ethnic diversity and economic performance

Journal of Economic Literature Vol. XLIII (September 2005), pp. 762800

Ethnic Diversity and Economic Performance

ALBERTO ALESINA and ELIANA LA FERRARA

We survey and assess the literature on the positive and negative effects of ethnicdiversity on economic policies and outcomes. Our focus is on communities of differ-ent size and organizational structure, such as countries, cities in developed countries,and villages and groups in developing countries. We also consider the endogenousformation of political jurisdictions and highlight several open issues in need of fur-ther research, in particular the endogenous formation of ethnic identity and themeasurement of ethnic diversity.

762

Alesina: Harvard University. La Ferrara: BocconiUniversity . We thank Roger Gordon, David Laitin, JohnMcMillan, and two anonymous referees for very usefulcomments. Angelo Mele provided excellent researchassistance. Alesina is grateful to the NSF for financialsupport through a grant to the NBER. This paper is partof the Polarization and Conflict Project CIT2CT2004506084 funded by the European CommissionDGResearch Sixth Framework Programme. This articlereflects only the authors views and the Community is notliable for any use that may be made of the informationcontained therein.

1. Introduction

New York and Los Angeles are amongthe two most troubled American citiesin terms of racial relations; at the same timethey are constant producers of innovation inthe arts and business. The United Statesitself is an economically successful meltingpot, but many of its social problems arerelated to racial and ethnic cleavages. Thetragedy of Africa is, according to many,largely a result of ethnic conflict, which isindeed pervasive in many parts of thedeveloping world. So, what are the pros andcons of diversity, being that racial, ethnic,religious, or linguistic?

The potential costs of diversity are fairlyevident. Conflict of preferences, racism, andprejudices often lead to policies that are atthe same time odious and counterproductivefor society as a whole. The oppression ofminorities may lead to political unrest oreven civil wars. But a diverse ethnic mix alsobrings about variety in abilities, experiences,and cultures that may be productive andmay lead to innovation and creativity. Inwhat follows, we try to highlight the trade-off between the benefits of diversity andthe costs of heterogeneity of preferences in adiverse multiethnic society.

In order to bring more evidence to bearon this question, we plan to examine jointlytwo strands of the literature that have pro-ceeded in a parallel way: one on cross-coun-try comparisons and one on localcommunities. The latter is itself split intotwo subareas with little communicationbetween the two, namely the public andurban economics literature on U.S. citieson the one hand, and the development lit-erature that focuses on groups and local

sp05_Article 3 8/18/05 9:46 AM Page 762

Alesina and La Ferrara: Ethnic Diversity and Economic Performance 763

communities on the other. Within bothstrands of the literature, one approach takesthe size and number of jurisdictions (coun-tries or localities) as given and studies theeffects of different degrees of diversity onquality of government, economic policies,growth, unrest, crime, civil wars, etc. A sec-ond and less developed approach focuses onthe fact that the number, size, and shape ofjurisdictions (countries or localities) is itselfendogenous, namely determined by theinteraction of individuals and groups.

So, is diversity good or bad?Fragmented societies are often more proneto poor policy management and pose morepolitico-economic challenges than homoge-nous ones; it is easy to find rather volumi-nous evidence on this point. However, to theextent that not all diverse societies are a fail-ure but in fact some work much better thanothers, and in fact rather well, it is importantto understand why and how. We propose asimple theoretical framework in which theskills of individuals from different ethnicgroups are complementary in the produc-tion process for a private good, implying thatmore diversity translates into increased pro-ductivity. On the other hand, individual util-ity also depends on the consumption of ashared public good and, since different eth-nic groups may have different preferenceson the type of public good to provide,increased diversity lowers the utility frompublic good consumption. The size of thepublic sector and the number of ethnicgroups are determined by the trade-offbetween these two forces. We verify theconsistency of this theory using repeatedcross-sectional data on countries and locali-ties in the United States and we find that,while ceteris paribus increases in ethnicdiversity are associated with lower growthrates, the interaction between diversity andthe income level of the community understudy is positive. This suggests that ethnicdiversity can be beneficial (or at least lessdetrimental) at higher levels of develop-ment. One potential explanation for this

1 For a series of case studies on work teams supportingthe notion that diverse groups function better when inter-group dynamics and allocation rules have been targeted tominimize the conflict between minority and majoritymembers, see Susan Jackson and Marian Ruderman(1996) and Thomas Kochan, Katerina Bezrukova, RobinEly, Jackson, Aparna Joshi, Karen Jehn, Jonathan Leonard,David Levine, and David Thomas (2002). For a moremacroeconomic survey on the interplay between ethnicityand institutions in the rational choice literature, seeAndreas Kyriacou (2005).

effect is that the productivity benefits of skillcomplementarities are realized only whenthe production process is sufficiently diversi-fied, as in advanced economies. Anotherpossibly complementaryexplanation isthat richer societies have developed institu-tional features that allow them to bettercope with the conflict element intrinsic indiversity and isolate or moderate its negativeeffects. From the micro to the macro level,in fact, the importance of adequate rules ofthe game to manage diversity is stressed byall disciplines.1

Before going any further, we need to clar-ify three points. First, we are aware thatAmerican cities are very different fromAfrican villages, but we believe that high-lighting similarities and differences in thefindings may shed some light on the ques-tion at hand, for instance how different lev-els of development and different types ofracial, linguistic, or religious conflict play outin the political economy of various parts ofthe world. Also, comparing microeconomiceffects at the team level with macroeconom-ic effects at the country or city level is usefulto get an idea on the mechanisms underlyingthe economic effects of diversity.

Second we need to clarify what we, andthe literature which we review, mean by var-ious terms like diversity, fractionalization,ethnicity, race, etc. The empirical literatureon cross-country studies has typically usedvarious measure of ethno-linguistic fraction-alization. An ethno-linguistic group (oftenreferred for brevity as ethnic group) isidentified by a language only in some casesand in other cases by language and skin coloror other physically attributes; a variety of


764 Journal of Economic Literature, Vol. XLIII (September 2005)

indexes have been suggested and we will dis-cuss below similarity and differences. In thecontext of the literature on American cities,racial groups are identified with the Censusdefinition of race, based on five categories:(1) white, (2) black, (3) American Indian,Eskimo, Aleutian, (4) Asian, Pacific Islander,and (5) other (including Hispanic). Somestudies also look at ancestry or ethnic ori-gin, most often defined in this context as thecountry of birth of the American individual(for in stance, Western European, EasternEuropean, Indian, etc.). In the developmentliterature on village communities, diversity ismeasured with reference to language/ethnicgroup and more seldom through member-ship in different clans or tribes. We will usethe terms fractionalization and diversitywhen we want to be generic and not refer toany particular type of identifying character-istics of the groups; we will use ethnic, racial,religious fragmentation, and diversity whenwe want to be more specific. With the termdiverse society (or city or country) wemean a nonhomogenous place. The termfractionalization, on the other hand, will bedirectly related to a specific measure ofnumber and size of groups: specifically, amore fractionalized place is one in which theprobability the two randomly drawn individ-uals belong to the same group is lower. Insurveying the existing literature, we do nottouch on the question of what identifies anethnic group and we take the classificationsadopted by the authors as given. We brieflydiscuss the issue of how ethnic identity isdefined in section 5, although we are awarethat this topic would deserve a much more indepth treatment (which however goesbeyond the scope of the present article).

Finally, and this is the third point, whenreviewing a broad strand of the literature atthe intersection of economics, political sci-ence, sociology, and history, it is important toset some boundaries. We limit ourselves todirect economic effects of diversity; weleave aside indirect effects that may gothrough civil wars, crime, revolutions, etc.

To put it differently, we do not review in anydetail the literature linking fractionalizationto civil wars or the literature on the effects ofwars (civil or otherwise) on the economy.However boundaries are never neatlydefined, and judgment calls are often neces-sary about which paper or issues fall onwhich side of the boundary. We tried ourbest, but we readily plead guilty of makingjudgment calls that may reflect our unevenknowledge of the literature.

We proceed in the following way. In sec-tion 2, we discuss the theoretical underpin-nings of the relationship between ethnicdiversity and economic performance. Wealso sketch a simple model, which has nopretence of being innovative but illustratesclearly the pros and cons of ethnic frag-mentation and sets the stage for the discus-sion of the literature (mostly empirical) thatfollows. Section 3 discusses the effects ofethnic and racial fragmentation in varioustypes of communities holding the numberand size of communities as exogenous. Weexamine evidence collected on three typesof communities: social groups, localities,and nations. Section 4 discusses the ques-tion of endogenous formation of groups,localities, and nations. Section 5 concludesby discussing several open questions in thisarea of research. The last section drawssome tentative conclusions and policyimplications.

2. Theories on Diversity

The goal of this section is to briefly high-light some basic economic forces underlyingthe relationship between ethnic diversityand economic performance. We begin bytrying to offer a coherent picture of themicrofoundations for this relationship. Wethen move to an analysis of the impact ofdiversity on public policies and productivitythrough a simple model.

2.1 Some Microfoundations

First diversity can affect economic choicesby directly entering individual preferences.



2 A business counterpart of the preference element indiversity may be seen in the theories of customer dis-crimination. According to these theories, businesseswhose employees reflect the ethnic mix of the communi-ties in which they operate perform better than those whodo not, as customer satisfaction increases from interactingwith service providers similar to themselves.

Early work on social identity theory hasestablished that patterns of intergroupbehavior can be understood considering thatindividuals may attribute positive utility tothe well being of members of their owngroup and negative utility to that of mem-bers of other groups (see e.g., Henri Tajfel,Michael Billig, Robert Bundy, and ClaudeFlament 1971). A recent formalization ofthis concept is the analysis of group partici-pation by Alesina and La Ferrara (2000),where the population is heterogeneous andindividual utility from joining a groupdepends positively on the share of groupmembers of ones own type and negativelyon the share of different types.2

Second, diversity can affect economic out-comes by influencing the strategies of indi-viduals. Even when individuals have no tastefor or against homogeneity, it may be opti-mal from an efficiency point of view to trans-act preferentially with members of ones owntype if there are market imperfections. Forexample, Avner Greif (1993) argues thattraders in Medieval times formed coalitionsalong ethnic lines in order to monitor agentsby exchanging information on their oppor-tunistic behavior. Ethnic affiliation helpedsustain a reputation mechanism in the pres-ence of asymmetric information. But strate-gies can be conditional on ones ethnicidentity also in the presence of perfect infor-mation. La Ferrara (2003a) shows that, whencontracts cannot be legally enforced (andtherefore have to be self-enforcing), mem-bership in ethnic groups allows an enlarge-ment of the set of cooperative strategies thatcan be supported. The reason is that bothpunishment and reciprocity can be directednot only at the individual but to other mem-bers of his/her group. A similar reasoning is

3 Note that all these studies indicate a role for ethnicdiversity in facilitating the enforcement of informal con-tracts, but this role becomes less and less important asmore transactions are mediated through the market, i.e.,as the level of economic development of the societyincreases. We return to this issue below.

proposed by James Fearon and David Laitin(1996) to explain interethnic cooperation.Using a social matching model, they showthat cooperation amongst different ethnicgroups can arise through either of two chan-nels. The first, which they label spiralequilibrium, occurs when conflict betweenindividuals is expected to spiral to the wholegroup and the fear of this induces coopera-tion on the equilibrium path. The second,labeled group-policing equilibrium, is onein which deviations by members of otherethnic groups are ignored and each groupsanctions deviations by its own members.Finally, an interesting application of thestrategic role of diversity concerns theincentives to innovate through individual ini-tiative. Two recent studies shed light on thispoint. Eli Berman (2000) uses a club goodmodel with social interactions to argue thatsmall communities can ensure the loyalty oftheir members by taxing activities outsidethe club (e.g., innovations). Tanguy Bernard,Alain de Janvry, and Elisabeth Sadoulet(2004) also study a context in which localcommunities try to restrain innovations bysubgroups, but once enough diversity existswithin a local community, differentiatingorganizations may actually emerge.3

Finally, diversity may enter the productionfunction. People differ in their productiveskills and, more fundamentally, in the waythey interpret problems and use their cogni-tive abilities to solve them. This can be con-sidered the origin of the relationshipbetween individual heterogeneity and inno-vation or productivity. An elegant formaliza-tion of this concept is provided by Lu Hongand Scott Page (1998), who prove two keyresults on this point. First, a group of cogni-tively diverse problem solvers can find opti-mal solutions to difficult problems; second,



under certain conditions, a more diversegroup of people with limited abilities canoutperform a more homogeneous group ofhigh-ability problem solvers. The intuition isthat an individuals likelihood of improvingdecisions depends more on her having a dif-ferent perspective from other group mem-bers than on her own high expected score. Ata more applied level, Alesina, EnricoSpolaore, and Romain Wacziarg (2000)employ a DixitStiglitz production structurewhere more variety of intermediate inputs,that can be interpreted as more variety ofindividual skills, increases total output.Andrea Prat (2002) raises related points inthe context of team theory. In teams wherejobs are complementary, homogeneity haspositive effects and the other way around.Gianmarco Ottaviano and Giovanni Peri(2003) also investigate the pros of diversity inproduction. Diversity and related amenitiesaffect the value of land, which enters theproduction function. These models, howev-er, do not identify a trade-off in the produc-tion function since more heterogeneity isalways better than less. The costs of hetero-geneity are outside the production function.Edward Lazear (1999a, 1999b) also discuss-es how different skills in a production unitmay increase overall productivity. He identi-fies a trade-off between the productive ben-efits of diversity and the possible costs thatmay arise due to difficult communicationbetween people with different languages,culture, etc. Purely from the point of view ofmaximization of productivity, there is anoptimal degree of heterogeneity that is iden-tified by the optimal point of this trade-offgiven also the nature of the production unitand its technology. This trade-off alsoemerges from a number of recent studies onorganization performance, surveyed amongothers by Jackson and Ruderman (1996),Katherine Williams and Charles OReilly(1998), and Orlando Richard, Kochan, andAmy McMillan-Capehart (2002). The major-ity of these studies rely on laboratory experi-ments to test the link between diversity and

performance, and generally find a positiveeffect of racial and gender diversity on cre-ativity and task completion. For example,OReilly, Williams, and Sigal Barsade (1997)analyze thirty-two project teams and findthat more diversity leads to more conflictand less communication, but controlling forthe latter it also leads to higher productivity.The few existing studies carried on realorganizations offer a more complex picture.Summarizing the findings of their recentresearch projects, Kochan et al. (2002) arguethat no significant direct relationshipbetween diversity and team performanceemerges unless the specific organizationalcontext and policies are accounted for. Theimportance of the organizational context tominimize conflict within the group is alsohighlighted in a model by MichelleGarfinkel (2004). She considers a setting inwhich individuals compete in a winner-take-all market. The formation of groups reducesconflict over the resource compared to thecase in which all individuals compete witheach other, but it also adds another dimen-sion of rivalry within the groups, as membersneed to agree on how to distribute theresource between themselves. The availabil-ity of institutional mechanisms to solve thewithin group allocation more or less easilythan the between groups one is a crucialfactor in understanding the role of diversityin teams.

2.2 Costs and Benefits of Diversity: ASimple Model

2.2.1 Private Goods, Public Goods, andDiversity

We provide here a simple model thathelps to clarify the pros and cons of ethnicdiversity and offers a useful perspective for areview of the empirical literature. Considera community, say a country, with K differenttypes of individuals, for a total population ofN individuals. For simplicity, every grouphas the same size s = N/K. Output producedin the country is given by:



4 This can be considered a reduced form simplificationof a production function with a variety of inputs a laDixitStiglitz as used by Alesina, Spolaore, and Wacziarg(2000). While we could derive our results using the fullyspecified DixitStiglitz production function, we find itmore useful for the reader to offer a simpler work horsemodel.

5 In Alesina and Spolaore (1997), there are multiplekinds of public goods to be supplied with fixed quantities.More generally, both the type and the quantity of publicgoods could change.

(1) Y = N(x;K),where x is the fixed amount of input, saylabor, equal for every person and type. Weassume that x> 0, xx< 0, where subscriptsdenote partial derivatives. If variety in production is good, then we haveK> 0, KK< 0. This is the simplest possibleway of capturing a benefit from variety inproduction, since per capita income isincreasing in the number of different typesin the population. We also assume comple-mentarity, i.e., xK> 0.4

Output can be either consumed privatelyor used to produce a public good, g.Individual utility is separable in the privateand public good and is given by:

(2) Ui = u(ci) + v(g,K),where uc> 0, ucc< 0, vg> 0, vgg< 0. We alsoassume vK< 0, vKK 0 and vgK< 0, implyingthat the enjoyment of the public good isdecreasing with the number of types in thepopulation. These preferences can berationalized in two ways. One is that sharinga public good implies contacts between peo-ple, and contacts across types produce nega-tive utility, as in Alesina and La Ferrara(2000). A different rationalization followsAlesina and Spolaore (1997). They distin-guish between different kinds of publicgoods in a context where the public goodchosen is the one preferred by the medianvoter. The larger the number of types in thepopulation, the larger the average distancebetween each type and the median one thatchooses the public good.5

The budget constraint implies:

(3) g = tN(x,K),

6 Note that the second order condition is always satis-fied under our assumptions.

where t is the income tax rate. Suppose thata benevolent government can choose thetax rate, for given number of types. Theproblem is:

max N[u(c) + v(g,K)]

s.t. Nc + g = N(x,K)

g = tN(x,K).

The first order condition that defines aninterior solution for this problem is:6

(4) Nvg(tN(x,K);K) = uc((1 t)(x,K)).

This equation implies that the marginal ben-efit of taxation in terms of production of pub-lic good (LHS) has to be equal to themarginal cost of taxation in terms of reduc-tion of private consumption (RHS).Distortionary taxes on, say, the labor supplywould not change the basic message.Standard applications of the implicit functiontheorem lead to the following result:

(5) sign{dt /dK} = sign{tNwvggK+ NvgK (1 t)uccK}.

Note that we are holding N constant toisolate the effects of more fragmentationwithout changing total population size.While the sign of (5) is generally uncertain,dt /dK < 0 as long as vgK is large enough inabsolute value. The intuition for this condi-tion is clear: as long as the marginal benefitof public consumption goes down substan-tially with an increase in ethnic fragmenta-tion, then a larger K means that the socialplanner will choose a smaller size of the pub-lic good in favor of more private good. Theonly force working against this effect is thedecreasing marginal utility of the privategood. In what follows, we refer to the casewhere dt /dK < 0 as our benchmark case.This benchmark implies that, as a countrybecomes more ethnically fragmented, it may



7 Note of course that if K< 0, then income per capitawould go down as fragmentation increases and the alloca-tion of this lower total output between private and publicconsumption would depend on the marginal benefits ofthe two.

8 Note that if there were no benefits in production fromvariety (K 0), then the solution would be at a corner withthe minimum number of groups, possibly 1, that is, a fullyhomogeneous society. The first order condition for thechoice of t is of course unchanged.

become more productive but it will chooseto have a smaller size of government(remember that t = g/Y, thus t represents thesize of government). More generally, privateconsumption will increase but public con-sumption will decrease. This is an empiricalimplication which we shall test below.7

Another application of the implicit functiontheorem leads to the following result:

(6) sign{dt /dx} = sign{tN2vgg (1 t)ucc}.

Note that if dt /dx < 0, then, a fortiori,dt /dK < 0 in expression (5). However onecould have dt /dx > 0 and dt /dK < 0, i.e., it isperfectly possible that the size of govern-ment is increasing with the level of individ-ual productivity x, and thus in GDP, butdecreasing in fragmentation.

We next allow the social planner to choosenot only the level of taxation but also theoptimal number of types, K, again holdingthe size of the country constant. The firstorder condition for an interior solution withrespect to K is:

(7) uc()(1 t)K + vg()tNK = vK()

and the second order conditions are satis-fied. Condition (7) equalizes the marginalbenefit of letting in an additional group interms of increased productivity and tax rev-enues (LHS) to the marginal cost of havingone more groups to share the public goodwith (RHS).8

An interesting comparative statics exer-cise regards the effect of an increase inx (individual level of input/productivity)on the optimal number of groups.Straightforward algebraic computationsapplying the implicit function and the

9 Intuitively, these conditions require that the indirecteffects of a change in t caused by a change in K do notoverride the direct effect of a change in x on K. Details areprovided in a theoretical appendix available from theauthors upon request.

10 For an earlier contribution on endogenous jurisdic-tions which is however not as focused on diversity of pref-erences, see David Friedman (1977). Also, the literatureon the size of nations has some connection with club the-ory (see in particular James Buchanan and Roger Faith1987). However, there are two key differences. One is thatclub theory generally does not consider the geographic dis-tribution of club members; second, club theory empha-sizes the issue of congestion which is absent from the sizeof nations literature.

envelope theorem establish, under fairlygeneral conditions, the following:9

Remark 1 If xK is positive and sufficientlylarge, then dK /dx > 0.

A higher level of per capita input raisesthe benefits of variety and increases the opti-mal number of groups if the cross partial xKis large enough. In this case, as the level ofindividual output increases, the productivitygains from variety go up as well, so the ben-efit from more ethnic fragmentation areincreasing with the level of per capita out-put. This is an empirically plausible implica-tion: the benefits of skill differentiation arelikely to be more relevant in more advancedand complex societies.

2.2.2 On the Number of Jurisdictions

The same theoretical framework can beextended to analyze the optimal number ofjurisdictions, along the lines of Alesina andSpolaore (1997, 2003).10 We can think of theoptimal size of a jurisdiction (say a country)as emerging from the trade-off between thebenefits from variety and the costs of het-erogeneity. In the language of our modelabove, we could think of a social plannerchoosing the size of the country with thegoal of maximizing total welfare subject tothe trade-off between benefits and costs ofheterogeneity. The larger the (positive)effect of variety in production and the lowerthe utility costs of heterogeneity, the largerthe size of the jurisdiction chosen by thesocial planner. More specifically, imagine



11 A large literature has documented this fact; seeAlesina, Robert Barro, and Silvana Tenreyro (2002) for areview and some new results.

that each place specializes in the productionof an intermediate good and assume thatcrossing the boundaries of political jurisdic-tions is costly. Then the more beneficial isvariety in production, the more efficient is toemploy many different intermediate goodsand the larger the total border crossingcosts. Should we then expect larger coun-tries to be more productive because theyhave more variety? The answer depends onthe structure of international trade. Tradeflows are generally hampered by countryborders even when the trade regime is freeand without explicit policy-induced barri-ers.11 With severe trade restrictions, the sizeof a country would be very important forproductivity; on the other hand with freetrade countries can be small, enjoy the ben-efit of homogeneity as far as public goodsprovision is concerned but take advantage ofdiversity in production (and consumption)by means of international trade. One impli-cation of this is that the effects of the size ofcountries on economic success is mediatedby the extent of freedom of trade, a resultempirically supported by Peter Katzenstein(1985), Ronald Rogowski (1987), AlbertoAdes and Edward Glaeser (1995), Alesina,Spolaore, and Wacziarg (2000), andFrancisco Alcala and Antonio Ciccone(2004), amongst others. Note that somediversity in a country may favor trade as well.For instance, a certain ethnic minority incountry A can be a link with a country Bwhere that ethnic group is a majority, there-fore facilitating trade between A and B. Theextent to which ethnic and cultural relationsfacilitate trade and more generally econom-ic integration is well established; see forinstance Samuel Huntington (1998) for aninformal discussion and James Rauch(2001), and Rauch and Alessandra Casella(2001, 2003) for models and empirical evi-dence. An important question is under whatconditions the optimal solution would be

12 Under certain conditions, Alesina and Spolaore(2003) show that, when borders are decided by majorityvoting, in equilibrium countries may be too smallbecause the benefits of size are not fully internalized bythe median voter.

13 Important references include Dennis Epple andThomas Romer (1991) and Stephen Calabrese, GlennCassidy, and Epple (2002) who also provide a good reviewof the previous literature.

reproduced by the market without a socialplanner, a question explored in depth byAlesina and Spolaore (2003). In general, theanswer is negative and the equilibrium sizeof jurisdictions varies as a function of thetype of political institutions and rules avail-able to change borders, a set of issues thatwe do not pursue here.12

The same ideas about endogenous borderformation can be applied to subnational gov-ernments; this is in fact the goal of Alesina,Reza Baqir, and Caroline Hoxby (2004) whoextend the AlesinaSpolaore framework andadapt it to localities with a special referenceto the United States. They produce a modelin which the number of localities depends onthe distribution of different groups and theirdensity in various geographical areas; onceagain the key is the trade-off betweeneconomies of scale, heterogeneity of prefer-ences, and in this case mediated by density ofthe population. The traditional literature onjurisdiction formation in the United States isoverwhelmingly focused on differences inincome, with the rich trying to isolate them-selves from the poor.13 We do not review thisliterature, and we focus instead on the muchmore limited empirical literature that con-centrates on jurisdiction formation basedupon racial and ethnic cleavages.

2.2.3 Summing Up the Implications of theTheory

The potential benefits of heterogeneitycome from variety in production. The costscome from the inability to agree on com-mon public goods and public policies. Onetestable implication is that more heteroge-neous societies may exhibit higher produc-tivity in private goods but lower taxation and



14 In principle, various ethnic groups could segregatewithin the same jurisdiction and use different publicgoods. However, segregation is often imperfect, may entailother costs, and some public goods are by nature jurisdic-tion wide.

lower provision of public goods (in relativeterms). The benefits in production fromvariety in skills are more likely to be rele-vant for more advanced societies. While inpoor economies ethnic diversity may not bebeneficial from the point of view of produc-tivity, it may be so in rich ones. The moreunwilling to share public goods or resourcesare the different groups, the smaller the sizeof jurisdictions.14 The larger the benefits inproduction from variety, the larger the size.If variety in production can be achievedwithout sharing public goods, differentgroups will want to create smaller jurisdic-tions to take advantage of homogeneity inthe enjoyment of the public good.

2.2.4 What Is Not in the Model

Many important aspects are missing inthis model, but a few in particular are worthpausing upon. First, we have not modeledexplicitly the possible benefits of informa-tion diffusion and enforcement of contractsrelated to ethnic diversity. These mecha-nisms are probably most likely to be impor-tant for developing countries or the poorerregions of middle income countries withvery rudimentary forms of market institu-tions and legal enforcement. MarcelFafchamps (2004) discusses a variety ofmodels of the pros and cons of ethnicallybased economic networks as way to substi-tute for western style markets, and reviewsmuch evidence with specific reference toseveral Sub-Saharan African countries inAfrica. We will return on this point below.

Second, the model is institution free so wecannot address the question of the interac-tion between political institutions and diver-sity. Certain types of institutions may bemore conducive to harmony than others.Paul Collier (2000, 2001) for instance argues

15 See Alesina, Baqir, and Easterly (2000) for someevidence on this point on U.S. cities.

that ethnic fragmentation is less disruptivein democracies because minorities feel rep-resented. Alesina and Glaeser (2004) discussat length how, in the United States, propor-tional representation both at the federallevel and at the local level was opposed inthe nineteenth and early twentieth centuryprecisely on the ground that it would allowmore representation of blacks, somethingthat would not please the white majority.More generally, the type of political institu-tion is endogenous to the nature of inter-group conflict, a point made in theory andalso tested by Philippe Aghion, Alesina, andFrancesco Trebbi (2004a). First of all, if agroup is politically dominant, it may imposea type of government that restricts freedomof the minority. On the other hand, a morefractionalized society in which no group isdominant may end up with a constitutionespecially careful to defend the rights ofminorities.

The third missing aspect in the model isthat, while pure public goods may be lower inmore fragmented communities, the amountof publicly provided private goodsespe-cially those that can be targeted to specificgroupsmay be larger. We can then have apositive correlation between fragmentationand ethnically based patronage.15

Finally, in the model an increase in diver-sity would simply lead to smaller jurisdic-tions. In practice this process may bepeaceful or not, leading to violent civil wars.This is an important topic that we do notinvestigate directly here; we refer the readerto Fearon and Laitin (2003), Fearon (2002),and the references cited therein.

3. The Consequences of Fragmentation

In this section, we review the main contri-butions that have linked ethnic fragmenta-tion to economic outcomes, going from themore aggregate level (country level frag-mentation and performance) to the moremicro level, i.e., local jurisdictions (cities,



16 An early and never published paper by David Canningand Marianne Fay (1993) used ethnic fractionalization as aninstrument for growth.

districts) down to the level of small groups(schools, associations, cooperatives).

3.1 Countries

3.1.1 Effects on Productivity and IncomeLevel

Economists have started to pay attentionto the effects of racial fragmentation acrosscountries at least since a paper by WilliamEasterly and Ross Levine (1997) who arguedthat, ceteris paribus, more racially fragment-ed countries grow less and that this factor isa major determinant of Africas poor eco-nomic performance.16 Several subsequentpapers confirmed these results in the contextof cross-country growth regressions. In theiroverview of Africas problems, Collier andJan Gunning (1999) also place much empha-sis on ethnolinguistic fractionalization (cou-pled with low political rights) as a majorexplanation for the lack of social capital, pro-ductive public goods, and other growthenhancing policies.

Easterly and Levines paper, and much ofthe literature that followed, used as a meas-ure of fragmentation the probability thattwo randomly drawn individuals from theunit of observation (say, country) belong totwo different groups. Their ethnolinguisticfractionalization (ELF) measure is aHerfindahl-based index defined as follows:

(8) ELF = 1i s2i ,

where si is the share of group i over the totalof the population. This index represents theprobability that two randomly drawn individ-uals from the population belong to differentethnic groups. The source used by Easterlyand Levine to construct the ethnolinguisticgroups is the Atlas Narodov Mira, originallycompiled by Soviet researchers. Apart fromissues of measurement (to which we returnbelow), the robustness of Easterly andLevines results has been called into questionby Jean-Louis Arcand, Patrick Guillaumont,

17 Arcand, Guillaumont, and Guillaumont Jeanneney(2000) note that African countries constitute only 27 of the172 observations in Easterly and Levines main regression,and highlight the potential sample selection bias generat-ed by the fact that the data is missing precisely for thosecountries (in Africa) that have experienced slower growth.

18 In fact, several countries in Latin America appear asmore fractionalized compared to Easterly and Levinesclassification using this more comprehensive index. SeeAlesina et al. (2003) for more details.

and Sylviane Guillaumont Jeanneney (2000)due to problems of data missingness.17

Despite the criticisms, subsequent estimateshave taken Easterly and Levines results as abenchmark, and have confirmed them.

Using the updated data set of Alesina,Arnaud Devleeschauwer, Easterly, SergioKurlat, and Wacziarg (2003), we now testwhether the negative correlation betweenethnic fragmentation and growth holds irre-spective of the level of economic develop-ment or, as our model suggests, is mitigatedwhen the benefits of heterogeneity for pro-ductivity are taken into account.Measurement and data issues are discussedbelow in section 5.2. A brief description ofthe data is contained in the appendix. Fornow it is enough to note that Alesina et al.(2003) construct two indices with the samestructure as above but using two different(although closely related) characterizationsof groups. One is more comprehensive, islabeled ELF, and extends the Easterly andLevine index by differentiating groups thatmay speak the same language but have dif-ferent ethnicity based upon certain physicalcharacteristics. A striking example would beblacks and whites in the United States, orvarious ethnic groups in Latin America allspeaking the same language, often that of aformer colonizer.18 The second index reliesexclusively on language spoken.

Table 1 shows some standard growthregressions adopting the baseline specificationof Alesina et al. (2003). The dependent vari-able is the growth rate of GDP per capitafrom 1960 to 2000 and we use a SUR methodin four ten-year periods. The first twocolumns use the more comprehensive indexof fractionalization (which we label ELF),



19 These results are very similar to those reported byAlesina et al. (2003). The only difference is that they useboth a linear and a quadratic term for initial per capitaincome. We use only the linear one because below weexplore interactions of the initial level of income withother variables and we want to keep a simpler specifica-tion. In any case results with a quadratic term for initialincome are very similar for our variables of interest.

while columns 3 and 4 use the one basedsolely on language. Columns 1 and 3 show abaseline regression with very few controls:regional dummies, initial income, andschooling. Columns 2 and 4 include addi-tional controls, such as measures of politicalstability and quality of policy. One may argue(and in fact we explore this point below) thatthe effect of fractionalization on growth maygo through exactly these variables; thereforeby controlling for these variables one mayunderestimate the effects of fractionalizationon growth. Overall, table 1 shows consider-able support for the negative effects of frac-tionalization on growth.19 In terms ofmagnitude, the estimates in column 1 sug-gest that, ceteris paribus, going from perfecthomogeneity to maximum heterogeneity(i.e., increasing ELF from 0 to 1) wouldreduce a countrys growth rate by 2 percent-age points per year. Increasing ethnic frac-tionalization by one standard deviation wouldreduce growth by 0.6 percentage points peryear. These are quite sizeable effects. All theother controls have signs consistent with thevast literature on growth.

An important question is whether or notthese negative effects from ethnic fraction-alization on growth depend on the level ofincome or other features of society. In themodel of section 2, we showed that, underreasonable conditions on technology, frac-tionalization may have positive (or less neg-ative) effects on output at higher levels ofdevelopment. Table 2 adds to all the regres-sions of table 1 an interaction term betweenfractionalization and GDP per capita. In allfour regressions, the interaction of initialGDP per capita and fractionalization hasthe expected (positive) sign, suggesting thatindeed fractionalization has more negative

20 This is a well known and common stumbling blockfor anybody who has tried to estimate empirically the costsand benefit of democracy on economic variables, a vast lit-erature that we do not review here; see Jose Tavares andWacziarg (2001) for one of the most recent and carefulcontributions.

effects at lower levels of income. In two outof four regressions, this effect is stronglystatistically significant.

Collier (2000) argues that fractionalizationhas negative effects on growth and productiv-ity only in nondemocratic regimes, whiledemocracies manage to cope better with eth-nic diversity. This is an important result worthexploring further. It is well known that percapita GDP and democracy are positively cor-related: richer countries are more democrat-ic. From a statistical point of view, this highcorrelation makes it quite difficult to disen-tangle the effects of democracy from theeffects of the level of income on any depend-ent variable that might be affected by eitherone or both.20 Table 3 considers the effects ofthe interaction of ethnic and language frac-tionalization with the Gastil index of democ-racy. This index is decreasing in the level ofdemocracy so the expected sign on the inter-action with fractionalization is negative. Theestimates in table 3 are consistent withColliers findings that fraction-alization hasless negative effects in democracies.

Table 4 uses the two basic specifications totry and disentangle the effects of income anddemocracy. Since we are adding several vari-ables with interactions, we use the simplerspecification. Overall, the effect of incomeseems more robust and more precisely esti-mated than the effect of democracy.However, these results have to be taken cau-tiously given the high correlation betweendemocracy and GDP per capita.

The punch line is that rich democracies aremore capable of handling productively eth-nic diversity. Note, however, that as arguedabove, the variable democracy may beendogenous to ethnic diversity. It may be thecase that racially fragmented societies thatchoose democratic institutions are also those



TABLE 1FRACTIONALIZATION AND LONG-RUN GROWTH

(DEPENDENT VARIABLE IS GROWTH OF PER CAPITA GDP)

ETHNIC LANGUAGEVariable 1 2 3 4

Dummy for the 1960s 0.059 0.153 0.065 0.156 (3.357) (5.144) (3.563) (5.248)

Dummy for the 1970s 0.057 0.158 0.062 0.161 (3.093) (5.222) (3.280) (5.333)

Dummy for the 1980s 0.036 0.141 0.042 0.145 (1.940) (4.601) (2.213) (4.725)

Dummy for Sub-Saharian Africa -0.008 -0.016 -0.009 -0.014

(-1.630) (2.853) (-2.026) (-2.595) Dummy for Latin America and the Caribbean -0.016 -0.011 -0.019 -0.018

(-4.458) (-2.923) (-5.252) (-4.201) Log of initial income -0.004 -0.018 -0.004 -0.018

(-1.499) (-3.767) (-1.660) (-3.724) Log of schooling 0.012 0.005 0.011 0.008

(2.767) (1.092) (2.627) (1.669) Assassinations -21.342 -13.988

(2.212) (-1.010) Financial Depth 0.012 0.010

(1.798) (1.652) Black Market premium -0.021 -0.022

(4.738) (-4.953) Fiscal Surplus/GDP (0.128) 0.132

3.369 (3.474) Log of telephones per worker

(0.006) 0.004 2.078 (1.488)

Fractionalization -0.020 -0.014 -0.019 -0.021 (-3.005) (-1.795) (-2.979) (-2.881)

No of Observations 82; 88; 94 40; 69; 66 82; 86; 92 39; 68; 65 R-squared .23; .17; .35 .32; .43; 54 .21; .21; .30 .36; .47; .52

(t-statistics in parentheses)Estimated using Seemingly Unrelated Regressions: a separate regression for each ten year period.

in which ethnic cleavages are less deep and/orthe power distribution of groups is such thatnone can impose a nondemocratic rule.

Related to the issue of how democracyinteracts with ethnic conflict and with thelevel of development is the role played byinstitutions in general. Easterly (2001) con-structs an index of institutional qualityaggregating Stephen Knack and PhilipKeefers (1995) data on contract repudiation,expropriation, rule of law, and bureaucratic

quality. He finds that the negative effect ofethnic diversity is significantly mitigated bythe presence of good institutions and themarginal effect of ethnic diversity at themaximum level of institutional developmentis actually zero. Again, the institutional vari-ables used as explanatory factors are likelynot exogenous and more work needs to bedone to assess the marginal impact of insti-tutional arrangements. Nonetheless, itseems important to take into account that,



TABLE 2FRACTIONALIZATION AND LONG-RUN GROWTH



Dummy for the 1960s 0.064 0.220 0.098 0.253 (2.522) (5.116) (3.910) (6.827)

Dummy for the 1970s 0.061 0.226 0.096 0.260 (2.369) (5.179) (3.735) (6.897)

Dummy for the 1980s 0.041 0.209 0.077 0.245 (1.542) (4.757) (2.951) (6.411)


(-1.574) (-2.479) (-1.478) (-2.138) Dummy for Latin America and the Caribbean -0.016 -0.013 -0.021 -0.019


(-1.297) (-4.253) (-2.420) (-5.523) Log of schooling 0.012 0.006 0.011 0.009

(2.775) (1.112) (2.599) (1.966) Assassinations -21.880 -16.919

(-2.311) (-1.303) Financial Depth 0.011 0.008


(-4.736) (-4.729) Fiscal Surplus/GDP 0.136 0.146

(3.618) (4.048) Log of telephones per worker

0.007 0.005 (2.532) (1.969)


Fractionalization log ofinitial income 0.001 0.015 0.008 0.025

(0.227) (2.084) (1.279) (3.977)

No of Observations 82; 88; 94 40; 69; 66 80; 86; 92 39; 68; 65 R-squared .23; .18; .35 .27; .48; .55 .22; .25; .28 .36; .55; .56

(t-statistics in parentheses) Estimated using Seemingly Unrelated Regressions: a separate regression for each ten year period.

whatever the mechanisms relating ethnicdiversity to economic growth, channelingdiversity toward productive uses may requirea particular set of rules of the game.

3.1.2 Effects on Public Policies

An important prediction of the modelsketched in section 2 is that the propensity to

supply true public goods should be lower inmore ethnically fragmented societies. Theempirical literature has focused more on thequality than on the quantity of publicgoods partly because of data availability. Inorder to carefully test the implication of themodel on the quantity of public goods provi-sion, one would need aggregate measures of



TABLE 3FRACTIONALIZATION, DEMOCRACY AND LONG-RUN GROWTH



Dummy for the 1960s 0.059 0.153 0.073 0.159 (3.290) (5.090) (3.897) (5.331)

Dummy for the 1970s 0.056 0.155 0.069 0.162 (2.869) (4.983) (3.418) (5.220)

Dummy for the 1980s 0.035 0.137 0.050 0.146 (1.790) (4.358) (2.420) (4.632)


(-1.628) (-2.493) (-1.371) (-1.805) Dummy for Latin America and the Caribbean -0.016 -0.012 -0.020 -0.017


(-1.619) (-3.933) (-2.274) (-4.029) Log of schooling 0.012 0.007 0.013 0.010

(2.842) (1.351) (3.108) (1.959) Assassinations -23.495 -14.057

(-2.423) (-1.045) Financial Depth 0.012 0.012


(-4.828) (-5.169) Fiscal Surplus/GDP 0.117 0.131

(3.060) (3.520) Log of telephones per worker

0.006 0.004 (2.185) (1.610)


Democracy 0.001 0.003 0.002 0.002 (0.867) (1.833) (1.390) (2.064)

Fractionalization

Democracy -0.002 -0.005 -0.003 -0.005 (-1.230) (-1.871) (-1.885) (-2.489)

No of Observations 82; 87; 93 40; 69; 66 80; 85; 90 39; 68; 65 R-squared .23; .19; .34 .33; .46; .53 .21; .26; .27 .35; .52; .52


the various components of the governmentbudget for a relatively large group of coun-tries. These data are notoriously of poor qual-ity and not disaggregated enough. Thereforeresults in this area have to be taken cautiously.

Rafael La Porta, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert Vishny

(1999) and Alesina et al. (2003) show that eth-nic fragmentation is negatively correlated withmeasures of infrastructure quality, literacy,and school attainment and positively correlat-ed with infant mortality. These correlationsare very strong in regressions without incomeper capita (that may be endogenous to ethnic



TABLE 4FRACTIONALIZATION, DEMOCRACY AND LONG-RUN GROWTH


ETHNIC LANGUAGEVariable 1 3

Dummy for the 1960s 0.118 0.138 (4.689) (5.593)

Dummy for the 1970s 0.115 0.135 (4.356) (5.197)

Dummy for the 1980s 0.096 0.117 (3.562) (4.426)

Dummy for Sub-Saharian Africa -0.005 -0.003

(-1.053) (-0.668) Dummy for Latin Americaand the Caribbean -0.017 -0.020

(-4.793) (-5.267) Log of initial income -0.012 -0.014

(-3.398) (-4.247) Log of schooling 0.012 0.012

(2.878) (2.979) Fractionalization -0.149 -0.170

(-3.510) (-4.135) Fractionalization log of initial income 0.017 0.020

(3.233) (3.769) Democracy 0.001 0.001

(0.665) (1.228) Fractionalization

Democracy -0.002 -0.003 (-1.067) (-1.944)

No of Observations 82; 87; 93 80; 85; 90 R-squared .21; .33; .30 .20; .39; .25


fragmentation). They lose some of their signif-icance in regressions where on the right handside one controls for GDP per capita.21 In anycase, neither of these studies argues that eth-nic fragmentation is the only cause of poorquality of government: La Porta et al. (1999),

for instance, argue that legal origins are atleast as important.

An interesting related question regardsthe size of public transfers rather than pub-lic goods. For a large sample of countries,Alesina, Glaeser, and Bruce Sacerdote(2001) show an inverse relationship betweenthe size of government social spending andtransfers relative to GDP on the one hand,and ethnic fractionalization on the other.One explanation is that altruism does nottravel well across ethnic lines. Relating thispoint to the model above, one can view

21 Another variable that is correlated with racial frag-mentation is latitude and this high correlation makes itsometimes difficult to disentangle the two effects sepa-rately, although it is unclear why latitude per se (leavingaside its possible effects on GDP per capita) should affectpublic policies. Often both variables (latitude and frag-mentation) used together are insignificant while they aresignificant if used separately.



redistributive policies as a public good in asociety that values equality as a public bene-fit. On this point, a comparison betweenUnited States and Europe seems especiallysuggestive. In the United States, welfarespending and redistributive policies aremuch smaller than in Europe, consistentwith the fact that the United States is muchmore racially and ethnically diverse thanmost countries in Continental Europe, apoint explored in much detail by Alesina andGlaeser (2004). One implication of thisanalysis is that, to the extent that WesternEuropean countries will become more eth-nically fragmented, their welfare systemswill be under stress.

3.2 American Localities: Counties andCities

3.2.1 Effects on Productivity and IncomeLevel

American localities are an ideal setting tostudy the effects of ethnic fragmentationbecause we have many observations andexcellent data, compared, say, to cross-coun-try data. Glaeser, Scheinkman, and Shleifer(1995) have examined the growth of U.S.cities using a similar structure to cross-coun-try growth regressions. They argue that themost appropriate measure of growth to usein this case is population growth. They notethat income growth is a natural measure forcross-country growth regressions becauselabor is relatively immobile across countries.Instead, within the United States, the highmobility of individuals suggests that popula-tion growth is the correct measure to use tocapture areas and cities that are becomingincreasingly more attractive economicallyand as a place to live in. As OlivierBlanchard and Lawrence Katz (1992) havenoted, migration within the United Statesresponds strongly and relatively quickly toincome opportunities.

Glaeser, Scheinkman, and Shleifer (1995)do not find any effect of racial fragmenta-tion on the growth of cities in the sample

1960 to 1990. Their only finding concerningracial composition is that population growthis positively correlated with racial segrega-tion in cities with large non-white commu-nities. This result suggests that growth ishigher when racial interaction is lowerbecause of segregation. A suggestive inter-pretation of this result that goes back to ourmodel may be that racial fractionalizationwith segregation may allow for diversity inproduction and lower interaction in publicgood consumption and social activity. JordanRappaport (1999) also studies populationgrowth in cities and counties in the UnitedStates. He controls for many more determi-nants of counties characteristics and ameni-ties and he finds that more raciallyfragmented counties grow less in terms ofpopulation.

In tables 5 and 6, we present some resultson population growth in counties that are inline with our cross-country results. For thereason discussed above, we follow the liter-ature in using population growth as ourdependent variable. Table 5 reproduces forcounties instead of cities and for a differentsample (19702000) the basic specificationof Glaeser, Scheinkman, and Shleifer(1995). We do not find any effect of frac-tionalization on population growth. In table6, we add an interaction of initial per capitaincome level and fractionalization and weexperiment with different samples, notingthat we do not have data on income percapita before 1970. Broadly speaking, theresults are consistent with the cross-countryresults: we find that fractionalization has anegative effect on population growth in ini-tially poor counties and a less negative (oreven positive) effect for initially richercounties. This result significantly corrobo-rates the cross-country evidence in a settingwhere institutional and political differencesshould be definitely lower than acrosscountries.

Two recent papers have looked directlyat the productivity enhancing effects ofdiversity in American cities. Ottaviano and



TABLE 5FRACTIONALIZATION AND POPULATION GROWTH IN U.S. COUNTIES

(DEPENDENT VARIABLE IS GROWTH IN LOG OF POPULATION 19702000)

Variable 1 2 3 4

Intercept -0.088 0.902 -0.088 0.906 (-1.600) -18.66 (-1.600) (18.690)

Log of population 1970 0.034 -0.034 0.033 -0.036 (5.610) (-6.860) (5.170) (-6.910) -

Income per capita 1970(a) .095 -0.071 0.100 -0.068 (6.300) (-5.630) (6.360) (-5.290)

Growth in log of population 19601970 1.619 1.620

(44.730) (44.740) Northeast -0.396 -0.273 -0.396 -0.271

(-12.480) (-10.960) (-12.410) (-10.800) Central -0.413 -0.318 -0.413 -0.316

(-19.740) (-19.330) (-19.610) (-19.080) South -0.115 -0.137 -0.116 -0.143

(-5.220) (-7.930) (-5.010) (-7.890) Fractionalization 1960 0.019 0.042

(0.370) (1.080)

No of Observations 3133 3120 3120 3120 Adj. R-squared .17 .50 .18 .50

(t-statistics in parentheses)(a) Coefficient multiplied by 103.

Peri (2003) use data on rents and wages inU.S. cities and find that U.S. born individ-uals living in more culturally diversecities in their terminology (i.e., cities with alarger share of foreign born people) payhigher rents than those living in morehomogeneous cities. In other words, diver-sity seems to have positive amenityeffects on production and consumption.Their findings are robust to instrumentingthe share of foreign born people with thedistance from the closest port of entryinto the United States. Along similar lines,Richard Florida (2002a, 2002b) argues thatamenities and diversity in U.S. citiesattracts human capital. He constructsindices of heterogeneity of a place that arenot directly related to ethnicity but involveproportions of gay households, diversity ofnight life, etc. and finds that places thatscore higher in these indices have alsohigher human capital. The direction of

causality is however unclear. Further work,possibly using firm and plant level data,would be useful in this area.

In subsequent work, Ottaviano and Peri(2004) find that the wage of white indi-viduals, after controlling for various otherdeterminants, are higher in more diversecities where diversity is measured with indexbased on main language spoken at home.They interpret this result as an indication ofhigher productivity with diversity. A possiblecriticism of this finding is that cities in whichthe language spoken at home is not Englishmay be cities of recent immigration andimmigrants may be attracted to faster grow-ing cities. While they try to correct for thisproblem of reverse causation, it is not clearthat there is a way of doing it convincingly.


A very large literature in political scienceand sociology examines the role of race in



TABLE 6FRACTIONALIZATION, INCOME AND POPULATION GROWTH IN COUNTIES(DEPENDENT VARIABLE IS GROWTH IN LOG OF POPULATION 19702000)

growth 19602000 growth 19702000 growth 19702000 growth 19802000Variable 1 2 3 4 5 6 7 8

Intercept -0.221 0.679 0.043 1.026 -0.096 0.974 -0.397 -0.229 (-2.500) (7.170) (0.660) (18.360) (-1.450) (16.930) (-9.550) (-7.160)

Log of population 1960 0.019 -0.038(2.250) (-4.510)

Log of population 1970 0.029 -0.039 0.038 -0.036(4.560) (-7.470) (6.020) (-6.710)

Log of population 1980 0.051 0.019 (13.630) (6.900)

Income per capita 1970(a) 0.216 0.049 0.059 -0.105 0.086 -0.094(8.810) (1.940) (3.060) (-6.790) (4.440) (-6.000)

Income per capita 1980(a) 0.028 0.003 (5.100) (0.740)

Growth in log of population 19501960 0.965

(20.880)Growth in log of population 19601970 1.618 1.622

(44.830) (44.720)Growth in log of population 19701980 1.060

(52.880) Northeast -0.378 -0.236 -0.385 -0.261 -0.405 -0.269 -0.253 -0.017

(-9.110) (-5.990) (-12.030) (-10.390) (-12.63) (-10.660) (-12.460) (-1.080) Central -0.446 -0.369 -0.410 -0.314 -0.420 -0.317 -0.252 -0.040

(-16.360) (-14.260) (-19.530) (-18.990) (-19.87) (-19.010) (-18.680) (-3.800) South -0.085 -0.048 -0.130 -0.155 -0.101 -0.138 -0.067 -0.028

(-2.500) (-1.670) (-5.530) (-8.490) (-4.410) (-7.720) (-4.990) (-2.890) Fractionalization 1960 -0.727 -0.906 -0.581 -0.505

(-3.270) (-4.280) (-3.380) (-3.770)Fractionalization 1960

Income per capita 1970(a) 0.415 0.471 0.297 0.271(3.950) (4.690) (3.650) (4.270)

Fractionalization 1970 -0.130 -0.335(-0.770) (-2.520)

Fractionalization 1970

Income per capita 1970(a) 0.001 0.151(0.020) (2.410)

Fractionalization 1980 0.068 0.015 (0.670) (0.190)

Fractionalization 1980

Income per capita 1980(a) -0.033 0.025 (-1.970) (1.740)

No of Observations 3120 3102 3120 3120 3133 3120 3137 3133 Adj. R-squared .19 .29 .18 .50 .18 .50 .20 .58

(t-statistics in parentheses)(a) Coefficient multiplied by 103.



the history of public policies in Americancities (for a recent contribution, see NancyBurns 1994 and the references cited there-in). Several papers within the economics lit-erature have argued that public goodprovision is lower and/or less efficient inmore racially fragmented American cities,results which are consistent with thoseobtained in cross-country samples and inmany ways follow similar procedures thatinvolve cross cities (rather than cross-coun-try) regressions. Alesina, Baqir, andEasterly (1999, 2000) show that, in morefragmented cities, the provision of produc-tive public goods (roads, hospitals, schools,etc.) is lower, while the types of expendi-tures that more closely resembles transferstargeted to ethnic and racial groups arelarger. In particular, spending on roads,schools, and other public goods are smallerin racially fragmented cities. Preferencesover roads are likely to be different if dif-ferent racial groups are located in differentparts of the city and preferences overschools are certainly ethnically sensitive.Instead, spending on public employment islarger in more ethnically fragmented cities,a result consistent with a use of public jobswith ethnically or racially motivated patron-age. Interestingly, they find that racial divi-sions have stronger effects than ethnic ones(where ethnicity is identified with countryof origin), a result consistent with evidencediscussed in the next section on theendogenous formation of localities in theUnited States.

A particularly important type of localpublic good is public education. JamesPoterba (1997) finds that in U.S. states, gov-ernment per child spending on K12 educa-tion decreases with the fraction of thepopulation aged 65 and above, and that thiseffect is strengthened when the differencebetween the fraction of nonwhite popula-tion aged 517 and the fraction nonwhiteaged 65 or more is included among the con-trols. This suggests an interplay of demo-graphic and racial composition effects, as if

older citizens were less inclined to spend onpublic goods that benefit younger genera-tions when these generations belong dispro-portionately to a different race. Usinghistorical data on U.S. states, ClaudiaGoldin and Katz (1999) find a similar rolefor heterogeneity, be it ethnic, racial, reli-gious, or economic. La Ferrara and AngeloMele (2003) investigate the relationshipbetween racial segregation and spending onpublic education across U.S. cities. JacobVigdor (2004) finds that the greater a com-munitys racial heterogeneity, the lower itsrate of response to the 2000 Census form.Response is interpreted as a local publicgood in that the amount of federal fundsallocated to the community depend on itsresponse rate. Alesina and La Ferrara(2002) also show how redistributive policiesare deeply affected by racial politics. Inmore racially fragmented communities,people are less willing to redistributeincome because the white majority feelsthat redistributive flows would favor a racialminority. Survey evidence suggests thatthose respondents who express attitudesless favorable to racial integration are alsomore averse to government intervention onredistributive matters. Erzo Luttmer (2001)finds that respondents in surveys show moresupport for welfare policies as the share ofpopulation of the same race of the respon-dent increases in his/her community.

One of the reasons why public policies inracially fragmented communities are worseis that social capital is lower. Two keyaspects of social capital are participation insocial activities or groups and trust.22 Usingdata from the General Social Survey,Alesina and La Ferrara (2000) provide evi-dence that, in American cities, individualsof different races are less willing to partici-pate in social activities in racially mixedcommunities. There are two non mutually

22 For survey on the relationship between ethnic diver-sity and social capital, see Dora Costa and Matthew Kahn(2003a).



exclusive explanations. One is that mem-bers of different racially identified groupshave different preferences on what a groupshould do or how it should be run, and theother is that there is a cost in sharing agroup with different races simply becauseof aversion to racial mixing. Alesina and LaFerrara (2002) show that, in Americancities, individuals living in more raciallyfragmented communities have a lowerpropensity to trust other people, while theydo not exhibit lower levels of trust towardinstitutions. Similar results were laterobtained by Costa and Kahn (2003b).Interestingly, all these authors also showthat income inequality reduces participa-tion and social capital but the effect ofracial conflict seems stronger.Experimental evidence on trust and partic-ipation included in Glaeser, David Laibson,Jose Scheinkman, and Christine Soutter(2000) is also consistent with these results:even in experimental settings and amongsta relatively homogeneous group of individ-uals (in terms of education), trust does nottravel well across racial lines.

In summary, looking at U.S. cities, thereseem to be two empirically relevant typesof diversity. One is ethnic fractionalization,where ethnic groups are defined accordingto the five Census categories (White;Black; American IndianEskimoAleutian;AsianPacific Islander; and Hispanics).This type of fractionalization seems to beassociated with poor public policies, lowtrust, and low city growth. On the otherhand, there are measures of cultural diver-sity based on language, life style, and atti-tudes. These measures seem to beassociated with a positive evaluation ofamenities and higher productivity. The twosets of results are not contradictorybecause the two indices of diversity arevery imperfectly correlated. Most notably,an index based on language does not distin-guish between whites and blacks and, inthis sense, may underestimate diversity inracially heterogeneous cities. But it is also

possible that, to the extent that a totallywhite city is very diverse in terms of lan-guage spoken at home, the language-baseddiversity index can overestimate hetero-geneity compared to the race-based one.The choice of the notion of cultural diver-sity to be used is ultimately a matter of con-text and of the particular problem to beanalyzed.

3.3 Village Communities in DevelopingCountries

3.3.1 Effects on Productivity

A particularly relevant setting in which tostudy the productivity effects of ethnic diver-sity is that of developing economies. Thereason is that a large share of economictransactions occurs outside the boundariesof the formal sector and need to be support-ed by enforcement schemes similar to thosedescribed in section 2.1. Although directempirical evidence is seldom available, anumber of recent studies on developingcountries allow to draw preliminary infer-ence on the impact of diversity on produc-tivity and economic performance at themicro level.

Manufacturing firms in Africa have beenstudied by several authors. Arne Bigsten,Peter Kimuyu, and Karl Lundvall (2000) usea data set on Kenyan firms in the food, wood,textile, and metal industries and examinewhat factors account for the choice of goingformal and for the degree of economic effi-ciency. They find that kinship and communi-ty ties among entrepreneurs of Asian originreduce the barriers to entry in the formalsector, so that even after accounting for differences in education, African firms aremuch more likely to be informal at start-up.23 In addition to the advantage that the

23 The distinction between the Asian and the Africanbusiness community in Kenya goes back to the colonialperiod, when the British organization rested upon a three-tier society in which traders and businessmen of Indianorigin occupied an intermediate position.



24 Further evidence on the relationship between ethnicnetworks and access to credit in the Kenyan manufactur-ing sector is provided by Tyler Biggs, Mayank Raturi, andPradeep Srivastava (2002).

formal status gives to Asian-managed firms(e.g., in terms of access to formal credit),Asian firms are shown to be more capitalintensive and more productive.24

Fafchamps (2000) focuses on the relation-ship between ethnicity and access to creditfor manufacturing firms in Kenya andZimbabwe and finds that, after controllingfor observable firm characteristics (e.g.,size) African firms are not discriminatedagainst in the allocation of bank overdraftand formal loans. The way in which ethnici-ty seems to make a difference is by offeringnetwork relationships that improve access tosupplier credit. This in turn affects produc-tivity and allows firms to remain in the mar-ket in the presence of negative shocks, as acommon way to absorb cash flow variationsis to delay payments to suppliers. The rela-tionship between trade credit and produc-tivity is further explored by RaymondFisman (1999, 2003), who shows that entre-preneurs of Asian and European origin aremore likely to obtain supplier credit, andthat firms that do not have access to suppli-er credit have a higher probability of facinginventory shortages and have lower rates ofcapacity utilization.

Fafchamps (2004) reviews more broadlythe evidence on the effects of ethnic net-works on economic performance and on theallocation of credit in several African coun-tries. He identifies two effects, one positiveand one negative. On the one hand, ethni-cally based networks can provide insurance,facilitate transactions, increase trust, andsimply substitute for rudimental marketinstitutions. On the other hand, the samenetworks may lead to bias against variousgroups. For instance, Fafchamps finds thepresence of an ethnic bias in trade creditusage amongst manufacturing firms in Sub-Saharan Africa. The direction of the bias isgenerally detrimental to entrepreneurs of

25 This obviously depends on the way in which networkstructure endogenously responds to the ethnic compositionof the community, a point we address in section 4.3.

African descent (p. 368). This author, goingbeyond some of his previous work, tries toidentify whether this amounts to active dis-crimination due to the lack of connectionwith appropriate networks. At least forZimbabwe and Kenya, he finds that this isindeed the case for black and female entre-preneurs. Interestingly, the same authorfinds virtually no effects of ethnically basednetworks on agricultural trade (chapter 19),a result that leads to be cautious about blan-ket generalizations across all Africa and alltrades on the role of ethnic networks.

Despite their focus on employer-level eth-nicity as opposed to ethnic fractionalization,the above studies potentially bear interestingimplications for the relationship betweenethnic diversity at the community level andfirm performance. In fact, for a given level ofcredit supply, the greater the number of eth-nic groups in the business community, thelower the chances that supplier credit is allo-cated efficiently if the criterion is purely eth-nic affiliation, which can ultimately harmeconomic productivity.25

An explicit focus on ethnic heterogeneityand economic performance is in the study byLa Ferrara (2002b). She uses an originaldata set on production cooperatives in theinformal settlements of Nairobi and hasinformation on all members of the surveyedgroups, which allows her to construct exactmeasures of group composition in terms ofincome, education, age, and ethnicity. Shefinds that ethnicity matters for gainingaccess to group resources, especially in theform of cheap loans: members who share thesame ethnicity as the chairperson are 20 to 25percentage points more likely to borrow fromthe group or from other members. Ethnicheterogeneity also seems to influence theorganization of production: members ofmore ethnically heterogeneous groups areless likely to specialize in different tasks and



more likely to all do the same job. Primafacie, this seems in contrast with theassumption of positive complementarities inproduction among different ethnic groups,which we made in our theoretical frame-work. However, the difficulty of allocatingdifferent tasks across different groups in theparticular context of La Ferraras studystems from the governance structure of thegroups, and on the lack of transparent allo-cation rules that characterizes most of thegroups under study. For example, ethnicallyfragmented groups more often adopt remu-neration schemes in which every worker getsthe same fixed amount, rather than beingpaid on the basis of the amount of work putin. These choices on division of labor andwage structure may be due to the relativedifficulty of reaching consensus onunequal task allocations and remunerationschemes in ethnically heterogeneous groups.In this case, the consequences of ethnicdiversity on differential access to inputs getreinforced by its impact on within-firmorganization of production. Viewed in thislight, our hypothesis that the beneficialeffects of diversity are stronger in more eco-nomically advanced societies is actually fullycompatible with these findings.

A recent application to lending groups isprovided by Dean Karlan (2003). He usesdata on members of a Peruvian microfinance organization and exploits the ran-dom selection of people into groups to esti-mate the effect of group composition onrepayment performance. He finds thatmembers of more homogeneous groups,both in terms of geographical proximity andof cultural affiliation, are more likely to saveand to repay their loans. Interestingly, cul-tural homogeneity is measured through ascore attributed by enumerators to eachrespondent on the basis of his/her language,dress, and hair style. These findings suggestthat monitoring and enforcement withingroups are easier the greater the socialaffinity among their members, as argued insection 2.1.

26 For a survey of this topic with a specific emphasis oneconomic inequality, see La Ferrara (2003b).

Finally, although very limited evidenceexists on the subject, ethnic diversity canalso have an impact on agricultural produc-tivity in developing countries. A recent studyby Karen Macours (2003) suggests thatinformal enforcement of property rights inthe land market creates incentives for rentaltransactions to remain within ethnic groups.In turn, in a highly fragmented environment,the exclusion of minority groups leads toethnic conflict, further weakening propertyrights and reinforcing segmentation.


Most of the literature on group hetero-geneity and collective action in small com-munities has focused on the role of incomeinequality as opposed to ethnic or racial het-erogeneity. The focus of that literature,exemplified by Mancur Olsons (1965) semi-nal contribution, has been on the relation-ship between inequality in the shares of thebenefits from the commonly provided goodthat accrue to different types and theirincentives to contribute.26 However, theextent to which this literature can be gener-alized to the impact of ethnic diversity is lim-ited, and requires a context in which types(e.g., ethnic groups) matter for public goodprovision only through their contributingcapacity, and a mapping between inequalityand ethnic fractionalization can be traced.

A recent literature, however, has lookedspecifically at the role of ethnic hetero-geneity with a particular emphasis on pub-lic good provision in poor communities. Anexample is the study by Cagla Okten andUna Okonkwo-Osili (2004), who use microlevel data from Indonesia to estimate howethnic diversity affects monetary and timecontributions to community organizations.They propose three ways in which ethnicdiversity can affect voluntary contributions:(1) diverse communities may have moredifficulty in defining common policies as



27 This author also discusses the effect of ethnic minor-itys control of municipalities and finds that it is associatedfor poorer tax collection.

their preferences diverge; (2) transactioncosts are higher in more diverse communi-ties, e.g., because of lower effectiveness ofinformal enforcement mechanisms; and (3)an altruistic orientation to contribute toones own ethnic group. Their empiricalresults suggest that increased ethnic het-erogeneity decreases both contributions tolocal community organizations, and theprevalence itself of such organizations. AdiBrender (2004) finds similar results forIsrael.27

Edward Miguel and Mary Gugerty (2004)also investigate the mechanisms throughwhich ethnic heterogeneity may harm publicgood provision. In particular, they focus onthe role played by social sanctions. As weargued in section 2, in environments withweak legal enforcement, most informal trans-actions rely on the availability of self-enforc-ing mechanisms related to repeatedinteraction and reputation, as well as on theimposition of social sanctions. Miguel andGugerty assume that such sanctions are moreeffective if imposed within ethnic groupsthan between groups. They test this hypothe-sis using data on 337 primary schools in ruralKenya. In addition to information on studentsand teachers, their data contains school com-mittee records which report the threat orapplication of sanctions and the fund raisingactivities of the school. They find that localethnic diversity is negatively correlated toschool funding and to the quality of schoolfacilities. According to their estimates, mov-ing from complete homogeneity to completeheterogeneity would reduce average localfunding by about 20 percent.

An insight into the motivations underlyingthe failures of collective action in heteroge-neous communities is offered by the recentwork of Abigail Barr (2003). She conductedfield experiments in Zimbabwe exploitingthe resettlement policies promoted by the

government, which generated a set of social-ly and ethnically heterogeneous villages(treatment) to be compared with nonreset-tled communities (control). From the resultsof her trust game, Barr concludes that thelower propensity to trust of resettled vil-lagers is due not to differences in altruism orin socially transmitted norms, but to thelower density of kinship ties. Again, this isconsistent with the hypothesis that informalenforcement does not travel well across kin-ship (and a fortiori ethnic) lines.

Finally, Jeff Dayton-Johnson (2000) andAsim Khwaja (2000) look at more production-oriented collective activities. The former usesdata on Mexican irrigation projects and findsthat canal maintenance is worse the moreunequal is the distribution of land and thehigher is social heterogeneity, proxied by thenumber of different farming communitiesrepresented in the same maintenance unit.Khwaja (2000) uses data on community-main-tained infrastructure projects in NorthernPakistan, the complexity of which ranges fromsimple irrigation channels to sophisticatedelectricity units. Again, he finds that socialheterogeneitymeasured as the fragmenta-tion into different clans, political, and religiousgroupsis negatively associated with projectmaintenance. An interesting differencebetween Khwajas work and the other existingstudies is its focus on institutional design. Infact, he finds that good task design can poten-tially compensate for fragmentation in allow-ing heterogeneous communities to succeed incollective action. This parallels the findings ofthe literature on organization behaviordescribed in section 2.1.

4. Ethnic Fragmentation and Endogenous Community Formation

In this section, we discuss how fragmenta-tion affects not just the economic policies andperformance of given communities, but theformation and composition of the relevantcommunities. In other words, what happenswhen community size and/or composition



28 Patrick Bolton and Gerard Roland (1997) explorehow income differences and redistribution may lead tobreak down of countries.

can be simultaneously determined with thepolicies?

4.1 Countries

A line of research by Alesina and Spolaore(1997, 2003), Alesina, Spolaore, andWacziarg (2000), and Spolaore and Wacziarg(2002) emphasizes the role of racial conflictas a determinant of the number and size ofcountries. The argument is as follows. Thesize of a country emerges from a trade-offbetween the benefits of scale (broadlydefined) and the cost of heterogeneity ofpreferences in the population. Benefits ofsize include economies of scale in the pro-duction of some public goods, internaliza-tion of policy externalities, the size of themarket, defense and protection from foreignaggression, and regional insurance schemes.The costs of heterogeneity arise because, inlarge and diverse countries, individuals withdifferent preferences have to share commonpolicies so the average utility of these poli-cies is decreasing with heterogeneity.Empirically, racial fragmentation is oftenassociated with differences in preferences,so racial cleavages are a major determinantof the determination of borders, secessions,and various centrifugal forces.28

A potentially testable implication of thisapproach is that, as the benefits of sizediminish, then it becomes more likely thatcountries can split into more homogenoussmaller political entities. One building blocof this argument is of course that opennessto trade is particularly beneficial for smallcountries. Results by Ades and Glaeser(1999), Alesina, Spolaore, and Wacziarg(2000), and Alcala and Ciccone (2004) sug-gest that, as freedom of trade increases, thebenefit of size for economic growth dimin-ishes. In a completely autarkic world, thepolitical size of a country also determines itseconomic size; in a world of free trade theybecome more disjoint. That is, from an

economic point of view (our production ofprivate goods in the simple model above),trade makes economic size larger. On theother hand, since countries can retain theirindependence while trading, they do nothave to share common public policies onwhich there are differences of opinions. Inethnically diverse societies, then, increasedeconomic integration should make it morelikely that conflicts are resolved with break-down of countries. Some insights on thisissue can be gathered from the political sci-ence literature on partition as a solution toethnic civil war, supported among others byChaim Kaufmann (1996, 1998). A criticalassessment of the view that separation is thebest solution for civil wars generated by eth-nic conflict is provided by NicholasSambanis (2000), who uses a cross sectionaldata set of all civil wars since 1944 and esti-mates the probability of partition as a func-tion of the type of civil war (ethnic/religiousas opposed to ideological) and of severalsocioeconomic factors, among which ethnicheterogeneity of the population.

The relationship between ethnic hetero-geneity and the likelihood of country break-downs is also mediated by the role of naturalresources, and this is a particularly relevantissue for developing countries. Naturalresource discoveries tend to be located inremote areas at the periphery of a country, asresources more centrally located have likelybeen discovered already. It is often the casethat people living in peripheral areas haveethnic identities that do not coincide withthe majority of the country as a whole. Theavailability of new natural resources makesthese regions more economically viable ontheir own and therefore increases pressurefor separation or autonomy.29

In addition to economies of scale, anotherbenefit of country size is defense and pro-tection from aggressions, so as the worldbecomes more peaceful one should observecentrifugal forces. Alesina and Spolaore

29 We are grateful to a referee for suggesting this point.



(2003) discuss historical evidence, arguingthat this implication is consistent with the dataconcerning the evolution of country size,international trade, and threats of conflicts.Recently, the collapse of the Soviet Union, byreducing the threat of and EastWest conflict,has certainly facilitated political separatism inEastern Europe. Huntington (1998) noteshow the end of the Cold War allowed therealignment of peoples into countries that bet-ter reflected homogenous civilizations. Inmost cases, this movement meant breakdownof countries and in a few cases movementtoward reunification.

Finally, an important issue is the relation-ship between ethnic heterogeneity, countryformation, and democracy. Alesina andSpolaore (2003) discuss the effect of authori-tarian systems on measured racial, linguistic,or religious fragmentation and country size.Dictators prefer large countries for severalreasons. One is that they can extract rentsfrom larger populations, another one is thatthey can support with size their bellicose atti-tudes. Historically, one of the main problemsof dictators has been to repress ethnic conflictin an attempt to create artificially homoge-neous countriesan issue to which we returnbelow when we discuss the endogeneity of thenotion of fragmentation. In fact, dictatorsoften use racial hatred to create support forthe dominance of one group over others, aresult consistent with models and empiricalevidence by Glaeser (2002). One of the impli-cations of this artificial repression of diversityis that centrifugal forces typically explodewhen dictators fall, as happened for examplein the Soviet Union,

ethnic diversity and economic performance

Documents

ethnic cleavages

result of ethnic conflict

potential costs of diversity

journal of economic

economic policies

pros andcons of diversity

diverse ethnic mix alsobrings

conflict of preferences