November 11, 2005

Communal Riots in India (Steven I Wilkinson)

(The Economic and Political Weekly
October 29, 2005)

Communal Riots in India
by Steven I Wilkinson

What causes communal riots in contemporary India? In my book Votes and Violence (2005), I argue that politicians both cause them and, more importantly, have the power to prevent them, through their control of the state governments responsible for law and order.1 Riots in India, I argue, are the product of incentives at the local level and the state level. At the local level, I argue that the most important cause of riots is the intensity of electoral competition. Where political competition is most intense, “parties that represent elites within ethnic groups” use anti-minority protests, demonstrations and physical attacks that precipitate riots in order to “encourage members of their wider ethnic category to identify with their party and the ‘majority’ identity to rather than a party that is identified with economic redistribution or some ideological agenda” (p 4).

However, the real issue in explaining riots, I argue, is not which local political or sociological factors increase the likelihood of violence, but whether the level of government responsible for law and order (the 28 states, in India’s federal system) chooses to prevent violence or intervene quickly to stop it when it does break out. Rioters are not heroes and the state police and paramilitary forces – even in states like Bihar where the general quality of administration is far from high – have shown themselves capable of preventing communal riots when given the right direction by their ministers.2

What then determines the reaction of state governments? I argue that state governments protect minorities when it is in their electoral interest to do so: “…politicians in government will increase the supply of protection to minorities when… minorities are an important part of their party’s current support base, or the support base of one of their partners in a coalition government; or when the overall electoral system in a state is so competitive – in terms of the effective number of parties – that there is therefore a high probability that the governing party will have to negotiate or form coalitions with minority supported parties in the future…” (pp 6-7). The growing competitiveness of state politics in India, in other words, much of it the result of backward caste mobilisation, is a positive thing for communal relations.3 An increase in the number of parties competing for Muslim votes increases the incentives for state governments to prevent anti-minority violence.

Many Indians, and scholars of Indian politics, will not be surprised by these arguments, and my work is in a rich tradition of studies on the connection between electoral competition and riots. But my book, I hope, makes a contribution in two ways: first, by explicitly theorising about the precise conditions under which electoral competition matters for riots at several levels, and secondly by using a great deal of new data (a joint 1950-95 dataset on Hindu-Muslim riots Ashutosh Varshney and I collected, as well as a great deal of other riot, election, census and socio-economic data that I collected independently) to test my hypotheses against other leading explanations for communal riots. The book uses some statistical analysis, but it also contains a great deal of qualitative evidence based on archival research, fieldwork, and extensive reading in primary and secondary sources.

Alternative Approaches

My focus on the importance of politics at the state level presents an alternative, however, to some recent scholarly work that identifies town level factors as key to explaining the prevalence of communal riots. Paul Brass (2003), for instance has focused on the importance of “institutional riot systems” in producing large-scale riots in towns in which they are endemic. And Ashutosh Varshney (2002:11) argues in his book that state and national-level politics are less important than town level civic organisations and networks: where a town’s civic organisations are interethnic and associational, “…polarising politicians either don’t succeed or eventually stop trying to divide communities by provoking and fomenting violence”.

The fact that several political science research projects have been looking at the issue of communal violence at the same time has been very productive in terms of helping each of us sharpen our arguments and explanations, and hopefully in terms of building wider cumulative knowledge about what causes riots. Brass’s work for instance, because it acknowledges the importance of electoral incentives at different levels and the role of the state, is largely compatible with my own. But of course differences in approach and emphasis remain. In a recent review of my work, (‘An Electoral Theory of Communal Riots’, EPW, September 24, 2005) Varshney draws attention to some of these differences, and I would like to turn now to a specific discussion of the criticisms he makes of my electoral arguments.

Electoral Incentives

The most important substantive question that Varshney raises about my book is how local and state-level incentives interact to lead to violence (p 4221). Why, he asks, should intense competition at the local level often lead to anti-minority polarisation while an increase in party competition at the state level leads to states acting more aggressively to prevent violence? The answer, clearly stated in the book, is that party competition is bipolar at the local level (p 23) but multipolar at the state level in most Indian states, meaning that different groups of voters are pivotal at each level (pp 138-46). Misunderstanding the fundamental nature of this argument, unfortunately, leads to further misunderstandings about the theoretical and analytical choices made in the book.

At the constituency level, as the work of Pradeep Chhibber and others has shown, electoral competition in India tends to follow the bipolar pattern that Maurice Duverger long ago predicted would obtain under a first-past-the-post electoral system. Given that upper caste Hindus and Muslims are disproportionately urban communities throughout much of the country (West Bengal is one notable exception), we often therefore have a situation in towns in which a party from the majority community confronts a Muslim-supported party, with the winner of the two-party race being decided by who can win over the remaining Hindu voters, something that I show in the book is frequently achieved through communal polarisation and riots.4

At the state level, however, because India is highly diverse, its state electoral constituencies relatively small, and the different social groups that support different parties are mobilised and distributed unequally across each state, the number of parties in most states now aggregates to considerably more than two.5 In 2002 for instance, my data show that the average number of effective parties in the 15 major states was 4.4, with only three of these 15 states (Andhra Pradesh, Himachal Pradesh and Gujarat) having fewer than three effective parties (p 143). I suggest that this multipolar competition among parties, many of which have strong caste or redistributional ideologies, is a good thing for communal relations because Muslims thus become an important swing vote for parties from the majority community. The advantages of multiparty competition for ethnic moderation have of course been raised by others, such as Donald Horowitz, but my book is, as far as I am aware, the first to show, using data on every Indian major state election from 1961-1995 (not just data from 2002, as Varshney implies on p 4220), that a greater number of parties at the state level leads to fewer communal riots, even when we control for socio-economic factors and a state’s previous level of violence.

The fact that I hypothesise different theoretical mechanisms at different levels explains why different measures of electoral competition are appropriate. At the town level, where competition is bipolar, I predict that violence will take place in seats with the tightest electoral races (measured by the margin at the previous election), the party representing the elite within an ethnic category that can potentially unite a majority of voters will have a greater incentive to foment polarising events and riots in order to solidify its vote. But at the state level, it is not the previous electoral margin that is important, but whether the government in power at the time of the riot might (i) break apart due to inter-party differences over the threat of communal violence and (ii) whether the overall competitive environment in a state is so intense that a substantial riot-induced shift in Muslim votes to other parties might cause the incumbent party or parties to lose ground to competitors. The appropriate measures at the state level, therefore, are the number of parties that are competing for office and the Muslim support for the major parties.

Theories and Non-Exceptions

Varshney cites a number of riots (pp 4222-23) that he thinks are “exceptions” to my general electoral arguments, and he argues that these exceptions are so frequent that they undermine the argument as a whole. Without getting into the interesting methodological issue of how many exceptions it might take to invalidate any particular thesis, let me simply make two points. First, there are not so many exceptions as to invalidate the significance of the main explanatory variables in my multivariate statistical analysis. Second, many of the “exceptions” that Varshney claims undermine my argument are not exceptions at all but are entirely consistent with my theory.

To reiterate, my book makes a general argument about the conditions under which democratic governments prevent riots. In the Indian context, the theory does not associate riots with government by any particular party – rather it predicts that democratic governments led by any party will prevent riots when they depend for votes directly or indirectly upon the minority groups. But Varshney says my theory is undermined because “one can think of not one or two, but many instances when the ruling party was not the anti-Muslim BJP, or its analytic equivalent, the Shiv Sena, but deadly Hindu-Muslim riots nonetheless took place” (p 4222). Having framed my argument in this way, he then points to riots that happened during Congress rule (e g, in Ahmedabad, Moradabad, Meerut, Hyderabad and elsewhere) as damning exceptions to my theory (pp 4222-23).

This portrayal unfortunately rests on a misunderstanding of my argument, which offers a general theory of the way in which party competition can affect governments’ response to riots, not the anti-BJP thesis that Varshney describes. In the book I highlight examples of Congress and Muslim League politicians’ complicity in partition-era riots in Bihar, UP, and in Calcutta (pp 5,74). I also point out that in the post-independence era Congress has at times benefited electorally from Hindu-Muslim violence (p 50) and I find that we can identify no robust statistical relationship between Congress rule and the level of riots, a result I attribute to the widely varying communal character of the party and its leadership across time and place (p 153). Lest anyone be in doubt about my position, I say on p 153 that “at one time or another, Congress politicians have both fomented and prevented communal violence for political advantage. Congress governments have failed, for example, to prevent some of India’s worst riots (e g, the Ahmedabad riots of 1969, the Moradabad riots of 1980, and the Meerut riots of 1987) and in some cases Congress ministers have reportedly instigated riots…and have blocked riot enforcement.” So of course I wholeheartedly accept Varshney’s point that riots can occur under governments of many political stripes, because that is my point as well.

Future Research Possibilities

The obvious direction for the future study of communal riots – at least from the perspective of my own discipline of political science – is to try to build explanations that work across many different levels (mohalla, town, state and national) and then test these explanations against plausible alternatives using good qualitative and quantitative data from at least several decades. This is a challenge, because the real world does not always provide us with the data we would wish to test our theories. I speak from experience, because even after spending many years working on my own book I was not able to gather material to test all the explanations I wished using the whole 45 years of data on which I had data on riots.6 Such testing is even more of a challenge for those scholars who focus on the role of inter-ethnic economic competition, town level riot networks or social capital, where our most systematic data tends to be cross-sectional rather than time-series.7

One point that Varshney makes about the data I use is the question of whether it is legitimate to match constituency level electoral data and town level riot data, because some towns have more than one constituency. “Wilkinson simply cannot be sure”, he says, “that riots at the town level are connected to the competitiveness of elections at the constituency level” (p 4221). The first response to this criticism is to say that the multiple constituency-single town issue applies to only a small minority of the towns in my sample. The average constituency in UP now has around 4,00,000 inhabitants and this population total is exceeded in only a dozen or so of the 167 towns in my sample and moreover we know that the Election Commission tries to draw boundaries to avoid crossing administrative boundaries wherever possible.8 Second, the procedure I use for the dozen remaining towns in the sample (seeing whether they contain constituencies that are highly competitive) is appropriate because we have good theoretical reasons to believe – on the basis of Brass’s study of Aligarh for instance – that competitive races in individual constituencies often lead to violence that ultimately affects the town as a whole.

Would my town level findings (reported in Table 2.1) about the importance of electoral competition in predicting riots at the local level go away if I were to drop the largest towns with multiple constituencies from my statistical analysis? I reran the statistical analysis to find out, and they would not.9 I do not of course disagree with Varshney that it would be better and more conclusive if we had data on riots that always perfectly matched the boundaries of our electoral units. In fact, after my book was sent to the publishers I was able to create such a dataset (for Gujarat, using electoral and census data I collected and constituency level riot-data obtained by Yogendra Yadav), and I have recently used it to analyse the pattern of electoral competition and riots in Gujarat before and after the 2002 riots to see, among other things, if the closeness of electoral competition there predicted violence. The findings – that competition at the constituency level predicts the location of riots – are entirely consistent with those in my book.

Ultimately the best long-term test of any explanation is whether it seems to work with new data, or be successful at predicting what will happen in new situations. Do other scholars looking at communal riots identify the same factors as important and does my explanation seem to make more sense than the alternatives in explaining patterns of violence? I am optimistic, because other scholars using their own data have found very similar patterns to the ones that I identify in my own work. Yogendra Yadav (2003) for instance, examined the patterns of violence and the electoral swings in Gujarat in 2002 and found that “the BJP managed to recover its eroding social base with carefully crafted and subtly executed politics of hatred…. Anti-Muslim violence played a crucial role in this process of recovery, damage control and acquisition.” I hope that my work, and the work of the many scholars who wrote in the pages of EPW and other Indian journals after the 2002 riots, will help to contribute to an emerging consensus on the crucial role of electoral competition and the actions of state in explaining patterns of peace and violence.


1 Votes and Violence: Electoral Competition and Communal Riots in India (Cambridge University Press, New Delhi, 2005). I am grateful to several colleagues who provided valuable comments on this piece.
2 Indian Express, March 26, 2002, p 1, New Delhi. Wilkinson pp 63-96 explores in-depth the issue of state effectiveness.
3 I point out (pp 8-9) that in systems where responsibility for law and order is shared across several levels of government such as national and local, the model can be extended “to incorporate electoral incentives and power asymmetries across these different levels”. Although in India states are by far the most important level to look at (national governments have in the past been loath to impose president’s rule on states in which they or their allies are in office) I have recently argued that the “changing dynamics between the central and state governments should also have a beneficial effect in reducing anti-Muslim violence” and in this and other papers I describe the ideological and pragmatic reasons why the current national Congress government is more likely to intervene against state governments that allow riots to break out. Steven I Wilkinson, ‘Elections in India: Behind the Congress Comeback’, Journal of Democracy, 16, 1 (January 2005), pp 153-67.
4 Wilkinson pp 21-26, Muslims are 18 per cent of the total population and 32 per cent of the urban population in UP, 6 per cent and 15 per cent in MP, 8 per cent and 18 per cent in Rajasthan, 11 per cent and 25 per cent in Maharashtra, and 9 per cent and 14 per cent in Gujarat according to the 2001 Census.
5 For the general link between ethnic heterogeneity, constituency size and the number of parties in a plurality system, see Gary Cox, Making Votes Count: Strategic Coordination in the World’s Electoral Systems, Cambridge, 1997, pp 205-21.
6 Varshney states that there is a “big silence” in my book on the issue of why my regressions were “run only for the truncated period 1961-1995…” and that “Wilkinson provides no assistance in the matter” (p 4221). In fact I would have liked to include data from 1950-60, and I explain clearly on p 147 why I was not able to do so: “I select 1961 as a starting date because state reorganisation was largely complete for major states by this date, and because the key demographic data for these new states is only easily available for 1961 onward”.
7 Achen (1982: 54-55) points out the very real problems with using cross-sectional data to sort out causality.
8 Varshney also raises the issue (p 4220) of cases with towns whose population is smaller than the total population of a constituency, but he provides no reason to convince us why it does not make sense to assume, as I do, that political life in a constituency of 4,00,000 probably revolves around the 2,00,000 population city that lies at its centre.
9 The statistical output from the replication and study referred to below is online at http://www.duke.edu/~swilkins/research.html.


Brass, R Paul (2003): The Production of Hindu-Muslim Violence in Contemporary India, Oxford University Press, India.
Varshney, Ashutosh (2002): Ethnic Conflict and Civic Life: Hindus and Muslims in India, Yale University Press, US.
Yadav, Yogendra (2003): ‘Gujarat: The Patterns and Lessons’, Frontline, December 21, 2002-January 3, 2003.