeneous regions within

these ethnically diverse states, and we know from other work that local level diversity is

negatively related to the size of the coethnic trust premium (?). As a result, the degree

to which national identification impacts the size of the coethnic trust premium may

differ outside areas with such diversity at the local level. Despite this, I find evidence

that the effects identified in this field experiment may generalize across Africa. In

nationally-representative survey data from sixteen sub-Saharan African states (?), the

average degree to which respondents prioritize their national identity vis-a-vis their

ethnic identity is negatively associated with the average degree to which coethnics

are trusted more than non-coethnics (Figure 8). Further, national relative to ethnic

identification is also negatively related to the size of the coethnic trust premium within

countries (Figure 9). While these data are observational rather than experimental, and

attitudinal rather than behavioral, they suggest that the micro-level findings reported

here are not unique to the specific location in which the data were collected.



The coethnic trust premium reported here is unlikely to account entirely for the un-

derdevelopment of economies in ethnically diverse African states, such as Malawi, or

for the concentration of trust within ethnic communities in those societies. Similarly,

the estimated impact of national identification and national identity salience reported

here are unlikely to offer a panacea to ethnic discrimination and ethnic-based trust.

However, these results do offer micro-level evidence that territorially-defined nations

in post-colonial Africa can form the basis of a group-based trust community, and that

increasing their relevance in the lives of citizens could help to reduce the degree to

which coethnicity determines interpersonal trust.

References

Afrobarometer (2006). Round 3 survey data. www.afrobarometer.org.

Afrobarometer (2008). Round 3 and Round 4 survey data. www.afrobarometer.org.

Alesina, A., Devleeschauwer, A., Easterly, W., Kurlat, S., and Wacziarg, R. (2003).Fractionalization. Journal of Economic Growth, 8(2):155–194.

Alesina, A. and La Ferrara, E. (2002). Who trusts others? Journal of Public Economics,85(2):207–234.

Alesina, A. and La Ferrara, E. (2005). Ethnic diversity and economic performance.Journal of Economic Literature, 43(3):762–800.

Anderson, B. (1991). Imagined Communities: Reflections on the Origin and Spread ofNationalism. Verso, London, UK, 2nd edition.

Ashraf, N., Bohnet, I., and Piankov, N. (2006). Decomposing trust and trustworthiness.Experimental Economics, 9(3):193–208.

Asiwaju, A. I. (1985). Partitioned Africans: Ethnic Relations Across Africa’s Interna-tional Boundaries, 1884-1984. St. Martin’s Press, New York, NY.

Bachman, B. (1993). An Intergroup Model of Organizational Mergers. Unpublisheddoctoral dissertation, University of Delaware, Newark, DE.

Bahry, D., Kosolapov, M., Kozyreva, P., and Wilson, R. (2005). Ethnicity and trust:Evidence from Russia. American Political Science Review, 99(04):521–532.

Banker, B. and Gaertner, S. (1998). Achieving stepfamily harmony: An intergroup-relations approach. Journal of Family Psychology, 12(3):310–325.

Bargh, J., Bond, R., Lombardi, W., and Tota, M. (1986). The additive nature of chronicand temporary sources of construct accessibility. Journal of Personality and SocialPsychology, 50(5):869.

Bargh, J. and Pratto, F. (1986). Individual construct accessibility and perceptualselection. Journal of Experimental Social Psychology, 22(4):293–311.

Barr, A. (2003). Trust and expected trustworthiness: Experimental evidence fromZimbabwean villages. The Economic Journal, 113(489):614–630.

Barr, A. (2004a). Kinship, familiarity, and trust: An experimental investigation. InFoundations of Human Sociality: Economic Experiments and Ethnographic Evidencefrom Fifteen Small-scale Societies, pages 305–334. Oxford University Press, NewYork.

www.afrobarometer.org

Barr, A. (2004b). Rational and biased trust. Center for the Study of African Economies,Working Paper Series, 2004-22.

Bates, R. (1983). Modernization, ethnic competition, and the rationality of politics incontemporary Africa. In Rothchild, D. and Olorunsola, V. A., editors, State VersusEthnic Claims: African Policy Dilemmas, pages 152–171. Westview Press, London,UK.

Bendix, R. (1964). Nation-Building and Citizenship. University of California Press,Berkeley, CA.

Berg, J., Dickhaut, J., and McCabe, K. (1995). Trust, reciprocity, and social history.Games and Economic Behavior, 10:122–142.

Beugelsdijk, S., De Groot, H., and Van Schaik, A. (2004). Trust and economic growth:A robustness analysis. Oxford Economic Papers, 56(1):118–134.

Bienen, H. (1983). The state and ethnicity: Integrative formulas in Africa. In Rothchild,D. and Olorunsola, V. A., editors, State Versus Ethnic Claims: African Policy Dilem-mas, pages 100–126. Westview Press, London, UK.

Billig, M. (1995). Banal Nationalism. Sage Publications, London.

Brambor, T., Clark, W., and Golder, M. (2006). Understanding interaction models:Improving empirical analyses. Political Analysis, 14(1):63–82.

Brewer, M. (1999). The psychology of prejudice: Ingroup love and outgroup hate?Journal of Social Issues, 55(3):429–444.

Brewer, M. and Gardner, W. (1996). Who is this “we”? Levels of collective identityand self representations. Journal of Personality and Social Psychology, 71(1):83–93.

Brewer, M. and Kramer, R. (1985). The psychology of intergroup attitudes and behav-ior. Annual Review of Psychology, 36(1):219–243.

Brewer, M. and Miller, N. (1984). Beyond the contact hypothesis: Theoretical perspec-tives on desegregation. In Miller, N. and Brewer, M. B., editors, Groups in Contact:The Psychology of Desegregation, pages 281–302. Academic Press, Orlando, FL.

Brewer, M. B. (1997). The social psychology of intergroup relations: Can researchinform practice? Journal of Social Issues, 53(1):197–211.

Brown, R. (2000). Social identity theory: Past achievements, current problems andfuture challenges. European Journal of Social Psychology, 30(6):745–778.

Brubaker, R. (2004). Ethnicity without Groups. Harvard University Press, Cambridge,MA.

Burnham, T., McCabe, K., and Smith, V. (2000). Friend-or-foe intentionality primingin an extensive form trust game. Journal of Economic Behavior & Organization,43(1):57–73.

Butz, D. (2009). National symbols as agents of psychological and social change. PoliticalPsychology, 30(5):779–804.

Butz, D., Plant, E., and Doerr, C. (2007). Liberty and justice for all? Implications ofexposure to the U.S. flag for intergroup relations. Personality and Social PsychologyBulletin, 33(3):396–408.

Calhoun, C. (1993). Nationalism and ethnicity. Annual Review of Sociology, 19:211–239.

Carter, T., Ferguson, M., and Hassin, R. (2011). A single exposure to the Americanflag shifts support toward Republicanism up to eight months later. PsychologicalScience, 22(8):1011–1018.

Collier, P. (2009). Wars, Guns, and Votes. Harper Perennial, New York, NY.

Connor, W. (1972). Nation-building or nation-destroying? World Politics, 24(3):319–355.

Connor, W. (1994). Ethnonationalism: The Quest for Understanding. Princeton Uni-versity Press, Princeton, NJ.

Cronk, L. (2007). The influence of cultural framing on play in the trust game: A Maasaiexample. Evolution and Human Behavior, 28(5):352–358.

Danielson, A. and Holm, H. (2007). Do you trust your brethren? eliciting trust attitudesand trust behavior in a Tanzanian congregation. Journal of Economic Behavior &Organization, 62(2):255–271.

Davidson, B. (1992). The Black Man’s Burden: Africa and the Curse of the Nation-State. James Currey Publishers, Oxford, UK.

Delhey, J. and Newton, K. (2005). Predicting cross-national levels of social trust: Globalpattern or Nordic exceptionalism? European Sociological Review, 21(4):311.

Deutsch, K. W. (1953). Nationalism and Social Communication: An Inquiry into theFoundations of Nationality. MIT Press, Cambridge, MA.

Easterly, W. and Levine, R. (1997). Africa’s growth tragedy: Policies and ethnic divi-sions. Quarterly Journal of Economics, 112(4):1203–1250.

Eckel, C. and Wilson, R. (2004). Is trust a risky decision? Journal of EconomicBehavior and Organization, 55(4):447–465.

Englebert, P. (2002). State Legitimacy and Development in Africa. Lynne Rienner,Boulder, CO.

Englebert, P., Tarango, S., and Carter, M. (2002). Dismemberment and suffocation.Comparative Political Studies, 35(10):1093–1118.

Ensminger, J. (2000). Experimental economics in the bush: Why institutions matter.In Menard, C., editor, Institutions and Organizations. Edward Elgar, London.

Ensminger, J. (2004). Market integration and fairness: Evidence from ultimatum,dictator, and public goods experiments in East Africa. In Henrich, J., Boyd, R.,Bowles, S., Camerer, C. F., Fehr, E., and Gintis, H., editors, Foundations of HumanSociality: Economic Experiments and Ethnographic Evidence from Fifteen Small-Scale Societies, pages 356–381. Oxford University Press, New York.

Fafchamps, M. (2004). Market Institutions in Sub-Saharan Africa. MIT Press, Cam-bridge, MA.

Fearon, J. D. (2003). Ethnic and cultural diversity by country. Journal of EconomicGrowth, 8(2):195–222.

Ferree, K. and Horowitz, J. (2010). Ties that bind? The rise and decline of ethno-regional partisanship in Malawi, 1994–2009. Democratization, 17(3):534–563.

Fershtman, C. and Gneezy, U. (2001). Discrimination in a segmented society: Anexperimental approach. Quarterly Journal of Economics, 116(1):351–377.

Gaertner, S., Dovidio, J., and Bachman, B. (1996). Revisiting the contact hypothesis:The induction of a common ingroup identity. International Journal of InterculturalRelations, 20:271–290.

Gaertner, S. L. and Dividio, J. F. (2000). Reducing Intergroup Bias: The CommonIngroup Identity Model. Sheridan Books, Ann Arbor, MI.

Geertz, C. (1963). The integrative revolution: Primordial sentiments and civil politicsin the new states. In Old Societies and New States, pages 105–157. Free Press, NewYork.

Gellner, E. (1983). Nations and Nationalism. Cornell University Press, Ithaca, NY.

Greig, F. and Bohnet, I. (2008). Is there reciprocity in a reciprocal-exchange economy?Evidence from a slum in Nairobi, Kenya. Economic Inquiry, 46(1):77–83.

Habyarimana, J., Humphreys, M., Posner, D. N., and Weinstein, J. M. (2009). Coeth-nicity and trust. In Hardin, R., Levi, M., and Cook, K. S., editors, Whom Can WeTrust? How Groups, Networks, and Institutions Make Trust Possible, pages 42–64.Russell Sage Foundation Publications, New York.

Hale, H. E. (2004). Explaining ethnicity. Comparative Political Studies, 37(4):458–485.

Harp, S. L. (1998). Larning to be Loyal: Primary Schooling as Nation Building inAlsace and Lorraine. American University Press, Washington, DC.

Hassin, R., Ferguson, M., Shidlovski, D., and Gross, T. (2007). Subliminal exposureto national flags affects political thought and behavior. Proceedings of the NationalAcademy of Sciences, 104(50):19757.

Henry, K. B., Arrow, H., and Carini, B. (1999). A tripartite model of group identifica-tion: Theory and measurement. Small Group Research, 30:558–581.

Herbst, J. (1989). The creation and maintenance of national boundaries in Africa.International Organization, 43(4):673–692.

Herrmann, R. K., Isernia, P., and Segatti, P. (2009). Attachment to the nation andinternational relations: Dimensions of identity and their relationship to war andpeace. Political Psychology, 30(5):721–754.

Hooghe, M., Reeskens, T., Stolle, D., and Trappers, A. (2009). Ethnic diversity andgeneralized trust in Europe. Comparative Political Studies, 42(2):198.

Horowitz, D. L. (1985). Ethnic Groups in Conflict. University of California Press,Berkeley, CA.

Jackson, R. H. and Rosberg, C. G. (1982). Why Africa’s weak states persist: Theempirical and the juridical in statehood. World Politics, 35:1–24.

Karlan, D. (2005). Using experimental economics to measure social capital and predictfinancial decisions. The American Economic Review, 95(5):1688–1699.

Knack, S. and Keefer, P. (1997). Does social capital have an economic pay-off? A crosscountry investigation. Quarterly Journal of Economics, 112:1251–288.

Kramer, R. and Brewer, M. (1984). Effects of group identity on resource use in a simu-lated commons dilemma. Journal of Personality and Social Psychology, 46(5):1044–1057.

Kymlicka, W. (2001). Politics in the Vernacular. Oxford University Press, Oxford.

Laitin, D. and Posner, D. (2001). The implications of constructivism for constructingethnic fractionalization indices. APSA-CP: Newsletter of the Organized Section inComparative Politics of the American Political Science Association, 12(1):13–17.

Melson, R. and Wolpe, H. (1970). Modernization and the politics of communalism: Atheoretical perspective. American Political Science Review, 64(4):1112–1130.

Michalopoulos, S. and Papaioanno, E. (2011). The long-run effects of the scram-ble for Africa. Working Paper 17620, National Bureau of Economic Research,http://www.nber.org/papers/w17620.

Miguel, E. (2004). Tribe or nation? nation building and public goods in Kenya versusTanzania. World Politics, 56(3):327–362.

Miles, W. F. and Rochefort, D. A. (1991). Nationalism versus ethnic identity in sub-Saharan Africa. American Political Science Review, 85(2):393–403.

Mosley, P. and Verschoor, A. (2005). The development of trust and social capital inrural Uganda: An experimental approach. Sheffield Economic Research Paper Series,2005011.

Nier, J., Gaertner, S., Dovidio, J., Banker, B., Ward, C., and Rust, M. (2001). Changinginterracial evaluations and behavior: The effects of a common group identity. GroupProcesses & Intergroup Relations, 4(4):299.

Petrie, R. (2004). Trusting appearances and reciprocating looks: Experimental evidenceon gender and race preferences. Unpublished manuscript, Georgia State University.

Piper, W., Marrache, M., Lacroix, R., Richardsen, A., and Jones, B. (1983). Cohesionas a basic bond in groups. Human Relations, 36(2):93.

Posner, D. N. (2004a). Measuring ethnic fractionalization in Africa. American Journalof Political Science, 48(4):849–863.

Posner, D. N. (2004b). The political salience of cultural difference: Why Chewas adnTumbukas are allies in Zambia and adversaires in Malawi. American Political ScienceReview, 98(4):529–545.

Posner, D. N. (2006). African borders as sources of natural experiments. UnpublishedManuscript. Unpublished Manuscript, University of California, Los Angeles.

Putnam, R. (2007). E pluribus unum: Diversity and community in the twenty-firstcentury. Scandinavian Political Studies, 30(2):137–174.

Riek, B., Mania, E., Gaertner, S., McDonald, S., and Lamoreaux, M. (2010). Does acommon ingroup identity reduce intergroup threat? Group Processes & IntergroupRelations, 13(4):403.

Robinson, A. L. (2009). National versus ethnic identity in Africa: State, group, andindividual level correlates of national identification. Afrobarometer Working Paper,No.112.

Robinson, A. L. (2013). Ethnic diversity, segregation, and ethnocentric trust in Africa.Unpublished Manuscript.

Sachs, N. (2010). Religion and Nation: Identity Manipulation in Diverse Societies.Stanford University, Dissertation.

Sahlins, P. (1989). Boundaries: The Making of France and Spain in the Pyrenees.University of California Press, Berkeley.

Schechter, L. (2007). Traditional trust measurement and the risk confound: An exper-iment in rural Paraguay. Journal of Economic Behavior & Organization, 62(2):272–292.

Schunn, C. D. (1999). Statistical significance bars (SSB): A way to make graphs moreinterpretable. Unpublished Manuscript, George Mason University.

Shelton, J. and Sellers, R. (2000). Situational stability and variability in African Amer-ican racial identity. Journal of Black Psychology, 26(1):27.

Smith, A. (1983). State and Nation in the Third World: The Western State and AfricanNationalism. Wheatsheaf Books, Brighton.

Smith, A. D. (1986). The Ethnic Origins of Nations. Basil Balckwell, Oxford.

Sniderman, P. M., Hagendoorn, L., and Prior, M. (2004). Predisposing factors and situ-ational triggers: Exclusionary reactions to immigrant minorities. American PoliticalScience Review, 98(1):35–49.

Spinner-Halev, J. and Theiss-Morse, E. (2003). National identity and self-esteem. Per-spectives on Politics, 1:515–532.

Tajfel, H., Flament, C., Billig, M., and Bundy, R. (1971). Social categorization andintergroup behaviour. European Journal of Social Psychology, 1:149–178.

Tajfel, H. and Turner, J. (1979). An integrative theory of intergroup conflict. In Austin,W. and Worchel, S., editors, The Social Psychology of Intergroup Relations, pages33–47. Brooks/Cole, Monterey, CA.

Tanis, M. and Postmes, T. (2005). A social identity approach to trust: Interpersonalperception, group membership and trusting behaviour. European Journal of SocialPsychology, 35(3):413–424.

Tracer, D. (2003). Selfishness and fairness in economic and evolutionary perspec-tive: An experimental economic study in Papua New Guinea. Current Anthropology,44(3):432–438.

Transue, J. E. (2007). Identity salience, identity acceptance, and racial policy attitudes:American national identity as a uniting force. American Journal of Political Science,51(1):78–91.

Uslaner, E. (2008). Where you stand depends upon where your grandparents sat. PublicOpinion Quarterly, 72(4):725–740.

Vail, L. and White, L. (1991). Tribalism in the political history of Malawi. In Vail, L.,editor, The Creation of Tribalism in Southern Africa, pages 151–192. University ofCalifornia Press, Berkeley.

Van Evera, S. (1994). Hypotheses on nationalism and war. International Security,18(4):5–39.

Watkins, S. (1991). From Provinces into Nations. Princeton University Press, Prince-ton, NJ.

Weber, E. (1979). Peasants into Frenchmen : The Modernization of Rural France,1870-1914. Stanford University Press, Stanford, CA.

Whiteley, P. (2000). Economic growth and social capital. Political Studies, 48(3):443–466.

Wilder, D. (1981). Perceiving persons as a group: Categorization and intergroup re-lations. In Hamilton, D., editor, Cognitive Processes in Stereotyping and IntergroupBehavior, pages 213–258. Erlbaum, Hillsdale, NJ.

Young, C. (2004). Revisiting nationalism and ethnicity in Africa. Paper presented atthe James S. Coleman memorial lecture at the University of California, Los Angeles,December 7.

Zahra, T. (2008). The minority problem and national classification in the French andCzechoslovak borderlands. Contemporary European History, 17(2):137–165.

Zak, P. and Knack, S. (2001). Trust and growth. Economic Journal, 111:295—321.

Table 1: National Group Identification Measures.

Class Variable

Affective

A1 We all belong to many different types of groups. Which of thefollowing statements is closest to your view?1. While I am proud of my Malawian identity, there are othergroups that I feel more proud to belong to.2. While I am proud of many of the groups to which Ibelong, I am most proud of my Malawian identity.

A2 Imagine that a story in the international media criticized Malaw-ians. Which of the following statements is closest to how you wouldfeel?1. I would not like it, but it would not feel like a personal insult.2. I would not like it, and I would feel personally insulted.

Behavioral

B1 Which of the following statements is closest to your view?1. How well other Malawians are doing does not really affect howwell I am doing.2. How well I am doing really depends on how well otherMalawians are doing.

B2 Which of the following statements is closest to your view?1. Malawians from different regions of the country cannotmanage without help from Malawians in other regions.2. Malawians from different regions of the country don’t really haveto rely on one another in order to manage.

Cognitive

C1 Which of the following statements is closest to your view?1. I see myself as quite similar to most Malawians.2. I see myself as quite different from most Malawians.

C2 Which of the following statements is closest to your view?1. Even though there is a lot of cultural variety amongMalawians, we are more the same than we are different.2. Because there is a lot of cultural variety among Malawians, thereis very little that makes us the same.

Note: Bolded items coded as 1 (stronger identification).

Table 2: Covariate Balance by Experimental Assignment.

Treatment Group Control Group(Flag Prime) (No Flag)

Age 31.25 31.34(0.63) (0.68)

Male (%) 59.05 53.62(3.40) (3.47)

Years of Education 6.00 5.75(0.19) (0.21)

Chewa (%) 51.40 48.31(3.42) (3.48)

National Identification -0.01 0.01(1.00) (1.00)

Means presented with standard errors in parentheses, N = 421.

Table 3: The impact of shared nationality and shared ethnicity on trust.

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

FEs FEs REs REs

CoNational 0.06 0.04 0.06 0.04(0.01)∗∗∗ (0.02)∗∗ (0.01)∗∗∗ (0.02)∗∗

CoEthnic 0.06 0.04 0.06 0.04(0.02)∗∗∗ (0.02)∗ (0.02)∗∗∗ (0.02)∗

CoNational & CoEthnic 0.04 0.04(0.03) (0.03)

Constant 1.10 1.11 1.10 1.11(0.02)∗∗∗ (0.02)∗∗∗ (0.02)∗∗∗ (0.02)∗∗∗

Round Fixed-Effects Yes Yes Yes Yes

OLS regressions with participant fixed-effects or participant random-effects.

Participant-clustered standard errors in parentheses.

N=1,671 game observations within 421 participants.∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Table 4: The impact of national identification, national identity salience, and theirinteraction on the size of the coethnic trust premium among conationals.

(1) (2) (3)Control Only Full Sample Full Sample

CoEthnic 0.14 0.13 0.14(0.05)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗

NatID (strength of) 0.06 0.06(0.03)∗ (0.03)∗

CoEthnic × NatID −0.07 −0.06(0.04)∗ (0.06)

Flag 0.01 0.02(0.06) (0.06)

CoEthnic × Flag −0.10 −0.11(0.05)∗ (0.06)∗

NatID × Flag −0.06(0.06)

CoEthnic × NatID × Flag 0.09(0.05)∗

Constant 1.03 1.04 1.03(0.05)∗∗∗ (0.05)∗∗∗ (0.05)∗∗∗

Round Fixed-Effects Yes Yes Yes

OLS regressions with participant random effects.

N = 841 games within 421 participants in the full sample.

N = 414 games with 207 participants in the control group.

Participant-clustered standard errors in parentheses.∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01

Figure 1: Schematic representation of the field site, located at the intersection of anethnic and a national boundary.

Figure 2: National Identity Prime: Discussion of the symbolism of the 1964 Malawiannational flag and the 2010 Malawian national flag.

Figure 3: Average Trust in Four Types of Partners, n=421. Error bars are StatisticalSignificance Bars (?).

-.10

.1.2

.3.4

Mar

gina

l Effe

ct o

f Coe

thni

city

on

Trus

t

-2 -1 0 1 2Strength of National Identification

As a Function of Strength of National Identification

Coethnic Trust Premium

Figure 4: The size of the coethnic trust premium as a function of strength of identi-fication with the nation. Bands represent 95% confidence intervals.

Figure 5: Average size of the coethnic trust premium by experimental national iden-tity salience condition, over different strengths of national identification.Averages differences in trust in coethnics relative to non-coethnics arepooled across participants.

* * * * * * * * * * * * * * * * * * * *

0.1

.2.3

Mar

gina

l Effe

ct o

f Coe

thni

city

on

Trus

t

-2 -1 0 1 2Strength of National Identification

Low (No Flag)

High (Flag)

National Identity Salience

* designates statistically significantly treatment effects at the 0.05 level

As a Function of National Identification and National Identity Salience

Coethnic Trust Premium

Figure 6: The marginal effect of coethnicity on trust by salience of the nationalidentity (experimentally manipulated) and pre-existing strength of nationalidentification.

.5

.5

.5.75

.75

.751

1

11.25

1.25

1.251.5

1.5

1.5Average Trust (0-2)

Aver

age

Trus

t (0

-2)

Average Trust (0-2)Low NatID

Low NatID

Low NatIDMed. NatID

Med. NatID

Med. NatIDHigh NatID

High NatID

High NatIDNo Flag

No Flag

No FlagFlag

Flag

FlagNo Flag

No Flag

No FlagFlag

Flag

FlagNo Flag

No Flag

No FlagFlag

Flag

FlagNon-Coethnic

Non-Coethnic

Non-CoethnicCoethnic

Coethnic

CoethnicTrust Game Partner

Trust Game Partner

Trust Game Partner

Figure 7: Average Trust by Partner Coethnicity, National Prime Treatment, andStrength of National Identification.

Figure 8: The bivariate correlation between average national relative to ethnic iden-tification and the average size of the coethnic trust premium for 16 Africancountries.

Figure 9: The average within country relationship between individual-level nationalrelative to ethnic identification and the size of the coethnic trust premiumacross 16 African countries. The relationship was estimated using a countryfixed-effects regression.

Appendices to:

Robinson, Amanda Lea. 2012. “National Identification and Interethnic Trust: Evi-dence from an African Border Region.” Working Paper, Stanford University.

Available at www.stanford.edu/~alrobins/Amanda_Lea_Robinson/Research

Contents

Appendix A Trust Game Instructions 2

Appendix B Summary Statistics 8

Appendix C Robustness 10C.1 Ordered Probit Estimates . . . . . . . . . . . . . . . . . . . . . . . . . 10C.2 Order Set Fixed-Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . 13C.3 National Identity Salience and the Coethnic Trust Premium in Non-

Conationals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16C.4 Alternative Measures of Trust . . . . . . . . . . . . . . . . . . . . . . . 17


Appendices to National Identification and Interethnic Trust 2

Appendix A Trust Game Instructions

The following scripts were developed using the game scripts provided by ? and ?.The general instructions for the entire group were given first in Chichewa and thenrepeated in Chitumbuka. Individual instructions were given to participants in theirnative language.

Instructions to Entire Group

Thank you all for taking the time to come today. Today’s activities may take three tofour hours. Before we begin I want to make some general comments about what weare doing here today and explain the rules that we must follow. We will ask each ofyou a few questions about yourself and your opinions, and then we will be doing someactivities with money. Whatever money you earn during the activities will be yours tokeep and take home. We will be supplying the money. This money was given to us byStanford University to use for research.

Today you will be participating in this activity with people from four different vil-lages, one of them your own. Fifteen residents from four different villages are heretoday. [Research supervisor lists the four participating villages in alphabetical order,and says whether each village is Chewa or Tumbuka, and Malawian or Zambian.] Thismeans that you will be doing this activity with Malawians and Zambians, Chewas andTumbukas.

Before we proceed any further, let me stress something that is very important. Manyof you were invited here without understanding very much about what we are planningto do today. If at any time you find that this is something that you do not wish toparticipate in for any reason, you are of course free to leave whether we have startedthe activity or not.

We will be asking you to do an activity with other individuals in this room today. Ifyou have heard anything about these types of activities, you should try to forget aboutthat because each activity can be completely different. It is important that you listenas carefully as possible, because only people who understand the way the activity workswill actually be able to participate.

We will run through some examples of how the activity works here while we are alltogether. You cannot ask questions or talk while here in the group. This is veryimportant. Please be sure that you obey this rule, because it is possible for one personto spoil the activity for everyone. If one person talks about the activity while sittingin the group, we would not be able to carry out the activity today. Do not worry ifyou do not completely understand the rules as we go through the examples here in thegroup. Each of you will have a chance to ask questions in private to be sure that youunderstand how the activity works.


After we have explained the activity, you will all go outside and wait while we call youin one at a time to participate. We will call you by the number on your ticket, so pleaselisten carefully for your number. While you are outside you can talk about football,the market, or anything else you want other than the activities here today.

The activity we are going to do today is sort of like a game, but because you can earnmoney depending on the decisions you make, it is much more serious than a game justfor fun. In fact, the activity is similar to real decisions you might make in your day today life. For example, imagine that you want to sell a bag of maize. You know that abag of maize is currently selling for 1500 MWK in this market (Chisinga/Chimaliro),but around 3000 MWK in Lilongwe (capital of Malawi). You’ve heard that someoneis taking a load of maize down to Lilongwe for a small fee, but you do not know thisperson personally. If you decide to send your bag of maize to be sold by this strangerin Lilongwe, a number of things might happen. The trader may return the followingweek with your 3000 MWK, he may never bring you the money you are owed, or hemay return some amount of money between 0 and 3000 MWK. So, if you do not trustthis stranger, you should just sell your maize locally, but if you trust that he will returnat least 1500 MWK to you, you should send your bag with him. The activity we aregoing to do today parallels this kind of decision.

Each activity is played by a pair of individuals. Each pair is made up of a Player 1 anda Player 2. Each of you will participate in this activity 8 times, four times as a Player1 and four times as a Player 2.

You will be Player 1 four times. Each time you are Player 1, you will have a differentpartner. In one activity, your partner will be from your own village. In each of theother three times you are Player 1, your partner will be from each of the other threevillages. We will tell you which village your partner is from, but we will not tell youwhich person from that village your partner is. It is important for you to rememberthat each time you play will be with a different person from a different village.

After everyone has been Player 1 four times, you will each have the chance to be Player2 four times. As Player 2, you will also be paired with four different people one fromeach of four different villages represented here today. Again, we will tell you whichvillage your Player 1 partner is from, but we will not tell you which particular personfrom each village you will be paired with. Remember that you will never be pairedwith the same person twice. Thus, you will do this activity 8 times total, always witha different partner from this room.

Heres how the activity works. Each activity will have two players: Player 1 and Player2.

When you are Player 1, we will give you 60 MWK for each activity. Player 1 then hasthe opportunity to send a portion of his 60 MWK to Player 2. He could send 60, or 40,or 20, or nothing. We will triple whatever amount Player 1 decides to give to Player2 and then Player 2 has the option of returning any portion of this tripled amount to


Player 1. Then the activity is over.

After every player has decided how much, if any, to put into the four different envelopesit will be time to be Player 2. Each of you will come into the room one at a time, andyou will see how much each of your Player 1 partners sent to you in your envelope,which will then be tripled. For each of the four envelopes sent to you, you will decidehow much (if any) of the money in the envelope you want to keep and how much (ifany) you want to leave in the envelope to be returned to the person who placed themoney there for you.

After every player has decided what to do with the money in the envelopes given tothem, we will call you one at a time one last time to open up your original envelopesand see how much, if any, money was returned to you by Player 2. Thus for each timeas Player 1, the player will go home with whatever he kept from his original 60 MWK,plus anything returned to him by Player 2. For each time as Player 2, he goes homewith whatever was given to him by Player 1 and then tripled by the research team,minus whatever he returned to Player 1. At the very end, you will go home with theamount you earned from four times as Player 1 and four times as Player 2.

Here are some examples.

[Research assistants acted out the following examples. When each hypothetical Player1 makes their choice, money was put in an envelope. Then, research assistants visu-ally showed Player 2 opening the envelope and visually showed the effect of tripling themoney. Then, Player 2 made his decision by putting the returned amount in an en-velope. The first two examples were acted out, and then the entire above instructionswere given in Chitumbuka, followed by the last two examples.]

1. Imagine that Player 1 gives 60 MWK to Player 2. We triple this amount, soPlayer 2 gets 180 MWK (three times 60 equals 180). At this point, Player 1 hasnothing and Player 2 has 180 MWK. Then Player 2 has to decide whether hewishes to give anything back to Player 1, and if so, how much. Suppose Player 2decides to return 120 MWK to Player 1. At the end of the activity Player 1 willgo home with 120 MWK and Player 2 will go home with 60 MWK.

2. Imagine that Player 1 does not send anything to Player 2. There is nothing forus to triple. Player 2 gets 0 MWK and so cannot return anything. At the end ofthe activity Player 1 will go home with 60 MWK and Player 2 will go home withnothing.

3. Imagine that Player 1 gives 40 MWK to Player 2. We triple this amount, soPlayer 2 gets 120 MWK (three times 40 equals 120). At this point, Player 1 has20 MWK and Player 2 has 120 MWK. Then Player 2 has to decide whether hewishes to give anything back to Player 1, and if so, how much. Suppose Player 2decides not to return any money to Player 1. At the end of the activity Player 1will go home with 20 MWK and Player 2 will go home with 120 MWK.


4. Imagine that Player 1 gives 20 MWK to Player 2. We triple this amount, soPlayer 2 gets 60 MWK (three times 20 equals 60). At this point, Player 1 has 40MWK and Player 2 has 60 MWK. Then Player 2 has to decide whether he wishesto give anything back to Player 1, and if so, how much. Suppose Player 2 decidesto return 40 MWK to Player 1. At the end of the activity Player 1 will go homewith 80 MWK and Player 2 will go home with 20 MWK.

Note that the larger the amount that Player 1 gives to Player 2, the greater the amountthat can be earned by the two players combined. However, it is entirely up to Player2 to decide what he should give back to Player 1. The first player could end up withmore than 60 MWK or less than 60 MWK as a result.

We will go through more examples with each of you individually when you come inone at a time. In the meantime, do not talk to anyone about the activity. Even if youare not sure that you understand the activity, do not talk to anyone about it. This isimportant. If you talk to anyone about the activity while you are waiting to play, wemust disqualify you from participating.

Now we will call in each person one by one to decide whether or not to send any moneyto each of your four different partners, and if so, how much. After all of you have playedas Player 1, then each of you will come in a second time to play as Player 2.

One-on-one Instructions to Player 1 Participant

[Participant is called in using his or her Game Identification Number, which he or shehas on piece of paper received upon registration. Informed consent was obtained, andthen the Market Survey was administered. After the market survey was completed, theresearch assistant gave the following instructions in the language of the participant:]

You will now do the activity as Player 1 four times. For each activity, you will have adifferent Player 2 partner. In one activity, your Player 2 partner will be from your ownvillage. In each of the other three times you are Player 1, your Player 2 partner will befrom each of the other three villages. I will tell you which village your Player 2 partneris from, but we will not tell you which person from that village is your partner. It isimportant for you to remember that each time you play will be with a different personfrom a different village.

For each activity, I will give you 60 MWK and you can decide how much to send ofthat 60 MWK to Player 2. You can send 60, or 40, or 20, or nothing. We will triplewhatever amount you decide to give to Player 2 before it is passed on to Player 2.Player 2 then has the option of returning any portion of this tripled amount to you.Then the activity is over.

Let me go through some examples:


[As research assistants went through these examples, the used real money and moved themoney around to illustrate how much each hypothetical player possessed at each stageof the game.]

1. Imagine that Player 1 gives 40 MWK to Player 2. We triple this amount, soPlayer 2 gets 120 MWK (three times 40 equals 120). At this point, Player 1 has20 MWK and Player 2 has 120 MWK. Then Player 2 has to decide whether hewishes to give anything back to Player 1, and if so, how much. Suppose Player 2decides to return 40 MWK to Player 1. At the end of the activity Player 1 willgo home with 60 MWK and Player 2 will go home with 80 MWK.

2. Imagine that Player 1 gives 20 MWK to Player 2. We triple this amount, soPlayer 2 is sent 60 MWK (three times 20 equals 60). At this point, Player 1 has40 MWK and Player 2 has 60 MWK. Then Player 2 has to decide whether hewishes to give anything back to Player 1, and if so, how much. Suppose Player 2decides to return 20 MWK to Player 1. At the end of the activity Player 1 willgo home with 60 MWK and Player 2 will go home with 40 MWK.

Now I want to ask you some questions, to see if you understand the game:

1. Imagine that you give all 60 MWK to Player 2. How much will you have left? (0MWK) How much will Player 2 receive? (180 MWK, three times 60 equals 180)If Player 2 returns 120 MWK to you, how much will you have total? (120 MWK)How much will Player 2 take home? (60 MWK)

2. Imagine that you give 20 MWK to Player 2. How much will you have left? (40MWK) How much will Player 2 receive? (60 MWK, three times 20 equals 60) IfPlayer 2 returns nothing to you, how much will you have total? (40 MWK) Howmuch will Player 2 take home? (60 MWK)

[If the participant answered these questions correctly, the research assistant proceeded toadminister the game. If the participant did not understand, the research assistant wouldexplain the game again until the participant was able to correctly answer the questionsabove.]

Now you will play as Player 1 four times, with four different individuals here today.

First, you will play with a person from Village.

Here are your 60 MWK.

[At this point 60 MWK were handed to the participant.]

Here is the envelope. Whatever you want to send to this person from Village,you should put in the envelope. Whatever you want to keep, you put in your pocket.

You can send them nothing, 20, 40, or 60 MWK, it’s up to you. Player 2 will receivethis amount tripled by me. Remember the more you give to Player 2 the greater theamount of money at his or her disposal. While Player 2 is under no obligation to give


anything back, we will pass on to you whatever he or she decides to return. I amgoing to turn my back, and I want you to put whatever money you want to send in theenvelope

[This was repeated for each of the three other partners from each of three other villages.]

One-on-one Instructions to Player 2 Participant

Now you are playing as Player 2 four times, with four different individuals here today.Four different individuals from four different villages decided how much money to sendto you. Remember, these are not the same individuals you were paired with when youwere Player 1.

We will open each of your four envelopes. For each envelope, you will decide how muchof the money in the envelope to return to the person that sent that money to you.Remember you can return nothing or keep nothing or anything in between.

Here is the envelope that was sent to you by a person from Village . Howmuch is in it? As you remember, I will now triple the amount sent to you. I will add

MWK to make it MWK total.

[Research assistant recorded on the envelope how much money it contained. He thenadded two times that amount in order to produce a total equal to triple the amount sentby Player 1.]

I will turn my back while you decide how much of that money you want to keep foryourself, and how much would you like to return to person from Village that sent youthat money. Whatever you want to send back, put in the envelope, and put the rest inyour pocket.

[This is repeated for each of the three other partners from each of three other villages.]

One-on-one Instructions to Player 1 Participant in Final Round

Now we are going to open each of the four envelopes you sent to other players whenyou were Player 1. We are going to see how much they returned of the money that yousent them, and then we tripled.

Here is the first envelope. You sent MWK to your partner from Village.We tripled that amount and made it MWK. How much did your partner re-turn to you?

[Research assistant recorded the amount returned on the envelope and the participantput the money in his or her pocket. This was repeated for all four envelopes.]


Appendix B Summary Statistics

This appendix provides summary statistics of the sample population and comparesthem to previous studies.

Table B.1 provides summary statistics for the primary variables used in the main analy-ses, including strength of national identification, experimental exposure to the nationalsalience prime, and behavioral trust.

Table B.2 provides additional summary statistics on trust behavior for the full studysample and for trust in partners of the same nationality and same ethnicity (SNSE)only. It also provides comparable statistics, when available, from other studies usingthe trust game in sub-Saharan Africa.

Table B.1: Descriptive statistics of key variables.

Mean Median Minimum Maximum N(SD)

National Identification 0.00 0.52 -3.50 1.33 421(1.00)

Exposure to Flag 0.51 1 0 1 421(0.50)

Trust 1.06 1 0 2 1671(0.41)


Tab

leB

.2:

Des

crip

tive

Sta

tist

ics

from

Tru

stG

ames

Acr

oss

Afr

ica

Fu

llS

am-

ple

SN

SE

On

lyC

ountr

yM

alaw

iM

alaw

iK

enya

Zim

babw

eG

han

aU

gan

da

S.

Afr

ica

Tan

zan

iaK

enya

Ken

yaS

tud

yP

opu

lati

on16

Vil

lage

s16

Vil

lage

s5

Vil

lages

24

Vil

lages

22

Fir

ms

2V

illa

ges

Stu

den

tsC

hu

rch

2V

illa

ges

Nair

ob

iT

rust

Gam

eP

airs

1671

421

20

141

212

67

64

63

25*

134

Dis

cret

eC

hoi

ces

44

11

55

511

11

11

6A

vg.

Entr

ust

ed0.

510.

560.4

40.4

30.4

50.4

90.4

50.5

60.3

50.3

0%

Tru

stin

gN

oth

ing

0.12

0.08

0.0

00.0

90.1

50.0

70.1

10.0

00.0

40.1

3%

Tru

stin

gA

ll0.

190.

230.0

00.0

60.1

1–

0.1

60.2

20.0

4–

Avg.

Ret

urn

ed0.

790.

831.5

41.2

81.4

90.9

9–

1.4

60.9

80.8

2N

ote:

*On

lysu

bje

cts

inC

ron

k’s

(200

7)u

nfr

amed

Tru

stG

am

eare

incl

ud

edh

ere.

Sou

rces

:?,?,?,?,

?,?,?,?.


Appendix C Robustness

This appendix reports results from four tests of robustness. The first set of analysesdeals with model specification, reporting results from ordered probit models. Thesecond set of analyses include order set fixed-effects instead of the round fixed-effectsreported in the main results. Third set of analyses show that the national primingeffect is specific to conationals. The fourth set of robustness tests deal with alternativemeasures of trust.

C.1 Ordered Probit Estimates

Table C.1 reports the results of an ordered probit model of the impact of shared na-tionality and shared ethnicity on trust behavior. These models include participantrandom-effects and round fixed-effects. Similar the main results, these models showthat participants condition their trust in roughly equal measure on both shared nation-ality and shared ethnicity. There is no trust advantage to sharing both identities aboveand beyond the additive effect of sharing each identity.

Table C.2 reports the results of models estimating the impact of national identificationstrength, national identity salience, and their interaction on the size of the coethnic trustpremium among conationals. The results are qualitatively similar to the OLS resultsreported in the main body of the paper: there is a positive coethnic bias, but thisbias is decreased by the presence of a national prime, especially among weak nationalidentifiers.


Table C.1: Ordered Probit Replication: The impact of shared nationality and sharedethnicity on trust game behavior.

(1) (2)Participant Participant

REs REs

CoNational 0.28 0.24(0.07)∗∗∗ (0.09)∗∗∗

CoEthnic 0.23 0.20(0.07)∗∗∗ (0.08)∗∗

CoNat & CoEth 0.07(0.13)

Cutpoint 1 −1.40 −1.42Cutpoint 2 1.52 1.50Log-likelihood −1231.23 −1231.10

Round Fixed-Effects Yes Yes

Ordered probit regression with participant random-effects.

Participant-clustered standard errors in parentheses.

N=1,71 game observations within 421 participants.∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01


Table C.2: Ordered Probit Replication: The impact of national identification, na-tional identity salience, and their interaction on the size of the coethnictrust premium among conationals.

(1) (2) (3)

CoEthnic 0.28 0.45 0.46(0.10)∗∗∗ (0.15)∗∗∗ (0.15)∗∗∗

NatID 0.12 0.19(0.09) (0.13)

CoEth × NatID −0.07 −0.19(0.10) (0.13)

Flag 0.05 0.07(0.19) (0.19)

CoEth × Flag −0.36 −0.38(0.21)∗ (0.21)∗

NatID × Flag −0.13(0.18)

CoEth × NatID × Flag 0.21(0.20)

Cutpoint 1 −1.81 −1.74 −1.74Cutpoint 2 1.38 1.49 1.51Log-likelihood −643.16 −640.25 −638.67


Ordered probit regressions with participant random-effects.

N=841 game observations within 421 participants.

Individual-clustered standard errors in parentheses.∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01


C.2 Order Set Fixed-Effects

The main analyses included game round fixed-effects – whether a particular game wasplayed first, second, third, or fourth in the series of four trust games – because par-ticipants tended to trust more in the first game played than in the subsequent threegames. However, such round fixed-effects do not account for the order in which thosegames were played. For logistical reasons, the order of game partners was not fullyrandomized: instead, participants were randomly assigned to one of four game ordersets:

Order 1: SNSE, SNDE, DNSE, DNDE26

Order 2: SNDE, DNSE, DNDE, SNSE

Order 3: DNSE, DNDE, SNSE, SNDE

Order 4: DNDE, SNSE, SNDE, and DNSE

While there is no impact of game order set on average trust (Kruskal-Wallis Test, χ2 =1.982, p = 0.58, DF = 3), it is possible that the conditional trust results reported inthe main body are driven by a particular order set. Thus, Table C.3 replicates the mainanalyses with order set fixed-effects, which identifies average effects within order sets. Inthe full specification (Model 3), there is a positive coethnic trust premium for the controlcondition at average levels of national identification. When the national identity is notmade salient (control condition), the size of the coethnic trust premium is decreasingas strength of identification with the nation is increasing. Exposure to the flag primereduces the coethnic trust premium, but the effect is not statistically significant forparticipants with average levels of national identification. However, as Figure C.1 shows,there is a positive and statistically significant coethnic trust premium under the lownational salience condition at all levels of national identification, while there is nostatistically significant coethnic trust premium under the treatment condition.

26SNSE = same nationality and same ethnicityDNSE = same nationality and different ethnicityDNSE = different nationality and same ethnicityDNDE = different nationality and different ethnicity


Table C.3: Order Set Fixed-Effects Replication: The impact of national identification,national identity salience, and their interaction on the size of the coethnictrust premium among conationals.

(1) (2) (3)

CoEthnic 0.09 0.12 0.12(0.03)∗∗∗ (0.04)∗∗∗ (0.04)∗∗∗

NatID 0.03 0.06(0.03) (0.03)∗

CoEth × NatID −0.02 −0.06(0.03) (0.04)∗

Flag −0.01 −0.01(0.06) (0.06)

CoEth × Flag −0.06 −0.05(0.06) (0.06)

NatID × Flag −0.05(0.06)

CoEth × NatID × Flag 0.08(0.05)

Constant 1.03 1.03 1.03(0.05)∗∗∗ (0.05)∗∗∗ (0.05)∗∗∗

Order Fixed-Effects Yes Yes Yes

Ordered probit regressions with participant random-effects. N=841



Figure C.1: Order Set Fixed-Effects Replication: The marginal effect of coethnicityon trust by national identity salience experimental condition and strengthof national identification.


C.3 National Identity Salience and the Coethnic Trust Pre-mium in Non-Conationals

The main results show that exposure to a national prime reduces the coethnic trustpremium among conationals. This finding is consistent with my argument that thisreduction is due to increased national identification. However, it is possible that theflag prime simply reduces the coethnic trust premium in general, in which case trust incoethnics across the border in Zambia should also be reduced. To rule out this possi-bility, Table C.4 reports the the marginal effect of coethnicity on trust in Zambians asa function of exposure to the Malawian national flag. The coefficient on the interactionbetween coethnicity and treatment status is not statistically significant and is actuallyin the opposite direction. This suggests that the reduction in the coethnic trust pre-mium among Malawians after exposure to the Malawian national flag operates throughtemporarily increased national identification.

Table C.4: National Identity Salience and the Marginal Effect of Coethnicity on Trustin Non-conationals.

(1)

CoEthnic 0.01(0.03)

Flag −0.02(0.06)

CoEth × Flag 0.08(0.07)

Constant 0.98(0.06)∗∗∗

Round Fixed-Effects Yes


N=830 games within 421 participants.

Participant-clustered standard errors in parentheses∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01


C.4 Alternative Measures of Trust

The measure of trust used in the main analyses maps the amount of money sent in thefirst stage of the trust game onto three values of the dependent variables, correspondingto no trust (Trust = 0), moderate trust (Trust = 1), and high trust (Trust = 2).Other potential operationalizations of trust behavior include a dichotomous indicatorof trusting anything, a dichotomous indicator of trusting everything, and the proportionof the original endowment entrusted to one’s partner. The main analyses are repeatedusing these three alternative dependent variables, and those results are reported inTables C.5 and C.6.

Table C.5 shows the impact of shared nationality and shared ethnicity on the threedifferent measures of trust. The impact of shared nationality is positive and statisticallysignificant for all three measure of trust, while shared ethnicity is statistically significantonly for trusting anything and the proportion trusted.

Table C.5: Alternative Measures of Trust: The impact of shared nationality andshared ethnicity on trust game behavior.

(1) (2) (4)Trust Trust Proportion

Anything Everything Trusted

CoNational 0.04 0.04 0.05(0.01)∗∗∗ (0.01)∗∗ (0.01)∗∗∗

CoEthnic 0.05 0.02 0.05(0.02)∗∗∗ (0.02) (0.01)∗∗∗

Constant 0.83 0.16 0.46(0.01)∗∗∗ (0.01)∗∗∗ (0.01)∗∗∗

Order Fixed-Effects Yes Yes Yes

OLS regression with participant random-effects. N=841


Table C.6 presents the results of impact of strength of national identification, nationalidentity salience, and their interaction on trust. In this fuller specification, there is astatistically significant coethnic trust premium for all three measures of trust for aver-age levels of national identification under low national identity salience. That coethnictrust premium is generally decreasing with stronger national identification under lownational identity salience, but this effect is statistically significant only for trustinganything. The impact of national identity salience on the size of the coethnic trustpremium is consistently negative, but is not statistically significant at average levelsof national identification for any measure of trust. To get a better sense of the im-


pact of the flag prime at different levels of national identification, Figures C.2, C.3,and C.4 graph the marginal effect of coethnicity by strength of national identificationand national identity salience. For the two dichotomous measures of trust – trustinganything and trusting everything – the coethnic trust premium varies by national iden-tification and national identity salience similar to the main results. In particular, thereis generally a positive coethnic trust premium under conditions of low national identitysalience, but decreasing with strength of national identification. In contrast, there isno coethnic trust premium when the national identity is made contextually salient. Forthe proportion trusted to one’s partner, though, the pattern looks slightly different. Inparticular, the size of the coethnic trust premium is actually increasing with strengthof national identification when the national identity is made salient, and for very strongnational identifiers, there is a statistically significant coethnic trust premium even un-der conditions of high national identity salience. Thus, for the proportion trusted, thereis only a negative treatment effect on the size of the coethnic trust premium for weaknational identifiers.


Table C.6: Alternative Measures of Trust: The impact of national identification, na-tional identity salience, and their interaction on the size of the coethnictrust premium among conationals.

(1) (2) (3)Trust Trust Proportion

Anything Everything Trusted

CoEthnic 0.06 0.07 0.09(0.03)∗∗ (0.03)∗∗ (0.02)∗∗∗

NatID 0.03 0.03 0.02(0.02) (0.02) (0.02)

CoEth × NatID −0.04 −0.02 −0.02(0.02)∗∗ (0.03) (0.02)

Flag −0.02 0.04 0.02(0.03) (0.04) (0.03)

CoEth × Flag −0.04 −0.06 −0.05(0.05) (0.05) (0.03)

NatID × Flag −0.01 −0.05 −0.05(0.03) 0.04 (0.03)

CoEth × NatID × Flag 0.06 0.03 0.06(0.03)∗ (0.04) (0.03)∗∗

Constant 0.89 0.14 0.47(0.03)∗∗∗ (0.04)∗∗∗ (0.03)∗∗∗



N=841 games within 421 participants.

Participant-clustered standard errors in parentheses.∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01


Figure C.2: Trust Anything: The marginal effect of coethnicity on trust by nationalidentity salience experimental condition and strength of national identi-fication.


Figure C.3: Trust Everything: The marginal effect of coethnicity on trust by na-tional identity salience experimental condition and strength of nationalidentification.


Figure C.4: Proportion Trusted: The marginal effect of coethnicity on trust by na-tional identity salience experimental condition and strength of nationalidentification.

Nationalism and Interethnic Trust: Evidence from an...

Documents

Transcript of Nationalism and Interethnic Trust: Evidence from an...