Ethnic inequality, political mobilization and the ethnification of political parties across states in India John D. Huber Pavithra Suryanarayan June 17, 2012 Abstract This paper examines the factors that lead to the ethnification of party systems by studying variation in the group basis of party support across the Indian states. We contrast two theories of why group-based support for parties emerges elite mobilization efforts or ethnic head counting to build winning coalitions, and levels of between group inequality. Using survey data from the Indian states in two national elections, the paper finds that the category of jati (sub-caste) causes the highest level of party system ethnification compared to religion or caste. The analysis also reveals a robust negative relationship between party mobilization efforts and party system ethnification, but that mobilization has a positive effect in highly competitive states, providing some evidence for the elite mobilization theories. Most significantly, the paper reveals a very large and robust effect of inequality between groups on party ethnification, suggesting that where identity politics is strong, it is often class politics in disguise. Professor, Department of Political Science, Columbia University, jdh39@columbia.edu. Ph.D student, Department of Political Science, Columbia University, ps2550@columbia.edu.
1 Introduction In ethnically divided democratic systems, political parties often seek to build support from specific groups, leading to the ethnification of party systems and electoral politics. The degree to which this occurs, however, varies considerably across political systems, with some characterized by parties with distinctive group orientations and others by political parties that lack any clear ethnic base. This paper examines the factors that lead to the ethnification of party systems by studying variations in the group basis of political support across the Indian states. We are specifically interested in exploring two distinct explanations for party system ethnification. The first and most common explanation for ethnically-oriented electoral politics emphasizes the utility of ethnic categories in political competition. This perspective, which has been emphasized in a wide range of research (Bates 1983, Horowitz 1985, Chandra 2004 and Posner 2005), holds that ethnicity provides a marker that elites can easily use in efforts to target voters and form electoral coalitions. Electoral politics is akin to an ethnic head count or a minimum winning coalition, where parties seek to represent winning coalitions of groups by strategically employing appeals to particular group identities, and where voters strategically invoke the particular identities that are most advantageous to them in efforts to become part of a winning coalition. The stakes of such ethnic competition are high, as the parties that win control of government distribute government resources to the groups that support them. From this perspective, the specific attributes of group members are essentially irrelevant. Ethnic identity creates a largely exogenous marker that simplifies targeting, coalition building, and exclusion from government resources. It does not matter if group members have a particular set of policy preferences or ideologies. Instead, what is important is that individuals can be identified for inclusion in government patronage (or exclusion from it) based on their group membership. An alternative explanation for the ethnification of party systems has received less attention in the ethnic politics literature. This explanation centers on the possibility that ethnified party systems emerge from the attributes of groups themselves, and in particular from a strong relationship between group identity and economic well-being, or class. Ethnicity and class can be strongly related, with some groups being much richer on average than others. When this is the case, party politics involving appeals based on income redistribution or class would result in ethnified party 1
systems, even if there was no overt group-based electoral appeals. If ethnicity provides a cue about economic well-being, then ethnicity also provides a clear indication to parties about the potential policy and redistributive demands of the groups members. Parties therefore compete for votes by pursuing policies and ideologies that appeal to the well-being of particular groups by virtue of the economic position of these groups. The substantive implications of party ethnification should depend on which explanation is most powerful. If the ethnic head count logic is the driving force behind party system ethnification, then high levels of such ethnification imply a high level of group-based identity politics, where a voter s perception of his or her place in society as well as of access to government resources is intimately tied to characteristics of the identity into which one is born. 1 By contrast, if high ethnification results largely from a strong correlation of identity with class or income, then the substantive impact of identity on political perceptions and outcomes is much less clear. Indeed, it become difficult to disentangle identity and class politics, and what we perceive as identity politics may largely be class-based politics in disguise. It is therefore important to understand the relative contributions of these two explanations to explaining party system ethnification in order to form clear expectations about the consequences of such ethnification. To explore empirically party system ethnification, we utilize a measurem Party voting polarization, described in Huber (2012). Using survey data from two national election studies in India (1999 and 2004), we are able to measure party system ethnification across the different states using the three central definitions of identity in India: religion, caste and subcaste, or jati. We have two specific goals. The first is to determine which definition of group identity religion, caste or jati is associated with the highest levels of party system ethnification. We find that party system ethnification is highest using jati to define group membership and is lowest using religion. The second goal is explore whether one of the two explanations described above has greater purchase in explaining cross-state variation in party system ethnification (using jati, the group identity associated with the greatest ethnification). We argue that a variable tapping elite mobilization efforts can be used to test ethnic head count arguments, and that a measure of between-group inequality ( BGI ) can be used to test arguments about the correlation of identity and class. The empirical analysis shows a very strong and robust relationship between BGI and ethnicfication of 1 Voters might belong to multiple ethnic categories and choose to vote on the basis of one type versus another. 2
party systems the more inequality there is between ethnic groups, the higher the levels of ethnification, suggesting ethnic politics in India has a stronger economic or class orientation than perhaps has been previously recognized. By contrast, the results for mobilization are more ambiguous. We find a robust negative relationship between ethnification and mobilization efforts by elites. But we also find that though the effect of mobilization is negative in most states, in the states where electoral competition is most intense, there is a positive relationship between mobilization and ethnification. The paper is organized as follows. The next section discusses our motivation for focusing on state politics in India. This is followed by an elaboration the two explanations for ethnification of party systems. We then present the measure of party system ethnification that will be used for the cross-state analysis. The two main substantive contributions follow. First we show that using jati groups rather than religion or caste yields the highest levels of party system ethnification. Then we present empirical models of variation in ethnification across the states. We conclude by discussing the implications of our results. 2 Why focus on state politics in India? We examine the utility of the two explanations for party system ethnification across the Indian states for several reasons. First, exploring these two pathways within a single country avoids some of the difficulties associated with cross-national studies of ethnicity. In particular, across countries, the institutional context (like electoral rules and decentralization) can vary considerably, influencing the ethnification of electoral politics (Huber 2012); the substantive meaning and importance of ethnic categories can vary considerably; and economic differences across groups can take on quite different meanings in different societies. Across the Indian states, the electoral rules and the menu and meaning of ethnic identities are largely the same (even though the categories themselves might vary), and while the states differ in their economic development, individuals across the states find themselves in essentially the same economic framework. Second, the Indian states are important and largely autonomous entities because of both the formal devolution of economic power to states as well as the discretion given to state actors in the implementation of centrally funded schemes (e.g., Chhibber and Nooruddin 2004). Even 3
national parties such as the Congress and the BJP modify their organizational structures, candidate nomination processes and electoral strategies to suit the state s political climate. Moreover, electoral constituencies are drawn and redrawn within state boundaries and elections are managed by the states administrative machinery under the guidance of a national election commission. The decline of the Congress party through the 1980s, the rise of regional and caste-based parties, the devolution of fiscal and economic planning in the post-1991 period and the ever increasing levels of discretion given to state administrative officials in managing and implementing federal projects has only intensified the degree to which politics is localized to states. It is therefore possible to examine variation in the ethnification of parties across states, and in particular whether factors associated with the ethnic head count or class-group relationships are particularly useful in explaining this variation. Third, a central challenge in studying party system ethnification is that individuals often have multiple identities, with political considerations playing a central role in determining which specific identities become politically salient in a given polity at a given point in time (e.g., Posner 2004). In Indian politics, scholars have focused on three different identities that have become salient political cleavages: religion (e.g., Brass 1993, Wilkinson 2004, Jaffrelot 2005 and Tachil 2010), caste (e.g., Srinivas 1957, Rudolph 1965, Krishna 2003 and Chandra 2004) and subcaste, or jati (e.g., Chhibber 1999 and Chandra 2004). 2 Although there exists this menu of identities around which parties and individuals can mobilize, previous studies have tended to focus on one type of group identity or another, leaving unanswered the question of whether one type of identity is in fact associated with the greatest degree of party system ethnification. Examining which group type leads to the greatest party system ethnification will therefore answer an important question about Indian electoral politics. Doing so will also shed light on the dynamics by which elites and voters employ ethnic identity in electoral politics. Understanding the effects of groups definition on the party system is also critical because ethnic categories in India are related to each other, with jati (hereditary groups based in large part on historical occupational categorization) nested within broad caste umbrellas (politicized social status categories), and caste nested within religion (even non-hindu citizens in many instances 2 Some have also focused on language (e.g., Fearon (2003), as part of a large cross-national study of language differences across groups), although in studies of India, language differences are seldom viewed as politically salient any more. 4
claim to have a jati or a caste). Moreover, given the way that the groups are defined, the level of economic differences across groups varies with group type, with these differences being smallest for religion and largest for jati. Thus, if we can identify which group identity leads to the most ethnified party system, we can get some insight into whether the size and economic conditions of groups are important factors in how party elites attempt to politicize ethnicity in electoral politics. Finally, the focus on India is important because of the the centrality of ethnicity in the study of Indian politics. Scholars have provided a range of explanations for why certain cleavages becomes activated. Brass (1993) argues that the vote for the hindu-nationalist Bharatiya Janat Party (BJP) is directly linked to the portion of muslims in an area. Rudolph and Rudolph (1993) argue that economic competition between hindus and muslims (as determined by their occupational patterns in the area) leads to greater salience of religion in politics. Chhibber (1999) argues that variations in the institutions of federalism, strong state autonomy and weak associational life lead to differences in the politicization of cleavages rather than in the structure of cleavages themselves. More recently, the instrumental view has gained traction with scholars emphasizing the strategic use of ethnicity by politicians in low information or highly competitive environments to win elections (Chandra 2004, Wilkinson 2004). While scholars have examined the historical, economic, associational, strategic and emotional appeal of ethnicity in politics, the alternative possibility that much of the ethnification of party politics maybe due to a correlation of group with class has been relatively under-studied. This paper therefore represents the first opportunity to carefully study the role of economic differences between groups in explaining variations in the ethnification of state party systems and contrasts this with the prevailing view of ethnic head counts. 3 Two possible explanations for the ethnification of party systems Most theoretical and empirical research on the ways in which parties develop an ethnic basis for support emphasize the intrinsic importance of ethnic categories in political competition rather than the link between ethnic group and class. Horowitz (1985), for example, argues explicitly against the idea that ethnicity is a mask for social class, or that ethnic affiliations simply provide a means for elites to satisfy class aspirations. Instead he maintains that ethnicity creates intrinsically compelling affiliations that are more important to individuals than class, which is why elites mobilize groups 5
using ethnic appeals instead of class appeals. He goes so far as to argue that ethnic voting occurs because of its expressive value, and given that ethnic identity is the over-riding factor shaping voting behavior, party elites must employ ethnic strategies to win elections. A key reason for the permeative role of ethnicity is that unlike class, ethnicity is rooted in descent. People are able to transcend their class in a lifetime or over a single generation in a way they are unable to do so with ethnicity. Consequently, while Horowitz acknowledges that ethnic societies can differ in the extent to which class and ethnicity coincide (i.e. ranked and unranked systems), he downplays the role of class in the politicization of ethnicity. Horowtiz s seminal work as been enormously influential, but scholars have also recognized that the argument is incomplete. If ethnicity is over-riding determinant of voting behavior, why do we see the ethnification of some party systems and not others? Political institutions clearly play an important role, but there remains widespread variation in the extent to which party systems have an ethnic foundation. Chandra (2004) provides an important step toward answering this in her study of patronage-based politics across Indian states. Her argument emphasizes not the expressive value of ethnicity (which she rejects), but rather its strategic value. Ethnic categories provide an essentially costless cue to party elites and voters about how electoral outcomes will map to material benefits from the state. Specifically, voters believe that access to government benefits will depend on having co-ethnics in power, and party elites commit to favoring particular ethnic groups by explicitly courting their support. Thus, electoral competition revolves around party elites mobilizing members of particular groups in efforts to win an ethnic head count. As with Horowitz, the attributes of group members themselves such as whether they tend to be poor or otherwise different or excluded are irrelevant. Instead, group identity alone is sufficient to generate cohesive voting behavior by groups, and the challenge to parties is to get their group members to the polls. Similarly, Posner (2005) argues that voters have an incentive to put just the right coalition of ethnic groups into power in order to maximize their returns from political office, creating rules of thumb for opting into and out of identities. In Zambia, the institutional configuration of the party system determines which ethnic cleavage becomes salient in the minds of the voters. During periods of multi-party rule, political competition takes on a national dimension and voters are encouraged to mobilize and to form minimum winning coalitions along linguistic lines. In periods of single-party rule, competition becomes localized and cleavages develop along tribal lines. While he provides an 6
instrumental logic for voter mobilization, he identifies the party system as the key institution and therefore parties play an important role as the first movers in mobilizing voters. If party politics follows the patterns suggested by authors such as Chandra or Posner through ethnic head counts or minimum winning coalitions, then the nature of substantive differences between groups such as class or other economic differences should explain little variation in the ethnification of parties. The main consideration for a group-member in deciding whether to support an ethnic party (and for party elites to compete for ethnic votes) should be whether the ethnic strategy is viable given a group s (or party s) size. Thus, the key to explaining cross-system differences in the ethnification of parties is to understand strategic incentives of party elites to mobilize voters. In the case of India, Chandra (2004) argues that such mobilization efforts are not based on general policy or issue concerns because voters are held to vote strictly based on group identity (caste). This reinforces the idea that parties mobilize along ethnicity as rule of thumb for sharing state patronage rather than by bringing transformative policy to bear on the state. Instead, efforts to mobilize voter support emphasize caste-based linkages between voters and parties. Quoting one BSP leader, Chandra (2004, 212) writes: In India, people do not cast their votes, they vote their caste. The central challenge, then, is to understand when parties find it in their interest to mobilize at all. In this regard, Chandra focuses on electoral competitiveness: for voters to believe it worthwhile to support the party of their group, they must believe that their support for their group s party will influence whether this party in fact wins control of government resources. If a party is sure to win or lose, an individual knows their support will not affect outcomes, but when the election is expected to be close, voters will have be most susceptible to mobilization efforts. Mobilization efforts should therefore lead to greater ethnication of party systems, particularly when elections are competitive. Most research on ethnic identity follows this idea that ethnicity is an independent force that trumps class, and we are aware of no research on the ethnification of party politics that emphasizes the importance of a link between group and class. But there is a growing body of recent research on ethnic politics that finds that the negative effects of ethnic divisions are greatest when these divisions are rooted in economic differences between groups. Baldwin and Huber (2010), for example, find that high ethnic diversity and general inequality are not correlated with levels of 7
public goods provision, but when there exists inequality between groups, public goods provision is lower. Cedarman et al (2011) show that economic differences between groups not simply group fractionalization or polarization contributes to inter-ethnic violence. And Alesina et al (2012) show that low levels of economic development are associated with high levels of economic differences between groups. This research on the effects of group-based inequality suggests the possibility that the intrinsic importance of group identity may not be as central to explaining ethnic electoral politics as many imagine. Instead, ethnic electoral politics may be due to the relative economic well-being of groups. Research on the US illustrates well this possibility that class might play an instrumental role in conditioning the political behavior of ethnic groups. A vast majority of blacks support the Democratic Party, the left-wing party in the US that advocates for greater redistribution from rich to poor than does the alternative party, the Republicans. Since blacks in the US have been poor relative to whites, black support for the redistributive Democrats makes sense for most blacks even if there is no explicit race-based targeting of government benefits. The Democrats have often been able to count on black support by making general redistributive arguments and without making explicit race-based appeals. And to the extent they do make race-based appeals, such appeals are in the context of affirmative action policies that often transcend a specific race. If ethnic categories and class are correlated for example because some groups are richer than others then a party system with substantial ethnification might simply reflect robust class politics. Voters may be making their decisions based on general redistribution, but such calculations result in what look like ethnically motivated voting. And explicit ethnic appeals in such contexts may have traction precisely because of the economic status of the group members to which such appeals are made. It is therefore important to examine empirically whether patterns of ethnification can be explained by the correlation of economic status and group identity. 4 Measuring the ethnification of party systems One challenge facing any effort to study the source of party system ethnification is to quantify the group basis of political support for parties across political systems under different definitions of group identity, even when such definitions results in a different number of parties or groups. 8
We should be able to say, for example, whether party ethnification in India is largest using an individual s religion, caste or jati, because we want to explain cross-state differences using the measure of identity that seems most important in producing such effects. Recent research on India has examined ethnification at the system-level. Heath (2005), for example, uses cluster analysis to create a polarization measures that gauges the connection of caste and party, and Chhibber, Jensenius and Suryanarayan (2012) create a measure that captures both the dependence of a party on social cleavages and the dependence of the social base on parties. The limitation of these measures in Heath and Chhibber et. al., which use the same six category measure of class, is that they do not allow valid comparisons across states using different definitions of groups, or when the number of parties varies. Other authors have taken a different approach. Chandra (2004), for example, focuses on the extent to which parties make explicitly ethnic-based appeals and uses that to categorize them as ethnic parties. While studying one party can be a very useful strategy for exploring and understanding micro incentives of parties and elites, it does not allow one gauge the overall extent to which a party system is ethnified. 3 Such system-level ethnification, which takes into account the ethnification of all parties, will often be central to determining the degree to which ethnic politics is shaping substantive outcomes of concern. To quantify the extent to which parties have distinctive ethnic bases of support, we draw on Huber s (2012) measure of Party Voting Polarization (PVP). The general goal of the measure is to quantify differences in the ethnic bases of support across parties, where such differences are weighted by party size. There are two steps involved in creating the measure. The first is to measure the distance, r ij, between the ethnic profiles of each pair of parties, a distance that is determined by comparing the proportion of support that each party receives from each ethnic group. The distance is zero for a pair of parties when the group-basis of support for each party is the same. Suppose Party A, for example, receives 30 percent of its support from the Green group and 70 percent of its support from the Purple group. Suppose that the same was true for Party B. Then there would be zero distance in the ethnic basis of support for these two parties. By contrast, suppose that all 3 In addition, classifying the ethnification of parties based on the nature of appeals may also have some disadvantages. One is that such classification requires subjective judgments that are difficult to replicate. In addition, the nature of appeals may mask the degree to which ethnicity is important in electoral politics. Voters may ignore appeals, or the may behave in an strongly ethnic fashion independent of such appeals. We therefore believe it is important to examine ethnification by explicitly measuring the ethnic basis of support for parties. 9
of Party A s support came from the Green group and all of Party B s support came from the Purple group. This would yield the maximal distance between the ethnic basis of support for each party. The distance r ij is formally defined as r ij = 1 G (Pg 2 i Pg j ) 2, (1) g=1 where Pg i and Pg j are the proportion of supporters of parties i and j who come from group g, and there are G groups. For two hypothetical parties, A and B, Table 1 provides two examples of the calculation of r AB. 4 In both examples, there are three groups, and we calculate the distance between the two parties using the composition of group support for each party. In the top example, Party A receives 40 percent of its support from Group 1, 35 percent of its support from Group 2, and 25 percent of its support from Group 3. Party B has a fairly similar profile, but a smaller proportion of its support comes from Group 1 and a larger proportion of its support comes from Group 3. Thus, r AB =.15. In the second example at the bottom of the table, the support base for each party is skewed toward a particular group, with Party 1 receiving 80 percent of its support from Group 1 and no support from Group 3, and Party 2 receiving 80 percent of its support from Group 3 and none from Group 1. Thus, r AB =.8, which is considerably larger than in the top example. The maximum r ij is 1, which occurs when all of party A s support comes from one group, and all of party B s support comes from another group. The minimum distance is zero, which occurs when the proportion of support that comes from each group for party A is the same as the proportion of support that comes from each group for party B. In general, as the distribution of groups supporting any two parties becomes more similar, r AB will decline. The second step is to aggregate these measures of ethnic differences between parties into a system level measure, something that requires one to take account of party size. Suppose that there are three parties, each with its own unique ethnic basis of support (so that the ethnic distance between each party is 1). Should the measure of ethnification be independent of the size of the parties? If one party received nearly 100 percent of the votes, would this be the same as if the parties were of equal size, or if two parties both received roughly half the vote (with the third 4 This example is taken from the supplemental materials in Huber (2012). 10
Table 1: Example of the calculation of electoral distance ( r AB ) between Party A and Party B Example 1 Pk A Pk B (Pk A P k B)2 Group 1 0.4 0.25 0.0225 Group 2 0.35 0.35 0 Group 3 0.25 0.4 0.0225 3 (Pk A P k B )2 =.0445 k=1 r AB =.0445 2 =.15 Example 2 Pk A Pk B (Pk A P k B)2 Group 1 0.8 0 0.64 Group 2 0.2 0.2 0 Group 3 0 0.8 0.64 3 (Pk A P k B )2 = 1.28 k=1 r AB = 1.28 2 =.8 party representing a small, negligible group)? As Huber (2012) describes, there are two theoretical frameworks for thinking about how to do address this aggregation issue. One is the fractionalization perspective, which is based on the idea that more equal sized ethnic parties should lead to more governance issues. Suppose that each party received support from one group, and it was a different group for each party. Then from a fractionalization perspective, we should aggregate the party distance measures in a way that yields a larger ethnic parties score as the number of parties proliferate (and, holding the number of parties constant, as the overall support for each party grows more similar). Polarization is the major alternative to fractionalization. It emerged in large part from Horowitz s (1985) argument that many ethnic groups is less a problematic for stable governance than two equal-sized groups. Many small groups, the argument goes, should not lead to substantial ethnification of politics because no group can impose its will. And one very large group should not lead to substantial problems because the dominant group can impose its will. The polarization perspective has gained the upper hand in studies of ethnic politics and civil conflict the intuitive 11
logic provided by Horowitz has considerable appeal, and empirical tests have been quite supportive. 5 If we apply the same logic to parties, we should expect the impact of ethnicity on politics to be greatest when there are two parties, each with different bases of ethnic support, and each roughly the same size. To implement the polarization perspective in our measure of party system ethnification, we can draw on Reynol-Querol s (2002) measure of Ethnic Polarization( EP ) (an adaptation of Esteban and Ray s (1994) measure of income polarization), which is defined as: EP = 1 G ( ) 1/2 2 si s i, (2) i=1 where s i is the proportion of individuals who belong to group i and there are G groups. EP implicitly assumes that the distance between all groups is one. We want to (a) adapt the measure to parties rather that groups, and (b) allow the distance between parties to vary (using r ij ). Following Desmet et al s (2009) adaption of EP to allow the distance between groups to vary between 0 and 1, Huber (2012) adapts the polarization perspective to party ethnification by defining Party Voting Polarization as 1/2 N N P V P = 4 p i p 2 j r ij, (3) i=1 j=1 where p k is the proportion of the vote received by party k. Holding the size of parties constant, the measures increase with increasing distance in the ethnic basis of support for parties. And holding the ethnic distance between parties constant (and assuming it is non-zero), the measures increase as the party system moves toward two parties of equal size. The polarization perspective applied to party systems is particularly attractive in the majoritarian electoral systems found across the Indian states. The first-past-the-post electoral systems will obviously disproportionately reward the top two parties, so the role or impact of ethically-based support for parties becomes less clear in situations where this vote support is fragmented across parties. In addition, in majoritarian systems, we should also expect that when parties have an ethnic basis in politics, electoral stakes will be largest when two competing parties are close to the threshold of victory. 5 See e.g., Reynol Querol 2002 and Montalvo and Reyno Querol 2005. 12
5 Party ethnification in India using different definitions of group identity If voters in India have multiple identities that can become relevant to electoral behavior, is there a specific identity that is associated with the highest level of party system ethnfication? We answer this question using the PVP measure to explore patterns of party system ethnification across the Indian states. The National Election Studies of 1999 and 2004, each conducted in the aftermath of national parliamentary elections, provide the core data for our analysis. 6 The NES surveys are distinctive for their large samples of voters across the Indian states. The surveys make every effort to ensure representation of social, cultural and political groups in the country. The final NES surveys are conducted face-to-face using a structured questionnaire and in the local dialect. 7 Consequently, the NES surveys are some of the most ambitious surveys undertaken on any single national polity. NES 2004, for instance, was conducted in 22 Indian languages and had 27,189 respondents. 8 We exclude two states from the analysis. Jammu and Kasmir has high levels of Hindu- Muslim tensions, and is the focus of international conflict between Pakistan and India. Manipur, in the North East, has had a complex separatist movement dominating politics there since the late 1970s, leading to a strong presence by the Indian army. Given the highly unique situations in these two states, we feel it makes sense to present results that exclude them. It turns out, however, that this decision does not affect the substantive results, which are essentially identical regardless of whether the two states are included or excluded. We measure PVP in the states using the three relevant definitions of ethnic group: religion, caste and jati. The measures are calculated using respondent s self-identification into these groups (when asked separate survey questions for each type of group identity). Figure 1 shows for each type of group the average number of groups (black bar) and the average effective number of groups (gray bar) across states. 9 In most states, religious diversity is relatively low. There are 2.7 religious 6 Elections to the 543 seats in the lower house of the national parliament in India are held every five years. The elections are held under the guidance of the Election Commission of India and according to single member, simple plurality rules. In the two elections under investigation in this paper, the 1999 and 2004 elections, the boundaries of the 543 constituencies remain unchanged in accordance to a freeze in constituency boundaries that had been in effect since 1976. 7 For more information on the NES and other surveys by CSDS, please go to http://lokniti.org. 8 In comparison, the World Values Survey (WVS) of India conducted in 2006 had 2000 respondents and was administered in ten major Indian languages. 9 The effective number of groups takes into account group size and is given by 1/1 ELF. 13
0 5 10 15 Caste Jati Religion Mean Eff. No. of Groups Mean No. of Groups Figure 1: The number of groups using different group identities groups and 1.5 effective religious groups. The state with the most religions has 7, and there are three states (CH, HR and HP) where respondents are all from the same religion (where ethification of parties by religion must be zero). By contrast, the definition of caste in India has seven categories, and each state has at least 6 caste categories named by survey respondents. Thus, the average number of groups using the caste category is 6.7 and the average effective number of groups is 4.2, nearly triple the number of effective groups compared with religion. Jati obviously produces the largest number of groups because the jati groupings are subcategories of caste. Thus, the average number of jati groups across the states is 11.4 and the effective number of jati groups is 7.9, nearly double the the average effective number of caste groups. The fact that the number of groups varies systematically with the definition of group identity is important for two reasons. First, at the extreme of ethnic voting behavior (where there is a oneto-one mapping from party to group) religious identity should lead to the greatest party system ethnification (because such ethnification is greatest when there are two equal-sized groups) and jati is the identity that should lead to the lowest such ethnification (because it is the definition of 14
group that typically leads to the greatest fractionalization in the number of groups). Thus, if religion yields relatively low ethnification (or jati relatively high) it must be because the voting behavior by religion is not particularly cohesive (or, in the case of jati, is particularly cohesive). Second, it reminds us that in determining which definition of group identity yields the highest level of party ethnifcation, we need to control for the underlying level of ethnic group polarization. Suppose, for example, that under each definition of group, there was one party for each group. Then, since jati has the most groups and religion the fewest, we might conclude that party ethnification is lowest using jati and highest using religion. Of course, such comparisons would not be particularly meaningful as they would be driven by the assumptions about how to aggregate scores of parties of different size and nothing else. To think about ethnification of parties under different definitions of groups, then, it is important to control for the underlying level of ethnic polarization in each state. Figure 3 plots PVP against the level of Ethnic Polarization using the three definitions of ethnic group. In addition, for each type of group, the model depicts the linear regression line. We see that PVP ranges from 0 to almost.4. Once states surpass a threshold of ethnic polarization (at about.2), there is considerable variation in the level of PVP at any particular level of EP. At EP equal to roughly.4, for example, PVP ranges from roughly.05 to.4. Similarly, at the highest levels of EP around.8 PVP ranges from almost 0 to nearly.4. Thus, variation in PVP is explained by a great deal more than variation in EP. To illustrate how voting patterns generate some of the differences, Tables 2 and 3 present patterns of party support in Gujarat using religion and jati. Congress and BJP received almost all the votes in Gujarat, both religion and jati have similar values of EP in the 2004 elections at 0.39 and 0.41 respectively, but these translated into a PVP of 0.23 for jati but only 0.11 for religion. Table 2, which describes support for the parties based on religion, indicates that although the BJP gets a larger proportion of support from the Hindus than the Muslims, both the BJP and Congress are supported primarily by Hindus, leading to a relatively low PVP score using religion. If we consider jati (Table 3), we see that several distinct differences across the parties in the ethnic basis of support occur. For Congress, 19 percent of their support is from Other Peasant OBC (vs. 7 percent for the BJP); 17 percent is from Other SC, Ashraf and Chamar/Jatav/Madigas (vs. 3 percent for the BJP). For the BJP, 49 percent of their support is from Rajput and Patel (vs. 19 percent for the BJP) and 5 percent is from Bhamins (vs. 1 percent for Congress). These differences 15
PVP 0.1.2.3.4 0.2.4.6.8 1 EP Religion Caste Jati Religion Caste Jati Figure 2: PVP using different group identities lead to relatively high scores using jati. Of course, we see that differences in PVP scores are not only driven by the definition of identity one uses. Using jati, for example, at particular levels of EP, there can be quite different levels of PVP. Table 2: The Ethnic Basis of Political Support by Religion for Parties Gujarat Religion Congress BJP Hindu 0.85 0.96 Muslim 0.15 0.04 Party Share 45.55 33.86 Which definition of group type yields the highest level of PVP at various levels of EP? Below EP of.8, at any level of EP, the PVP scores using jati (the triangles) are typically greater than the scores using other definitions of group, and the PVP scores using religion (the solid circles) 16
Table 3: The Ethnic Basis of Political Support by jati for Parties Gujarat Jati Congress BJP Other Peasant OBC 0.19 0.07 Other STs 0.19 0.17 Rajput 0.10 0.16 Patel 0.09 0.33 Other SC 0.07 0.02 Ashraf 0.06 0.00 Balmiki/Bhangi 0.05 0.02 Bunkar/Padmasali 0.04 0.03 Chamar/Jatav/Madigas 0.04 0.01 Nai/Mangli 0.03 0.02 Bhil/Yeruluka/Karbi 0.03 0.05 Other Service OBC 0.03 0.04 Oher Muslim OBC 0.03 0.02 Kumhar 0.02 0.03 Bhamin 0.01 0.05 Party Share 45.55 33.86 are typically less than those for caste (the open circles). We can also see this by looking at the regression line, where the solid line for jati is higher than the lines for religion or caste, except at the highest levels of EP (where almost all the data is from religion). And though the highest PVP exists for religion when EP is greater than.8, at this high level of ethnic polarization, religion produces the highest and lowest levels of party system polarization. In addition to this graphical depiction, we can see that jati produces the highest level of ethnic polarization on average across the Indian states by considering how often particular definitions of group identity yield the highest PVP scores. In the 41 state-specific surveys that we have across the two elections, jati produces the highest PVP score 24 times (or 59 percent of the time), caste produces the highest score 14 times (34 percent of the time) and religion produces the highest score only 3 times (7 percent of the time). Finally, when we regress PVP on EP, and indicator for election year, and indicator variables for group type (with caste as the omitted category), the coefficient on the religion indicator is -.03 (p-value.05) and the coefficient on the jati indicator is.05 (p-value.006). This implies that controlling for EP, using religion to tap group identity yields 17
PVP scores that are on average.03 lower than those resulting from using caste, and that are.08 lower than those obtained using jati. And using jati results in PVP scores that are on average.05 greater than those resulting from using caste. These are fairly large differences given that the mean of PVP across all groups is.16 with a standard deviation of.09. The evidence strongly suggest, then, that jati is the definition of group identity that is associated with the highest level of party system ethnification across the Indian states, whereas religion is the definition of group that is associated with the lowest levels of party system polarization. This is particularly striking given that religion has on average the smallest number of groups and jati the largest number. This means that voters within and across jati are coordinating their votes to a very large extent. Since jati produces the highest levels of party system ethnification, our focus will be on understanding cross-state variation in party ethnification using the jati definition of groups. 6 Variation in party system ethnification across the Indian states We now turn to the main objective of our analysis, which is to explore empirically party system ethnification across the Indian states. The ethnic head count perspective posits that ethnicity (rather than class or ideology or preferences) drives vote choice, and that variation in ethnification is driven by elite efforts to mobilize voters. Since voters are looking to elect co-ethnics, when party elites make mobilization efforts, it should be along ethnic lines. Thus, we should see a positive correlation between mobilization efforts and the ethnification of the party system. Alternatively, individuals might respond to appeals based on income or class. If so, then ethnification should be strongest when income and ethnic identity are most strongly intertwined, so that pursuit of one s class interest and pursuit of one s ethnic group interests are difficult to separate. Our goal is to identify variables that tap these two possibilities in order to learn their relative contributions to explaining cross-state differences in party system ethnification (based on jati group identity) in OLS models where the dependent variable is the state-level PVP score. To measure efforts by elites to mobilize voters, we draw on responses to the following question asked in both the 1999 and 2004 NES surveys: Did any candidate, party worker or canvasser come to your house during the campaign to ask for your vote? 18
Mobilization is defined as the proportion of respondents in the state who answer yes to this question. If elite mobilization efforts emphasize ethicity, and if voters are responsive to these ethnic appeals, then the coefficient for Mobilization should be positive. To measure the degree to which group identity proxies for class, we use the NES to calculate between-group inequality (BGI). BGI is a component of the Gini index that is based on the average income differences between groups, weighted by group size. When BGI is large, there is a strong correlation of group and class because the average incomes of group members are strongly differentiated from each other. When BGI is small, there is very little correlation of class and group because the average incomes of groups are roughly the same. The formal definition is BGI = 1 2ȳ ( k m=1 n=1 k p m p n ȳ m ȳ n ), (4) where m and n index groups, p m is the proportion of the population in group m, ȳ m is the average income of group m, and there are k groups in society. To calculate BGI, we need a measure of individual income. In India, many individuals do not have a cash income. We therefore follow the strategy used in a number of previous studies in developing countries that use various asset indicators to gauge economic well-being (see, for example, Filmer and Pritchett (2001), McKenzie (2005) and Huber, Orgazalek and Gore (2012)). Specifically, the Indian election surveys of 2004 and 1999 ask individuals if they own particular items. Both surveys contain the following 7 variables that are associated with economic well-being: Car/Jeep/Van Tractor Television/Color Television/Cable Television Scooter/Motorcycle/Moped Telephone/Mobile telephone Bicycle Pumping set Respondents were given a score of 1 for each asset listed above. To measure a respondent s income we used factor analysis with each respondent receiving a factor score of asset ownership 19
by comparing her level of asset ownership with other respondents in the same state. Respondents factors scores were then rescaled to their percentile rank (thus ranging from 0 and 100), which allowed us to perform the Gini decomposition on the income values. Figure 3 shows a scatter plot of PVP and the two independent variables of central interest. As noted above, in examining the relationship between PVP and any right-hand side variable, it is important to control for the underlying level of Ethnic Polarization. The values of PVP plotted in the figure are therefore the residuals from a simple regression where we control for EP. The top panel shows the bivariate scatter plot of PVP and Mobilization. There is a reasonably strong negative relationship between the two variables, with states that have a high level of pre-electoral party mobilization efforts having a lower level of party system ethnification. The bottom panel shows a strong positive relationship between PVP and BGI. This is consistent with the idea that ethnic voting behavior is proxying for class interest in states where an individual s jati identity distinguishes him or her from others economically, support for political parties typically has a clear ethnic basis. In states where jati is most weakly correlated with economic well-being, the support bases of parties typically do not have a clear ethnic orientation. Do these relationships depicted in Figure 3 survive when controlling in an OLS regression framework for other variables that could affect PVP? Table 4 presents a number of statistical models where PVP is the dependent variable, BGI and Party Mobilization are the right-hand side variables of central interest, and a number of other controls are included as well. All of the variables are standardized to have a mean of zero and standard deviation of 1 in order to facilitate comparisons of the coefficients. 10 As noted, it is important to control for EP (so that we know variation in EP is due to variation in voting patterns rather than variation in the number and size of groups). Model 1 presents results where the right-hand side variables include only BGI, Party Mobilization, EP and an indicator variable for the 2004 survey to explore whether there were systematic differences across the two elections. We see that the coefficient of BGI is the largest in magnitude and the most precisely estimated a one standard deviation increase in BGI is associated with an increase in PVP of 0.53 standard deviations. This effect is even larger than that of EP itself, the coefficient for 10 A coefficient of 1 therefore indicates that an increase of one standard deviation in a right-hand side variable is associated with a 1 standard deviation increase in PVP. 20
PVP, residual controlling for EP -.2 -.1 0.1.2 Party Contact and PVP using Jati groups Controlling for EP AS99 GU99 OR99 BH99 MP99 MG04 PU99 RJ99 TN99 GU04 HR99 UP99 MH04 JH04 AS04 AP99 KT99 WB99 TN04 OR04 TR04 MH99 NG04 MZ04 HR04 HP04 UP04 RJ04 MP04 BH04 PU04 AR04 KR99 KR04 GO04 AP04 SK04 UT04 WB04 1 1.5 2 2.5 Party Contact CH04 KT04 PVP, residual controlling for EP -.2 -.1 0.1.2 TR04 NG04 MZ04 BGI and PVP using Jati groups Controlling for EP SK04 HP04 KR04 MG04 GO04 MH04 UT04TN99 MH99 TN04 AS99 KR99HR04 GU99 GU04 KT04 AS04 RJ04 MP04 BH04 JH04 WB04 PU04 AR04 UP04 WB99 UP99 RJ99 MP99 AP99 KT99 CH04 AP04 OR04 0.1.2.3 BGI HR99 PU99 OR99 BH99 Figure 3: Party ethnification, BGI and mobilization 21