‘unfinished business’: historic complementarities, political competition and ethnic violence in...

G ModelJ

‘c

SS

a

ARRAA

JNOZF

KTIPCRR

1

iwbt

LERa

0

ARTICLE IN PRESSEBO-3310; No. of Pages 19

Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Journal of Economic Behavior & Organization

j ourna l h om epa ge: w ww.elsev ier .com/ locate / jebo

Unfinished business’: Historic complementarities, politicalompetition and ethnic violence in Gujarat

aumitra Jha ∗,1

tanford Graduate School of Business, 655 Knight Management Center, Stanford, CA 94305, United States

r t i c l e i n f o

rticle history:eceived 28 June 2013eceived in revised form 15 February 2014ccepted 28 February 2014vailable online xxx

EL classification:25181210

eywords:radenstitutionsolitical polarizationultureeligionegression discontinuity

a b s t r a c t

I examine how the historical legacies of inter-ethnic complementarity and competitioninteract with contemporary electoral competition in shaping patterns of ethnic violence.Using local comparisons within Gujarat, a single Indian state known for both its non-violentlocal traditions and for widespread ethnic pogroms in 2002, I provide evidence that wherepolitical competition was focused upon towns where ethnic groups have historically com-peted, there was a rise in the propensity for ethnic rioting and increased electoral supportfor the incumbent party complicit in the violence. However, where political competitionwas focused in towns that historically enjoyed inter-ethnic complementarities, there werefewer ethnic riots, and these towns also voted against the incumbent. These historic legaciesproved to be important predictors of the identity of the winner even in very close electoralraces. I argue that these results reflect the role local inter-ethnic economic relations canplay in altering the nature and the benefits of political campaigns that encourage ethnicviolence.

© 2014 Elsevier B.V. All rights reserved.

. Introduction

On February 27th, 2002, a carriage of the Sabarmati Express carrying Hindu activists caught fire at Godhra railway stationn the western Indian state of Gujarat. At least 58 people were burnt alive. The burning of the Sabarmati Express precipitated

Please cite this article in press as: Jha, S., ‘Unfinished business’: Historic complementarities, political competition andethnic violence in Gujarat. J. Econ. Behav. Organ. (2014), http://dx.doi.org/10.1016/j.jebo.2014.02.023

eeks of violent ethnic rioting between Hindus and Muslims throughout the state that claimed at least a thousand livesetween February and the end of April, 2002, and forced at least 98,000 people into refugee camps. Many of the riots appearedo have been planned in advance, targeting local ethnic minorities, mainly Muslims.

∗ Tel.: +1 650 721 1298.E-mail address: [email protected]

1 I owe much thanks to Ken Arrow, Susan Athey, Prashant Bharadwaj, Avner Greif, Dan Hungerman, Lakshmi Iyer, Kimuli Kasara, Asim Khwaja, Davidaitin, Jessica Leino, Aprajit Mahajan, Atif Mian, Ken Shotts, Steven Wilkinson and seminar participants at AALIMS, Berkeley, Chicago, Clio, Davis, therasmus University Rotterdam, Harvard, the Hebrew University of Jerusalem, Northwestern, Stanford, LiCEP, LSE, MIT, Princeton, the NBER Conference oneligion and Culture, NEUDC, NYU, Rochester, UBC, USC, UWO, the all-UC economic history group and the World Bank for useful comments and suggestionsnd to SIEPR, the Harvard Academy, and CSDP and the Niehaus Center at Princeton for support.

http://dx.doi.org/10.1016/j.jebo.2014.02.023167-2681/© 2014 Elsevier B.V. All rights reserved.

dx.doi.org/10.1016/j.jebo.2014.02.023

dx.doi.org/10.1016/j.jebo.2014.02.023

http://www.sciencedirect.com/science/journal/01672681

http://www.elsevier.com/locate/jebo

mailto:[email protected]

dx.doi.org/10.1016/j.jebo.2014.02.023

ARTICLE IN PRESSG ModelJEBO-3310; No. of Pages 19

2 S. Jha / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx

Fig. 1. Patterns of ethnic riots and attempted violence in Gujarat, February–April 2002.

The complicity of the ruling Bharatiya Janata Party (BJP) state government, an organization with strong historic ties toHindu nationalists (Hansen, 1999), was suspected, as accounts of pogroms led by local members of the legislative assembly(MLAs) and close advisors to the Chief Minister rapidly surfaced.2 The ruling party dissolved the 187 seat state assembly inJuly 2002, ten months early, and in the subsequent state elections, the BJP vote share rose by an average 4.84 percentagepoints, gaining them 10 seats. One high-ranking police official, who chose to be anonymous, told the author in 2007 that theGujarat violence was ‘unfinished business’ from India’s long history of inter-ethnic relations. He claimed that many in thegovernment believed that by allowing the violence to happen once, it would not happen again.

The incentives for political figures to encourage mobilization on ethnic lines in order to swing close elections has beenobserved in settings around the world, including Africa (e.g. Eifert et al., 2010), the United States (Olzak, 1992) and SouthAsia (Blakeslee, 2013; Wilkinson, 2004; Brass, 1997). Ethnic heterogeneity and polarization has also been associated withreduced public goods provision, lowered growth and increased civil conflict (e.g. Easterly and Levine, 1997; Alesina and LaFerrara, 2005; Montalvo and Reynal-Querol, 2005). And in the many developing country environments where ethnic partiesfavor their own with patronage (e.g. Chandra, 2007; Banerjee et al., 2008), ethnic cleansing may itself act as a response toanticipated inter-ethnic electoral competition (Jha and Wilkinson, 2012).

Yet, much less is known about the conditions under which poor ethnically diverse societies remain peaceful even in thepresence of strong political incentives for violence. In this paper, I examine how different types of “unfinished business” in theform of historical legacies of inter-ethnic complementarity and competition interact with electoral incentives in generatingethnic violence. I focus on local comparisons of towns within districts of the western Indian state of Gujarat. Gujarat providesa useful environment to study these questions as it could easily fit within a square of 340 miles and enjoys recordedhistorical interactions between Hindus and Muslims going back at least to the eighth century, but despite being one ofIndia’s wealthiest states and harbouring remarkable traditions of peaceful coexistence and non-violent political mobilisation,towns throughout Gujarat have also become notorious for terrible ethnic pogroms. The wave of ethnic rioting betweenFebruary–April 2002, among South Asia’s worst since the Partition, was also almost entirely confined within Gujarat’s


borders (Fig. 1). In towns in other states, ethnic violence occurred but rarely escalated or exhibited contagion (see AppendixFigure 4).

2 In 2012, the BJP MLA and state minister Maya Kodnani was ultimately convicted for her role in the massacre of Muslims in Naroda Patiya and sentencedto 28 years in jail. Testimony collected by the People’s Union for Democratic Rights (2002) also implicated local BJP MLAs in Delol, Sanjeli and Visnagar. Atthe same time, while Muslim-majority areas were placed under strict curfew, state law enforcement appears to have shown an asymmetric unwillingnessto intervene during mob violence led by Hindus (People’s Union for Democratic Rights, 2002).

dx.doi.org/10.1016/j.jebo.2014.02.023


S. Jha / Journal of Economic Behavior & Organization xxx (2014) xxx–xxx 3

Fa

pecarae2

rMhftsss(

telcs

seatf

ig. 2. Ruling party (BJP) vote change, 1998–2002. Relative to other towns where the BJP gained vote share following the riots, medieval ports swanggainst the BJP and became more competitive following the 2002 elections.

I draw upon work by Jha (2013b) who describes how overseas trade in the Indian Ocean was coordinated by Muslimilgrimages, particularly the Hajj, inducing a thousand years of inter-ethnic complementarities between Hindu produc-rs and Muslim traders at small indentations in the coast that formed ‘natural harbours’ on the medieval coastline. Iomplement Jha’s sub-continent wide study of riots over a 150 year period by restricting the sample geographicallynd culturally to nearby towns drawn from a single state, and temporally to a single two month period of ethniciots between February and April 2002. Using a novel set of data on Gujarati towns spanning its medieval, colonialnd modern electoral history, I find that despite being larger and more ethnically diverse, Gujarati medieval portsxperienced 30% fewer days of rioting than otherwise similar towns during the riots between February and April,002.

I next go beyond this work to study the interaction between local institutions and electoral competition. I find thatiots are more likely when close political races are concentrated in towns which were founded by Muslims, houseduslim mints or political capitals, or faced a longer history of Muslim political rule – factors which, I argue, generated

istorical incentives for inter-ethnic competition. Thus, ethnic riots appear to be encouraged when political incentivesor organizing violence intersect with inter-economic competition. I show, in contrast, that a historic legacy of inter-ethnicrade reduces the duration of riots the most in towns that housed competitive races. Furthermore, and unlike otherwiseimilar towns where the ruling party enjoyed vote gains in the elections that followed the violence, medieval port con-tituencies exhibit a vote swing of around six percentage points against that party following the 2002 pogroms. Thesewings were pivotal enough to shape the outcome of close races, leading to a loss to the ruling party of three seatsFig. 2).

I interpret these results as reflecting the fundamentally conditional nature of the relationship between electoral compe-ition and ethnic violence. In environments where groups compete economically, close electoral competition may providenhanced incentives to local politicians to exacerbate pre-existing inter-ethnic tension that can readily lead to local vio-ence. However, the presence of inter-ethnic complementarity or institutional mechanisms that support inter-ethnicooperation appears to have a stronger mitigating effect where the incentives for political violence are also otherwisetronger.

Beyond the papers already mentioned, this paper builds upon well-established literatures in economics and politicalcience, including those examining the determinants and the effects of close elections, the effects of religious intensity and


thnic identification on voting, and the corresponding incentives by political figures to foster ethnic mobilization. Estebannd Ray (2008) present a model where elites from ethnic majorities benefit from providing resources for poor members ofheir group to riot against minorities to avoid class conflict. This resonates with Glaeser (2005)’s analysis of how incentivesor politicians to send hate messages in the US South increased with calls for redistribution. More generally, Canes-Wrone

dx.doi.org/10.1016/j.jebo.2014.02.023

G Model
ARTICLE IN PRESSJEBO-3310; No. of Pages 19

et al. (2001) show that in competitive elections, local politicians have greater incentives to ‘pander’ to voters’ sentiments bypublicly advocating popular policies that they personally know will not be socially beneficial.3

In the South Asian context, Brass (1997) describes how politicians in old Indian cities develop ‘institutionalized’ riotsystems to mobilize their base on ethnic lines.4 Wilkinson (2004) provides state-level evidence that states with a largernumber of “effective parties” competing – settings where Muslims are more likely to be electorally pivotal – are less likelyto exhibit electoral violence. At the same time, Wilkinson and Haid (2009) and Dhattiwala and Giggs (2012), also looking atconstituencies in Gujarat in 2002, find a correlation between close races in the state legislature elections in 1998 and thesubsequent propensity for riots. Similarly, Pathania and Tandon (2011) take a sample of riots in India between 1989 and1995, and find that the killing was more intense where the BJP won in contested national (Lok Sabha) elections between1989–91. Blakeslee (2013) finds that the BJP was more likely to direct the ‘Rath Yatra’ – a procession originating in Gujarataimed at religious mobilization – at marginal constituencies that subsequently experienced greater rioting. Varshney (2002)describes how towns with ‘civic engagement’ – the development of organizations and social norms conducive to peace – alsoreveal a lack of violence. This paper complements and unifies these works by establishing the exacerbating role of politicalcompetition on violence when it intersects with inter-ethnic economic competition but the mitigated incentives for violencewhen political competition occurs among communities that enjoy traditions of ethnic tolerance.

This paper also builds upon and contributes to an important literature that exploits local geographical comparisonsand finds that historical exposure to different political institutional environments, particularly empires, past incidents ofviolence or propensities for inter-ethnic trade, can have lasting effects on traits as diverse as financial development (Guisoet al., 2008; Grosjean, 2012), trust in bureaucracy (Becker et al., 2012) contemporary attitudes towards religion, voting anddemocracy (Grosfeld and Zhuravskaya, 2012), and differences in ethnic assimilation, hatred, tolerance and trust (Voigtländerand Voth, 2011; Nunn and Wantchekon, 2011; Jha, 2008b, 2013b; Diaz-Cayeros and Jha, 2013). This paper adds to theseworks by showing how modern electoral incentives can strengthen or undermine the effects of such historical cultural andinstitutional legacies.

This paper also suggests a direction for additional nuance to an important series of papers that exploits proposed quasi-random variation in close races to identify the effects of winner identities. Based as they are upon such local comparisons,these studies have naturally been ahistorical for the most part. Yet, just as the outcome of close races in Gujarat can be pre-dicted by local economic incentives for inter-ethnic violence that interact with political campaign platforms, the measuredeffects of slight victories in developing countries may depend and interact in important ways with pre-existing patterns ofethnic polarization and historical institutions.

I begin by providing a simple theoretical framework and some historical context that together motivate the empiricalapproach. I next introduce the data and provide the main empirical results before discussing avenues for further research.

2. Theoretical framework

A simple theoretical framework can help illuminate the relationships between economic complementarity and compe-tition and the incentives for ethnic violence that emerge both in the presence and absence of political competition. First letus consider an environment where there are just two types of agents: local and non-local (please see Jha (2008a) for detailsof the formal theory).5 Non-locals differ from locals only in that they have better outside options.6 Suppose that individualsfrom either group have the following choices every period: to stay or leave town, to produce a good for exchange, and toattack any other agent that they encounter. Attacks are destructive, but may be useful for seizing the victim’s property and todeter or punish the victim’s actions. “Strong” individuals may exist who are more likely to prevail in a violent attack againstweaker opponents.

In the environment above, an important condition that favours ‘peaceful co-existence’ – an equilibrium with a mixedpopulation of locals and non-locals, full production, no out-migration and no violence – is that non-locals and locals producecomplementary goods or services. To see this, consider first the alternative: that locals and non-locals provide substitutegoods and thus are competitors. Then, with repeated interactions, a strong local will have an incentive to attack weak non-locals, as this allows that local not only to seize the non-local’s property but also to encourage non-locals to leave, reducingthe future competition the local faces. In fact, non-local competitors provide more attractive targets of violence than weaklocals, as local competitors are harder to encourage to leave due to their lower outside options. Thus rather than classviolence, societies where local and non-local groups compete are likely to exhibit greater ethnic violence.


In contrast, when ethnic groups provide complementary goods or services to one another, then the incentive to attacknon-locals falls over long time horizons. If non-locals leave if attacked, locals will face reduced supply and higher futurerelative prices for goods that only non-locals can provide. Not only may there be incentives for reduced violence, there

3 Close races and religiosity also appear to affect mobilisation. Cox and Munger (1989) show that higher expenditures on mobilization efforts occur inclose races in the US Congress. At the same time, Gerber et al. (2010) show that shocks that lower the costs of engaging in religious practice raise voterturnout.

4 See also Jha (2013a) who describes how electoral incentives were a likely precipitant of riots against Tata’s Nano plant in West Bengal.5 This initial discussion of the framework draws significantly upon Jha (2013b).6 This is arguably a reasonable assumption for groups who originally immigrated from elsewhere or whose group membership confers superior access to

trade, migration or other outside opportunities than for non-immigrants whose networks, endowments and information may also be concentrated locally.

dx.doi.org/10.1016/j.jebo.2014.02.023

G ModelJ

wi

wmoSwvmr

lmpl

beep(pof

wotc

riscic

claecissio

iaoHwtmt

c



ill also be incentives to invest in further complementary organizations and ‘institutions’ that may further reduce thesencentives (Milgrom et al., 1991; Jha, 2013b).

It is interesting to consider how incentives that emerge from local ethnic complementarities or competition then interactith political incentives. Suppose there are two types of political agents. A state-level incumbent ‘ruler’ cares about winningajority support among different constituencies, and can adopt a state-wide policy platform that he or she will ‘crackdown’

r be ‘complicit’ against ethnic violence should any occur. Local politicians care about winning their local political contest.uppose that state-level crackdowns on violence raise the expected costs of violence by civilians. Note that in a constituencyhere ethnic groups produce substitutes, the ethnic majority (which is likely to include the median voter) prefers ethnic

iolence to be cheap and thus the state-level ruler to be ‘complicit’. This is because lowering the costs of violence makes it aore credible strategy both as a means to deter entry and in-migration and to encourage out-migration by minorities that

educes subsequent economic competition.In contrast, in a constituency where ethnic groups produce complements, the majority is more likely to prefer the state-

evel ruler to ‘crackdown’ on ethnic violence. A ‘cracking down’ platform will reduce the likelihood of minority exit. If residentinorities are more likely to enjoy a costly-to-replicate source of complementarity, having the resultant cheaper minority

roducts should provide greater benefit for members of the majority. Thus both ethnic minorities and the majority are moreikely to prefer the crackdown platform when they provide complements to one another.

Now consider the incentives due to local electoral preferences. Suppose that ex ante electoral preferences are shapedy historical inter-ethnic relationships and by other endowed constituency characteristics. Suppose that there is also anxogenous component to electoral preferences that can lead to shocks that also change the relative competitiveness oflections. This assumption is consistent with an important literature that documents the role played by factors beyondoliticians’ control in shaping incumbent electoral support, such as wins in football games, natural disasters and lotteriese.g. Huber et al., 2012; Healy et al., 2010). Suppose that local politicians, having observed previous election results andre-election polls, invest resources in their campaigns – i.e. in mobilizing their base and reducing the mobilization of thepposition. One form of such an investment lies in organizational capacity that overcomes the costs of collective action andacilitates group-level violence (Beaman et al., 2003; Wilkinson, 2004).

In models with this simple setup, a robust and intuitive finding is that both local incumbent and challenger politiciansill tend to devote more resources to the most competitive elections as these are likely to yield the greatest return (in terms

f seats) for every dollar of campaign resources.7 Furthermore, the state-level incumbent, ceteris paribus, will also choosehe state-wide ethnic violence platform that puts greater weight on the preferences of the median voter in electorallyompetitive constituencies.

Incentives for both local politicians and individuals in close races to invest resources to mobilize and be mobilized in ethniciots, however, will again differ according to the nature of economic competition and complementarity. Local politiciansn close races are likely to invest in mobilizing ethnic violence most where it is likely to be most effective – i.e. whereuch mobilization coincides with pre-existing ethnic competition, and less where individuals have greater incentives toheck ethnic violence – i.e. where there is ethnic complementarity. In this way, close races should magnify the differencesn violence and subsequent voting behaviours in societies where groups compete economically and where they providedomplements.

It is useful to summarise the empirical implications of this simple discussion. First, the adoption of a state-wide platformomplicit with ethnic violence should naturally raise the incidence of violence throughout the state, but should exhibitess contagion beyond the state’s borders. Second, the ethnic violence that results from a ‘complicit’ platform should beccentuated in places where ethnic groups compete economically for pre-determined reasons, and be lower in places wherethnic groups provide complements. Third, electoral competition should have magnified effects on violence when it occurs inonstituencies with historical conditions that favour inter-ethnic competition and reduced effects on violence when it occursn constituencies that favour inter-ethnic complementarity. Fourth, voters in constituencies with economic competitionhould respond favourably a state-wide ‘complicit’ platform, while voters in constituencies with economic complementarityhould oppose it. Fifth, differences in the effectiveness of state and local political campaigns based upon ethnic polarizationn environments with ethnic complementarities and competition should lead there to be systematic differences in theutcomes even of close political races.

My empirical strategy uses the geographical focuses of medieval incentives for inter-ethnic economic competition andnter-ethnic economic complementarity (as well as organizations that reinforced inter-ethnic complementarity and trust)s pre-determined proxies for the presence or absence of such relationships in contemporary towns. I compare the incidencef rioting between February and April 2002 and changes in voting behaviour in the subsequent elections in towns whereindus and Muslims enjoyed robust complementarities due to Muslim pilgrimages to geographically proximate townsithin the same district where Hindus and Muslims were likely to compete for medieval political patronage and to other


owns that were otherwise similar in their geographical and medieval characteristics. My identification of the effects of aedieval overseas trade legacy rests on the assumption that the incentives that led some geographically proximate towns

o be focuses of trade to the Middle East in the medieval period – such as their location at medieval-era natural harbours

7 See e.g. Erikson and Palfrey (2000) and Grimmer et al. (2011). Note also that campaign resource investments by challengers and incumbents are strategicomplements in this setup.

dx.doi.org/10.1016/j.jebo.2014.02.023

G Model

that subsequently silted up or ceased to trade – are otherwise irrelevant for contemporary inter-ethnic relations. I controlfor other initial geographical dimensions through which the selection of historical port locations may have created localadvantages, as well as show the irrelevance of contemporary port activity as a useful placebo comparison. I also comparethese towns to towns that were focuses of Muslim political patronage historically. These towns also emerged in the medievalperiod, were also historically attractive for Muslims, but the incentives for inter-ethnic interaction were historically thatof inter-ethnic competition rather than complementarity. A little context may help clarify why such historical incentivesexisted and why they may have persisted.

3. Context

Gujarat is one of India’s wealthiest states, its modern borders demarcated to leave it linguistically homogeneous andmade mostly of flat plains and marsh that could fit easily within a square of 340 miles. The riots that followed the burning ofthe Sabarmati Express at Godhra, though initially concentrated near that city, rapidly spread to towns throughout the state asit became clearer that the state government and police were either complicit or at least unlikely to intervene. In Porbandar,a medieval port on the western coast, a respondent mentioned that attempts were made to instigate violence by outsiders,who sent women’s bangles to the members of the Hindu community as a symbol of their lack of masculinity in not joiningthe riots [author’s field observations, 2006–7]. Bangles were delivered to other communities in the state as well.8

Yet, not every community responded to the bangles with violence. Gujarat shows great diversity in its inter-ethnicrelations, both historically and as historical incentives manifest themselves in modern political behaviour. With 14% ofIndia’s coastline, Gujarat has housed centers of commerce trading to the Middle East even before the rise of Islam (Casson,1989), and has a history of political rule by Muslims that goes back to the 11th century. Hindus and Muslims within Gujaratitowns have long faced incentives to interact and to develop means to mitigate, and sometimes also propagate, ethnic violence.

I build upon work by Jha (2008b, 2013b), who documents that due to the coordination of overseas trade throughout Mus-lim pilgrimage, including the Hajj, which made Mecca the largest textile market in the world, Muslim seafarers throughoutthe Indian Ocean – who naturally had preferred access to their own pilgrimage routes – enjoyed exogenous, non-replicablecomplementarities with local communities in inter-ethnic trade. Furthermore, because trade was coordinated by the pil-grimage, it was relatively easy for any Muslim to enter, and intra-Muslim competition limited the potential for monopsonisticrents. Mutual incentives existed among local communities in medieval ports to develop systems of norms, beliefs and orga-nizations that reinforced inter-ethnic trust and tolerance. Jha (2013b) shows that medieval ports that were the geographicalfocuses of these historic complementarities between Hindus and Muslims benefited from lasting reductions in ethnic vio-lence across the towns of the sub-continent between 1850 and 1995. Jha provides case and household level survey evidencein support of a number of institutional mechanisms: Muslims in medieval ports have continued to be specialised in comple-mentary roles coordinated by their religious leadership, show enhanced propensities to join religious organizations, showenhanced propensities to vaccinate their sons against polio – a behavioural measure of trust in the majority community –and report less exposure to ethnic conflict in their communities.9

Thus norms, organizations and beliefs of syncretism and ethnic tolerance, inspired in part by medieval inter-ethnic trade,appear to continue to persist in medieval ports, despite the decline of international trade, and to have spread in varyingdegrees throughout the state.10 Years later, when asked about the local response to the bangles used to incite violence duringthe 2002 riots, my Porbandari respondent remarked: “This did not work. We didn’t want [violence], our police and officialsdid not want it. Nobody wanted it.” [author’s field observations, 2006–7]. Porbandar, despite a reputation for gang violencecentered around caste-based organized crime, remained peaceful during the Hindu–Muslim riots of 2002.


But Gujarat’s history was not just one of inter-ethnic complementarity between Hindus and Muslims due to trade, but alsoone of considerable ethnic violence and potential grievance. Though Arabs had conquered neighbouring Sind by amphibiousinvasion in the early eighth century, one of the first encounters that contemporary India faced of Muslim military powerwas a raid by the Afghan ruler, Mahmud of Ghazni, on the temple city of Somnath, approximately 70 miles from Porbandar,

8 See, for example, the description of riots in the village of Sanjeli, Dohad district, by the People’s Union for Democratic Rights (2002).9 These patterns have shown spillovers throughout the state. In the non-medieval port town of Santrampur, near Godhra, Muslims in the Sant locality

witnessed the arrivals of Hindu mobs of 150 on the 28th of February and 2500 on the 3rd of March, containing many that they recognized from amongtheir neighbours. While most Muslims were able to flee, two were hacked to death by swords while trying to escape, and two mosques, a madrassa and adargah (a tomb of a local Muslim saint) were destroyed (People’s Union for Democratic Rights, 2002).Yet, approximately one mile from these events, another mixed Hindu–Muslim area housed a temple of the Pranami sect of Hinduism. The Hindu communityoffered protection to their Muslim neighbours, who chose not to flee and were not attacked (Burman, 2005)[pg. 106]. The Pranami sect was establishedin 1651 by Mehraj Thakur, the son of the Diwan (Vizier) of Jamnagar, a medieval port in western Gujarat, who had followed the trading and pilgrimageroutes to Basra in Iraq and had returned to organize a syncretic sect that focused on the essential similarities of Islam and Hinduism and the desirabilityof social interactions between the two. Later called Mahamati Prannath, the sect he founded gained around 3 million adherents around northern India(Burman, 2005)[pg. 103], among these the mother of Mahatma Gandhi. Gandhi was raised in the medieval port of Porbandar and Rajkot, and would laterget his start as a lawyer for Porbandari Muslim merchants in South Africa. He would draw upon Gujarati Hindu and Jain ideals of ahimsa – or non-violence– and arguably also the social traditions of inter-ethnic tolerance among Muslims and Hindus in Gujarati medieval ports in crafting his political messageof non-violence.

10 Please see Jha (2013b) for detailed case study comparisons of institutions of tolerance developed by Muslim trading communities, such as the Bohras,in Surat relative to Ahmedabad, as well as discussion of other medieval ports.

dx.doi.org/10.1016/j.jebo.2014.02.023

G ModelJ

ioe

iatfel

mw

ppteicis

4

Sstvt“g

ops

loVbtf

ca

clpd

r

t

wibm



n 1026. From the time of the chronicler, Alberuni (1030), who accompanied Mahmud, to the modern day, the destructionf the temple has been seen as a Kosovo Pole-like event in Indian history, one that has polarized Hindus and Muslims againstach other ever since.11

In fact, in direct contrast to the robust complementarities visible at medieval trading ports were the incentives presentn towns that were the centres of Muslim political authority, where Hindus and Muslims often acted as substitutes for onenother and competitors for patronage. Following the Somnath raid, Muslim rule began to spread into India, and eventuallyo Gujarat too. With the conquest came control of patronage and land revenue systems, based upon control of the surplusrom India’s mainly agricultural wealth. These patronage systems were concentrated in towns, like Ahmadabad, which werestablished by the fiat of the Muslim rulers themselves, as well as existing cities that became the administrative capitals ofocal rulers, such as Palanpur and the medieval port of Cambay (Raychoudhari, 1998).

In medieval Indian kingdoms, political and religious patronage played a very important role. Though not necessarilyembers of the royal household themselves, the majority of the city’s population was often tied by client relations to peopleho were.12 Once flourishing cities that lost their roles as political centres rapidly became ghost towns.

Following the Muslim conquest, it is likely that Muslim clients, both converts and immigrants, substituted for and com-eted with Hindu clients for patronage. Though “vertical” inter-ethnic links existed between Hindu artisans and Muslimatrons, such ties were often in competition to intra-Muslim patron-client relations. Though the Hindu and Muslim artisanshat constituted the majority of the populations of these cities lived side by side, there was limited incentive for inter-thnic exchange between these groups. Thus, despite the fact that, like medieval ports, political centres provided historicalncentives for conversion to Islam and enjoyed historical wealth, patronage centres were the historical focus of inter-ethnicompetition rather than inter-ethnic complementarity and exchange. Thus patronage centres were less likely to developnstitutions to support to support such exchange, and instead were likely to continue to be loci for competition in economicpheres.13

. Data

To construct the dataset for the analysis, I went through news reports on Gujarat from the day of the burning of theabarmati Express on February 27th until April 15th. News sources include rediff.com and the Times of India. These wereupplemented from an amicus curiae brief by the Concerned Citizens Tribunal and a compilation of eyewitness refugeeestimonial by the People’s Union for Democratic Rights (2002), which provided information about less-widely publicisediolence in smaller towns and villages. Following Varshney and Wilkinson (2004), I coded a riot as occurring in a town ifhere was evidence of violence by communally-identifiable “mobs” or other large groups in that town. I also coded a day ofviolence” as having occurred in a town if there was an isolated incident, such as a stabbing, without any evidence of broaderroups being involved.

Following Jha (2008b, 2013b) I identified a town as a medieval trading port if it exhibited substantive evidence of directverseas trade, prior to the 18th century and independent of European involvement. This definition eliminates most riverorts and those ports founded by Europeans. Of the 68 confirmed medieval trading ports in undivided India, 13 are in ourample of Gujarati towns (Fig. 1).

The Periplus Maris Erythraei (Casson, 1989), an ancient Greek navigator’s handbook (ca 1st–4th century) provided theocations of a number of pre-Muslim and early Muslim ports. These records were then supplemented by the accountsf contemporary Muslim, Christian and Chinese observers, including Chau Jukua (1225), Ibn Battuta (1355), Ludovico dierthema (1503), Duarte Barbosa (1519) and Zayn al-Din al Malibari (1528). These contemporary narratives were augmentedy secondary sources (Yule, 1866; Subrahmanyam, 1990; Chaudhuri, 1995; Chakravarti, 2000). Finally, every town listed inhe Imperial gazetteers of India from 1907 was examined for evidence of contemporary and medieval trade and to trace itsoundation and historic role as an administrative, as well as a trade, center.


Medieval ports were then linked to their geographical location using ArcGIS. This enabled me to gather data on initialonditions that might shape the initial selection of locations for medieval ports, including the presence of navigable riversnd the presence of medieval-era indentations on the coast which may have served as natural harbours.14

11 It is not an accident that the BJP leader L.K. Advani, had chosen Somnath to begin his Rath Yatra procession – aimed at ethnic mobilization across theountry – in 1990 (Blakeslee, 2013). Yet, records exist that in 1262, the authorities of the rebuilt Somnath temple made a large-scale land grant of templeands to a Muslim trader, Nur-ud-din Firuz of Hormuz to settle in the adjacent medieval trading port of Veraval, aware of the commercial taxation androsperity that a colony of Muslims could bring (Sircar, 1962; Thapar, 2004; Jha, 2013b). This ethnically-mixed community still exists, despite Veraval’siminished economic prospects as a fishing port in a largely vegetarian state. Yet, Veraval too remained peaceful in 2002.12 Thus, when the Mughal Emperor Aurangzeb went South to campaign in the late 17th century, it was not unusual that four of five of Delhi’s 400,000esidents left with him (Blake, 1991).13 Consistent with this interpretation,Field et al. (2008) find evidence that violence within the city of the Muslim capital of Ahmadabad occurred withinhe context of competition between the different ethnic groups over property.14 To categorize medieval era “natural harbours,” I used the US Geological Survey Digital Atlas of South Asia 2001 to identify water bodies that wereithin 10 km of the modern Indian coastline, including non-perennial ponds and streams and those without an outlet to the sea. If these water bodies

ntersected the coast in the medieval period, they would have produced minor inlets, or sheltered harbours. I define towns within 10 km of those waterodies as having had access to a “medieval natural harbour.” This approach identifies major irregularities and inlets that are likely to have existed in theedieval period.

dx.doi.org/10.1016/j.jebo.2014.02.023

G Model

Looking at effects over long periods of history raises a separate challenge that deviates from a canonical experiment:even controlling for initial conditions, towns under study were subject to differences in external political influences bothduring and after the treatment that might also influence subsequent religious relations. Some component of these politicalinfluences – e.g. the expansion of Muslim or European political rule – might have been in part a result of a desire to occupyregions with active medieval trade. To account for these political channels, I compare the effect of medieval trade both withand without a rich set of controls for these political factors, and assess sensitivity of the baseline estimates.

I also examine the effect of Muslim political rule. If, as I have argued, Muslim political rule is likely to have renderedintensified historic and persistent ethnic competition for patronage, then a greater exposure to Muslim political rule mightbe associated with a greater propensity for contemporary ethnic mobilization. The presence of state protectors of localminorities may also have crowded out mutual incentives to develop local institutions of trust in towns with inter-ethniccomplementarity (Jha, 2013b). However, it might also be that longer exposure to Muslim minority rulers has a beneficialeffect by reducing psychological biases, as in Beaman et al. (2009). To derive such a local measure of the historic stock ofMuslim political rule, I used historical maps of the extent of Muslim sultanates and empires over time, particularly drawing onSchmidt (1999) and the medieval chronicler Ferishta (1620). Layering these maps over time gives a measure of the durationevery location was under Muslim rule prior to the death of the Emperor Aurangzeb in 1707. Since, our preferred estimationinvolves 1981 district level fixed effects, we will be exploiting the local variation of towns within the same district that actedas patronage centers due to their genesis as Muslim administrative cities, or had different extents of Muslim rule.15 I alsoadd separate controls for the religion of native rulers in the colonial period, relative to those districts – such as Ahmadabad,Broach, Panch Mahals and Surat – that were controlled directly by the British.

Another potential reason mooted for group cooperation (e.g. Wade, 1988) or conflict (e.g. Miguel et al., 2004) is thepropensity for natural disasters that require people to work together or may lead to greater attempts to extract. Economicshocks due to rainfall (Bohlken and Sergenti, 2010), or relative income changes between Hindus and Muslims have alsobeen suggested as potential precipitants of violence (Mitra and Ray, 2010), while economic shocks may also encourage theadoption of religious behaviours that provide entry into ‘clubs’ that provide social insurance (Chen, 2010). While the twomonth period of the ethnic riots in Gujarat did not exhibit much weather variation, long-term differences in propensitiesfor such shocks might also be important. Thus I collected data from the Imperial gazetteers on a number of different naturaldisasters from 1850 to 1900, including droughts, earthquakes, locust infestations, floods and cyclones.

To examine the extent to which past ethnic cleansing may have also driven contemporary conflict, I use the Partition-erameasure of ‘minority outflows’ developed by Bharadwaj et al. (2008). It is the reduction of the Muslim population relativeto that expected given local trends in Muslim fertility and local mortality rates of those not targeted during the time of thePartition. As Jha and Wilkinson (2012) argue, this measure is arguably a good measure of ethnic cleansing, as it gauges howmany members of the target ethnic minority departed, were removed, were forcibly converted as well as were killed duringthe Partition period. I also build on Bharadwaj et al. (2008) to examine whether a key potential source of grievance – theextent to which a district or native state territorial unit16 received Hindu and Sikh refugees from Pakistan – also may haveinfluenced the extent of contemporary Hindu–Muslim violence and political polarization.

I merged the town data to the Electoral Commission’s data on each Vidhan Sabha constituency using the spatial pointlocation of the centre of the town. While some urban areas, particularly for the larger towns, such as Ahmadabad, Surat,Vadodara and Rajkot span multiple constituencies, the old precincts that are of particular interest largely correspond to thecentral constituencies. I supplement this electoral data with data on town-level religious composition from the 1981 Censusand other town-level data on potential factors that might also influence political behaviour – such as worker participationrates, the caste composition and literacy rates – from the town directories of the 1991 Census.

Since I argue that medieval trade incentives due to the Hajj induced complementarity between Hindus and Muslimsthat later were undermined with increased European intervention, modern ports can act as a useful placebo comparison,one which also allows us to control for any modern port effects. For this reason, I also gathered data on contemporary portactivity from the Gujarat Maritime Board.

5. Results

Table 1 provides summary statistics for our sample, by status as medieval port. Notice that apart from being signif-icantly more likely to be located at medieval natural harbours, and to be coastal towns, medieval ports are statistically


indistinguishable to other Gujarati towns along a range of other initial conditions and subsequent medieval and colonialcharacteristics, including in their propensity for natural disasters, distance to navigable rivers, history under Muslim politicalrule and Partition-era flows. In terms of modern indicators, Gujarati medieval trading ports seem to have equivalent levels ofemployment and literacy, though a higher proportion are active ports. I control for the presence of active modern ports, while

15 It is likely, however, that Muslim conquerors – like those that resisted them – contested those with the highest wealth potential the most. Insofar aswealth decreases ethnic violence (e.g. Fearon and Laitin, 2003) this is likely to lead to downward bias on using these measures as a proxy for the historicallegacy of inter-ethnic competition.

16 Gujarat was divided into many territorial units prior to Independence, largely due to the many princely states. Thus data observed at the level ofhistorical territorial units can create within-modern district variation that we will exploit.

dx.doi.org/10.1016/j.jebo.2014.02.023



Table 1Summary statistics, Gujarat riots data, February–April 2002.

Other Gujarati towns (N = 167) Medieval Ports (N = 13) 2-Sided t (Welch)

Mean SD Mean SD

Electoral and Violence measures:H-M riot occurred, February–April 2002 0.180 0.385 0.231 0.439Days of H-M rioting, February–April 2002 0.413 1.658 0.615 1.193Any H-M violence occurred, February–April 2002 0.305 0.462 0.308 0.480Vote swing to BJP 1998–2002 3.328 12.523 −0.706 10.807BJP vote %, 2002 49.143 11.144 47.705 7.164BJP vote %, 1998 45.815 10.902 48.411 12.451BJP vote %, 1995 41.990 13.397 41.748 14.828Vote Margin of Victory %, 1998 15.044 10.553 18.889 14.262Vote Margin of Victory %, 2002 14.585 12.007 8.889 8.589 **

Distance to Godhra measures:Within 100 km of Godhra 0.168 0.375 0.000 0.000 ***

Within 200 km of Godhra 0.407 0.493 0.308 0.480Within 300 km of Godhra 0.623 0.486 0.385 0.506Geographical initial conditions:Town with Medieval Natural Harbour 0.168 0.375 0.769 0.439 ***

Coastal town 0.162 0.369 1.000 0.000 ***

Log (Dist. to Navigable River) 13.022 0.273 12.964 0.372Natural disasters, 1850–1900 2.060 2.203 1.308 2.250Historical measures:Medieval Muslim City 0.018 0.133 0.231 0.439Centuries Muslim Rule 3.465 1.020 2.998 1.176Town under Hindu ruler 0.635 0.483 0.462 0.519Town under Muslim ruler 0.084 0.278 0.231 0.439% Hindu/Sikh Partition Inflows 1931–51 1.150 1.031 1.160 1.017% Muslim Outflows 1931–51 1.486 1.818 2.462 2.512Demographic measures:Log(population 1991) 10.156 0.989 11.342 1.267 ***

Class 1 Town (100,000+) 0.078 0.269 0.385 0.506 *

Class 2 Town (50–100,000) 0.162 0.369 0.231 0.439Class 3 Town (25–50,000) 0.287 0.454 0.308 0.480Proportion Muslim 17.283 0.130 24.534 0.142 *

Prop. Scheduled Tribe in Town 4.936 8.736 3.433 5.412Prop. Scheduled Caste in Town 7.521 3.813 5.763 1.701 ***

Economic measures:Active port in 2001 0.072 0.259 0.692 0.480 ***

Worker’s participation rate 30.142 7.475 29.001 2.272Literacy rate 72.887 9.570 71.670 6.365

All measures based upon 1991 values unless otherwise mentioned. Welch t-test allows unequal variances:* Significant at 10%.

aI

tten

2vml(id

p

** Significant at 5%.*** Significant at 1%.

lso exploiting these as a useful placebo comparison to see whether it is in fact historical inter-ethnic complementarities as argue, or instead, modern trade that might explain the results. I show that modern port activity appears irrelevant.

Other dimensions in which medieval ports appear different is that they continue to have a larger proportion of Muslimshan the average Gujarati town (24% compared to 17.3%), and on average are significantly larger in population than otherowns in our sample.17 In the absence of inter-ethnic complementarity, these factors are also likely to raise the likelihood ofthnic rioting. Yet, despite these increased potential risks, medieval ports do not appear a priori statistically different fromon-ports in their probability of experiencing ethnic riots in 2002.

Table 2 examines the determinants of the log-normalized number of days of Hindu–Muslim rioting during February–April002. This measure arguably captures the relative intensity of ethnic rioting that is more straightforward to observe anderify than other measures of intensity, such as riot deaths or injuries (for these, please see Appendix Table 2). I first matchedieval ports to other towns by the initial geographical factors mentioned above that might influence initial medieval port

ocation (Col 1), then by medieval and colonial historical factors that might influence contemporary inter-ethnic relationsCol 2). I then compare medieval ports to other towns with similar modern (but pre-riot) economic and social characteristics,


ncluding Muslim population (Col 3), and compare towns within the same district (Col 4). All regressions control flexibly foristance to Godhra and town size.

17 The relatively small share of scheduled castes in medieval ports can be accounted for by the relatively high share of Muslims – it appears that theroportions of Hindus that are scheduled castes are fairly similar in medieval ports and other towns.

dx.doi.org/10.1016/j.jebo.2014.02.023



Table 2Regression (OLS): log. days of Hindu–Muslim riots, February–April 2002.

N = 180, clusters = 18 districts (1) (2) (3) (4) (5) (6) (7) (8)

Medieval Port −0.186 −0.208* −0.289** −0.292*** −0.331** −0.310** −0.323*** −0.323***

[0.136] [0.100] [0.110] [0.099] [0.121] [0.110] [0.103] [0.105]Coastal town 0.027 0.05 −0.003 −0.014 0.029 −0.032 −0.002 −0.041

[0.046] [0.050] [0.078] [0.126] [0.076] [0.120] [0.068] [0.109]Log (Dist. to Navigable River) 0.097 0.082 0.288** 0.035 0.283** 0.007 0.324*** 0.022

[0.102] [0.084] [0.106] [0.228] [0.102] [0.284] [0.097] [0.226]Natural disasters, 1850–1900 0.008 0.001 −0.01 −0.007 −0.015 −0.004 −0.014 −0.009

[0.019] [0.021] [0.018] [0.018] [0.017] [0.015] [0.017] [0.018]Medieval Muslim City 0.246 0.237 0.188 0.232 0.218 0.355 0.317

[0.393] [0.389] [0.373] [0.364] [0.359] [0.322] [0.325]Centuries Muslim Rule (to 1707) 0.068* 0.113** 0.064 0.118** 0.074 0.103** 0.06

[0.035] [0.044] [0.052] [0.043] [0.046] [0.045] [0.051]Hindu Colonial Ruler 0.095 0.053 0.144** 0.039 0.149** 0.019 0.103

[0.079] [0.086] [0.066] [0.076] [0.062] [0.076] [0.066]Muslim Colonial Ruler 0.173 0.087 0.240** 0.025 0.189 0.009 0.184*

[0.135] [0.141] [0.108] [0.095] [0.114] [0.119] [0.105]% Partition Majority Inflows (1931–51) 0.021 −0.007 0.03 0.003 0.037 −0.009 0.019

[0.033] [0.032] [0.029] [0.031] [0.029] [0.030] [0.028]% Partition Muslim Outflows (1931–51) −0.029 −0.025 −0.066 −0.027 −0.055 −0.016 −0.044

[0.017] [0.019] [0.041] [0.017] [0.046] [0.016] [0.040]Prop. Scheduled Tribe in Town −0.014*** −0.017** −0.013*** −0.019** −0.013*** −0.018**

[0.004] [0.007] [0.004] [0.007] [0.004] [0.007]Prop. Scheduled Caste in Town 0.002 0.004 0.001 0.002 0.001 0.004

[0.008] [0.008] [0.008] [0.009] [0.008] [0.008]Active port in 2001 0.133 0.128 0.031 0.079 0.041 0.063

[0.145] [0.174] [0.171] [0.211] [0.161] [0.185]Workers’ Participation Rate 0.003 0.002 0.003 0.002 0.003 0.002

[0.008] [0.008] [0.007] [0.007] [0.008] [0.008]Literacy Rate 0.002 −0.001 0.002 −0.002 0.001 −0.003

[0.005] [0.005] [0.005] [0.005] [0.005] [0.005]Prop. Muslim in Town, 1981 1.158** 0.811 1.133** 0.859* 1.065** 0.806*

[0.513] [0.469] [0.471] [0.463] [0.480] [0.453]Prop. Muslim in Town2, 1981 −1.232* −1.026 −1.089* −0.988 −1.005 −0.95

[0.625] [0.636] [0.610] [0.633] [0.613] [0.606]Vote Margin <5% 98 −0.112 −0.125**

[0.079] [0.057]Med. Port × Vote Margin <5% −0.473* −0.328

[0.264] [0.263]Med. Muslim City × Vote Margin <5% 1.433** 0.916

[0.538] [0.750]Log Vote Margin 98 −0.011 −0.014

[0.028] [0.026]Med. Port × Log. Vote Margin 98 0.204 0.202

[0.127] [0.126]Med. Muslim City × Log. Vote Margin 98 −0.532*** −0.459**

[0.177] [0.173]Joint-F Med Port vars. 6.70 7.83 6.32 5.85Prob > F 0.01 0.00 0.01 0.01Incremental Controls Geog Colonial Modern Modern Modern Modern Modern Modern1981 District FE N N N Y N Y N YR-squared 0.46 0.49 0.53 0.61 0.55 0.62 0.56 0.62

Notes: Geographical controls include log. distances from navigable rivers, coastal town, Natural Disasters, log. population (1991), Class of town (I,II,III) anddistance to Godhra (100, 200, 300 km). Colonial controls add: Centuries Muslim rule, Muslim-founded or capital, % Partition Hindu/Sikh Inflows and MuslimOutflows. Modern controls add: Proportion SC/ST, Modern active port, Literacy Rate, Worker’s Participation Rate, Proportion Muslim 1981 (and a quadraticterm). All values observed in 1991 unless noted. All interactions are demeaned. Robust standard errors in brackets (clustered at 1981 district):

* Significant at 10%.** Significant at 5%.

*** Significant at 1%.

As Table 2 suggests, medieval ports experienced between 18.6% and 29.2% fewer days of Hindu–Muslim rioting betweenFebruary and April 2002, comparing both between and within districts. This effect strengthens when comparing townswith similar political and colonial histories, modern economic and social characteristics and comparing towns within thesame district. Note that political histories and contemporary ethnic composition also increased the intensity of the rioting:


comparing across districts, towns that spent longer under Muslim rule had more days of rioting (Cols 2–3), as did townswith a greater share of Muslims in their population. However, despite having more Muslims that otherwise similar towns,medieval ports exhibited less rioting. Note also that a number of modern economic indicators, such as the literacy rate, the

dx.doi.org/10.1016/j.jebo.2014.02.023

G ModelJ

w(

IrabbAei

lpcwi

tiabrgplt

tirst

cmftpmc

cco

s2BcI

bhc

p

ae



orker’s participation rate or our placebo comparison – whether a town was an active modern port – seem to be irrelevantCols 3–4).

Columns 5–8 examine how the effect of historical inter-ethnic incentives differ with the extent of political competition. use two different measures of political competition: a discrete measure – whether the vote margin between the victor andunner up was less than 5% in 1998 – and a continuous measure – the logarithm of the vote margin in 1998. Notice that theverage effect of a medieval port legacy strengthens with these controls (Cols 5–8), and the point estimate of the interactionetween medieval port is consistently negative on the 5% vote margin dummy and positive on the log vote margin measure,oth suggesting that medieval ports that were the focuses of greater political competition exhibited fewer days of rioting.s the F-tests reveal, the average medieval port effect and the interaction are jointly significant in all specifications. Theseffects are consistent with the role of historic inter-ethnic complementarities and supporting organizations in mitigatingncentives for violence.

Columns 5–8 also provide an interesting comparison with medieval Muslim patronage towns, which as we have arguedacked such incentives and were more likely to be centers of economic competition. Notice that though medieval Muslimatronage towns do not appear to reveal significant increases on average in the days of rioting in 2002, and our politicalompetition measures themselves appear to have weak average effects on violence, medieval Muslim patronage towns thatere the focuses of political competition saw much greater rioting. This is again consistent with our theoretical discussion:

t is where political competition and inter-ethnic economic competition coincide where ethnic rioting is more likely.One concern may be that though the (log. normalised) number of days of rioting in the two month period captures

he relative intensity of ethnic rioting in a relatively observable manner, such measures may still be susceptible to over-nfluence by outlier towns where the rioting was most severe. Table 3 examines the extensive margin – the probability thatny riot occurs in a town during this period. Notice that a medieval trade legacy reduces the probability of a riot occurringy between 13.1 percentage points and 21.7 percentage points. The determinants of the extensive margin of riots show aobust consistency with those of the intensive margin in other dimensions as well. Once again, medieval ports that experiencereater political competition show lowered probabilities of rioting, while medieval patronage towns that experience greaterolitical competition exhibit greater probabilities of an ethnic riot occurring. Once again, a history of Muslim rule raises the

ikelihood that any riot occurs, and once again, our placebo comparison – modern ports – seem as likely as any other towno have an ethnic riot.18

Table 4 examines the determinants of the vote share change of the incumbent ruling party, the BJP, in the elections thatook place immediately after the riots. Notice that despite an average increase of 4.84 percentage points across the staten BJP support following the violence, medieval ports swung 6.28 percentage points against the BJP in the aftermath of theiots, comparing towns within the same district (Col 4, 6, 8). Towns within the same district that were exposed to greaterpells of Muslim rule also appear to have swung towards the BJP following the violence, consistent with the interpretationhat rather than reducing biases, longer Muslim rule may have generated greater inter-ethnic competition.19

The BJP also appears to have made particular gains in close elections, consistent with a successfully targeted electionampaign (Cols 5–8). Though medieval ports experienced an average swing against the BJP, and politically competitiveedieval ports also experienced a lowered propensity for ethnic rioting (Tables 2 and 3), the BJP’s losses in vote share

ollowing the riots does not seem to have been greater when medieval ports were more politically competitive. In con-rast, the BJP’s gains appear to have been accentuated in Muslim medieval patronage cities that had previously beenolitically competitive. As we have seen, these towns also were both more likely to have experienced an ethnic riot andore days of rioting (Tables 2 and 3), and are arguably locations where political competition and inter-ethnic competition

oincided.A related approach is rather than look to the 1998 elections for a measure of subsequent political competition, we

an examine whether the ultimate winner of the 2002 elections was predictable by historic incentives for inter-ethnicomplementarity and competition even in those races that ended up being very close in 2002. Table 5 presents the probabilityf a BJP seat win looking only at those towns in which the winner won by less than 5% (Cols 1–3) and 10% (Cols 4–6).20

In fact, as Table 5 suggests, even in the Gujarat elections that followed the 2002 riots, there appears to have been notrong incumbency effect for BJP candidates (Cols 3, 6). Yet, even in these close races, the probability that the winner of the002 election was from the BJP was heavily influenced by the nature of historical incentives. Relative to other close races, the


JP was less likely to win in medieval ports following the riots, and significantly more likely to win in medieval patronageities. These results suggest that the outcome of Gujarat’s closest elections following the riots were not ‘as good as random’.nstead, these findings appear consistent with the reduced effectiveness of the BJP’s broader campaign to win close elections

18 Appendix Table 2 examines the determinants of another measure of riot intensity: the log. number of deaths and injured per capita that occurredetween February and April 2002. Though this measure is particularly susceptible to outliers, such as Ahmadabad, that experienced more than threeundred deaths alone, the sign of both the average effect of a medieval port legacy and the interactions with our political competition measure areonsistent with the patterns above, and in some specifications marginally significant.19 Ironically, as Table 4 suggests, the BJP lost vote share in towns in districts which had already experienced greater ethnic cleansing of the Muslimopulation during the Partition of 1947, and but made gains in towns that had sustained a greater subsequent Muslim population.20 The limitations of the regression discontinuity approach applied to elections are still an open research area, with a measurable incumbency advantageppearing in very close US elections (Caughey and Sekhon, 2011; Grimmer, 2011) but less evidence for incumbency differences in non-US elections (Eggerst al., 2013).

dx.doi.org/10.1016/j.jebo.2014.02.023



Table 3Regression (OLS): probability of a Hindu–Muslim riot, February–April 2002.


Medieval Port −0.131* −0.140** −0.186** −0.168** −0.217** −0.179** −0.214*** −0.191**

[0.075] [0.051] [0.064] [0.063] [0.075] [0.080] [0.073] [0.083]Coastal town 0.024 0.024 0.008 0.008 0.034 −0.005 0.011 −0.011

[0.037] [0.046] [0.054] [0.084] [0.054] [0.083] [0.047] [0.077]Log (Dist. to Navigable River) −0.050 −0.118 0.018 −0.075 0.010 −0.122 0.040 −0.107

[0.071] [0.090] [0.080] [0.287] [0.078] [0.291] [0.065] [0.306]Natural disasters, 1850–1900 0.005 −0.006 −0.014 −0.006 −0.019* −0.005 −0.018 −0.008

[0.014] [0.015] [0.013] [0.018] [0.010] [0.017] [0.011] [0.017]Medieval Muslim City 0.066 0.051 −0.002 0.045 0.020 0.138 0.085

[0.229] [0.229] [0.251] [0.240] [0.253] [0.169] [0.228]Centuries Muslim Rule (to 1707) 0.046 0.072* 0.002 0.077** 0.011 0.067* 0.002

[0.029] [0.036] [0.045] [0.034] [0.038] [0.035] [0.042]Hindu Colonial Ruler −0.022 −0.049 0.037 −0.057 0.048 −0.075 0.012

[0.079] [0.082] [0.057] [0.075] [0.053] [0.076] [0.063]Muslim Colonial Ruler 0.108 0.046 0.207 0.002 0.187 −0.007 0.192

[0.134] [0.145] [0.154] [0.111] [0.148] [0.132] [0.157]% Partition Majority Inflows (1931–51) 0.001 −0.018 0.018 −0.011 0.025 −0.020 0.009

[0.028] [0.028] [0.018] [0.029] [0.022] [0.027] [0.019]% Partition Muslim Outflows (1931–51) −0.019 −0.017 −0.035 −0.018 −0.030 −0.011 −0.021

[0.017] [0.020] [0.043] [0.019] [0.043] [0.019] [0.043]Prop. Scheduled Tribe in Town −0.009** −0.010 −0.009* −0.012* −0.009* −0.011

[0.004] [0.006] [0.004] [0.006] [0.004] [0.007]Prop. Scheduled Caste in Town 0.005 0.010 0.004 0.008 0.005 0.010

[0.008] [0.009] [0.008] [0.009] [0.008] [0.009]Active port in 2001 0.060 0.038 −0.014 0.004 −0.011 −0.013

[0.092] [0.117] [0.114] [0.150] [0.099] [0.126]Workers Participation Rate 0.002 0.002 0.002 0.003 0.002 0.002

[0.007] [0.008] [0.006] [0.007] [0.007] [0.008]Literacy Rate 0.003 0.001 0.003 0.001 0.002 0.000

[0.003] [0.003] [0.003] [0.003] [0.003] [0.003]Prop. Muslim in Town, 1981 0.952* 0.488 0.931* 0.519 0.873 0.473

[0.516] [0.569] [0.500] [0.554] [0.509] [0.563]Prop. Muslim in Town2, 1981 −0.967 −0.426 −0.854 −0.388 −0.784 −0.354

[0.618] [0.694] [0.602] [0.688] [0.585] [0.667]Vote Margin <5% 98 −0.098 −0.125**

[0.060] [0.052]Med. Port × Vote Margin <5% −0.327* −0.259

[0.188] [0.198]Med. Muslim City × Vote Margin <5% 1.042** 0.625

[0.379] [0.459]Log Vote Margin 98 −0.002 0.000

[0.026] [0.029]Med. Port × Log. Vote Margin 98 0.173* 0.179

[0.094] [0.106]Med. Muslim City × Log. Vote Margin 98 −0.400*** −0.314*


Notes: Geographical controls include log. distances from navigable rivers, coastal town, Natural Disasters, log. population (1991), Class of town (I, II, III) anddistance to Godhra (100, 200, 300 km). Colonial controls add: Centuries Muslim rule, Muslim-founded or capital, % Partition Hindu/Sikh Inflows and MuslimOutflows. Modern controls add: Proportion SC/ST, Modern active port, Literacy Rate, Worker’s Participation Rate, Proportion Muslim 1981 (and a quadraticterm). All values observed in 1991 unless noted. All interactions are demeaned. Robust standard errors in brackets (clustered at 1981 district):



by adopting a platform complicit with ethnic violence in towns with close races that enjoyed inter-ethnic complementarityand the higher dividends of such an approach in similar towns where ethnic groups have historically competed.

6. Robustness and mechanisms


So far we have argued that the effect of medieval ports reflects historical incentives for inter-ethnic complementarityand the accompanying community organizations that emerged to support such inter-ethnic exchange. Another possibilityis that medieval ports simply just happened to have less violent individuals or that attempts to instigate violence in this

dx.doi.org/10.1016/j.jebo.2014.02.023



Table 4Regression (OLS): change in ruling party (BJP) vote %, 1998–2002.


Medieval Port −1.665 −1.200 −4.508* −6.280** −4.785* −6.365** −4.549* −6.339**

[2.745] [1.740] [2.442] [2.799] [2.350] [2.763] [2.508] [2.722]Medieval Muslim City −1.943 −1.866 −1.544 −2.155 −1.952 −0.777 −1.407

[3.018] [2.659] [3.679] [3.445] [3.828] [3.438] [3.752]Centuries Muslim Rule (to 1707) 1.886 3.669*** 3.860*** 3.394*** 3.424** 3.225*** 2.675**

[1.257] [0.836] [1.179] [0.872] [1.240] [0.887] [1.125]% Partition Muslim Outflows (1931–51) −1.533** −1.462*** −2.171 −1.355** −1.850 −1.244** −1.620

[0.602] [0.487] [1.906] [0.472] [2.012] [0.504] [1.704]Prop. Muslim in Town, 1981 39.174** 26.089 39.996** 26.097 39.438** 24.744*

[18.520] [17.574] [18.666] [16.814] [17.780] [13.826]Prop. Muslim in Town2, 1981 −48.177 −34.989 −48.752 −37.050 −47.948 −34.613

[28.976] [25.976] [29.579] [26.347] [28.900] [24.194]Vote Margin <5% 98 2.403 6.857**

[3.454] [3.159]Med. Port × Vote Margin <5% −0.154 5.721

[5.133] [5.008]Med. Muslim City × Vote Margin <5% 10.639* −3.483

[5.937] [7.302]Log Vote Margin 98 −1.598 −3.613***

[1.456] [0.977]Med. Port × Log. Vote Margin 98 0.656 0.001

[2.509] [2.067]Med. Muslim City × Log. Vote Margin 98 −4.771** −1.342



pePte

TR

Np



eriod just happened to be less common. Table 6, Panel A presents the determinants of whether a town experienced anythnic violence (both riots and isolated incidents of attempted violence) during the two months of the Godhra riots, while


anel B examines the probability a town experienced attempted violence but no escalation that resulted in riots. Noticehat despite significantly less rioting than otherwise similar towns, medieval ports appear to be as likely as other towns toxperience violent incidents. In fact, medieval ports that were the focuses of greater political competition appear, if anything,

able 5egression (OLS): BJP winner, close elections sub-sample, 2002.

Clusters = 16 districts (1) (2) (3) (4) (5) (6)

2002 Vote Margin Within 10% 10% 10% 5% 5% 5%Medieval Port −0.510*** −0.543*** −0.527*** −0.402** −0.426** −0.425**

[0.151] [0.140] [0.118] [0.185] [0.181] [0.184]Medieval Muslim City 0.814*** 0.810*** 0.771***

[0.197] [0.223] [0.243]Log (population), 1991 0.049 0.048 0.042 0.017 0.024 0.022

[0.079] [0.082] [0.079] [0.096] [0.092] [0.096]Centuries Muslim Rule (to 1707) −0.038 −0.027 −0.061 −0.062

[0.084] [0.086] [0.120] [0.126]Prop. Muslim in Town, 1981 0.796 1.060 0.481 0.402

[1.151] [0.991] [1.203] [1.029]Prop. Muslim in Town2, 1981 −1.145 −1.436 −0.919 −0.823

[1.445] [1.241] [1.438] [1.151]BJP incumbent 1998 0.148 −0.042

[0.180] [0.302]Observations 83 83 83 55 55 55R-squared 0.09 0.1 0.12 0.05 0.07 0.07

otes: Robust standard errors in brackets (clustered at 1981 district): * Significant at 10%; ** Significant at 5%; *** Significant at 1%. No Medieval Muslimatronage city fell within the 5% vote margin in 2002.

dx.doi.org/10.1016/j.jebo.2014.02.023



Table 6Regression (OLS): ethnic violence, February–April 2002.


Panel A: any ethnic violence (including isolated incidents)Medieval Port −0.015 −0.048 −0.057 −0.106 −0.073 −0.104 −0.067 −0.088

[0.150] [0.134] [0.134] [0.104] [0.146] [0.114] [0.117] [0.082]Medieval Muslim City 0.088 0.051 0.080 0.069 0.088 0.041 −0.007

[0.242] [0.249] [0.244] [0.251] [0.246] [0.223] [0.153]Vote Margin <5% 98 −0.110 −0.086

[0.073] [0.050]Med. Port × Vote Margin <5% −0.481 −0.137

[0.327] [0.218]Med. Muslim City × Vote Margin <5% 0.583* 0.068

[0.289] [0.312]Log Vote Margin 98 0.058* 0.041

[0.030] [0.025]Med. Port × Log. Vote Margin 98 0.043 −0.096

[0.147] [0.121]Med. Muslim City × Log. Vote Margin 98 0.036 0.312**

[0.165] [0.137]Joint-F Med Port vars. 3.84 1.29 0.62 0.68Prob > F 0.04 0.30 0.55 0.52R-squared 0.40 0.45 0.48 0.66 0.49 0.67 0.49 0.68Panel B: ethnic violence that did not escalate to a riotMedieval Port 0.116 0.092 0.129 0.061 0.144 0.074 0.148 0.104

[0.107] [0.122] [0.132] [0.108] [0.136] [0.107] [0.115] [0.100]Medieval Muslim City 0.023 0.000 0.082 0.023 0.068 −0.097 −0.092

[0.351] [0.382] [0.406] [0.397] [0.410] [0.258] [0.233]Vote Margin <5% 98 −0.012 0.039

[0.050] [0.043]Med. Port × Vote Margin <5% −0.154 0.121

[0.282] [0.206]Med. Muslim City × Vote Margin <5% −0.458 −0.557

[0.355] [0.530]Log Vote Margin 98 0.060** 0.041**

[0.025] [0.017]Med. Port × Log. Vote Margin 98 −0.130 −0.275**

[0.086] [0.098]Med. Muslim City × Log. Vote Margin 98 0.436** 0.626***



* Significant at 10%.
** Significant at 5%.

more likely to experience violence that nevertheless did not escalate into riots (Panel B, Cols 7–8). Rather than stopping anyprecipitating events, the major effect of a medieval trade legacy appears to be to reduce the escalation of religious violenceinto broader mob confrontation. This is consistent with the presence of community-level incentives provided by institutionalmechanisms that mitigate shocks to inter-ethnic peace, rather than the presence of inherently peaceful individuals or thelack of agents provocateur in medieval ports.

If, as I have argued, medieval ports have less support for political platforms complicit with ethnic violence, then loweredpolitical support for the BJP might also be expected in the aftermath of the preceding major wave of ethnic violence between1990 and 1993. This wave of violence culminated in widespread riots following the destruction of the Babri Mosque in Ayod-hya in Uttar Pradesh in December 1992, with Gujarati towns relatively less affected (the previous peak of 35 Hindu–Muslimriots in Gujarat in 1990 compares to 105 riots in 2002 (Varshney and Wilkinson, 2004)). Yet, as Table 7 (Cols 1–4) suggest,medieval ports also gave the BJP around 6–7 percentage points less support than otherwise similar towns in the subsequent1995 elections.


Another potential concern is that perhaps the vote swing against the BJP visible in medieval ports in 2002 is simply due toregression to the mean, perhaps due to higher pre-2002 support for the BJP in medieval ports. As Table 7 (Cols 5–8) reveals,despite lowered support in 1995, medieval ports were not significantly more likely than towns with similar geographical,

dx.doi.org/10.1016/j.jebo.2014.02.023



Table 7Regression (OLS): ruling party (BJP) vote %, 1995 and 1998.

Obs = 180, clusters = 18 districts (1) (2) (3) (4) (5) (6) (7) (8)

BJP vote share in: 1995 1995 1995 1995 1998 1998 1998 1998Medieval Port −7.692** −6.360** −4.363 −6.095 2.264 0.821 1.507 2.386

[2.693] [2.912] [4.389] [4.213] [3.279] [2.927] [3.117] [3.136]Medieval Muslim City 0.488 −1.58 −2.473 11.349** 9.426* 9.616*

[6.870] [6.769] [8.069] [5.366] [5.179] [5.416]R-squared 0.15 0.21 0.26 0.44 0.22 0.26 0.30 0.43Incremental Controls Geog Colonial Modern Modern Geog Colonial Modern Modern1981 District FE N N N Y N N N Y


*

h2ie

ciaeeap

pepe

arafb

iecartc

7

oeruw

* Significant at 10%.** Significant at 5%.** Significant at 1%.

istorical or modern characteristics to oppose or support the BJP in the elections in the 1998 elections that preceded the002 violence.21 In contrast, medieval Muslim patronage centers, which, as we have seen, showed more support for the BJP

n close elections in 2002, were also more likely to support the BJP in 1998. Both of these patterns appear inconsistent withffects driven by regression to the mean.

The focus of our analysis has been on the effects of historical patterns of inter-ethnic complementarity and economicompetition on the incentives for ethnic riots and the subsequent political support of agents of the party seen to be complicitn that riots. A separate question is whether ethnic violence itself affects political support directly, or both ethnic violencend political support are chiefly a manifestation of historic inter-ethnic incentives and other factors. As I have argued, thescalation of precipitating acts of ethnic violence to full-scale mob riots depends directly on local support and thus is clearlyndogenous to local incentives in the towns involved. Further, it is likely, as the theoretical framework describes, that politicalgents will have incentives to target the organization of violence more in areas where it is more likely to be successful as aolitical strategy.

However, as Table 6 suggests, the locations of violent events, including isolated acts, appear less clearly selected. Medievalorts were as likely as other towns to experience such violence, and there were a number of ‘failed experiments’: inter-thnic violent acts which did not escalate into rioting. Insofar as there is a random component to the extent to which theserecipitating events were attempted or occurred in otherwise similar towns matched along a range of observables, I canxamine whether towns that isolated actors may have ‘intended to treat’ with riots show differential voting patterns.

Table 8 (Cols 1–5) examines whether a town that exhibited any ethnic violence (including isolated incidents) showed greater vote swing towards or away from the ruling party following the violence. Notice that there appears to be noesidual correlation of violence on the ruling party’s vote share on average. However, medieval ports that experiencedn act of ethnic violence actually reduced their support for the BJP by an additional 7 percentage points. This providesurther suggestive evidence that is consistent the framework: attempting ethnic violence in medieval ports may have in factackfired electorally, by instead mobilizing those groups opposed to ethnic pogroms.

Table 8 (Cols 6–10) examines the residual correlation between the number of days of the ethnic riots that occurredn a town and its support for the ruling party. Note that, on average, towns with similar pre-independence histories thatxperienced more extensive rioting do show increased subsequent support for the ruling party (Cols 6–7). The residualorrelation is weakened by controls for ethnic demography, modern economic factors and district fixed effects, however,nd the interactions with proxies for historical incentives are insignificant. This may be because of the spillover effects ofiots: riots may alter voting behaviour both where they occur and elsewhere. But the results are also consistent with bothhe propensity for riots and subsequent voting being functions of local support that itself may reflect deeper historical andontemporary economic and political incentives.

. Discussion

This paper argues that there exists a fundamentally conditional relationship between political competition and the naturef inter-ethnic economic incentives for exchange in generating incentives for ethnic violence. While it is likely that marginallectoral constituencies in general are likely to be the focus of more intensive political campaign efforts than safe seats, the


esults suggest that the extent that such intensive campaigns manifest themselves in ethnic riots can depend importantlypon the legacy of historical incentives for trade between communities. Where marginal electoral constituencies coincidedith or reflect pre-existing historic inter-ethnic economic competition, as occurred in medieval Muslim patronage cities in

21 Between 1995 and 1998, Gujarati towns experienced four Hindu–Muslim riots in total.

dx.doi.org/10.1016/j.jebo.2014.02.023



Table 8Regression (OLS): change in ruling party (BJP) vote %, 1998–2002: violence interactions.

N = 180, clusters = 18 districts (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

Medieval Port −1.597 −1.062 −4.386* −6.318** −6.398** −0.568 −0.183 −3.853 −5.985** −5.994**

[2.493] [1.695] [2.520] [2.800] [2.365] [2.524] [1.839] [2.480] [2.665] [2.727]Ethnic Violence (incl. Isolated) 4.426 2.852 2.137 −0.362 −0.622

[2.799] [2.887] [2.505] [2.502] [2.423]Med. Port × Ethnic Violence −6.982*

[3.741]Med. Muslim Cty × Ethnic Violence 6.273

[5.080]Log Riot Days 5.900** 4.877* 2.269 1.008 0.692

[2.469] [2.465] [2.446] [1.817] [1.985]Med. Port × Log Riot Days 0.439

[2.976]Med. Muslim Cty × Log Riot Days 1.616

[1.910]Medieval Muslim City −2.195 −1.974 −1.515 −3.842 −3.142 −2.405 −1.733 −2.98

[3.015] [2.802] [3.724] [5.179] [2.631] [2.654] [3.589] [3.694]Joint-F Med Port vars. 6.18 2.87Prob > F 0.01 0.08Incremental Controls Geog Colonial Modern Modern Modern Geog Colonial Modern Modern Modern1981 District FE N N N Y Y N N N Y YR-squared 0.36 0.4 0.48 0.61 0.62 0.38 0.42 0.48 0.62 0.62




Gujarat from 1998 to 2002, politicians appear to have gained local incentives to foster ethnic mobilization and violence. Incontrast, where political competition occurred in societies with historic incentives for inter-ethnic trade and tolerance, suchas Gujarat’s medieval ports, attempts to foment riots appear to have failed. These towns also appear to have sanctioned theparty seen as complicit in the violence at the polls.

The findings suggest that local historical incentives for inter-ethnic exchange and competition can have important elec-toral consequences, particularly in environments where ethnic violence can be used as a campaign strategy. The framework,however, raises a further intriguing possibility. State politicians seeking to win marginal constituencies where ethnic groupscompete should emphasize the preferences of the majority in these communities. This may lead them to adopt platformscomplicit with violence state-wide, leading to a ‘contagion’ of ethnic rioting. Yet, the corollary may also be true: state politi-cians seeking to win marginal constituencies where ethnic groups enjoy complementary economic relations should also faceincentives to emphasise these voters’ preferences more relative to their more polarized constituents, leading potentially toa de-escalation in state-wide platforms of ethnic hatred. While proper testing of this conjecture would require multiplestate-elections, and is a topic of on-going research, the Gujarat case does provide some corroboration.

Indeed, five years after the Gujarat riots, in 2007, the state prepared itself for another election. As Fig. 3 and Table 1suggest, the average margin of victory in medieval ports fell from 18.8% in 1998 to 8.89% in 2002, with 40% of seats inGujarati medieval ports won by margins of less than 5%. Although there was an increase in electoral competition on average,which might be expected to increase the intensity of local campaigning, there was no surge in electoral rioting in the stateand the incidence of riots remained below the levels of rioting that had occurred even in non-election years prior to thepogrom. Narendra Modi, the Chief Minister, campaigned explicitly on a platform of economic development, with ethnicmobilization conspicuously absent from the ruling party platform.22 Dissident elements in the state ruling party, includingalleged stalwarts of the 2002 riots like the then state home minister, rallied against the party leadership, accusing the ChiefMinister of betraying the Hindu community and no longer being worthy of the honorific “Hindu Hriday Samrat” (Emperor


of Hindu Hearts).23 Instead, observers noted that Modi explicitly reached out in his election campaign to Muslims from theMemon, Khoja and Bohra communities, each of whom had their origins in India as medieval traders (Engineer, 2008).24 In aremarkable reversal, de-emphasizing ethnic mobilizing appears to have helped the BJP win an unprecedented third term in

22 In an interview to the Hindu newspaper, when “asked where Muslims figured in his vision of Gujarat, he flared up: I don’t like this thinking. I workfor five-and-a-half crore Gujaratis. For me, anyone who lives here is a Gujarati, and I will not allow politics to come into this.” Vidya Subramaniam, “TheMuslim Question in Gujarat” The Hindu, October 9th, 2007.

23 See footnote 22.24 While accounts agree that these groups were viewed as marginal constituencies, accounts differ on why these communities were plausibly persuadable

despite the pogroms in which they also were targeted. Engineer (2008), himself a prominent left-wing dissident Bohra, argues that Bohras and other GujaratiMuslim trading communities “are rich traders and they will be as much attracted by the development discourse as upper caste Hindu Gujaratis.” In contrast,

dx.doi.org/10.1016/j.jebo.2014.02.023



Fa

2easc

wlt–tdthec

A

j

R

a

AAB

B

B

th

ig. 3. Change in Ruling Party Vote Share, 1998–2002. Relative to other towns where the BJP gained vote share following the riots, medieval ports swanggainst the BJP and became more competitive following the 2002 elections.

007 and a fourth term in 2012, with a significant degree of Muslim support, particularly among trading communities, andven winning in Muslim majority constituencies, without having fielded a single Muslim candidate. Like Uhuru Kenyatta,nother alleged orchestrator of ethnic violence, Narendra Modi has gone from being internationally-reviled and potentiallyubject to criminal prosecution for his alleged role in state-sponsored pogroms to being seen as a credible Prime Ministerialandidate.

Could it be that the anonymous police officer in Ahmadabad was right, that the ethnic violence in 2002, arguably India’sorst since the Partition, was ‘unfinished business’ that would result in less ethnic tension thereafter? That officer’s view,

ike that of some in the Jim Crow South (Bleakley and Owens, 2010) was that the minority would ‘learn its place’, and ceaseo compete for political patronage or resources. However, it does not appear that opponents of violent ethnic domination

whether among the ethnic minority or the majority – ‘learned their place’: towns with a long history of inter-ethnicolerance voted strongly against the ruling party in the aftermath of the violence, when such action might yet have provenangerous and costly. By doing so, they may have also subsequently rendered themselves electorally more competitive andhus more influential in shaping state-wide political platforms on the issue of ethnic mobilization. Electoral competitionas long been seen as a source of ethnic mobilization and violence. Yet, even in highly ethnically-polarized states, whenlectoral competition becomes located among communities with local traditions of tolerance and inter-ethnic trust, suchompetition may actually generate broader beneficial spillovers of peace.

ppendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/.jebo.2014.02.023.

eferences

l-Din al Malibari, Z., 1528. Tohfut-ul-mujahideen. Oriental Translation Fund of Great Britain, London, Translated from the Arabic by Rowlandson, M.J. in1833.

lberuni, A.-R.M.I.A., 1030. Indica, 2005 ed. Munshiram Manoharlal, Delhi. Edited by Sachau, Edward C.lesina, A., La Ferrara, E., 2005. Ethnic diversity and economic performance. Journal of Economic Literature 43 (3), 762–800.anerjee, A., Iyer, L., Somanathan, R., 2008. Public action for public goods. In: Paul Schultz, T., Strauss, J. (Eds.), Handbook of Development Economics, vol.

4. Elsevier.


arbosa, D., 1519. A Description of the Coasts of East Africa and Malabar in the Beginning of the Sixteenth Century. Hakluyt Society, London, Translatedfrom the Portuguese by Stanley, Henry E. J. in 1866.

eaman, L., Chattopadhyay, R., Duflo, E., Pande, R., Topalova, P., 2009. Powerful women: does exposure reduce bias? Quarterly Journal of Economics 124(4), 1497–1540.

he Bohra activist J.S. Bandukwala claims that: “The Syedna [the Bohra’s spiritual leader] will always seek a cordial relation with the party in power. It is inis interests, and as he sees it, in the interests of the community”. See Raheel Dhattiwala “Muslim as BJP supporter in Gujarat”, The Hindu, 2013.

dx.doi.org/10.1016/j.jebo.2014.02.023

http://dx.doi.org/10.1016/j.jebo.2014.02.023

http://dx.doi.org/10.1016/j.jebo.2014.02.023

http://refhub.elsevier.com/S0167-2681(14)00068-7/sbref0005
































































































G Model

Becker, S.O., Boeckh, K., Hainz, C., Woessmann, L., 2012. The empire is dead, long live the empire! Long-run persistence of trust and corruption in thebureaucracy. In: CEPR Discussion Paper 8288.

Bharadwaj, P., Khwaja, A., Mian, A., 2008. The big March: migratory flows after the partition of India. Economic and Political Weekly 43 (August (35)).Blake, S.P., 1991. Shahjahanabad: The Sovereign City in Mughal India, 1639–1739. Cambridge University Press.Blakeslee, D., 2013. Propaganda and Ethno-Religious Politics in Developing Countries: Evidence from India, May. mimeo, Columbia.Bleakley, H., Owens, E., 2010. Violence Beyond Reason: Prohibition of Alcohol and the Decline of Lynching in the US South. mimeo, Chicago GSB.Bohlken, A.T., Sergenti, E., 2010. Ethnic violence and economic growth: an empirical investigation of Hindu–Muslim riots in India. Journal of Peace Research

47 (September (5)), 589–600.Brass, P.R., 1997. Theft of an Idol. Princeton University Press, Princeton, NJ.Brass, P.R., 2003. The Production of H indu–Muslim Violence in Contemporary India. University of Washington Press, Seattle.Burman, J.J.R., 2005. Gujarat Unknown: Hindu–Muslim Syncretism and Humanistic Forays. Mittal, New Delhi.Canes-Wrone, B., Herron, M., Shotts, K., 2001. Leadership and pandering: a theory of executive policymaking. American Journal of Political Science 45 (July

(3)), 532–550.Casson, L. (Ed.), 1989. The Periplus Maris Erythraei. Princeton, Princeton.Caughey, D., Sekhon, J.S., 2011. Elections and the regression discontinuity design: lessons from close U.S. House Races, 1942–2008. Political Analysis 19,

385–408.Chakravarti, R., 2000. Nakhudas and Nauvittakas: ship-owning merchants in the West Coast of India (C. AD 1000–1500). Journal of the Economic and Social

History of the Orient 43 (February (1)), 34–64.Chandra, K., 2007. Why Ethnic Parties Succeed: Patronage and Ethnic Head Counts in India. Cambridge University Press.Chaudhuri, K.N., 1995. Trade and Civilisation in the Indian Ocean: An Economic History from the Rise of Islam to 1750. Cambridge University Press,

Cambridge.Chen, D., 2010. Club goods and group identity: evidence from Islamic Resurgence during the Indonesian Financial Crisis. Journal of Political Economy 118

(2), 300–354.Cox, G.W., Munger, M., 1989. Closeness, expenditures, and turnout in the 1982 U.S. House Elections. American Political Science Review 83 (March (1)),

217–231.Dhattiwala, R., Biggs, M., 2012. The political logic of ethnic violence: the anti-Muslim pogrom in Gujarat 2002. Politics and Society 40 (December (4)),

483–516.di Verthema, L., 1503. The navigation and voyages of Lewes Vertomannus, gentleman to the citie of Rome. Hakluyt, transl. from the Latin by Eden, Richard,

1576.Diaz-Cayeros, A., Jha, S., 2013 February. Can Contract Failures Lead to Ethnic Assimilation? Evidence from Cochineal in Mexico. mimeo, Stanford Graduate

School of Business.Easterly, W., Levine, R., 1997. Africa’s growth tragedy: policies and ethnic divisions. Quarterly Journal of Economics 111 (4), 1203–1250.Eggers, A., Folke, O., Fowler, A., Hainmueller, J., Hall, A., Snyder, J., 2013. On the Validity of the Regression Discontinuity Design for Estimating Electoral

Effects: New Evidence from over 40,000 Close Races. June, mimeo, Harvard.Eifert, B., Miguel, E., Posner, D.N., 2010. Political competition and ethnic identification in Africa. American Journal of Political Science 54 (April (2)),

494–510.Engineer, A.A., 2008. Gujarat Elections and Aftermath. In: Secular Perspective, January 1–15.Erikson, R., Palfrey, T., 2000. Equilibria in campaign spending games: theory and evidence. American Political Science Review 94 (3), 595–609.Esteban, J., Ray, D., 2008. On the salience of ethnic conflict. American Economic Review 98 (December (5)).Fearon, J.D., Laitin, D.D., 2003. Ethnicity, insurgency and civil war. American Political Science Review 97 (1), 75–90.Ferishta, M.K., 1620. The History of Hindustan, 2nd ed. Dahinam Publishers, New Delhi, Translated from the Persian by Alexander Dow, 1770, and edited

by B.S. Dahiya, 2003.Field, E., Levinson, M., Pande, R., Visaria, S., 2008. Segregation rent control and riots: the economics of religious conflict in an Indian city. American Economic

Review: Papers and Proceedings 98 (May (2)), 501–509.Gerber, A., Gruber, J., Hungerman, D., 2010. Does church attendance cause people to vote? Using blue laws’ repeal to estimate the effect of religiosity on

voter turnout. mimeo, Notre Dame.Glaeser, E.L., 2005. The political economy of hatred. Quarterly Journal of Economics 120 (January (1)), 45–86.Grimmer, J., 2011. Appropriators Not Statesmen: The Distorting Effects of Electoral Incentives on Congressional Representation. September, mimeo, Stanford

Political Science.Grimmer, J., Hersh, E., Feinstein, B., Carpenter, D., 2011. Are Close Elections Random? mimeo, Stanford.Grosfeld, I., Zhuravskaya, E., 2012. Persistent Effects of Empires: Evidence from the Partitions of Poland, February. mimeo.Grosjean, P., 2012. The institutional legacy of the Ottoman Empire: Islamic rule and financial development in South Eastern Europe. Journal of Comparative

Economics 39 (1), 1–16.Guiso, L., Sapienza, P., Zingales, L., 2008. Long Term Persistence, working paper 08-11, Chicago GSB.Hansen, T.B., 1999. The Saffron Wave: Democracy and Hindu Nationalism in Modern India. Princeton University Press.Healy, A.J., Malhotra, N.A., Mo, C.H., 2010. Irrelevant events affect voters’ evaluations of government performance. Proceedings of the National Academy of

Sciences 29 (12), 804–809.Hearn, P., 2001. Global GIS database. Digital atlas of South Asia, electronic resource. US Geological Survey digital data series.Huber, G.A., Hill, S.J., Lenz, G.S., 2012. Sources of bias in retrospective decision-making: experimental evidence on voters’ limitations in controlling

incumbents. American Political Science Review 106 (November (4)), 720–741.Ibn, B., Abdullah, A., 1355. The Travels of Ibn Battuta, 1971 ed. Lenox Hill, New York. Translated from the Arabic by Lee, Samuel, 1829.Jha, S., 2008a. A theory of trade, ethnic cronyism and tolerance. Stanford GSB (mimeo).Jha, S., 2008b. Trade, institutions and religious tolerance: evidence from India. In: GSB Research Paper 2004. Stanford Graduate School of Business, Stanford,

CA, January.Jha, S., 2013a. Analyzing political risk in developing countries: a practical framework for project managers. Business and Politics 15 (April (1)), 117–136.Jha, S., 2013b. Trade, institutions and ethnic tolerance: evidence from South Asia. American Political Science Review (November).Jha, S., Wilkinson, S.I., 2012. Does Combat experience foster organizational skill? Evidence from ethnic cleansing during the partition of South Asia. American

Political Science Review (November).Jukua, C., 1225. His work on the Chinese and Arab trade in the twelfth and thirteenth centuries, entitled Chu-fan-chi. Paragon Books, New York. Translated

from the Chinese by Hirth, Friedrich and Rockhill, W.W. in 1966.Miguel, E., Satyanath, S., Sergenti, E., 2004. Economic shocks and civil conflict: an instrumental variables approach. Journal of Political Economy 112 (4),

725–754.Milgrom, P., Qian, Y., Roberts, J., 1991. Complementarities, momentum and the evolution of modern manufacturing. American Economic Review: Papers

and Proceedings 81 (May (2)), 84–88.


Mitra, A., Ray, D., 2010. Implications of an Economic Theory of Conflict: Hindu–Muslim Violence in India.Montalvo, J.G., Reynal-Querol, M., 2005. Ethnic polarization, potential conflict and civil wars. American Economic Review 95 (June (3)), 796–816.Nunn, N., Wantchekon, L., 2011. The Slave Trade and the Origins of Mistrust in Africa. American Economic Review 101 (December (7)), 3211–3251.Olzak, S., 1992. The Dynamics of Ethnic Competition and Conflict. Stanford University Press, Stanford, CA.Pathania, V., Tandon, S., 2011. Divisive Politics and Violence: Evidence from the Success of the Bharatiya Janata Party in India. mimeo, LSE.

dx.doi.org/10.1016/j.jebo.2014.02.023























































































































































































































































































































































































































































































































































































































































































































































































































































































































G ModelJ

PR

SSSTVVVWWW

Y



eople’s Union for Democratic Rights, 2002. ‘Maaro! Kaapo! Baalo’! State, Society and Communalism in Gujarat. May, mimeo.aychoudhari, T., 1998. The agrarian system of Mughal India: a review essay. In: Alam, M., Subrahmanyam, S. (Eds.), The Mughal State,. Oxford University

Press, Delhi, chapter 8.chmidt, K.J., 1999. Atlas and Survey of South Asian History. Vision Books, New Delhi.ircar, D.C., 1962. Veraval Inscription of Chaulukya-Vaghela Arjuna. Epigraphia Indica XXXIV, 141.ubrahmanyam, S., 1990. The Political Economy of Commerce: Southern India, 1500–1650. Cambridge University Press, Cambridge.hapar, R., 2004. Somanatha: The Many Voices of a History. Penguin, New Delhi, India.arshney, A., 2002. Ethnic Conflict and Civic Life: H indus and Muslims in India. Yale University Press, New Haven.arshney, A., Wilkinson, S.I., 2004. Varshney Wilkinson Dataset on Hindu–Muslim Violence in India, Version 2, October.oigtländer, N., Voth, J., 2011. Persecution Perpetuated: The Medieval Origins of Anti-Semitic Violence in Nazi Germany, working paper 17113, NBER June.


ade, R., 1988. Village Republics: Economic Conditions for Collective Action in South India. ICS Press, Oakland.ilkinson, S., Haid, C., 2009. Ethnic Violence as Campaign Expenditure: Riots, Competition, and Vote Swings in India. mimeo.ilkinson, S.I., 2004. Votes and Violence: Electoral Competition and Ethnic Riots in India Cambridge Studies in Comparative Politics. Cambridge University

Press, Cambridge.ule, Sir Henry (Ed.), 1866. Cathay and the Way Thither. Hakluyt Society, London.

dx.doi.org/10.1016/j.jebo.2014.02.023



















































































































































































‘unfinished business’: historic complementarities, political competition and ethnic violence in...

Documents