Economic Policies vs. Identity Politics - The Rise of a Right-wing, Nationalist Party in India

Economic Policies vs. Identity Politics: The Rise of a RightwingNationalist Party in India Abstract As in many other parts of the world, India has witnessed a surge in the popularity of the Bharatiya Janata Party (BJP), a right wing nationalist party. This paper examines the respective roles of economic policy and “identity politics” for the sharp rise in support for BJP in a region where it had a negligible vote share until 2014. Using household level panel data from 3500 rural households in West Bengal, we examine the effect of different welfare benefits delivered by state and national governments, on support for the regional incumbent (Trinamool Congress (TMC)) and the BJP after 2014. We find that receipt of these benefits was associated with increased support for the respective party controlling either level of government. However, changes in scale, effectiveness, composition or targeting of these benefits cannot account for the large observed increase in the relative popularity of the BJP. We also find no evidence of any association with post-2014 economic distress of the household, or political violence attributed to TMC party activists in the 2018 local government elections. On the other hand, religion, tribal and immigrant status of households are strong and robust predictors of changes in political support, irrespective of controls for welfare benefits and household incomes. Hence the results indicate that the BJP’s rise in West Bengal reflected the growing importance of identity politics per se, rather than economic policies. 1 1 Introduction Across the world, right wing nationalism is on the rise. It has spread rapidly across high and middle income countries across the globe, including Austria, Brazil, France, Germany, Hungary, Italy, Netherlands, Russia, Turkey, UK and US where right-wing or extreme-right wing parties characterized by populist, nationalist platforms have gained the votes of large sections of poorer and less educated citizens. Both demand and supply side factors are argued to be relevant. Demand-side factors appear to have been driven by a combination of economic and cultural grievances (Golder, 2016, Muis and Immerzeel, 2017, Guriev and Papaioannou, 2022, Bardhan, 2022). Economic grievances include rising unemployment among the unskilled in an era of rising globalization and immigration. Cultural grievances arise from transition to a postmodern and increasingly multi-cultural society that threaten traditional values, combined with “social identity” concerns that generate homophobic, racist and anti-immigrant attitudes. Considerable empirical evidence is becoming available both of the role of economic (Alabrese et al., 2019, Autor et al., 2020, Oliveira, 2022) and cultural grievances conditioned by racial and historical factors (Kaplan and Mukand, 2014, Ochsner and Roesel, 2017, Enos, 2016). Political scientists and sociologists have also pointed to the role of supply-side factors such as electoral systems, ideological positions, organizational strengths, and the media (Golder (2016), Muis and Immerzeel (2017)). However, despite widespread speculation concerning the relative importance of economic issues vis-a-vis the conjunction of cultural grievances, social identity and supply side factors,1 we are not aware of any empirical analyses of this question. In the last decade right-wing nationalist parties have also become significantly more popular in India, giving rise to debates concerning the respective importance of economic policies and shifting identity politics centered around religion, caste and nationhood. In 2014, the nationalist Bharatiya Janata Party (BJP), under the leadership of Narendra Modi, attained an absolute majority in the lower house (Lok Sabha) of the Parliament of India for the first time in its 34-year old history.2 In the 2019 parliamentary elections, the BJP’s vote share increased further, with large gains in states that had been previously dominated by regional parties. Some scholars of Indian politics argue that it is difficult to explain this as a rejection of regional incumbents by voters owing to economic distress. This is because since the BJP acquired control of the national government in 2014 it introduced policies that disproportionately hurt the poor and underprivileged.3 Despite 1 For instance an opinion article on the recent Turkish election by Bret Stephens in the New York Times (May 31, 2023) states: “...the importance of Erdogan’s Islamism as the secret of his success. It’s a rebuke to James Carville’s parochially American slogan, ‘Its the economy, stupid’. Actually, no: It’s also God, tradition, values, identity, culture and the resentments that go with each. Only a denuded secular imagination fails to notice that there are things people care about more than their paychecks.” 2 BJP was the leading member of the coalition that was in power during the period 1999–2004 (the Vajpayee government), but they won only 182 of the 543 seats in the Lok Sabha. 3 These policy changes included reduced outlays for the national rural employment guarantee (NREGS) workfare program, demonetization and the implementation of the new value-added tax (Goods and Services Tax, or GST). 1 this, BJP’s vote share among these sections of the population actually grew. Accordingly, Jaffrelot (2019) and Chhibber and Verma (2019) emphasize the growing role of identity politics and supplyside factors, e.g. the BJP’s nationalist ethno-majoritarian ideology, strong political organization, Modi’s personal popularity and the party’s choice to field SC/ST candidates in order to appeal to them as a voting bloc. An alternative viewpoint emphasizes the role of changes in welfare programs since 2014 (Deshpande et al., 2019, Wilkinson, 2020, Joshi et al., 2022). These authors argue that clientelistic welfare programs such as NREGS, the rural employment program administered by state and local governments’ shrank, and became less effective in generating votes for regional incumbents. At the same time, the new benefit programs that the BJP-led national government introduced relied less on local intermediaries for distribution, thereby reducing the scope for distortions in delivery and corruption as well as the capacity of local incumbents to claim credit. These programs include the new “zero-balance” bank accounts (Jan Dhan Yojana), construction of in-house toilets (Swachchh Bharat Yojana) and new methods of disbursing cooking gas or LPG subsidies (Ujjwala Yojana).4 These authors provide evidence that households receiving these benefits were more likely to switch their vote to the BJP. This effect could have been compounded further if some voters selectively blamed their state governments for economic distress caused, for example, by inflows of low-skilled immigrants, or insufficient government aid after adverse weather shocks. These contrasting viewpoints gives rise to the question of the relative importance of economic policies and identity politics in accounting for the rise of the BJP. This constitutes the motivation of this paper. We rely on a longitudinal survey of 3500 household heads on political attitudes, welfare benefits received and economic well-being conducted in two districts (Hugli and West Medinipur) in West Bengal, a state in eastern India. Such a dataset enables us to disentangle the role of changes in different kinds of benefits received by the household and its economic circumstances, and markers of identity (religion, caste, tribal and immigrant status), while controlling for a large range of household characteristics (including assets owned, education and occupation). West Bengal is a particularly interesting case study because of the dramatic shift from support for left-wing to right-wing parties that has taken place over a relatively short period. From 1977 until 2009, the state was dominated by an alliance of left-wing parties (Left Front), and the competition in each election cycle was mainly between the centrist Indian National Congress and its allies, and the Left Front. Through the 2000s, the state experienced increasing political competition between the Left Front and the All India Trinamool (Grass roots) Congress, TMC, a breakaway faction of the Indian National Congress. In the 2009 Lok Sabha elections the TMC won the majority of seats in the state; TMC then won the majority of seats in the state assembly elections in 2011 and 2016. Until the 2019 Lok Sabha elections, in West Bengal the BJP was considered 4 See Table 2 for a list of the programs included in the set of central benefits. 2 to be a non-contender. The 2019 elections changed all that: in 2019, across the state the vote share of the BJP increased by 23% over its 2014 vote share.5 The Left-center parties that have dominated West Bengal politics since 1977 have systematically engaged in clientelistic distribution of state benefits.6 See, for example, Bardhan et al. (2015, 2021), Dey and Sen (2020), Shenoy and Zimmerman (2020), Mahadevan and Shenoy (2023). It is also a state bordering Bangladesh, with a significant proportion of Muslim households and immigrants. TMC, the principal regional incumbent party, has often been blamed for engaging in ‘appeasement’ of these groups, which may have generated a backlash among native Hindus. The household surveys were conducted in the districts of Hugli and West Medinipur in which changes in vote share for different parties mirrored outcomes in the rest of the state. The first round was conducted over several months in 2013-14, shortly before the 2014 national parliamentary elections. The second round was analogously conducted over 2018-19, a few months before the 2019 national elections. In each round we asked questions about the household’s demographic characteristics, socio-economic status, landholding, cultivation of potatoes (the major cash crop in this area) and the benefits they had received from various government programs. In addition, we conducted a “straw poll”, where the respondent simulated casting a secret ballot for their preferred political party. This allows us to examine if households change the political party they support, and how the changes vary with the benefits received, their economic circumstances, and markers of social identity such as religion, caste, and immigrant status. To identify the causal role of benefits received, we address concerns of endogeneity of benefit receipt with the help of a “leave-out” shift-share instrument, where a household’s benefits are instrumented by time-invariant household characteristics interacted with per-household benefits received in other local government (gram panchayat) jurisdictions in the same district (Levitt and Snyder Jr, 1997, Bardhan et al., 2021).7 We use a probabilistic voting framework in which citizens’ support for a political party depends on both ideological (such as the party’s ideological platform, combined with organizational strength, advertising and media outreach), and utilitarian considerations (such as welfare benefits and perceptions of how their economic well-being are affected by policy choices). With retrospective voting, households’ future expectations of respective policies of competing parties is based on their recent observations and experiences. The role of changes in identity politics and supply-side factors is inferred from the effects of household characteristics such as religion, caste and immigrant status, 5 In our surveys we find a 18% increase in the BJP vote share. See Figure 4. These include the National Rural Employment Guarantee Scheme (or NREGS), Below Poverty Line (or BPL card), Panchayat Toilet Building Programs, and Kanyashree a cash transfer program to households with teenage girls conditional on not dropping out of school and not getting married. See Table 2 for the list of programs. However, it should be noted that state benefits does not necessarily mean that the state government funded these programs; rather they were in charge of distributing the benefits, including identifying households that were to receive these benefits. 7 These two papers show that the IV estimate is less biased compared to the OLS estimates, and that the former bias is negligible if the number of villages in each district is sufficiently large. 6 3 after controlling for benefits, changes in household income, household demographics, occupation, education and landownership. Our main findings are as follows. First, we find no evidence of a reduction in the aggregate scale of state benefits during 2014–2018 compared to the 2010–2013 period. While certain programs such as NREGS shrank (as found by Deshpande et al. (2019) at the all-India level), there were other state benefit programs that expanded in West Bengal, ensuring that the proportion of households that received at least one state benefit did not fall. Second, we do not find evidence that state benefits (aggregating across different programs) became less effective in generating votes for the TMC. If anything, receipt of state benefits was associated with a statistically significant net 1– 14 percent increase in the likelihood that the household head voted for the TMC. Analogous to Deshpande et al. (2019) the effectiveness of NREGS benefits alone as a generator of votes for the TMC declined after 2014, but this was accompanied by rising effectiveness of other state benefit programs. Third, we also find some evidence that households that received the new benefits introduced by the BJP-led central government switched to voting for the BJP in 2019. But this effect is only marginally statistically significant. It is quantitatively negligible, accounting for less than 2 percent of the increase in the BJP’s vote share observed for sample households. Moreover, the estimated effect is not robust to alternative specifications. Fourth, the combined effect of changes in scale of state and central benefits and their respective effectiveness in generating votes for either party, predicts a net decline in the BJP vote share between 2014 and 2019, and a rise in the TMC vote share. Hence the evidence does not support the claim that the rising vote share of the BJP can be explained by the changing structure or effectiveness of benefit programs. These results turn out to be robust when we additionally incorporate changes in the composition of state and central benefits, by examining the different impacts of the three most important programs of either kind. They are also robust if we incorporate respective targeting patterns of state and central benefits. In particular, we find that the state programs were better targeted to the poor than the central benefits, and that the pro-poor targeting of state benefits became significantly more pronounced after 2014. Hence the evidence runs counter to claims that the central benefits were better targeted to poorer households than state benefits. Moreover, we find no evidence that clientelistic benefits were preferentially delivered to recent non-Hindu immigrants and disadvantaged socio-economic groups (Scheduled Castes (SC) or Scheduled Tribes (ST)). On the contrary, these groups were significantly less likely to receive state benefits, and this pattern became more even more pronounced after 2014. Hence the increased tendency for households belonging to the dominant group of native Hindu non-SC-STs to vote for the BJP cannot be rationalized by resentment generated by increased “appeasement” of minorities by the TMC. We subsequently investigate whether voters experiencing rising economic distress in the form of declining household incomes since 2014 were more likely to switch to supporting the BJP. Our data includes household income from employment and from the cultivation of potatoes, the major cash 4 crop in the area. Controlling for the receipt of welfare benefits and household characteristics, we find that post-2014 changes in these income sources had no effect on the likelihood that a household voted for the BJP. We also find no evidence for the role of violence perpetrated by TMC party activists during the recently concluded 2018 village council (panchayat) elections. Since the violence intimidated opposition candidates from withdrawing from the panchayat elections, we examine whether the vote share of the BJP was higher in constituencies where the TMC candidate ran unopposed in the 2018 election. We find no such patterns in the data. Turning finally to the role of changing identity politics, we examine how shifts in support for the BJP varied across household demographics and socio-economic status. The raw data shows the rise in BJP vote share was remarkably uniform (around 20%) across landowning status, SCs and Hindu upper or other backward castes (with a slightly lower increase among the landless (14%) and STs (18%)). It was substantially lower among non-Hindus (7%), and especially among recent non-Hindu immigrants (where the BJP vote share fell). The same patterns emerge when we control for other household characteristics, welfare benefits received and changes in household economic circumstances. It is striking to observe that despite experiencing greater discrimination in the distribution of state benefits since 2014 (compared either to central benefits or pre-2014 levels), minority groups such as non-Hindus and ST groups were significantly less likely to switch their votes from the TMC to the BJP. In summary, the evidence indicates that the BJP’s rise in West Bengal reflected the growing importance of identity politics per se, rather than voter responses to changes in income or receipt of government welfare benefits.8 Of course, data limitations prevent us from investigating all possible hypotheses for why voters may have reduced their support for the TMC. Claims we cannot investigate include perceptions of local corruption, or dissatisfaction with governance that is not reflected in the quality of government services or benefit distribution. Also, since our empirical exercise uses data from only two districts of the state, our findings may not apply to other districts or Indian states. There are many distinctive aspects of West Bengal’s politics and economy compared to other Indian states. Nevertheless our analysis is of interest for a number of different reasons. West Bengal is a state with a population of 90 million, comparable to many countries in the world such as the UK or France. So understanding factors underlying the rise of the BJP in this state is of interest for its own sake. Moreover, it may be premature to declare that our results would not extend to other Indian states. For instance, our analysis for West Bengal replicated the main results of Deshpande et al. (2019) at the all-India level: the scale and effectiveness of the NREGS program in generating votes for the regional incumbent declined, while new central benefits introduced after 2014 helped raise the BJP vote share. The main 8 These results are consistent with the claims of Choudhary et al. (2020), who, use national level survey data to argue that in 2019 demographics and identity (caste and religion) were the key determinants of voting for BJP, in contrast to 2014 when it was able to strike the “issue-based electoral chord”. 5 reason for the discrepancy between our respective interpretations of these findings is twofold: (a) our analysis included the scale and effectiveness of all other state programs which expanded and became more effective after 2014, in contrast to the NREGS program; and (b) we estimated the combined impact of all state and central benefits on the TMC and BJP vote share and found that the relative quantitative magnitudes of predicted changes in these vote shares were the opposite of what was actually observed. Hence whether or not our results would extend to the all-India level must await an extension of our research methodology, once suitable data for such an exercise becomes available. The paper is organized as follows. Section 2 of the paper describes the institutional context, details of the data, and a range of descriptive facts relating to changes in voting patterns. Section 3 contains the main empirical results, while Section 4 concludes. 2 Context and Data 2.1 Background: West Bengal During the period 1977–2011, West Bengal’s state legislature was dominated by the Left Front, a coalition of left leaning parties led by the Communist Party of India (Marxist), or the CPI(M). The Left Front secured an absolute majority in each successive election during this period, both for the state legislative assembly seats, and for the national parliamentary seats. After Mamata Banerjee, a veteran member of the major opposition party, the Indian National Congress, defected in 1997 to form the Trinamool (grass roots) Congress (TMC) party, the TMC became the primary opposition party in the state. It often fought elections in alliance with the Indian National Congress. The Bharatiya Janata Party (BJP) was a minor player throughout this 35-year period. During the 2009 national elections, it received only 6% of the vote share and won one of the 42 parliamentary seats in West Bengal. This increased to 17% vote share and two seats in 2014, and then a spectacular increase to over 40% vote share and 18 seats in the 2019 national elections, so that it became the second-most important political party in West Bengal. Although the TMC won the majority of the parliamentary seats in 2019, its vote share did not change much from 2014. On aggregate, it would appear that the BJP’s vote share increased at the expense of the Left Front, whose vote share sank from 30% in 2014 to below 10% in 2019. In Figure 1, we can look across the different parliamentary constituencies in West Bengal, and see how the vote share of the three main electoral factions changed over the period 2009 to 2019. Comparing the maps for 2009 and 2019, it appears as if areas the Left Front was dominant in 2009 have now switched over to support the BJP.9 However, as will become evident when we use 9 A commonly heard term describing this change in the voting pattern in West Bengal is Baam theke Ram (from 6 7 Notes: Party winning each seat in Parliamentary elections in 2009, 2014 and 2019 presented. Figure 1: West Bengal National Parliamentary Election Results: 2009–2019 Figure 2: Vote Shares and Seat Shares of Different Parties in Parliamentary Elections. All India and West Bengal. 2004–2019 Panel A: Vote Shares Panel B: Seat Shares Notes: Vote share of a particular party defined as a number of votes received by all candidates of the party in a particular election as a proportion of total number valid votes cast in the particular election. Seat share of a particular party is defined as the number of seats won by the party as a proportion of the total number of seats (543 for All India and 42 for West Bengal). our longitudinal household data to examine vote-switching patterns, the reality is more nuanced. Figure 2 shows that in 2019 the BJP registered an uptick in vote share across India as a whole, but that the increase was more dramatic in West Bengal. Until 2009 the BJP’s vote share in West Bengal was below 10%, whereas it was around 20% in the country as a whole. Between 2009 and 2014 it increased by about 10% both in West Bengal and overall, and then between 2014 and 2019 it increased by 23% in West and by 7% overall. The corresponding contrast between West Bengal and the country as a whole in the evolution of seat shares was even more drastic. 2.2 Data and Descriptive Statistics We use longitudinal survey data from a sample of 3500 households residing in 72 randomly selected villages in the potato-growing talukas (subdivisions) of the Hugli and West Medinpur districts of West Bengal.10 Panel A of Figure A1 shows the location of the two districts within the state and Panel B marks the locations of the villages. We analyse data from two survey rounds, conducted in 2013 and 2018. We collected extensive data on the households’ demographic and socio-economic characteristics, transfers they received from the government and their engagement in civil society and political activities such as attending the the Left to Ram, the Hindu God). 10 Villages were required to be at a minimum of 8 kilometers apart from each other. The surveys were conducted for other unrelated projects: see Maitra et al. (2017), Mitra et al. (2018), Maitra et al. (2022b,a). 8 Figure 3: Vote Shares. Survey Data Notes: Proportion of Households voting for Left, BJP and TMC in strawpoll conducted in 2013 and 2018 and 90% confidence interval presented. village gram sabha, and political rallies. We also conducted a straw poll where respondents cast a simulated “secret ballot” for the political party they supported.11 Note we had no attrition in our sample across the two rounds. Our analysis for this paper uses a longitudinal household-level dataset of 3507 households, with data from 2013 and 2018. 2.2.1 Vote shares Figure 3 presents the vote shares of the Left Front, BJP and TMC elicited in our 2013 and 2018 surveys. The Left Front’s vote share in our sample dropped from 29% to 10%. This is matched by a large increase in the BJP’s vote share, from 4.4% to 22.5%, and a slight drop in the vote share 11 The households were given a paper ballot and assured that their votes would remain confidential. Above 97% of households consented to participate. They received the ballot paper and a locked box, privately selected their preferred political party and folded and placed the ballot in the locked box. They then returned the box to the interviewer. This approach has been widely used in this literature. Casey (2015), in Sierra Leone and Bardhan et al. (2021) in West Bengal, India used a similar process to elicit political preferences and the elicited vote share in their studies were significantly correlated with actual vote shares at the electoral constituency level in the subsequent election (0.73 and 0.57 respectively). The National Election Surveys that have been widely used to measure political preferences in India also use the same process. 9 Figure 4: Vote Shares Changes at Different Levels Notes: Polling booth data obtained from the published Form 20 data made available by the Election Commission of India. of the TMC from 61.6% to 57.3%. Figure 4 compares these vote shares calculated from elicited survey data in the bottom-right panel, with changes in the actual vote shares in the electoral constituencies that span the sample villages (bottom left panel). These are aggregated for the two districts in the top-right panel, and for the entire state in the top-left panel. Although the patterns are very similar on the whole, our survey underestimates the changes relative to the actual changes that finally occurred. In our sample the BJP’s vote share rose by 18% points, while they increased by 23% in West Bengal in the actual 2019 election. Similarly in our surveys the Left Front’s share fell by 19% points whereas in the elections it decreased by 22% points across West Bengal. TMC’s share fell by 4% points in our sample, in contrast to a 4% point increase in its vote share across the state. The Sankey diagram in Figure 5 allows us to portray how individual voters switched their support across political parties, or remained loyal to the same party, between 2013 and 2018. It is evident that the major flow is not just the result of Left Front supporters switching to the BJP. In fact, a 10 Figure 5: Vote Switching Patterns. Survey Data Notes: Vote switching patterns of household heads between the two rounds presented. large number of households that voted for the TMC in 2013 switched to the BJP in 2018. At the same time, a larger number of 2013 Left Front voters switched in favour of the TMC in 2018.12 Therefore, in what follows we analyze factors that could explain four key variables: the switch in households’ political support toward the TMC, away from the TMC, away from the Left Front, and toward the BJP. However Figure 5 corroborates the patterns observed in Figure 3: there was a large increase in votes for BJP, a large drop in votes for the Left, while votes for TMC remained roughly unchanged across the two polls. 2.2.2 Household Characteristics Table 1 provides details of household characteristics in the sample. Only 10% of the households had female heads.13 Fifteen percent of household respondents report their religion as non-Hindu; the majority of these are Muslim. Six percent of households report they belong to the scheduled castes, and 10 percent report they belong to the scheduled tribes; the remaining 70% are general caste Hindus. On average 1.4% of respondents report their household relocated to the village after 12 One could argue that this figure does not strictly correspond the patterns suggested in footnote # 9. As Figure 1 suggests, it is likely that many of the erstwhile Left voters had moved to the TMC as an intermediate step. So some 2013 TMC voters that vote for BJP in 2018 were erstwhile left voters. If we assume that all of these voters were traditionally Left voters, the number of Left voters that voted for BJP in 2018 was 726. The number of Left voters in 2013 that voted for TMC in 2018 was 542. 13 These are mainly widowed women. 11 2003, however there is considerable variation in proportion across villages (SD 12%). The median household head had not completed primary school. Fifty four percent of households report that the head’s main occupation is cultivation, and 21% of households report casual labour. Close to one-fifth of households do not own any agricultural land, and 60 percent own less than one acre. Close to 60% live in a non-permanent (or kuchcha dwellings, built with materials such as mud or tin). Thus the majority of households in our sample are of low socio-economic status. Table 1: Selected Descriptive Statistics. Household Characteristics Survey NSS Mean (1) SD (2) Mean (3) SD (4) Male Headed Household Non Hindu Household SC Household ST Household General Caste/OBC Household Recent Immigrant Household Household Size Head Married Head more than Primary Schooling Head Occupation: Cultivator Head Occuption: Labour Landless Landholding 0-0.5 acres Landholding 0.5-1 acres Landholding 1-1.5 acres Landholding 1.5+ acres Kuchha House 0.9090 0.1543 0.0576 0.0955 0.6926 0.0143 5.6179 0.8891 0.4916 0.5421 0.2133 0.1950 0.3992 0.2295 0.1762 0.0927 0.5697 0.2876 0.3613 0.2330 0.2940 0.4615 0.1186 2.5734 0.3141 0.5000 0.4983 0.4097 0.3963 0.4898 0.4206 0.3811 0.2900 0.4952 0.9022 0.1589 0.1985 0.1146 0.6869 0.2969 0.3656 0.3889 0.3185 0.4638 4.6276 0.8988 0.4404 0.1807 0.3862 0.1638 0.4922 0.1166 0.0679 0.1594 2.0605 0.3018 0.4964 0.3848 0.4869 0.3701 0.4999 0.3210 0.2516 0.3660 Change in Employment Income Proportion experiencing decline in Employment Income Change in potato value-added Proportion experiencing decline in Potato value-added -0.2278 0.6875 0.2530 0.0858 1.3857 0.4636 0.4582 0.2802 Notes: Employment income and Potato Value added in real terms (using the All India Price Index Number for Agricultural Labourer, General Index, 1986−87 = 100). Employment Income and Potato value added in Rs. (’0000). Change in employment income = Employment income (2018) – Employment income (2010–2013). Change in potato value-added = Potato valueadded (2018) – Potato value-added (2010–2013). Decline in Employment income = Change in Employment income < 0. Decline in Potato value-added = Change in Potato value-added < 0. For NSS, pooled data from the rural sample of the 66th and 68th rounds used. The bottom panel of Table 1 provide statistics about the households’ income from employment and from the cultivation of potatoes, the leading cash crop in this region.14 More than 70% of the households reported that they cultivated potatoes, and the majority of them cultivated both in 2013 and 2018. On average, employment income decreased over the period (69% of households 14 Employment income and Potato Value added in real terms (using the All India Price Index Number for Agricultural Labourer, General Index, 1986 − 87 = 100). 12 experienced a drop in employment income), though average value added in potato increased over the period (only 8.5% of households experienced a fall in potato value added). We also categorize households into different income quantiles as follows. We regress average total household income during the period 2010–2013 on dummy variables indicating that the household does not own cultivable land, that it lives in a kuchcha house; that the household head is female, has less than primary schooling, and reports agricultural labour as the primary occupation. Table A1 in the Appendix presents the results from this regression (we report separately results that include and exclude village fixed effects). Unsurprisingly, landless households, those living in a kuchcha house, with less-educated household heads and those whose primary occupation is labour are predicted to have lower household income. We then classify households into four different quantiles on the basis of their predicted household income: Q1 (below 25%), Q2 (25–50%), Q3 (50–75%) and Q4 (more than 75%). 2.2.3 State and Central Benefits Received As discussed, it has been argued that the BJP’s vote share increased over 2014-19 partly because of new benefit programs that they introduced after they came to power at the centre in 2014. In particular, we focus on four programs that were aggressively marketed and prominently featured Mr Modi’s image in advertisements on billboards, television and social media. These include grants to construct in-house toilets (Swachchh Bharat Yojana), free “zero-balance” bank accounts (Jandhan Yojana), direct transfers of subsidies for cooking gas into beneficiaries’ bank accounts (Ujjwala Yojana) and credit cards for farmers (Kisan Credit Card.) While the last-mile implementation of these schemes required the involvement of the state and local governments, it has been argued that beneficiary eligibility for these schemes was determined through objective criteria, and hence the delivery of benefits was less vulnerable to distortion by local intermediaries or officials. They were also advertised as more universal, better targeted and more efficiently administered than pre-existing welfare benefit programs. We refer to the services delivered through these four schemes as “central” benefits, and to all other welfare benefit schemes as “state” benefits. Households are coded as having received the state benefits if they report accessing work through the NREGS workfare program, having a below-poverty-line (BPL) card, having a toilet constructed with funds from the gram panchayat, access to piped drinking water provided by the gram panchayat, medical help accessed with the help of the gram panchayat, housing construction funded through the state government’s scheme (Aawaas Yojana), flood relief and the conditional cash transfer for girls who delay marriage until the age of 18 (Kanyashree), also financed by the state government. Note that by construction, central benefits did not exist prior to 2014, while state benefits could have been distributed either before or after 2014. In Panel A of Table 2 we present the proportion of households that reported receiving state benefits 13 Table 2: Fraction of households who received central and state benefits in 2010–2013 and 2014–2018 2010–2013 (1) 2014–2018 (2) 0.523 (0.500) 0.036 (0.187) 0.060 (0.237) 0.303 (0.460) 0.039 (0.194) 0.125 (0.331) 0.047 (0.212) 0.455 (0.498) 0.374 (0.484) 0.206 (0.404) 0.259 (0.438) 0.091 (0.287) 0.143 (0.350) 0.142 (0.349) 0.090 (0.286) 0.035 (0.185) 0.749 (0.434) 0.762 (0.426) Panel A: State Benefits Workfare scheme (NREGS) Below poverty line card Toilet (Gram Panchayat) Drinking Water (Gram Panchayat) Housing / House Construction Flood Relief Self-help group Medical Help (Gram Panchayat) CCT for delayed marriage (Kanyashree) Any State Benefit Panel B: Central Benefits In-house Toilets 0.155 (0.362) 0.187 (0.390) 0.161 (0.367) 0.033 (0.180) Bank accounts (Jandhan) Cooking gas subsidy ((Ujjwala) Kisan Credit Card Any Central Benefit 0.427 (0.495) Notes: As explained in the text, central benefits were not defined prior to 2014. For each benefit we report the average number of households that report receiving at the benefit at least once in the periods 2010–2013 (column 1) and 2014–2018 (column 2). “Any State Benefit” and “Any Central Benefit” denote the proportion of households that received at least one state (Panel A) or central (Panel B) benefit. Figures in parentheses denote standard deviations. 14 at least once during the four-year periods 2010–2013 (column 1) and 2014–2018 (column 2). Panel B presents the proportion of households that reported receiving central benefits at least once during 2014–2018. Between the two periods, the proportion of beneficiaries who reported they had received workfare through the NREGS decreased by 7 percentage points, and the proportion who reported receiving drinking water through the gram panchayat (GP) decreased by 5%. These results resemble findings of Deshpande et al. (2019) at the all-India level. However, for all other schemes, there was an increase in the fraction of households who reported receiving the benefit. The increase is particularly striking for the below-poverty line cards that grant households an expanded volume of food and other household necessities through the public distribution system: in the 2013 survey only 4% of households reported they had a BPL card, but in 2018, 37% of households had it. Similarly, before 2013 only 6% of households reported their neighbourhood had new toilets but in 2018 this was a substantially higher 21%. The proportion reporting they had received the housing benefit increased 4 to 9%, and those participating in women’s self-help groups increased from 4 to 14%. Overall, however, the proportion of households who received at least one state benefit remained unchanged, around 75%, indicating that the proportion of households receiving state benefits had not changed.15 On the other hand, 43% of households reported receiving at least one new central benefit: between 15 and 19% reported receiving either in-house toilets, bank accounts or the cooking gas subsidy (Ujjwala). If we count the number of benefits a household received, we see that the average household in our sample received 1.2 benefits during 2010–13, but this increased to 1.8 in 2014– 2018. Thus there is no evidence that households became less likely to receive state benefits overall, although clearly the reach of some benefits such as NREGS and drinking water shrank while others expanded. There is also no evidence that the new central benefits had a larger coverage rate than state benefits: in 2018 the proportion of households who reported receiving at least one central benefit was smaller than the proportion who reported receiving work through the NREGS alone. 3 Voting Patterns 3.1 Underlying Voting Model and Regression Specification We adapt a standard probabilistic voting model (Grossman and Helpman, 1996, Dixit and Londregan, 1996) where voting decisions are based on a comparative evaluation of competing parties on a combination of utilitarian and ideological dimensions. The utilitarian dimension incorporates the voter’s perception, or expectation, of how their own material well-being is, or will be, affected by policies of the different political parties. The ideological dimension incorporates an intrinsic 15 Of course it could well be that the remaining 25% were better-off households who did not rightfully qualify for the benefits. 15 preference for political parties based on ideological or identity affinity, loyalty, the personal characteristics of the candidates running for election on the party’s platform, or the voter’s exposure to the party’s electoral campaign. Applied to the context in hand, material well-being depends on the value the household attaches to the welfare benefits they receive from the government. While utility-based considerations are commonly assumed to apply differentially to voters of different economic classes, the ideological component varies with indicators of social identity such as religion, gender, caste or tribe, as well as idiosyncratic personal preferences. Suppose there are two competing parties p = k, l in an election held in year t. Voter i receives the benefit bundle Bipt at some point between election year t and the previous election year t − 1, and this gives her material utility βp Bipt , where βp is the marginal utility vector of the benefit bundle from party p. Voter i’s ideological preference for party p equals Γs(i)pt δs(i) + ipt , where s(i) denotes a vector of social identity characteristics of voter i, Γs(i)pt is the common systematic component of ideological preference of voters with social identity s for party p at date t, δs is a dummy for group s, and ipt is an idiosyncratic ideological preference of i for party p which is identically and independently distributed across the population. Let Mp denote the number of benefits that party p distributes, so Bipt ≡ {Bimt }m=1,...,Mp , βp ≡ {βpm }m=1,...,Mp . Then voter i who belongs to social groups s(i) votes for party k instead of l 6= k if Mk X βkm Bimt + Γs(i)kt δs(i) + ikt > Ml X βlm0 Bim0 t + Γs(i)lt δs(i) + ilt (1) m0 =1 m=1 Use γs(i),t to denote (Γs(i)kt − Γs(i)lt ), the relative ideological affinity of voters in social group s(i) with party k compared with party l at date t. Then Condition (1) implies that i votes for party k in election year t if [ Mk X βm Bimt − Ml X βm0 Bim0 t ] + γs(i),t δs(i) > ilt − ikt (2) m0 =1 m=1 and the probability of this event is increasing in [ Mk X m=1 βm Bimt − Ml X βm0 Bim0 t ] + γs(i),t δs(i) (3) m0 =1 where the first expression represents the utility-based component and the second term the ideological component. Condition (3) leads us to the regression specification Vote Partyipt = β0 + β1 State Benefitsit + β2 Central Benefitsit + γt Xi + εIpt 16 (4) where Vote Partyipt denotes a binary variable, which takes the value 1 if household i votes for party p in election year t and 0 otherwise; Xi is a vector of household socio-economic characteristics for household i. State Benefitsit and Central Benefitsit denote the total number of state and central benefits that household i received between years t−1 and t. For the 2013 round, central benefits are set equal to zero. The time-varying coefficient γt on household characteristics provides an estimate of how identity politics per se are changing over time owing to changes in party ideology, campaign mobilization efforts and intervening events affecting voters’ sensitivity to identity issues. The effect of benefits on votes (β1 and β2 ) is assumed to not vary over time in the specification given by equation (4); we shall relax this assumption in the analysis below to examine the hypothesis of declining effectiveness of clientelistic benefits. The corresponding regression for the change in the likelihood that i votes for party p is given by ∆Vote Partyip = β1 ∆State Benefitsi + β2 ∆Central Benefitsi + ∆γXi + ∆εIpt (5) where ∆ denotes the difference if the value of the variable between the two rounds. Thus, the likelihood that voter i switches allegiance from one party to another can be decomposed into the part caused by a change in the flow of state and central benefits, and the change in ideological or identity-based affinity with the party between the two rounds. This specification addresses selection concerns because unlike (4), in equation (5), household fixed effects wash out time-invariant drivers of political support for a particular party. On the other hand however, specification (5) does not allow for any changes in the effectiveness of state benefits at generating votes. 3.2 Role of Welfare Benefits In this section we focus on the role of state and central benefits in explaining changes in voting patterns of household heads. We start with estimates of the benefit regression coefficients from equation (4), estimated separately for the two different rounds, to allow for changes in the votegenerating effectiveness of state benefits across the two elections that have been argued by previous authors (Deshpande et al., 2019). An obvious concern with the specification in equation (4) is possible endogeneity of distribution of benefits which makes it difficult to give a causal interpretation to the OLS estimates. Higher level officials at the block or district level that control the allocation of these benefit programs across different GPs might strategically target them to GPs depending on anticipated voting propensities of their residents in the next election.16 Targeting to “loyal” areas would give rise to a positive bias, 16 For example, Mahadevan and Shenoy (2023) show that in West Bengal the ruling party channels disproportionate funds to politically-aligned jurisdictions in water-stressed areas and consequently gains votes in subsequent elections. 17 and to “swing” areas would generate a bias depending on how successful the benefit distribution might be in inducing voters to switch their allegiance. Moreover, the incentives of the incumbent party to deliver benefits might vary with their perceptions of its competitive standing in the next election. Less secure incumbents would be inclined to deliver more benefits to swing areas, giving rise to a negative bias. To reduce these biases, we use a supply-side or “leave out” instrument for the scale of programs at the GP level: the average number of benefits of either kind distributed in other GPs in the same district, as in Levitt and Snyder Jr (1997). These are interacted with fixed household characteristics such as caste, landownership, education, and religion (significant determinants of within-GP targeting) to predict the delivery of benefits to individual households. These characteristics are retained as controls in the regression, to incorporate their direct effect on voting propensities. We also include district fixed effects in the regression, to removes biases owing to possible unobserved heterogeneity across districts. The leave out instrument reflects factors that are less likely to be correlated with GP-specific unobserved political attributes. As Levitt and Snyder Jr (1997) and Bardhan et al. (2021) show, under plausible assumption, the resulting IV estimate is less likely to be biased and this bias converges to zero as the number of GPs in the district increases.17 As it turns out, the instrument turns out to have significant predictive power, with large values of the first stage F-statistic for the first-stage regressions (which are presented in Tables A2—A3). Nevertheless we evaluate the significance of the regression using a weak-instrument-robust Anderson-Rubin χ2 test. Panels A and B of Table 3 show the OLS and IV regression results for voting patterns in 2018 and 2013 respectively. These regressions control for a range of household demographic and socioeconomic characteristics. In any given round, the IV benefit coefficients are larger in magnitude compared to the OLS results, suggesting the OLS estimates are biased downwards (possibly reflecting the stronger motivation of insecure incumbents to provide more benefits and target areas with swing voters). In 2018, the OLS estimate indicates receipt of a state benefit raised the likelihood of the recipient to vote for the TMC by 5.9%, while the corresponding IV estimate of this effect is 9.3%. This increase came largely at the expense of the BJP (compare columns 2 and 4). The corresponding OLS and IV estimates of the effect of receiving a central benefit on the likelihood of voting for BJP were 2.7% and 6.3% respectively (see columns 3 and 4), though the latter is imprecisely estimated and is not statistically significant at the 10% level. Hence the vote-generating effectiveness of state benefits for the TMC was stronger compared to that of central benefits in favor of the BJP. 17 Recently this approach has been adopted widely in developing country contexts: see for example Lamichhane and Mangyo (2011), Bai et al. (2019), Dang and La (2019), Sedai et al. (2020), Bardhan et al. (2021), Maitra et al. (2023). The arguments of Levitt and Snyder Jr (1997) and Bardhan et al. (2021) imply that the recent criticism of this approach in settings with peer effects (Betz et al., 2018, McKenzie, 2021) does not apply to our setting. However, in the presence of heterogeneous treatment effects, the IV estimate should be interpreted as a local average treatment effect on the sub-population that is ‘treated at the margin’ as a consequence of variation in the instrument, in contrast to the OLS estimate which is a biased estimate of the average treatment effect applying to the entire population. 18 Table 3: Voting Patterns in 2018 and 2013, using Aggregate State and Central Benefits OLS IV OLS IV OLS IV OLS IV (1) (2) (3) (4) (5) (6) (7) (8) Panel A: Voting for Party in 2018 Central Benefits 2014–2018 State Benefits 2014–2018 Constant Number of Observations R-squared TMC BJP Left 0.007 (0.019) 0.059*** (0.009) 0.508*** (0.059) -0.035 (0.050) 0.093*** (0.021) 0.458*** (0.066) 0.027* (0.016) -0.046*** (0.009) 0.328*** (0.056) 0.063 (0.046) -0.090*** (0.018) 0.394*** (0.062) -0.004 (0.009) -0.008* (0.005) 0.069** (0.029) -0.014 (0.031) -0.009 (0.010) 0.072** (0.034) 3,507 0.052 3,507 0.044 3,507 0.067 3,507 0.049 3,507 0.025 3,507 0.024 First Stage F Central Benefits 146.32 [0.000] 165.50 [0.000] State Benefits Anderson-Rubin χ2 (32) 151.54 [0.000] 141.04 [0.000] 90.81 [0.000] Panel B: Voting for Party in 2013 TMC BJP Left State Benefits 2010–2013 Constant Number of Observations R-squared 0.018* (0.010) 0.474*** (0.049) 0.055** (0.026) 0.433*** (0.057) 0.003 (0.003) 0.039* (0.021) 0.018* (0.010) 0.023 (0.025) -0.031*** (0.010) 0.463*** (0.047) -0.088*** (0.025) 0.526*** (0.053) 3,507 0.366 3,507 0.361 3,507 0.060 3,507 0.055 3,507 0.268 3,507 0.254 First Stage F State Benefits Anderson-Rubin χ2 (32) Panel C: Switch in Voting Patterns Difference in Central Benefits Difference in State Benefits Constant Number of Observations R-squared 257.64 [0.00] 81.07 [0.00] 77.11 [0.00] 47.42 [0.00] Against TMC To TMC To BJP -0.016 (0.013) -0.019*** (0.006) 0.227*** (0.044) 3,507 0.091 0.012 (0.039) -0.043*** (0.013) 0.236*** (0.046) 3,507 0.083 0.005 (0.011) 0.018** (0.007) 0.307*** (0.039) 3,507 0.159 -0.005 (0.035) 0.014 (0.014) 0.310*** (0.039) 3,507 0.159 0.016 (0.015) -0.027*** (0.007) 0.260*** (0.053) 3,507 0.060 0.050 (0.047) -0.059*** (0.014) 0.274*** (0.055) 3,507 0.045 Against Left 0.018 (0.012) 0.011 (0.008) 0.391*** (0.041) 3,507 0.212 0.014 (0.040) 0.011 (0.016) 0.391*** (0.040) 3,507 0.212 First Stage F Difference in Central Benefits 114.69 [0.000] 279.90 [0.000] Difference in State Benefits Anderson-Rubin χ2 (17) 33.21 [0.011] 30.77 [0.021] 90.62 [0.000] 22.87 [0.154] Notes: In Panel A the dependent variables are support for TMC (Columns 1 and 2), BJP (Columns 3 and 4) and Left Front (Columns 5 and 6) as elicited in the 2018 survey, and in Panel B they are as elicited in the 2013 survey. In Panel C the dependent variables indicate a switch in political support toward and against a particular party from 2013 to 2018. In columns 2, 4, 6 and 8, the benefits variables are instrumented by the “leave-out” instrument described in the text. Regressions also control for religion, caste and the migration status of household, gender, marital status, education and occupation of the household head, household size and and indicator for residence in Hugli district. Standard errors in parentheses are clustered at the village level. Significance: ∗∗∗ p < 0.01,∗∗ p < 0.05,∗ p < 0.1. The full set of regression results corresponding to Panel A are presented in Table ??. 19 Did state benefits become less effective at generating votes for the TMC between 2013 and 2018? Alternatively, did voters in 2018 give the BJP (rather than the TMC) credit when they received state benefits? Our data suggest that the answer to both questions is no. The IV estimates presented in Table 3 show that in 2013 each additional state benefit that an average household received increased the likelihood that they voted for the TMC by 5.5% percentage points, decreased the likelihood that they voted for the Left Front by 8.8% (recall that the Left Front rather than the BJP was the main competition for the incumbent TMC in 2013) and increased the likelihood that the household voted for the BJP by 1.8%. Rather than having a smaller effect, in 2018 the state benefits were even more effective at influencing how households voted: each additional state benefit increased the likelihood that the household voted for TMC by 9.3% and decreased the likelihood that they voted for BJP by 9%. The use of the leave-one-out instrument above addresses the concern that other unobserved characteristics of the local village councils may influence which village councils deliver more versus less government benefits. One may also worry that within village councils, benefits may be strategically delivered to households with particular characteristics that the researcher cannot observe. For example, council officials may favor either loyal voters who can be depended on to vote for the incumbent party, or to swing voters who may change how they vote if they receive the benefit. As explained in Section 3.1, one way of dealing with this concern is to use the specification with household fixed effects (given by equation (5)), which washes out time-invariant characteristics that could influence which households received benefits. The specification allows us to examine whether and in which direction the household head switched party allegiance between 2013 and 2018, and how this switch was affected by changes in their receipt of benefits. In Table 3 Panel C, we present results from regressions for the probability that voters switched their support to the TMC (Column 1), and for the probability that they switched against the TMC (Column 2). This is followed by the likelihood that they switched towards the BJP (Column 3), and away from the Left Front (Column 4).18 We present both the OLS and the IV estimates of these switching likelihood regressions, where the instrument for changes in state benefits is the change in the leave-out instrument for per capita benefits in a given GP, interacted with the same household characteristics. The instruments are strong predictors of the change in state benefit flows (see the first stage estimates presented in Table A4). The IV results in Panel C show that an additional state benefit received reduced the likelihood of switching against TMC by 4.3 percentage points (column 2) and reduced the likelihood of switching to the BJP (by 5.9 percentage points, column 7). An additional state benefit received also increased the likelihood of switching to TMC, but this effect is small (1.4%) and not statistically significant. On the other hand, an increase in central benefit had no effect on switching to or against TMC. It induced a 5% increase in the likelihood of switching to the BJP, but the IV 18 Since at baseline very few voters supported the BJP and over time very few voters switched toward the Left Front, we do not run regressions for switching against BJP and towards the Left Front. 20 Table 4: Predicted Effect of Aggregate State and Central Benefits on Votes State Benefits 2010–2013 2014 – 2018 Average Per Household (Q) Unit Effect on TMC Vote Share (β̂T M C ) Unit Effect on BJP Vote Share (β̂BJP ) Total Predicted Effect on TMC Vote Share (β̂T M C × Q) Total Predicted Effect on BJP Vote Share (β̂BJP × Q) Central Benefits 2014 – 2018 Estiimated/ Actual Effect (4) (1) (2) (3) 1.16 0.055 0.018 0.064 0.016 1.79 0.093 -0.09 0.166 -0.161 0.54 -0.035 0.063 -0.019 0.034 Predicted Effect on Change in TMC Vote Share † State Benefits Central Benefits 0.103 -0.019 Total Predicted Effect of Benefits (Central + State) Actual Change in TMC Vote Share 0.084 -0.05 Predicted Effect on Change in BJP Vote Share ‡ State Benefits Central Benefits -0.177 0.034 Total Predicted Effect of Benefits (Central + State) Actual Change in BJP Vote Share -0.143 0.18 Notes: Actual Change in TMC and BJP vote shares from the strawpoll, reported in Figure 3. Q2013 and Q2018 denote 2018 − the average number of benefits received during the period 2010–2013 and 2014–2018 respectively. † : β̂T2018 MC × Q 2013 2013 ‡ 2018 2018 2013 2013 β̂T M C × Q ; : β̂BJP × Q − β̂BJP × Q . No Central Benefits in 2013. estimate is not statistically significant (column 7). In summary, even in 2018, clientelistic state benefits continued to help local incumbents. While central benefits helped the BJP, the effect is not precisely estimated. The results in Table 3 provide estimates of effectiveness of state and central benefit flows on voting propensities. These need to be combined with data on the actual flows to generate corresponding predicted effects on changes in vote shares. Table 4 carries out this calculation. The rise in state benefit flows in conjunction with rising effectiveness of these benefits in generating votes for the TMC yields an estimated rise of TMC vote share of 10.3% (which drops slightly to 8.4%, when we include the central benefits). Clearly the 5% reduction in this vote share cannot be explained by the hypothesis that clientelism was becoming less effective at generating votes. Conversely, the new central benefits predict a rise in BJP vote share by 3.4%. When we also include the effect of the state benefits we predict a net 14.3% decline in the BJP’s vote share. Since in fact the BJP’s vote share increased by 18%, it seems clear that their success is not driven by changes in welfare benefit flows. 21