Does multiparty competition increase ethnic security fears? Evidence from North India

Abstract

This article explores the political foundations of ethnic security fears and examines the relationship between the structure of party competition at the constituency level and security fears at the individual level in Uttar Pradesh, India’s largest state. The analysis shows that security fears are higher in constituencies characterised by multiparty competition. The central implication of this finding is that rather than leading to ethnic moderation, multiparty competition in ethnic party systems tends to be more ethnically divisive than in two party systems, increasing collective fears of the future.

Keywords

ethnic conflict party competition security fears

Introduction

Ethnic security fears are one of the main preconditions for ethnic violence (Brass, 2003; Lake and Rothchild, 1998). Understanding the factors that are associated with security fears therefore has the potential to provide insights into the conditions under which attempts to mobilise violence are likely to be successful or not. Yet relatively little is known about the determinants of security fears. Which types of people tend to feel most vulnerable about riots and mob violence? Is it to do with majority or minority ethnic status? And under what conditions are security fears heightened? Reliable answers to these questions are currently lacking.

In this article we provide a clear evidential basis on which to evaluate some key claims that have been made about the political foundations of ethnic security fears. We show that the structure of political competition at the local level in ethnic party systems is closely associated with collective fears for the future. In particular, we show that in North India security fears are very much higher for people living in constituencies characterised by multiparty competition than for people living in constituencies characterised by two party competition. The reason for this, we suggest, is that multiparty competition in ethnic party systems tends to be more ethnically divisive than in two party systems, making groups more uncertain about their position and prospects for the future.

In generating our hypotheses we draw on two clear implications from the logic of ethnic identification and political competition. First, ethnic identities can be fluid and endogenous to competitive politics (Chandra, 2005; Eifert et al., 2010). Accordingly the extent to which ethnic divisions are salient depends at least in part upon the extent to which political parties mobilise along ethnic lines (see also Posner, 2004). Second, multiparty competition provides electoral incentives for parties to make narrow social appeals. When a party faces at least two other viable competitors, building broad social coalitions becomes risky, as it is possible for one or both of the other competitors to undercut one’s support base by making more direct overtures to narrower segments of society (Chhibber and Nooruddin, 2004). This means that parties are not only more likely to target voters with club goods, but in doing so political conflicts over access to state resources may be reproduced at the local level as social conflict and antagonism between the different parties’ main support groups. In multiparty contests social relations between different communities are thus more likely to be tense, and accordingly security fears are more likely to be high.

This article differs from previous research in that it focuses on ethnic security fears rather than actual ethnic conflict. This is important for a variety of reasons. According to Brass (2003) most instances of ethnic violence and riots are anticipated. They are not spontaneous events but tend to be preceded by an uneasy suspense: ‘a kind of smouldering fire that can erupt into flames over any kind of incident’ (Brass, 2003: 357). Similarly Lake and Rothchild (1998) argue that ethnic conflict is most commonly caused by collective fears of the future. As groups begin to fear for their physical safety, a series of dangerous and difficult-to-resolve strategic dilemmas arise that contain within them the potential for violence. Accordingly, ethnic activists and political entrepreneurs reinforce these fears of physical insecurity and cultural domination and polarise society, driving groups further apart. When fears of physical insecurity emerge, violence can and often does erupt. As Lake and Rothschild observe, these within group and between group interactions can ‘produce a toxic brew of distrust and suspicion that can explode into murderous violence’ (1998: 4). It is thus important to understand the conditions under which these security fears emerge. Thus, rather than looking just at Hindu-Muslim relations specifically we examine the structure of security fears more generally.

This article also differs from many previous studies in that it focuses on local level factors rather than state level or country level factors. This too is important in a number of respects. Ethnic violence in India is ‘highly locally concentrated’ (Varshney, 2001: 371). As Varshney explains, state (and national) politics provides the context within which the local mechanisms linked with violence are activated. To understand the causes of ethnic security fears, we must investigate these local mechanisms. Moreover, in first past the post (single-member simple plurality, SMSP) electoral systems, the effective arena of contestation is the constituency. The number of parties competing at the constituency level affects the size of the minimum winning coalition needed to win office, and as the size of the minimum winning coalition changes parties’ strategic incentives also change. Examining the local level thus provides an opportunity to explore how these different types of contest indirectly influence social relations and security fears on the ground.

In order to study these issues we draw on data collected in Uttar Pradesh (UP), North India. Uttar Pradesh is India’s largest state, with a population of close to 200 million and represents an important case study to test our hypotheses for a number of reasons. Firstly, India – and UP in particular – occupies a prominent position in much of the literature on ethnic conflict. Uttar Pradesh has a long history of ethnic violence, and much has been written about the Ramjanmabhumi issue in Ayodhya (the liberation of Ram’s birthplace) which led to the demolition of the Babri Masjid in December 1992 and widespread violence (Nandy et al., 1995; Van der Veer, 1994). The issue of ethnic security is thus a salient one, and even today the threat of ethnic violence is never very far from the public consciousness. A recent report in India Today documented the occurrence of 364 communal incidents, 64 deaths and 1298 injuries in Uttar Pradesh alone between 2009 and 2012 (Sharma and Menon, 2012).

Secondly, Uttar Pradesh is also ethnically diverse and represents a fairly clear cut example of what Horowitz (2000: 291) has labelled an ethnic party system, where ‘parties derive their support overwhelmingly from an identifiable ethnic group (or cluster of ethnic groups) and serves the interests of that group’. Ethnic outbidding is a well-known dynamic in ethnically divided democracies (Horowitz, 2000) and parties compete with each other to be the best defender of their ethnic group. This competition can cause groups to fear one another (Saideman et al., 2002).

Thirdly, despite using a first past the post (single-member simple plurality, SMSP) electoral system, Uttar Pradesh is also characterised by multiparty competition.¹ The principal competition in the state involves the SP (Samajwadi Party: Socialist Party), which gets strong support among Yadavs (traditionally a low- to middle-ranking cluster of agricultural-pastoral castes) and Muslims (Michelutti, 2008); the BSP (Bahujan Samaj Party: Popular People’s Party), which gets strong support among the Scheduled Caste (former untouchable) communities (Chandra, 2004); the BJP (Bharatiya Janata Party: Indian People’s Party), a Hindu nationalist party, which gets strong support among Brahmins (traditionally one of the high status priestly castes) and other Hindu Upper Castes (Heath, 1999); and Congress, which is the only party to lack a distinctive ethnic base (Heath and Yadav, 1999).²

Uttar Pradesh thus provides a good opportunity to explore the link between variations in the structure of party competition at the local level and security fears in an ethnic party system, while holding other important factors constant that may be associated with security fears, such as electoral system, incumbent government, level of economic development, historical background and other state level factors. These conditions have the potential to offer valuable insights on the impact of local level political factors on ethnic security fears, which may have wider relevance for understanding the political production of ethnic security in other ethnic party systems in India and around the world.

The article proceeds as follows. In the first section we review the relevant literature on issues related to ethnic security fears and how this has been applied to the Indian context. From this discussion, we derive a set of hypotheses that are tested in the last section using data from the 2009 Indian National Election Study, conducted by the Centre for the Study of Developing Societies (CSDS). Finally, we conclude by presenting the implications of our findings and by suggesting avenues for further research.

Multiparty competition and ethnic security fears

Previous research on ethnic conflict suggests two main ways in which the structure of political competition may also be related to ethnic security fears. In recent years, a growing body of research has examined the making of ‘violent entrepreneurs’ and the ways in which political elites stoke ethnic fears and orchestrate violence and the institutional incentives that shape their behaviour (see in particular Brass, 2003; and also Hansen, 2001; Tambiah, 1996; Wilkinson, 2004). In one influential account, Wilkinson (2004: 4) argues that competitive political contests stimulate political elites to mobilise anti-minority protests, demonstrations and physical attacks in order to ‘encourage members of their wider ethnic category to identify with their party and the majority identity.’ This implies that in competitive contests local level campaigning may stoke up security fears. But secondly, and more importantly, Wilkinson argues that the real issue in explaining riots is not whether local political elites try to mobilise violence, but whether State level authorities try to prevent violence or intervene quickly to stop it when it does break out. Accordingly, at the state level, multiparty contests stimulate political elites to prevent violence. In particular, he argues that multiparty competition leads to ethnic moderation, and that a greater number of parties at the state level lead to fewer ethnic riots (2004).

As party competition increases…majority politicians will have greater incentives to appeal to Muslim voters who can provide them with the margin of victory…. [and so] the increasing party competition for minority voters has led to a reduction in Hindu-Muslim violence, as politicians are forced by electoral incentives to take firm action to prevent Hindu-Muslim riots. (Wilkinson, 2004: 170).

There are thus two different but slightly overlapping processes at work: multiparty competition at the state level moderates ethnic conflict since parties are sensitive to minority voters; and the closeness of the race at the constituency increases ethnic conflict since competition is bipolar and so parties try to mobilize majority voters.³ Building on Wilkinson’s (2004) electoral incentives model, we examine whether local level political conditions – particularly to do with the structure of political competition – also influence ethnic security fears among the public more generally. Although elite incentives to orchestrate violence may indeed be lower in multiparty states as Wilkinson argues, the indirect effects of multiparty competition at the local level in ethnic party systems may have a rather different effect on security fears among ordinary people.

This proposition differs from Wilkinson’s electoral incentives model in a number of important respects. Whereas Wilkinson’s model assumes that competitive contests at the constituency level encourage parties to chase after majority voters and stoke up security fears directly; our approach suggests that multiparty competition at the constituency level encourages parties to make narrow social appeals, which increases security fears indirectly. That is, security fears are not higher because parties are directly trying to precipitate acts of violence, but rather when ethnic groups are political rivals rather than political allies both groups become more fearful of the other and fearful of the future. By contrast ethnic groups will be more secure if they have access to decision makers, if they can block harmful government policies and if they can veto potentially damaging decisions. Accordingly, the extent to which different ethnic groups are incorporated within political parties may serve to limit or enhance the threat felt by various groups. Local party systems that incorporate different ethnic groups and provide collective representation for different groups may therefore be more likely to limit security fears.

Whereas in two party contests parties have to reach out at the local level and appeal to a broad cross section of society in order to secure a winning majority, in multiparty contests the threshold for victory is much lower, sometimes as little as 25%, and so parties can secure victory by mobilising and appealing to a much narrower section of society. In other words, as the size of the minimum winning coalition changes, parties’ strategic incentives also change. Chhibber and Nooruddin (2004) argue that when a party faces at least two other viable competitors, building broad coalitions becomes risky, as it is possible for one or both of the other competitors to undercut one’s support base by making more direct overtures to narrower segments of society. Political parties in multiparty contests therefore need to make appeals to ‘vote banks’ and particular support groups. Chhibber and Nooruddin (2004) show that in multiparty states in India, needing only a plurality of votes to win, parties use club rather than public goods to mobilise smaller segments of the population. They also note that in multiparty states, voters are more likely to say that parties represent a specific caste, and that local caste relations are conflictual rather than consensual (Chhibber and Nooruddin, 2004: 179).

This has clear implications for thinking about how local level campaigning may indirectly influence social relations since the size of the minimum winning coalition is determined at the constituency level. When there is a strong link between ethnic identity and partisan support, the distinction between political identities and social identities becomes more blurred. For example, the SP is frequently referred to as a ‘Yadav’ party (the Yadav parivar – or Yadav family), and the BSP is frequently referred to as a Dalit or Scheduled Caste party. In order to secure symbolic and material gains people tend to vote for caste-based parties by conducting ethnic head counts (and voting for parties which contain members of their ethnic community) rather than by comparing policy platforms (Chandra, 2004). Accordingly, elections are frequently seen as a good way to get what you can for your own community, and – importantly – as a good way to try and prevent other communities from taking away from you what you already have (see Michelutti and Heath, 2013). Under these conditions, there is thus greater potential for political conflicts over access to state resources to be reproduced at the local level as social conflict and antagonism between the different parties’ main support groups. In multiparty contests social relations between different communities are thus more likely to be tense, and accordingly security fears are more likely to be high.

Hypotheses

As discussed above, the main hypothesis that we test is whether in ethnic party systems, security fears are higher for people living in multiparty constituencies (H1). Since the threshold required for victory is lower in multiparty seats, parties can appeal to and mobilise a narrow section of society. Accordingly, harmonious ethnic relations are harder to sustain when ethnic groups are political rivals rather than political allies, and so security fears will tend to be higher.

Of course, to examine the independent impact of the structure of political competition on security fears it is necessary to take into account other factors that might be of potential causal significance, and which may be either directly (or indirectly) associated with the effective number of parties. Drawing on previous work we therefore also test a number of additional hypotheses. Following Horrowitz (2000), Wilkinson (2004) and others, we examine whether security fears are associated with social demography, and specifically the size of the Muslim population in the Constituency (H2a). Several studies have argued that the relative balance between the majority and minority community is an important predictor of conflict, because the size of the minority community increases the majority’s sense of threat (Wilkinson, 2004) which may in turn heighten ethnic security fears. In addition, according to Posner (2004) the political salience of ethnic difference is contingent on the size of the group in question. To investigate these claims we examine the extent to which security fears are higher in constituencies where there is a large Muslim population.

In addition, following the large literature on whether social diversity influences a range of outcomes, such as social trust (Putnam, 2007), democracy (Fish and Brooks, 2004) and civil war (Ellingsen, 2000), we also examine whether security fears are higher in socially diverse constituencies, based on the size and number of the principal caste-community formations in the state (H2b).

Following Brass (2003), Varshney (2001, 2002) and numerous others, we also examine whether security fears tend to be higher among people living in constituencies which have experienced high levels of violent unrest in the recent past (H3). It is well known that violence begets violence, and, as Varshney (2001: 370) observes, the geographical dimension of violent outbreaks constitutes ‘a puzzling empirical regularity’, where ‘despite ethnic diversity, some places (regions, nations, towns, villages) manage to remain peaceful, whereas others experience enduring patterns of violence.’ Barry Posen (1993) makes a similar point and argues that if violence has occurred in the past, then affected groups will be more likely to think that their security is at risk. Chaim Kaufmann (1996: 173) goes even further and doubts whether it is in ‘anyone’s power to resolve ethnic hatreds once there has been large-scale violence.’ Accordingly, we would expect security fears to be much higher in ‘riot prone’ places.

Following Wilkinson (2004) and Eifert et al. (2010), we examine whether security fears tend to be higher among people living in constituencies with close political contests (H4). Regardless of whether political elites directly try to orchestrate violence, if – as Eifert et al. (2010) argue – the role that ethnicity plays is to secure an advantage in the competition for power, then it is likely to be most useful, and to become most salient as a social identity, during elections that are closely fought. We would therefore expect that if ethnic identities are more salient, then ethnic divisions may also be more pronounced, which could in turn lead to higher security fears.

Lastly, drawing on Varshney (2001, 2002), we also examine the extent to which security fears are related to the ethnic character of civil society. Varshney distinguishes between interethnic and intraethnic associations, and whereas the former help to build bridges and manage tensions, the latter do not. Interethnic networks of engagement can help to regulate and manage tensions and conflicts because they connect individuals from different ethnic backgrounds, build trust and encourage reciprocity, which – as Varshney (2001: 378) argues – help to ‘nip rumors, small clashes, and tensions in the bud.’ Building on this idea we examine whether people active in associational forms of engagement (such as sports clubs, reading clubs and other civic associations) exhibit lower security fears than people who are not members of such organisations. We might anticipate that security fears will not only be lower for people active in these forms of associational engagement themselves, but that fears will, on average, also be lower for all people living in areas characterised by high levels of associational engagement.

We also examine whether members of intraethnic caste-based and religious organisations (such as caste or religious associations) exhibit higher security fears than people who are not members of such organisations. Caste associations have been active channels through which political mobilisation has taken place, and accordingly those people who have been directly exposed to such caste-based mobilisation may feel more threatened by and suspicious of other communities.

Measuring ethnic security fears

Our first task is to examine the structure of security fears in the Indian context in order to ascertain whether they contain a specifically ethnic component, related to the sort of within group and between group interactions which may lead to collective fears of the future discussed above. To test these expectations of whether attitudes toward security fears are uni-dimensional, or whether we can identify specific ethnic security concerns, we need a dataset with a large number of items that tap into these issues. To this end we draw on data from the 2009 Indian National Election Study (INES), conducted by the Centre for the Study of Developing Societies (see Appendix in Shastri et al., 2009 for an overview of the series). The INES is one of the largest political surveys carried out anywhere in the world, and is unique of developing countries. The sample size for the 2009 survey is 34,365 adults, and the survey is representative at both the national level and state level and is based on sophisticated methods of random sampling and face-to-face interviewing. We use state level data on Uttar Pradesh from the 2009 Survey, which has a sample size of 2869, and which was based on interviews from 78 out of the 80 Lok Sabha Constituencies. The INES is a particularly valuable data source for our purpose since it also contains a special split-sample module which asks an unusually high number of questions regarding security fears – and, importantly, provides numerous survey items that (a priori) plausibly invoke either security fears about ethnic conflict, violence and intimidation or other non-ethnic fears about domestic violence and state-society relations. The exact wordings of the questions we use are set out below:

I will read out some things that people feel insecure about these days. Tell me how insecure do you personally feel about…

Riots and mob violence

Terrorism

Physical attack and harassment

Theft and robbery

Domestic violence

Police, army and security forces

These questions were put to 569 people. These items, we argue, prime individuals to think about within group and between group interactions, which may activate security fears that have a specifically ethnic character, and domestic interactions and interactions with state officials, which we would expect to have a weaker ethnic dimension. The most clear-cut examples of issues related to ethnic security fears are the items on riots and mob violence and domestic terrorism. As already mentioned, Uttar Pradesh has a long history of ethnic violence between Hindus and Muslims and low level communal clashes are an endemic feature of day to day life. The threat of domestic terrorism is also a real and pressing concern. According to Piazza (2010), there are over 74 active ethnic and religious terrorist groups in India, which is second only to Iraq in number.⁴

In addition to these high-impact, episodic forms of ethnic violence, we also examine more mundane everyday concerns to do with physical attack, harassment and robbery. Are concerns about these everyday episodes of violence and intimidation related to concerns about large scale acts of ethnic violence? There is some reason to suspect that they be. Everyday episodes of violence and intimidation frequently occur along ethnic lines, and are a ready feature of social life (Michelutti, 2008). According to Brass (2003) these grievances, frustrations and discriminations are an important part of the ‘riot system’ and episodes of assault, violence and robbery may stoke up local level security fears, making groups more fearful about the future. By contrast, we do not anticipate that the other items, which emphasise interactions within the family and the state, will load on to the same ethnic security dimension.

Our measurement strategy has two stages. Firstly, using exploratory factor analysis (EFA) of all available questions, we establish that these items load on two different attitude dimensions.⁵ To further confirm the existence and nature of these two dimensions and check the robustness of the EFA results, we perform a confirmatory factor analysis (CFA). This is a more stringent test of the existence of two dimensions. We directly test whether constraining specific items to load on one dimension and not the other produces important misfits between our model and the data. The first latent variable, which we identify as referring to ethnic security fears, includes items on large scale episodic acts that have a specific ethnic dimension such as mob violence and terrorism, as well as mundane everyday acts of physical attack, harassment and robbery that may be interpreted as forming part of the ‘riot system’. The second latent variable lacks such a clear meaning and includes the items on domestic violence and the police, army and security services. To reflect this ambiguity we label this latent variable as referring to other security fears, though we should note that the factor loadings are not particularly strong, which further suggests that the latent variable is not particularly well defined. However, for the purposes of our analysis the important point is that these items do not load on to the ethnic security dimension and we can therefore be confident that our variable of interest is measuring security fears that have a specific ethnic dimension rather than just security fears in general.

We test several alternative models, with the best fitting model presented in Figure 1.⁶ Inspection of the modification indices reveals that there may be additional sources of co-variation between the indicators that load on to the ethnic security latent variable. These suggest the presence of two potential sub-dimensions of ethnic security fears, relating to fears about large scale episodic events (riots and terrorism) and mundane everyday events (harassment, physical attack and robbery). Allowing the error variances of these indicators to co-vary significantly improves the fit to the data and the resulting fit indices are all good, comfortably falling within an acceptable range.⁷

Figure 1.

Security fears in Uttar Pradesh, Confirmatory Factor Analysis. Fit statistics: Chi-square = 14.2; df = 6; p > 0.01; CFI = 0.99; RMSEA = 0.05; SRMR = 0.0169.

These factor analyses show that a uni-dimensional view of security fears does not hold up in the data. Instead, and in accordance with our expectation, we can identify security fears that have a specific ethnic dimension, though it is also possible to sub-divide these ethnic security fears into episodic and everyday fears. For our main dependent variable we use the overall measure of ethnic security fears (which has a mean of zero and a standard deviation of one). However, as a robustness check we also replicate our analysis on the different sub-dimensions of ethnic security fears to see if there is any difference between the effect of party competition on episodic security fears and everyday security fears. By way of comparison we also examine the residual category of general security fears (examining each item separately), which according to our theory should be empirically unrelated to party competition since they lack a clear ethnic dimension. We also replicate our analysis on each of the individual items that comprise the ethnic security latent variable.

In order to examine the determinants of these security fears, our key independent variables at the individual level are education,⁸ age, sex, economic status,⁹ caste-community and associational involvement in civic and caste/religious organisations. We control for caste-community to investigate whether security fears are higher among minority ethnic groups, specifically Muslims. Our measure of caste-community takes into account the major politically salient ethnic groupings in the state, distinguishing between Upper Castes, Yadavs, Other Backward Classes (OBCs), Scheduled Castes and Muslims. Our key independent variables at the constituency level are the Muslim population size, the effective number of ethnic groups, the incidence of previous riots, the closeness of the electoral contest and the effective number of parties. To measure Muslim population at the constituency level we use aggregated survey estimates from the 2009 INES.¹⁰ On average, Muslims represent 18 percent of the population at the constituency level, though this varies quite substantially across constituencies (see Appendix). To measure social diversity at the constituency level we again use aggregated survey estimates from the 2009 INES, and calculate the effective number of ethnic groups, where ethnic groups are defined in terms of the most politically salient social groupings in the state mentioned above. To measure the previous incidence of riots at the constituency level we use the Crime in India annual data series published by the Government of India, Ministry of Home Affairs, National Crime Records Bureau and compiled and made available by Marshall and Marshall (2008). We use the mean number of riots reported per year (i.e., a single incident involving 5 or more persons) at the district level between 2000 and 2006, matched to electoral constituencies and distinguish between those constituencies which are prone to riots (more than 70 riots per year) and those which are not (fewer than 70 riots per year).¹¹ The average number of riots per year across this period was 66, with a high of 439 recorded in Aligarh in 2000. To measure the structure of party competition we use official election results from the 2009 Lok Sabha election (compiled and made available by the Data Unit at the CSDS) and compute the effective number of parties at the constituency level. The average effective number of parties in a constituency is 3.94, but there is a substantial amount of variation between constituencies (standard deviation = 0.85), with a range from a low of 1.82 to a high of 5.79. The closeness of the contest is simply the difference in vote share between the first two parties in the constituency.

Analysis

What factors, then, are associated with ethnic security fears? To what extent is variation structured by individual-level factors, and to what extent is it structured by ecological factors? Do Muslims feel more insecure than Hindus? Are security fears higher in areas with a sizeable Muslim population and in areas which are riot prone? And taking into account these factors, what impact if any does the structure of party competition have? To answer these questions we specify a linear model (OLS), with standard errors clustered at the constituency level.¹² The full model that we specify is as follows:

\begin{aligned} E t h n i c S e c u r i t y f e a r s & = β 0 + β 1^{*} (C a s t e - c o m m i n i t y) \\ + β 2^{*} (E c o n o m i c p o s i t i o n) \\ + β 3^{*} (A g e) + β 4^{*} (G e n d e r) \\ + β 5^{*} (C i v i l s o c i e t y a s s o c i a t i o n) \\ + β 6^{*} (C a s t e a s s o c i a t i o n) \\ + β 7^{*} (M u s l i m p o p u l a t i o n) \\ + β 8^{*} (R i o t h i s t o r y) \\ + β 9^{*} (V i c t o r y m a r g i n) \\ + β 10^{*} (E N P) + ϵ \end{aligned}

Table 1 displays the results of five models, with different specifications of the dependent variable. In Model 1 the dependent variable is the overall measure for ethnic security fears. Models 2 and 3 present alternative specifications of ethnic security fears, distinguishing between large scale episodic fears and mundane everyday fears, respectively. By way of comparison the dependent variables in Models 4 and 5 are domestic security fears and state security fears respectively, which we hypothesize are unrelated to the structure of party competition.

Table 1.

Determinants of ethnic security fears, OLS.

	Ethnic security fears		Episodic security fears		Everyday security fears		Domestic security fears		State security fears
	Coef.	SE	Coef.	SE	Coef.	SE	Coef.	SE	Coeff.	SE
(Constant)	–1.62***	0.53	–1.55***	0.51	–1.65*	0.64	1.10*	0.49	1.79***	0.44
Age	0.04	0.03	0.04	0.04	0.04	0.03	0.01	0.03	–0.03	0.03
Education	0.08	0.04	0.06	0.05	0.09*	0.04	0.01	0.04	–0.09	0.04
Upper caste	–0.10	0.12	–0.13	0.14	–0.07	0.12	–0.18	0.12	–0.08	0.12
Yadav	–0.03	0.13	–0.19	0.14	0.17	0.14	–0.23	0.13	–0.07	0.14
OBC	–0.04	0.12	–0.04	0.13	–0.05	0.12	–0.11	0.10	–0.12	0.11
Muslim	0.01	0.12	–0.09	0.14	–0.10	0.12	–0.04	0.13	–0.10	0.13
Other	–0.14	0.18	–0.07	0.18	–0.21	0.22	0.06	0.16	0.17	0.16
Female	0.22**	0.07	0.23*	0.09	0.21**	0.08	0.18*	0.08	0.11	0.07
Econ status	–0.04	0.03	–0.04	0.04	–0.05	0.04	–0.03	0.03	–0.01	0.04
Civil assoc	–0.08	0.17	–0.04	0.19	–0.13	0.18	–0.37*	0.17	–0.01	0.16
Caste assoc	0.40*	0.15	0.44**	0.16	0.34	0.17	–0.01	0.14	0.19	0.14
Muslim %	0.42	0.29	0.55	0.31	0.26	0.32	–0.14	0.33	–0.03	0.29
Soc diversity	–0.05	0.05	–0.07	0.06	–0.02	0.06	0.06	0.05	–0.07	0.05
Riot prone	0.34***	0.12	0.35***	0.12	0.34**	0.13	0.15	0.12	0.16	0.09
ENP	0.26**	0.09	0.27***	0.08	0.24*	0.10	0.13	0.09	0.15	0.08
Victory margin	0.98	0.66	0.83	0.65	1.14	0.72	0.98	0.67	0.77	0.48
N	551		551		551		434		452
R-square	0.116		0.113		0.100		0.070		0.053

Notes: Standard errors are clustered by constituency. Reference categories are: Male, Scheduled Caste, N = 560.

*** p < 0.005; ** p < 0.01; * p < 0.05.

From Model 1 we can see that ethnic security fears are significantly higher among women than men and significantly higher among people who are members of caste associations than those who are not. However, none of the other individual level variables, to do with age, education, caste community, economic status or membership of civil associations are significant. These findings provide partial support for Varshney’s civil society hypothesis.¹³ Although membership of civil society organisations, as operationalised here, is not well equipped to explain variation in ethnic security fears this may in part reflect difficulties in measuring non-ethnic civil associations, since many of the associations in question may be ethnically exclusive in character. By contrast, the finding that membership of ethnic organisations, such as caste and religious associations, is associated with security fears is much more consistent with Varshney’s general argument. The results also suggest that even if Muslims are frequently the target of ethnic violence (see Wilkinson, 2004), they do not have greater security fears than other communities. However, women, who are also often targets of ethnic violence – see Appadurai (1998) on the renewed salience of rape in ethnic violence – do exhibit greater security fears than men.

Turning to the constituency level variables we can see that the riot history term is positive (b = 0.34 in Model 1) and highly significant. Understandably, people who live in places where there have been serious outbreaks of violence tend to be much more concerned about security issues than people who live in relatively peaceful areas. However, even taking this into account, the effective number of parties has a significant and substantively strong impact on security fears. There is thus strong support for our core hypothesis that ethnic security fears are shaped by the structure of party competition. The ENP term is positive (b = 0.26) and highly significant. This indicates that ethnic security fears are much greater among people living in multiparty constituencies than among people living in two party constituencies. The effects of a 1 to 5 change in the number of effective parties on the party competition variable yields an increase of 1.4 points in the level of ethnic security fears.

However, there is little support for the idea that ethnic security fears are higher when electoral contests are competitive. The coefficient for victory margin is small and does not have a significant association with security fears.¹⁴ Similarly, the social composition of the constituency does not have much of an impact on ethnic security fears, and although the direction of the coefficients for Muslim population size are in the expected direction they do not achieve statistical significance.¹⁵ The social diversity term is not significant either and the coefficient is very close to zero. Security fears are not any higher among people living in socially mixed areas than they are among people living in more socially homogenous areas.¹⁶

As a robustness check we replicate our model with different specifications of the dependent variable, distinguishing between ethnic security fears related to large scale episodic events, such as mob riots and terrorism (Model 2), and more mundane forms of everyday harassment and intimidation (Model 3). If security fears are mobilised by elites directly, then the effects of multiparty competition may exert a stronger influence on security fears related to riots and terrorism since these acts of ethnic violence are more often politically motivated, and so fears may be more prone to political manipulation. However, if security fears are a bi-product of multiparty competition and are an indirect consequence of different social groups being mobilised in opposition to each other, which then makes people more fearful of everyday social interactions, we would also expect a strong and significant effect on the more mundane forms of everyday fears as well. By contrast, multiparty competition should exert a somewhat weaker impact on domestic and state security fears since these issues are less closely related to ethnic mobilisation.

We can see that the parameter estimates for each of the first three models are very similar. Caste association, riot history and multiparty competition all have a similar effect on the different specifications of ethnic security fears and the magnitude of the coefficients are very similar across the models (although the coefficient for caste association is imprecisely estimated in Model 3 and does not reach conventional levels of significance). By way of comparison the structures of domestic and state security fears are somewhat different. They are not related to membership of caste associations – so lack the ethnic dimension; they are not related to riot history – and so lack the conflict dimension; and are not related to party competition – and so lack the political dimension. Interestingly however, fears about domestic violence are somewhat lower amongst people who participate in civil society associations, perhaps indicating that some of these organisations act as a valuable source of support.

As one additional robustness check, we also replicate our model on each of the separate indicators that make up our ethnic security variable to see if there are any important differences between them (see Table 2). Once again we can see that the main findings hold up well, and although there is some slight variation in the magnitude and significance of the main predictors, the effect of ENP is remarkably consistent and is significant and in the expected direction in each of the models. By contrast there is some slight variation between the models in terms of the significance of the riot history variable and the caste association variable. Although in the expected direction, the coefficient for riot history is imprecisely estimated and does not reach conventional levels of significance in Model 4, where the dependent variable is fears about theft and robbery; and the coefficient for caste association does not reach significance in Model 1, where the dependent variable is fears about riots and mob violence. In addition there are some modest caste-community effects in Model 3: Yadavs are significantly less concerned about physical attack and harassment than Scheduled Castes. This is perhaps not surprising, as Scheduled Castes still – even today – suffer caste discrimination and are frequently the victims of day to day harassment. Lastly, the margin of victory term is significant and positive in Model 2: people living in less competitive constituencies tend to be more concerned about terrorism. This is slightly counter-intuitive, and given that it is only a one-off finding it is probably worth not reading too much into it. Indeed, given that the single items are more likely to be affected by measurement error than the latent constructs, we should treat the results with some caution. However, the main point is that even when we disaggregate ethnic security fears in this way the main finding of ENP still holds.

Table 2.

Security fears, robustness check, OLS regression.

	Riots and mob violence		Terrorism		Physical attack and harassment		Theft and robbery
	Coef.	SE	Coef.	SE	Coef.	SE	Coef.	SE
(Constant)	1.24***	0.40	1.61***	0.45	1.24***	0.40	1.02**	0.37
Age	0.02	0.03	0.02	0.03	0.02	0.03	0.03	0.02
Education	0.06	0.03	0.05	0.04	0.03	0.03	0.05	0.03
Upper caste	–0.10	0.10	0.10	0.10	–0.17	0.11	–0.06	0.10
Yadav	0.10	0.12	0.24	0.13	–0.22*	0.10	–0.05	0.12
OBC	–0.13	0.11	0.06	0.12	–0.12	0.10	0.01	0.10
Muslim	0.11	0.12	0.10	0.10	0.17	0.11	0.17	0.11
Other	–0.01	0.11	–0.18	0.16	–0.05	0.11	0.03	0.10
Female	0.19**	0.07	0.05	0.04	0.12	0.07	0.15*	0.06
Econ status	–0.04	0.03	–0.04	0.03	–0.05	0.03	–0.02	0.03
Civil assoc	–0.02	0.13	–0.11	0.15	0.03	0.13	–0.04	0.15
Caste assoc	0.13	0.13	0.28*	0.14	0.25*	0.12	0.30*	0.12
Muslim %	0.42	0.28	0.16	0.29	0.51	0.27	0.55*	0.26
Riot prone	0.22**	0.08	0.28**	0.10	0.29***	0.10	0.17	0.10
Victory margin	0.39	0.48	1.31*	0.62	0.50	0.51	0.84	0.49
ENP	0.15*	0.07	0.22*	0.10	0.19**	0.07	0.26***	0.07
N	437		443		479		496
R-square	0.10		0.11		0.12		0.13

Notes: Standard errors are clustered by constituency. Reference categories are: Male, Scheduled Caste.

*** p < 0.005; ** p < 0.01; * p < 0.05.

These results support our central hypothesis that party competition influences ethnic security fears specifically, rather than just feelings of insecurity more generally. One final consideration though is to what extent variations in the structure of party competition are actually a cause of ethnic security fears as opposed to being a response to ethnic security fears. An alternative causal narrative is that party systems are more fragmented in constituencies characterized by high levels of insecurity. Thus, in places where there are high levels of ethnic security fears it is harder for parties to build the broad social coalitions that underpin two party contests. There is thus an issue of endogeneity that we must address in order to have confidence in our results.

When reciprocal causation is suspected, a common approach is to use Two-stage least-squares (2SLS) estimation and to instrument the endogenous variable (Sovey and Green, 2011). In effect this involves replacing the problematic independent variable with a variable that is unaffected by the outcome. A good instrument therefore needs to be correlated with the endogenous variable but not caused by the dependent variable. A common strategy is to use lags of the endogenous variable (Sovey and Green, 2011). Although not ideal, the idea is that while levels of ethnic security fears may influence the effective number of parties and vice versa it is less likely that current levels of ethnic security fears can influence past effective number of parties.¹⁷

The use of a lagged instrument is particularly appropriate for the task at hand since, due to the large scale delimitation exercise undertaken in India between the 2004 and 2009 elections, the constituency boundaries in 2004 are somewhat different to the constituency boundaries in 2009. There is thus less likely to be a direct relationship between past effective number of parties and current levels of security fears, since the political context for many voters will have changed.¹⁸ However, past effective number of parties should still be correlated with current effective number of parties since there is at least some overlap between the constituency boundaries.¹⁹

In order to estimate the 2SLS model we aggregate individual level ethnic security fears to the level of the constituency. This allows us to explore whether the level of security fears in a constituency influences the effective number of parties and vice versa. For the 2SLS models, we estimate two structurally related equations:

\begin{aligned} E t h n i c S e c u r i t y f e a r s = β 0 + β 1^{*} (E N P) \\ + β 2^{*} (M u s l i m %) + β 3^{*} (R i o t s) \\ + β 4^{*} (S o c i a l d i v e r s i t y) + ϵ \end{aligned}

\begin{aligned} E N P = γ 0 + γ 1^{*} {(E N P}_{t - 1}) + γ 2^{*} (M u s l i m %) \\ + γ 3^{*} (R i o t s) + γ 4^{*} (S o c i a l d i v e r s i t y) + ϵ \end{aligned}

Our key coefficient of interest is β1, which represents the average change in the level of ethnic security fears in a constituency expected from competition involving one effective party more, controlling for the other factors in the model. Table 3 reports the results from an OLS and a 2SLS model. The OLS estimates from Model 1 replicate the findings from our previous analysis and show that ENP is a significant predictor of ethnic security fears. As we mentioned above though, endogeneity poses a problem for these results since contemporaneous ethnic security fears may be jointly determined by party competition.

Table 3.

Effect of party system format on ethnic security fears, Two Stage Least Squares.

	Model 1: OLS ethnic security fears		Model 2: 2SLS ethnic security fears
	Coeff.	SE	Coef.	SE
(Constant)	–1.33**	0.48	–1.67**	0.71
ENP	0.27**	0.09	0.37**	0.18
Muslim %	0.80*	0.42	0.79*	0.41
Social diversity	–0.04	0.08	–0.05	0.08
Riots	0.42***	0.15	0.44***	0.15
R²	0.22		0.20
First stage F-test			6.22
ENP			22.91
Weak identification test
Anderson canon. corr. LM stat			18.49 (1)
Cragg-Donald Wald F stat			22.91

Notes: *** p < 0.01; ** p < 0.05; *p < 0.10. N = 75.

We thus turn to 2SLS estimates. For an instrument to be valid, it must satisfy the conditions of relevance and exogeneity. The instruments pass the standard tests.²⁰ The 2SLS estimation shows the marginal effects of party competition to be both statistically significant and substantively important. It is also worth noting that when we use the IV in Model 2 the coefficient on parties becomes larger, which is what we would expect to happen when we filter out the contamination from the reverse causation. These estimates signify a process in which levels of ethnic security are responding to the structure of political competition rather than the other way round.

Conclusion

In this contribution we provide the first large scale investigation into the sources of ethnic security fears in India. We show that individual level variation in ethnic security fears has a strong geographical structure. In line with previous theoretical work we find that people living in ‘riot prone’ areas tend to exhibit substantially higher collective fears of the future. This finding is reassuring for the validity of the study, though not wholly unexpected. However, even taking this factor into account, our findings also show that the structure of party competition has a significant impact on ethnic security fears. In particular, in ethnic party systems security fears are much higher among people living in constituencies characterised by multiparty competition. This finding goes against prior theoretical expectations which have tended to regard multiparty systems as having a moderating effect on ethnic divisions – at least insofar as they relate to ethnic violence (Wilkinson, 2004).

However, higher security fears in multiparty systems should not be surprising. Whereas in two party contests parties have to reach out at the local level and appeal to a broad cross section of society in order to secure a winning majority, in multiparty contests the threshold for victory is much lower and so parties can secure victory by mobilising and appealing to a much narrower section of society. Accordingly harmonious ethnic relations are harder to sustain when ethnic groups are political rivals rather than political allies, and so collective fears tend to be higher.

These findings help to develop existing theory in a number of ways. Previous research on ethnic violence has emphasised how the incentives for political elites to orchestrate violence can vary according to the structure of political competition. Our findings show that in a different way the structure of party competition also has an indirect influence on the potential responsiveness of the public to such appeals. Security fears are widely seen as a precondition for ethnic violence, and although multiparty competition may reduce elite incentives to orchestrate violence directly, it does not appear to do so without a cost. In the process multiparty competition divides communities locally, making harmonious relations harder to sustain.

This point has wider implications for our understanding of how different institutional arrangements influence the production of ethnic violence. For a long time electoral systems which encouraged multipartyism, such as PR, were seen as an effective means of reducing ethnic conflict as they ensured group representation which could then lead to power sharing arrangements. However, one drawback with this system is that parties that represent only one ethnic group have clear incentives to play up ethnicity and harden ethnic divisions at election time rather than to build cross-ethnic coalitions (Selway and Templeman, 2012). For this reason, scholars such as Donald Horowitz (2004) and Benjamin Reilly (2001) have argued instead for the use in divided societies of electoral systems with centripetal effects, such as the alternative vote in single-member districts, which reward candidates who take moderate positions and can attract support from a broad cross section of the electorate. Our results lend support to this notion, and electoral systems which promote ethnic moderation at the constituency level may help to alleviate security fears on the ground. This possibility suggests an interesting avenue for further research.

Footnotes

Acknowledgements

I am grateful to Sanjay Kumar in particular for generously sharing the CSDS survey data, and to Lucia Michelutti, Mike Spagat, Lara Nettelfield and Sandra Halperin for their help and advice on earlier drafts.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Indian-European Research Networking Programme in the Social Sciences (ANR-DFG-ESRC-NWO with ICSSR: Grant number ES/K005936/1).

Supplemental material

The online appendices are available at .

Notes

References

Appadurai

(1998) Dead certainty: Ethnic violence in the era of globalization. Journal of the International Centre for Ethnic Studies 2(2): 1–32.

Brass

(2003) The Production of Hindu–Muslim Violence in Contemporary India. Seattle: University of Washington Press.

Chandra

(2004) Why Ethnic Parties Succeed: Patronage and Ethnic Hounds Counts in India. Cambridge: Cambridge University Press.

Chandra

(2005) Ethnic parties and democratic stability. Perspectives on Politics 3(2): 235–252.

Chhibber

Murali

(2006) Duvergerian dynamics in the Indian states: Federalism and the number of parties in the State Assembly elections. Party Politics 12(1): 5–34.

Chhibber

Nooruddin

(2004) Do party systems count? The number of parties and government performance in the Indian states. Comparative Political Studies 37(2): 152–187.

Diwakar

(2007) Duverger’s law and the size of the Indian party system. Party Politics 13(5): 539–561.

Eifert

Miguel

Posner

(2010) Political competition and ethnic identification in Africa. American Journal of Political Science 54(2): 494–510.

Ellingsen

(2000) Colorful community or ethnic witches’ brew? Multiethnicity and domestic conflict during and after the Cold War. Journal of Conflict Resolution 44(2): 228–249.

10.

Fish

Brooks

(2004) Does diversity hurt democracy? Journal of Democracy 15(1): 154–167.

11.

Hansen

(2001) Wages of Violence: Naming and Identity in Postcolonial Bombay. Princeton, NJ: Princeton University Press.

12.

Heath

Yadav

(1999) The united colours of Congress: The social profile of Congress voters, 1996 and 1998. Economic and Political Weekly 34(34–35): 2518–2528.

13.

Heath

(1999) The BJP’s rise to power: Social, regional and political expansion in the 1990s. Economic and Political Weekly 34(34–35): 2511–2517.

14.

Horowitz

(2000) Ethnic Groups in Conflict. Berkeley, CA: University of California Press.

15.

Horowitz

(2004) The alternative vote and interethnic moderation: A reply to Fraenkel & Grofman. Public Choice 121: 507–516.

16.

Huber

Suryanarayan

(2013) Ethnic inequality and the ethnification of political parties: Evidence from India. Unpublished manuscript.

17.

Kaufmann

(1996) Possible and impossible solutions to ethnic civil wars. International Security 20(4): 136–175.

18.

Lake

Rothchild

(1998) Spreading fear: The genesis of transnational ethnic conflict. In: Lake

Rothchild

(eds) The International Spread of Ethnic Conflict: Fear, Diffusion, and Escalation. Princeton, NJ: Princeton University Press, pp. 3–32.

19.

Marshall

Ramsey Marshall

(2008) Crime in India: Riots, murders and dacoity, 1954–2006. Available at: http://www.systemicpeace.org/inscr/inscr.htm (accessed 24 January 2012).

20.

Medsker

Williams

Holahan

(1994) A review of current practices for evaluating causal models in organizational behavior and human resources management research. Journal of Management 20(2): 439–464.

21.

Michelutti

(2008) The Vernacularisation of Democracy: Politics, Caste, and Religion in India. Delhi: Routledge.

22.

Michelutti

Heath

(2013) The politics of entitlement: Affirmative action and strategic voting in Uttar Pradesh, India. Focaal - Journal of Global and Historical Anthropology 65: 56–67.

23.

Nandy

Trivedy

Mayaram

. (1995) Creating a Nationality: The Ramjanmabhumi Movement and Fear of the Self. Delhi: Oxford University Press.

24.

Piazza

(2010) Terrorism and Party systems in the States of India. Security Studies 19(1): 99–123.

25.

Posen

(1993) The security dilemma and ethnic conflict. Survival 35(1): 27–47.

26.

Posner

(2004) The political salience of cultural difference: Why Chewas and Tumbukas are allies in Zambia and adversaries in Malawi. American Political Science Review 98(4): 529–545.

27.

Putnam

(2007) E Pluribus Unum: Diversity and community in the twenty-first century. The 2006 Johan Skytte Prize Lecture. Scandinavian Political Studies 30(2): 137–174.

28.

Reilly

(2001) Democracy in Divided Societies: Electoral Engineering for Conflict Management. Cambridge: Cambridge University Press.

29.

Saideman

Lanoue

Campenni

. (2002) Democratization, political institutions and ethnic conflict: A pooled time-series analysis, 1985-1998. Comparative Political Studies 35(1): 103–129.

30.

Selway

Templeman

(2012) The myth of consociationalism? Conflict reduction in divided societies. Comparative Political Studies 45(12) 1542–1571.

31.

Sharma

Menon

(2012) Seven riots in Akhilesh Yadav’s six months tenure as Uttar Pradesh CM, IB sets off riots alarm. India Today, 19 September. Available at: http://indiatoday.intoday.in/story/ib–warns–centre–of–volatile–situation–in–uttar–pradesh/1/217850.html (accessed 16 July 2013).

32.

Shastri

Suri

Yadav

(eds) (2009) Electoral Politics in Indian States: Lok Sabha Elections in 2004 and Beyond. Delhi: Oxford University Press.

33.

Sovey

Green

(2011) Instrumental variables estimation in political science: A readers’ guide. American Journal of Political Science 55(1): 188–200.

34.

Tambiah

(1996) Levelling Crowds: Ethnonationalist Conflicts and Collective Violence in South Asia. Berkeley, CA: University of California Press.

35.

van der Veer

(1994) Religious Nationalism: Hindus and Muslims in India. Berkeley, CA and Los Angeles, CA: University of California Press.

36.

Varshney

(2001) Ethnic conflict and civil society: India and beyond. World Politics 53: 362–398.

37.

Varshney

(2002) Ethnic Conflict and Civic Life: Hindus and Muslims in India. New Haven, CT: Yale University Press.

38.

Wilkinson

(2004) Votes and Violence: Electoral Competition and Ethnic Riots in India. Cambridge: Cambridge University Press.