ISSN (Print) - 0012-9976 | ISSN (Online) - 2349-8846
Reader Mode
-A A +A

Growing Cleavages in India?

Evidence from the Changing Structure of Electorates, 1962–2014

Abhijit Banerjee ( is with the Massachusetts Institute of Technology, Cambridge, Massachusetts. Amory Gethin ( and Thomas Piketty ( are with the Paris School of Economics, Paris.

This paper combines surveys, election results and social spending data to document a long-run evolution of political cleavages in India. The transition from a dominant-party system to a fragmented system characterised by several smaller regionalist parties and, more recently, the Bharatiya Janata Party, coincides with the rise of religious divisions and the persistence of strong caste-based cleavages, while education, income and occupation play a diminishing role (controlling for caste) in determining voters’ choices. More importantly, there is no evidence of the new party system of being associated with changes in social policy, which corroborates the fact that in India, as in many Western democracies, political conflicts are increasingly focused on identity and religious–ethnic conflicts rather than on tangible material benefits and class-based redistribution.

What governs the choice of who to vote for in India? How has it changed over time? A claim that is often heard is that the traditional cleavages of caste and religion have been shrinking over time and that this process accelerated because of Narendra Modi’s leadership of the Bharatiya Janata Party (BJP), which placed it on a broad and inclusive platform around the theme of development. Milan Vaishnav (2015), while summarising the 2014 Indian elections for the Carnegie Endowment for International Peace, writes:

Economic factors played an unusually large role in shaping voting behavior. Traditional patterns of caste-based voting were much less evident, and regional parties, often thought to be gaining ground, suffered a setback. A slightly deeper look, however, reveals that these changes were not necessarily unique to the 2014 general election. There is evidence to suggest that many of these trends have been percolating beneath the surface for some time. What 2014 has done is to bring these trends to the fore of public consciousness.

This emphasis on economic factors over social factors (such as caste and religion) makes a certain amount of theoretical sense in the context of India’s rapid growth accompanied by growing inequality (Chancel and Piketty 2017; Bharti 2018) which opens a space for competition over different types of government interventions. Indeed, the BJP’s 2014 campaign emphasised an aspiration to change the relation between the citizen and the state. But is it actually happening—are people in different economic positions voting very differently? And are social factors becoming less predictive of voting patterns? To answer these questions, we make use of post-electoral surveys both for national and the state elections over the period of 1962 to 2014 (though the data coverage is often patchy). The evidence shows some very clear patterns.

Sectarian Divide of the Indian Electorate

The role of caste in predicting support for what are conventionally described as parties of the right (the BJP, Shiv Sena, and Akali Dal) has not diminished over 1999 to 2014. The upper castes were always much more likely to support these parties than the rest of the electorate and this continues to be the case. There was a sharp increase in the upper caste bias in the support for these parties in the mid to late 1990s, which has come down somewhat since its peak, but if anything, the bias was higher in 2014 than in 2009. By contrast the Scheduled Castes and Scheduled Tribes (SCs/STs) have always had a bias against the right, but that bias has not really changed very much since the end of the 1990s. The only group where we see a sharp change in their support for the right are the Muslims, who, interestingly, were moving closer to the non-Muslim population in terms of their support for the right between 1998 and 2009, but that trend was sharply reversed in 2014. These results are robust to just focusing on the BJP (as against the right) and including controls for voter characteristics other than caste (education, occupation, state of residence, etc).

The story of the Congress is partly the flip side of this. In the 1960s and early 1970s, it was the hegemonic party, but this changed by the 1990s, when both the right and centre-left parties (including a number of caste-based parties) ate into their support base. Nonetheless, the bias against the Congress among upper castes became less sharp since the late 1990s due to a shift in their support away from the centre-left and left parties, which recovered a bit since. Likewise, the positive bias among the SCs/STs has declined slightly over the last 60 years. in part, due to a shift in their support towards the centre-left and left parties.

While there was a sharp economic cleavage among right voters in the 1960s and 1970s—they were more educated and richer than the rest of the population, even after controlling for their location, their caste, religion and other demographics—this effect became weaker in the 1990s and disappeared in the recent years. Correspondingly, the support for the Congress among university graduates has risen relative to its overall support in the population since 1970s, and in 2014 was indistinguishable from that of the BJP. In other words, the caste and religion-based schisms remain sharp, but the economic cleavages seem to have mostly disappeared over the last decades. It is only in this limited sense that the support for the right is now more broad-based, as a number of commentators have suggested (Vaishnav 2018).

These results are broadly confirmed by the results from state assembly elections. While our data here only starts in 1996 for Uttar Pradesh (UP) and after 2000 for all the other states, the patterns are very similar. Upper castes favour the right in general, though there are lots of ups and downs in the extent of the upper caste bias. There is no clear pattern in whether the economically better off favour the right.

That there is no clear division of economic interests across the parties, comes from looking at changes in the spending patterns of the states when the state government shifts to the right. While the right-wing ruled states have lower social spending, in general, there is no evidence that switching to a right-wing party reduces social spending. In fact, there is some evidence in the opposite direction. The one caveat here is that we do not have data for state-level policies with respect to caste-based reservations for jobs, where the differential caste basis of these parties may make a big difference.

Taken together these results suggest that the main driver of political differentiation in India are the nature of caste and religious identification and the related variation in cultural and ideological positions. This is consistent with the findings by Piketty (2018) that the straight economic model of the rich and the more educated voting for the right and the less educated and less wealthy voting for the left, which emerged in countries like France, the United Kingdom and the United States, immediately in the post-war period, started breaking down after 1990, with the well-educated now voting more and more for the left and the others aligning with the right.1 Similarly, in India, voters seem to be gradually less driven by straightforward economic interests, and more by sectarian interests and cultural priorities. But, unlike in the West, this does not seem to require any sacrifice of economic interests for Indian voters, since all the parties support similar policies.

Again, unlike in the West, caste identity is highly specific to India. One might be tempted to argue that in some cases caste provides a better proxy for permanent income and economic position than other indicators (that is, income, education and especially asset ownership that are not particularly well measured in surveys). One natural interpretation of the fact that income, education and occupation seem to play so little autonomous role in explaining political cleavages (controlling for caste) is that India’s political conflict has given unusual importance to caste-based reservation policies and relatively little importance to income-based, education-based or wealth-based redistributive policies. The analogy would be the rising importance of the migration-based and the religious-ethnic conflicts in the West, in the context when policies aiming at reducing inequalities in income, education and wealth have lost strength. The decline of class-based redistribution in the West (and the fact that it never really took shape in India) can also be related to the changes in global ideology since the 1980s–90s, an evolution which might possibly reverse itself in the future.

These results are also consistent with the general position taken by Chhibber and Verma (2018), who argue that the Indian voter is motivated by ideology as much as anything else and that the main ideological divide has its roots in the national movement and its immediate aftermath. They may also be consistent with theories that give a central place to politics of patronage that is targeted towards specific groups (Chandra 2017; Wilkinson 2009; Dunning and Nilekani 2013), under the assumption that the caste-based cleavages reflect different positions on the affirmative programmes or other mechanisms to deliver patronage to specific social groups.

In terms of the overall agenda the paper that comes closest to us is Chakrabarti (2017). She shows that the fraction of the state legislature that is from the upper castes is strongly correlated with the share of total state development expenditure that goes to the social sector and this relationship survives a range of specifications, including those that include state fixed effects. By contrast our focus is on the ruling party or coalition, which may be identified with the upper castes but put up candidates from the lower castes for strategic reasons—during the 2014 election the BJP, for example, was at pains to emphasise that its prime ministerial candidate, Narendra Modi, was not from the upper castes. Perhaps for this reason we find that once we control for a fixed effect for the state, the identity of the ruling party has no discernable effect on social spending.

Classifying the Political Parties

Given the sheer number of parties on the Indian political landscape, we need to classify them to make the analysis tractable. We start from the two main parties, the Indian National Congress (INC) and the BJP. Congress has held power most of the time since the first general post-independence elections in 1952, except briefly in 1977–80 (following the Emergency period and the short-lived anti-Congress alliance) and 1989–91, and most importantly in 1998–2004 and 2014–19, when BJP was heading the government. In most of our results on the structure of the electorates, we include the Congress vote in a “centre” alliance together with the vote for centre parties like the Nationalist Congress Party (NCP), the Dravida Munnetra Kazhagam (DMK) or the Telangana Rashtra Samithi (TRS) that have usually allied with Congress. Likewise, we include the BJP vote in a “right” alliance together with parties with the Shiv Sena (SHS), the Shiromani Akali Dal (SAD) or the Telugu Desam Party (TDP). Finally, we include the various communist parties, the Bahujan Samaj Party (BSP) and the Janata Dal in a “centre-left and left” alliance. We refer to Table 1 for the full classification that we use for India’s main contemporary political parties.

We should stress that although we rely on the conventional usage of the terms right, left and centre to refer to political alliances in the Indian context, we fully recognise that these terms originated in Europe, dominated by a class politics that is very different from today’s India. Our purpose in this research is precisely to investigate the changing meaning of such classifications, by looking at the changing structure of the corresponding electorates.

To ensure that this classification is accurate we also asked a set of economists and political scientists working on Indian politics as well as some senior journalists in prominent newspapers and some active politicians to classify 18 major Indian political parties based on a left, centre-left, centre, centre-right, right scale. There is an impressive amount of concordance both between the views of these experts and our classification. Our main results, however, do not change if we slightly alter their classification (in particular if we look separately at Congress and BJP votes, or if we exclude the CPI and the CPI(M) from the centre-left and left coalition).

Vote Banks of Different Parties

The data we use in this section combines surveys and official election results. Data on both state and general (Lok Sabha) election results since 1947 are available at the constituency level from reports made public by the Election Commission of India, and were recently digitised and harmonised by Jensenius (2016). In order to study the electoral behaviours at the individual level, we rely on surveys conducted by the Lokniti Institute and by the Centre for the Study of Developing Societies (CSDS). These include the National Election Studies (NES), available from 1996 to 2014, as well as a number of other state election surveys. We complete our database with surveys from the Indian National Election studies conducted by the Inter-University Consortium for Political and Social Research in 1967, 1971 and 1979.

Given important variations in the definitions of available variables, we focus on a restricted set of characteristics which could be harmonised across the surveys. We divide caste/minority affiliations into five groups: SCs/STs, Other Backward Classes (OBCs), Brahmins, other Forward Castes (FCs) and Muslims. Education takes four values corresponding to illiteracy, primary education, secondary education and university degree. We decompose age into four groups (25–34, 35–49, 50–64 and 65+) and use dummy variables for gender and rural/urban areas.

The harmonisation of income is more challenging given that only income brackets were reported for the earlier years. Following Piketty (2018), we approximate income deciles by expanding surveys and reweighting observations so as to attribute individuals to their multiple potential income groups. This is equivalent to assuming that the voting patterns are constant within brackets. We also construct a social class variable by following Chakrabarti’s (2018) classification of different occupational groups. Lower classes are composed of low-skilled workers and illiterate individuals, the middle class includes small businessmen, craftsmen and skilled workers, and upper classes comprise mid-level and high-level civil servants, politicians, business-owners, medium and large landowners and higher educated voters.2 Finally, we combine retrospective questions from the 1979 and 1967 surveys to get information on electoral behaviour in 1962.

National Elections Results

We first use the long time-series data for the national elections to investigate the caste composition of support for right-wing parties, left-wing parties and centrist parties. Figure 1 shows the decline of the Congress from its original hegemonic state to its 2014 nadir and the rise of the BJP. Between 1962 and 1984, Congress was supported by between 40% and 50% of Indian voters. Starting in 1989, this share decreased steadily until today, reaching only 19% in 2014. Correspondingly, the BJP took an increasingly important place in India’s political spectrum, receiving an unprecedented vote share of 31% in the last Lok Sabha election.

Figures 2, 3 and 4 decompose support for centrist parties, right-wing parties and left-wing parties by caste group. Electoral politics in India have always been characterised by strong caste cleavages. The Muslims and the SCs/STs have always been more likely to vote for the Congress and other centrist parties, while Brahmins and other upper castes are most biased in favour of the BJP and other right-wing parties. Over time, the Congress’s popularity has declined among all the groups while that of the BJP has mainly been going up, except among SCs/STs and the Muslims where there is no clear long-run trend. The 2014 election was an exception: for the first time, nearly one-third of SCs and STs supported the BJP and other right-wing parties. However, support for the right among other caste groups increased in similar proportions, leaving the voting gaps between upper castes and lower castes essentially unchanged.

The centre-left and left parties have gone up and down, with a peak in the late 1990s. The groups that are most likely to support them are the Muslims, the SCs/STs and the OBCs. Overall, caste cleavages appear to be remarkably strong and persistent since the beginning of the 2000s: between 50% and 60% of Brahmins have voted for right-wing parties in all national elections, compared to less than 15% of Muslims.

Framework of estimation: To summarise this data in a single number, we look at the difference between the average vote share of these party groupings in that year’s national election and the share they got from the upper castes/Brahmins on one side and the SCs/STs on the other. Specifically, we estimate:

where = 1 if individual i voted for a party belonging to group p (centrist, right-wing, centre-left to left or other), and = 0 otherwise. is a dummy for individual i belonging to caste c. Xi is a vector of controls (state, social class, income, education, gender and rural/urban) and εi is the error term. Without controls, we have:

If refers to belonging to upper castes, for instance, then β corresponds to the difference between the proportion of upper castes and the proportion of other castes voting for p. It can be estimated by ordinary least squares (OLS) using a linear probability model of the form:

Adding controls preserves the intuitive meaning of the indicator: all other things being equal, upper castes are more likely to vote for party p than other castes by β percentage points. Since control variables are only categorical, the linear probabi­lity model is saturated and can be estimated by OLS using heteroscedasticity-robust standard errors (Wooldridge 2002). In the main text of this section, we present results for right-wing parties. We extend this main specification to centrist parties, centre-left and left parties. Our main conclusions are robust to considering these different party groupings and to restricting the sample to Hindus alone.

Drivers of electoral support: Figure 5 (p 38) describes the evolution of support for right-wing parties among upper castes before and after controls. Without any controls the upper castes have always been more likely to vote for the right by about 20 percentage points on average. However, this apparent stability conflates within state trends with the fact that the BJP may be growing faster in states where its support base is more or less biased towards the upper castes. The dotted line controls for state effects and the solid black line controls in addition for other individual characteristics. This reduces the level of ­upper caste bias but the slope over time, if anything, goes up: while upper castes were more likely to support right-wing parties by 5 to 10 percentage points in the 1960s and 1970s, the gap has risen to 15–20 points in recent years.

Figure 6 shows similar results for the SCs/STs, who have always been less favourable to the right than the rest of the population. Lower castes’ opposition to right-wing parties seems to have remained stable over time, both before and after controlling for individual characteristics. The voting gap between SCs/STs and other castes has always ranged between 5 percentage points and 10 percentage points. By contrast, Muslims have become increasingly inclined to vote for centrist or left-wing parties (Figure 7). In the early 1960s, they were about as likely to support the right as other religious groups, while in 2014 non-Muslims were more likely to do so by 30 percentage points. This extreme and rising religious polarisation is due to Muslims being the only social group who has not become more supportive of the BJP. While a rising share of upper castes, OBCs, and more recently SCs and STs have been attracted towards the right, a stable 85% to 90% of Muslims have continued to vote for centrist, left-wing or other parties.

We do the same exercise for education, with a focus on university graduates (Figure 8). Strikingly, while it is always true that the graduates are biased in favour of right-wing parties, the bias appears to go down over time. Once we control for state effects and respondent fixed effects, the bias is significantly reduced, boiling down to zero in recent years. This
suggests that education has become less and less important to understand political cleavages in India: state specificities and caste affiliation, which are strongly correlated to education, have remained much more fundamental.

The same conclusion holds for income (Figure 9) and social class (Figure 10), which are both generally strongly correlated to caste affiliation. While belonging to top 10% earners seemed to have an effect on vote choice in 1971, it has come close to zero in recent elections. Similarly, upper classes are generally more supportive of right-wing parties, but the effect is purely driven by the fact that they are more likely to belong to upper castes. Once one controls for available socio-demographic characteristics, upper classes are about as likely to support right-wing parties as middle or lower classes.

Regression analysis: Table 2 (p 39) reports regression results on the main determinants of support for the BJP or other right-wing parties between 1962 and 2014.3 In line with what previous figures suggested, caste identity appears to be the strongest factor for understanding electoral behaviours. In 2014, Muslims were less likely than OBCs to support the right by more than 30 percentage points, while Brahmins were more likely to do so by more than 10 percentage points. Education was not significantly associated with vote choice before 1996, but this effect decreased again until 2014. Age and gender do not have any significant effect. The last column shows the difference between the 2014 and the 1962 estimated coefficients. Except for the Muslim bias against right-wing parties, which has dramatically increased over time, the caste gradient does not seem to have changed significantly during the past decades.

The same exercise can be done for the centrist parties and centre-left to left parties. Upper castes are less likely to vote for these parties than the rest of the population, and once we control for state and respondent characteristics, there is no stable long-run trend. Support for the centre among SCs/STs has been going down relatively to other caste groups, even though it remains slightly higher than that in the entire ­population. Among all groups of parties, Muslim support is increasingly biased towards centrist parties and in particular the Congress: in 2014, they were more likely to vote for Congress alone by more than 15 percentage points compared to other voters. Centre-left and left parties have essentially ­attracted a large share of the Congress’s former electoral base among lower castes, while the BJP has been more successful among upper castes, so that only Muslims have remained faithful to Congress.

In summary, India’s national party system has substantially changed since the 1960s as the once hegemonic Congress became increasingly challenged by the BJP and the often caste-based parties of the centre-left. Despite these structural changes, caste status has remained the most important social cleavage materialised in national elections. Even after accounting for other state-level and individual-level specificities, upper castes appear to be significantly biased towards the right-wing parties, while centrist and left parties receive higher support among the lower castes and the Muslims. While these caste divisions have remained more or less stable over time, religious cleavages have increased dramatically. Muslims have been the only social group to not become more likely to support the BJP, remaining faithful to the Congress and other centrist parties. Strikingly, while education and income has played a role in some specific elections, we find no evidence of the emergence of a new cleavage linked to economic or  human capital over time.

State Elections Results

India’s transition from the Congress dominance to a multiplicity of fragmented party systems opposing the INC to regional parties and, more recently, to the BJP in state elections, has been associated with a progressive diversification of the nature of political competition in state elections (Figure 11, p 40). We have used a set of surveys conducted by Lokniti–CSDS during state elections to study how varieties in state party politics translate into specific cleavage structures.

Our sample covers 28 elections which took place between 1996 and 2016 in nine major states: Bihar, Gujarat, Jharkhand, Maharashtra, Rajasthan, Tamil Nadu, UP, Uttarakhand and West Bengal. There are large variations in the relative vote shares received by regional parties, Congress and the BJP. In some states (such as Gujarat), Congress hegemony was gradually replaced by a two-party system which opposed the INC to the BJP. In another group of states (such as Maharashtra), both Congress and the BJP have had to build coalitions with other smaller parties. Finally, the Congress has almost completely disappeared in some states and has been replaced by one or more regional parties. In Tamil Nadu, for instance, the AIADMK and the DMK have essentially alternated in holding power since the beginning of the 1970s.

Statewise drivers of electoral support: We start by looking at the caste and religious basis of support for right-wing parties in states where the BJP is a major competitor. Figure 12 decomposes the vote shares of right-wing parties by caste group, pooling all surveys available in each state. Support for the right is strongly differentiated by caste: in all states for which we have data, it is always the case that upper castes are more likely to support the right than SCs/STs or Muslims. The relationship between social class and right-wing affiliation is also positive, but substantially weaker (Figure 13). Lower classes are always less likely to support the BJP than upper classes by about 10 to 150 percentage points.

However, given that income, wealth, social class and caste are strongly correlated, these plots of the unconditional correlation are potentially misleading. Figure 14 (p 41) shows the difference between the share of upper castes and the share of other caste groups voting BJP/right, after controlling for social class, age, gender and locality size (rural/urban). Right-wing bias towards upper castes survives the inclusion of controls in more surveys, and while state-specific dynamics are visible, there does not seem to be any long-run common trend. Figure 15 plots the difference in vote shares for right-wing parties ­between the upper class and the middle/lower classes, after controls. While there is evidence that the right tends to be slightly biased towards the upper class, the relationship is much weaker: in most elections, the gap does not exceed 5 percentage points.

In states where Congress is still a key competitor, the caste basis of centrist parties is less clear-cut and depends upon the nature of the state party system (Figure 16). Centrist parties tend to receive stronger support among upper castes when they face a strong left-wing competitor (Bihar and West Bengal). When they face the BJP, on the other hand, they tend to attract a higher proportion of voters among lower castes and Muslims (as in Gujarat, Maharashtra or Rajasthan).

Regression results: In Table 3, we pool all state election surveys over the 1996–2016 period and run models equivalent to those used in the previous section. In line with our previous findings at the national level, centrist parties tend to be strongly biased towards Muslims, while leftist parties’ and right-wing parties’ electoral bases are more concentrated among lower castes and upper castes respectively. Social class is significant, but its role is much smaller: upper class individuals are more likely to support right-wing parties by only 3 percentage points. Finally, centrist and right-wing parties tend to receive greater support in cities, while independents and other small parties are more common in rural areas.

In line with our results in national elections, caste and religion appear in most cases to be strongly significant. Social class does seem to play a role in some elections, but voting differences, if anything, seem to have decreased over time. Our analysis of voting patterns in Indian states therefore suggest that caste has continued to structure local politics since the end of the 1990s, as in the case of national elections.

Social Spending and Party Affiliation

The results suggest that the main dividing factor between the political parties is social rather than economic, except perhaps in their views of affirmative action quotas, which are both social and economic. A plausible implication of this is that the shift in which party governs a state should not affect its economic decisions. We investigate this by asking whether social spending goes up when there is a shift in political  power away from the right, which is what one would expect to find in the West.

Our data on social spending comes from the Reserve Bank of India, which has released a set of documents providing information on the allocation of state budgets in recent years. We digitise these reports to obtain a measure of total social spending covering the 2003–17 period. The reports distinguish between revenue and capital expenditures and provide detailed information on the allocation of these expenditures to different sectors. We compute social expenditures by aggregating revenue and capital expenditures for education, sports, art and culture, medical and public health, water supply and sanitation, housing, welfare of SCs, STs and OBCs, social security and welfare, and labour and labour welfare.

Methodology for estimation: One issue we have to deal with is which denominator to choose. One possibility is to measure social spending as a fraction of gross state domestic product (GSDP). To the extent that states build fiscal capacities and allocate tax revenues to different sectors, this contains information about how states decide on whether or not to expand the social sector in the long-run. Another possibility is to divide social spending by total developmental expenditures, defined as the sum of expenditures dedicated to both the social and the economic sector.4 This measure corresponds better to short-run motives: given a fixed budget allocated to development, governments choose which sector to prioritise. Since government terms tend to be relatively short, we choose to focus on the latter measure.

In addition, we use both state surveys and national election studies to compute a measure of the relative representation of different caste groups and social classes in state governments.5 More specifically, we define government bias towards group c as:

A value higher than 1 indicates that caste or class c was overrepresented in voters supporting the party in power, while a value lower than 1 means that the ruling party was relatively more supported by other groups. For reasons of data availability and sample sizes, we restrict our analysis to 18 major states: Andhra Pradesh, Bihar, Chhattisgarh, Goa, Gujarat, Haryana, Himachal Pradesh, Jharkhand, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Odisha, Punjab, Rajasthan, Tamil Nadu, UP and West Bengal.

We run regression models of the form:

Socialit is the share of developmental expenditures dedicated to the social sector in state i at time t. Ideology is a measure of the representation of different ideologies in state governments, such as the total vote share received by right-wing parties in the last election or the ruling party’s ideological orientation. X is a vector of controls in which we include the logarithm of real state GSDP per capita as well as the overall electoral turnout in the last state election.6 Finally, μi and λt are state and year fixed effects, and εit is the error term. Notice that our explanatory variables are all lagged to account for the fact that changes in social expenditures are decided by governments for the next year.

Cross-sectional evidences: Figure 17 reveals a strong negative correlation between the average vote share received by right-wing parties in state elections and the average share of developmental expenditures dedicated to the social sector during the 2003–17 period. In Gujarat, where the BJP has won every election since 1995 with large popular support, state budgets allocated less than 40% of developmental expenditures to the social sector on average. In Kerala and West Bengal, both states with strong centre-left or leftist parties and no significant right-wing contestant, the corresponding figure was higher than 55%. Figure 18 shows a similar negative link between social spending and the caste basis of the party in power.

One problem with interpreting this evidence is that the difference could reflect any state characteristic—the political culture, the economy, the level of poverty, etc. We therefore include state effects and present changes in social spending as a function of right-wing vote shares and caste biases in previous years. Figure 19 shows that states with strong right-wing parties do not significantly decrease social expenditures more than other parties during their term. The absence of correlation between political representation and changes in the share of developmental expenditures allocated to the social sector is also visible when looking at upper caste representation (Figure 20). Governments supported by a larger relative proportion of upper castes are not more or less likely to expand the social sector.

Regression analysis: Table 4 presents our main regression results. We use all states-years for which we have data and we cluster standard errors by election periods to account for correlated unobserved heterogeneity within election periods. Columns (1) to (6) show the effect of popular support for different party groups on social spending before and after controls, without state fixed effects. The results point to a significant link between the ideology of states’ main parties and social expenditures: a one percentage point increase in popular support for the BJP or other right-wing parties is associated with social expenditures lower by 0.2 percentage points on average. Columns (7) to (12) confirm that these  effects are purely driven by interstate heterogeneity. After accounting for state fixed effects, the vote shares received by different party groups have no significant effect on the evolution of social spending within states. While the cross-sectional evidence is robust and significant, evolutions over time therefore suggest more complex and unclear patterns. Our findings therefore point to the importance of long-run historical trajectories (rather than switching party labels) to understand variations in welfare regimes across Indian states. The fact that neither caste-based cleavages, nor class-based divides have had measurable social policy consequences suggests that political conflict in India has not been primarily focused on the redistribution of economic resources or the redesign of service delivery. Rather, divisions between social groups have essentially been based upon symbolic claims.


Our results show that political cleavages are strong in India. The view that the main parties now speak to the same electorates is not corroborated by evidence. However political cleavages in India’s party system have developed mostly along the lines of caste identity and religious conflict. Inequality in education, income or occupation seems to have a limited impact on political preferences (after controlling for caste, religion and other attributes). The BJP and right parties are characterised by the fact that they disproportionally attract voters from upper castes. Congress and centre parties are relatively more successful among lower caste Hindus and especially Muslims. Centre-left and left parties make their stronger score among lower castes (SCs, STs and OBCs).

Our results might also provide some insight into why the Indian state has not been under more pressure to improve the delivery of social services, to raise more revenue through greater and more progressive taxation, or to carry out the reforms necessary for improving the environment or the employment landscape. The big political fights seem to be about caste and religious identity in its many forms and the caste quotas in educational institutions and government jobs, the one place where the identity and economic dimensions intersect. Interestingly, the amount of redistribution that actually happens through the quota system is quite limited, just because there are not so many government jobs and not that many high-quality educational institutions. But it is possible that in a world of multidimensional competition, the fact that quotas and fights over symbolic aspects of identity (cow slaughter, Ram Mandir, triple talaq, etc) are so salient means that all the other, potentially very important dimensions of political competition (better schools and health facilities, cleaner air, land redistribution, etc), tend to get lost. One of the key challenges might be to develop policy instruments that address issues such as effective access of lower and middle classes to high-quality public services (irrespective of caste or religious identity), the reduction of income and wealth inequality, or the effectiveness of progressive taxation, that are sufficiently salient and verifiable that they can help move India’s political cleavages in a more productive direction.


1 Relatedly Gethin and Morgan (2018) have shown that rising class cleavages in Brazil can be explained not only by poorer voters’ support for the Workers’ Party welfare policies, but also by upper classes’ disappointment with the political system’s corruption.

2 For a full list of occupations and education levels included in different social classes, see Chakrabarti (2017).

3 Income is excluded from this analysis since it was unfortunately not available in the 1996, 1998 and 1999 surveys. Social class is also excluded since occupation categories could not be harmonised before 1996.

4 Development expenditures directed to the economic sector include nutrition, relief on
account of natural calamities, agriculture and allied activities, rural development, special area programmes, irrigation and flood control, energy, industry and minerals, transport and communications, science, technology and environment, and general economic services.

5 When state election surveys are available, the computation of the social basis of ruling parties is straightforward. For states and years where no dedicated survey is available, we use the closest national election studies available to match voters with their corresponding parties or coalitions at the state level.

6 Our GSDP data come from the National Institution for Transforming India ( We obtain GSDP per capita by dividing total GSDP by state populations obtained from Ministry of Statistics and Programme Implementation (, and we deflate our series using India’s national CPI obtained from the Federal Reserve Bank of St Louis (


Bharti, Nitin (2018): “Wealth Inequality, Class and Caste in India, 1961-2012,” Working Paper 2018/14.

Chakrabarti, Poulomi (2017): “One Nation, Many Worlds: Varieties of Developmental Regimes in India,” Working Paper.

Chancel, Lucas and Thomas Piketty (2017): “Indian Income Inequality, 1922-2015: From British Raj to Billionaire Raj?” Working Paper 2017/11.

Chandra, Kanchan (2017): Why Ethnic Parties Succeed: Patronage and Ethnic Head Counts in India, Cambridge University Press.

Chhibber, Pradeep K and Rahul Verma (2018): Ideology and Identity: The Changing Party Systems of India, Oxford University Press.

Dunning, Thad and Janhavi Nilekani (2013): “Ethnic Quotas and Political Mobilization: Caste, Parties, and Distribution in Indian Village Councils,” American Political Science Review, 107(1), pp 35–56.

Gethin, Amory and Marc Morgan (2018): “Brazil Divided: Hindsights on the Growing Politicization of Inequality,” Issue Brief 2018/3.

Jensenius, Francesca R (2016): “Competing Inequalities? On the Intersection of Gender and Ethnicity in Candidate Nominations in Indian Elections,” Government and Opposition, 51(3), pp 440–63.

Piketty, Thomas (2018): “Brahmin Left vs Merchant Right: Rising Inequality and the Changing Structure of Political Conflict,” Working Paper 2018/7.

Vaishnav, Milan (2015): “Understanding the Indian Voter,” Carnegie Policy Brief,

— (2018): “What Is the Secret to the Success of India’s Bharatiya Janata Party?,” Carnegie Policy Brief,

Wilkinson, Steven I (2009): Votes and Violence: Electoral Competition and Ethnic Riots in India, Cambridge University Press.

Wooldridge, Jeffrey M (2002): Econometric Analysis of Cross Section and Panel Data, Massachusetts Institute of Technology.

Updated On : 20th Mar, 2019


(-) Hide

EPW looks forward to your comments. Please note that comments are moderated as per our comments policy. They may take some time to appear. A comment, if suitable, may be selected for publication in the Letters pages of EPW.

Back to Top