Why Dominant Parties Decline: Evidence from India’s Green Revolution*

Aditya Dasgupta†

Abstract

Political scientists disagree whether economic growth strengthens or weakens dominant party regimes, the most common form of authoritarianism today. This paper reconciles competing arguments and develops a new theoretical explanation on the basis of a histor- ical natural experiment: the impact of the green revolution on single-party dominance in India. In contrast to modernization theory’s focus on rising incomes, I argue that changes in the economy which provide incentives for regime outsiders to capture the state can con- tribute to democratization. Fixed effects and instrumental variable analyses of panel data on high-yielding variety (HYV) crop adoption connected to 22,000 state and 3,000 parlia- mentary election races between 1957 and 1987 suggest that the green revolution accounted for half of the dominant Congress party’s long-run decline. The democratizing effect of the green revolution was due to the increasing value of agricultural subsidies, which agri- cultural producers sought to capture by supporting agrarian opposition parties. By con- trast, the green revolution caused no shift in the caste of elected MPs and income increases on their own resulting from rainfall shocks improved the dominant party’s election perfor- mance, ruling out modernization theory. The findings highlight a new class of explanations for why dominant parties decline.

*For valuable feedback, I thank Sam Asher, Rikhil Bhavnani, Dipak Dasgupta, Daniel Esser, Devesh Kapur, John Marshall, Nathan Nunn, James Robinson, Prerna Singh, Steven Wilkinson, George Yin, Daniel Ziblatt, Adam Ziegfeld and seminar participants at Harvard, Yale, and the American Political Science Association Annual Meet- ing, 2013. For financial support I thank the US-India Fulbright Scholar Program and IQSS. All errors are my own. †PhD Candidate, Department of Government, Harvard University. 1737 Cambridge Street, Cambridge, MA, 02138. Website: http://scholar.harvard.edu/adasgupta. Email: [email protected].

1 1 Introduction

Dominant party regimes – political systems in which a single party monopolizes political power despite competetive elections – are the most common form of authoritarianism today (Mag- aloni and Kricheli, 2010).1 A growing literature recognizes that competitive authoritarian and dominant party regimes are viable regime types in their own right as opposed to “halfway houses" on the path between authoritarianism and democracy (Brownlee, 2007; Levitsky and Way, 2010;

Schedler, 2006).

Given the prevalence and durability of dominant party regimes, the puzzle of of when and how dominant parties decline is among the most pressing in political science. Political scien- tists agree that economic growth plays a crucial role in dominant party survival and decline

(e.g. Pempel (1990)). However, political scientists have suggested two directly opposing logics.

On the one hand, some argue that economic growth can strengthen dominant parties by en- hancing their performance-based legitimacy (Huntington, 1991; Reuter and Gandhi, 2011). On the other hand, an influential strand of scholarship loosely based around modernization the- ory suggests that economic growth has a positive effect on democratization (Boix and Stokes,

2003; Lipset, 1959), and can reduce the effectiveness of clientelism (Dixit and Londregan, 1996;

Robinson and Verdier, 2013), an electoral strategy upon which dominant parties are especially reliant (Greene, 2007; Magaloni, 2006).

Cross-national empirical evidence on the matter is generally inconclusive.2 However, two stylized facts are clear, providing a motivation for the present study. First, dominant parties have proven to be extremely durable across the income spectrum, ranging, for example, from

1The literature sometimes makes an additional distinction between “authoritarian" dominant party regimes versus “democratic" dominant party regimes. As the distinction between authoritarian and democratic dominant party regimes is non-central to the argument made in the paper, I refer to dominant party regimes as a whole. 2Greene (2010) finds that across countries short run economic growth has a positive effect on authoritarian dominant party vote share but that per capita income levels have a negative impact in the long run. Reuter and Gandhi (2011) find that economic growth decreases the probability of elite defections from hegemonic parties but that per capita income levels have no effect. Moreover, serious concerns about endogeneity bedevil cross-national correlations between income levels and democratization (Acemoglu et al., 2008).

2 the Congress party which ruled for roughly 40 years in low-income India, to the PRI which ruled for roughly 70 years in middle-income Mexico, to the LDP which ruled high-income Japan for over five decades. Second, despite significant variation across settings in income levels, episodes of economic growth in each of these settings have corresponded to major periods of democratization. Magaloni (2006) argues that rising incomes and urbanization in Mexico contributed to the rise of the major oppposition party, PAN. Similarly, it has been argued that

Japan’s ‘economic miracle’ contributed to the emergence of a multi-party system. This paper makes the case that agricultural economic growth due to the green revolution played a pivotal role in the decline of single-party dominance in India.

This paper develops a new theory of when and how dominant parties decline. It argues that changes in the economy which provide incentives for regime outsiders to mobilize politically in order to capture the state contribute to democratization. The argument provides a new the- oretical explanation for how economic growth may contribute to dominant party decline and democratization, in a way that differs entirely from modernization theory. It clarifies puzzling cross-national empirical patterns by showing that while income increases on their own are un- likely to lead to democratization, economic growth can contribute to dominant party decline through an alternative channel – incentives for opposition mobilization – that the existing liter- ature has overlooked.

This argument is developed and tested in the context of the historical decline of single-party dominance in India, the world’s largest democracy. India is a crucial and dramatic case for un- derstanding why dominant parties decline. Though India successfully maintained democratic electoral institutions in the decades following independence in 1947, the “dark side" of this suc- cess was that it was achieved on the basis of electoral dominance by a single party, the Indian

National Congress, which won all national elections for three decades continuously until losing power for the first time in 1977. Over the course of the 1960s, 1970s and 1980s, the Congress party gradually lost its political monopoly in state-level elections and then, by the 1990s, in

3 national-level elections as well. Today, state- and national-level politics in India are character- ized by highly competitive, indeed fragmented, multi-party political competition.

To investigate the role played by economic growth in this transition, this paper analyzes a historical natural experiment: the massive increase in agricultural productivity due to the in- troduction of high-yielding variety (HYV) crops to India in the late 1960s, a transformation com- monly known as the “green revolution". Building on existing scholarship (Brass, 1980; Hasan,

1989b; Jaffrelot, 2003; Rudolph and Rudolph, 1987; Varshney, 1998), this paper contends that the green revolution played an important role in the decline of single-party dominance in India.

However, it adds to existing scholarship in three ways. First, it develops a systematic theoretical account of competing channels through which the green revolution may have contributed to dominant party decline, each corresponding to a distinct theory of how economic growth con- tributes to democratization. In particular, it differentiates an implicit version of modernization theory based upon rising incomes that is widespread in the historical literature from an alter- native, and I argue more compelling, theory which highlights capture of increasingly valuable state-controlled resources as a motive for opposition mobilization and democratization. Sec- ond, it provides empirical evidence for the causal effect of HYV crop adoption on weakening single-party dominance in India, using fixed effects and instrumental variable identification strategies based on sub-national variation in the adoption of HYV crops. Third, it tests compet- ing theoretical channels for this democratizing effect, utilizing a variety of detailed subnational historical agricultural and political data.

Theoretically, this paper identifies two plausible channels through which the green revolu- tion may have contributed to dominant party decline. The first is an income effect, the chan- nel typically posited by modernization theory. Since the Congress party sustained support in rural areas largely through clientelism and machine politics (Bailey, 1970; Weiner, 1967), the green revolution may have weakened the Congress party by raising incomes among farmers and lower-caste voters, making clientelism less effective (Dixit and Londregan, 1996; Robinson

4 and Verdier, 2013). It suggests as a more plausible alternative a second potential channel, a mo- bilization effect. Since the introduction of highly productive and input responsive HYV crops increased the value to agricultural producers of crop price supports and input subsidies con- trolled by the state, the green revolution plausibly provided economic incentives for agricultural producers to mobilize politically in order to seek greater subsidy levels. Because agricultural producers were under-represented in the urban-biased Congress party (Lipton, 1977; Rudolph and Rudolph, 1987) this mobilization took the form of support for agrarian opposition parties and contributed to the Congress party’s long run decline.

Empirically, this paper assembles a new panel dataset that connects district-level HYV crop adoption data to nearly 22,000 state assembly and 3,000 parliamentary election races across In- dia’s major states between 1957 and 1987, the key decades of the green revolution as well as of the Congress party’s initial decline. The first step of the empirical analysis estimates OLS regres- sions which control for district and year fixed effects, thereby absorbing any any time-invariant omitted variables and common shocks and trends. The second step of of the empirical analysis takes an instrumental variable identification strategy, exploiting the timing of the introduction of HYV crops to India in 1967 together with cross-sectional variation in suitability for HYV crop cultivation based on pre-existing irrigation coverage and the presence of naturally occurring aquifers.3 Both the OLS and IV estimates show that HYV crop adoption caused a large decline in the Congress party’s vote and seat share in state and parliamentary elections, accounting for roughly half of the long-run decline of single-party dominance in India. Robustness tests show that the estimates are not driven by selection bias in the diffusion of HYV crops nor by pre-trends preceding the introduction of HYV crops. The estimates are also robust to control- ling for region-specific time trends as well as for major alternative explanations, including local

3The IV analyses take advantage of the fact that HYV crop cultivation requires access to controlled irrigation (e.g. “a feature of the Green Revolution in India is that the ability to exploit the new seeds profitably was substan- tially different across India because of exogenous differentials in local soil and weather conditions” (Foster and Rosenzweig, 1996, p. 932)).

5 strength of Congress party organization and the fractionalization of opposition parties (Kohli,

1990; Ziegfeld and Tudor, 2013).

Tests of competing channels provide evidence that economic incentives for political mobi- lization by agricultural producers, rather than increased incomes or the erosion of traditional social hierarchies, were the primary channel for the democratizing effects of the green revolu- tion. Dis-aggregating the effects of HYV crops by opposition party type reveals that that specif- ically opposition parties with significant support from agricultural producers were the main electoral beneficiaries of the green revolution. Analysis of agricultural data shows that HYV crops placed downward pressure on the market price of crops and intensified the usage of in- puts such as fertilizer and tractors, changes which increased the value to agricultural producers of state-controlled agricultural subsidies and provided a motive for political mobilization. By contrast, analyses utilizing rainfall shocks as a placebo test – in the sense of missing the “active ingredient" necessary for democratization – reveal that rising incomes on their own did not re- duce but improved the dominant Congress party’s election performance. Moreover, analyses of the digitized biographies of Indian Members of Parliament (MPs) show that HYV crop adoption did not cause a shift in the caste, occupation, or educational attainment of elected politicians, suggesting that the erosion of traditional social hierarchies in political representation was not a salient channel either.

The findings show that the green revolution played a pivotal role in the decline of single- party dominance in India by providing agricultural producers with an incentive to mobilize po- litically. It differs from existing accounts which focus on the rise of lower caste groups in politics

(Jaffrelot, 2003), economic liberalization (Chhibber and Kollman, 2009), weakening party insti- tutions (Kohli, 1990), and opposition coordination (Ziegfeld and Tudor, 2013). The findings also provide within-country evidence that helps to reconcile competing theories about the effects of economic growth on the decline of dominant party regimes and on democratization more generally (Acemoglu et al., 2008; Boix and Stokes, 2003; Greene, 2007; Limongi and Przeworski,

6 1997). Contrary to the predictions of modernization theory, income increases on their own en- hance the performance legitimacy of the incumbent regime. However, economic growth which increases the value of political control and provides incentives for political mobilization by in- terest groups excluded from the incumbent regime can contribute to democratization. This is a novel theoretical lesson that, to my knowledge, has not been systematized and tested before.

In the conclusion, this paper contends that many episodes of democratization and dominant party decline around the world may be usefully viewed through the theoretical framework de- veloped and tested in this paper.

The remainder of the paper is organized as follows. The paper first provides historical back- ground on the dominance and subsequent decline of the dominant Congress party in India. It then develops the theoretical argument about the role played by the green revolution in this de- cline. It then describes the data before presenting the empirical strategy and reporting results. It concludes by placing the democratizing effects of the green revolution in India in comparative perspective.

2 Background

Congress Dominance and Decline

The first two decades following India’s independence in 1947 were a period of electoral dom- inance at both the state and national level by the Congress party, which had spearheaded the movement for India’s independence. Kothari describes this period as the “Congress system"

(Kothari, 1964), in which political competition was limited to rival factions within the Congress party, rather than originating in serious electoral threats from opposition parties.

The Congress party utilized patronage in order to dominate rural elections. Weiner (1967) and Bailey (1970), in fieldwork in different regions of India, found that the Congress party se- cured its electoral dominance on the basis of machine politics and clientelist ties to rural voters.

7 In Wilkinson’s words, “at the local level, Congress politicians in the 1950s and 1960s typically contracted for votes through upper-caste local intermediaries, who used their social status and control of land, credit and muscle power to deliver local upper as well as lower-caste votes to the Congress candidate (Kitschelt and Wilkinson, 2007, p. 113)." Thus despite a primarily urban, upper caste leadership (Jaffrelot, 2003), and an economic agenda focused upon state-supported industrial growth (Lipton, 1977; Rudolph and Rudolph, 1987), the Congress party was remark- ably successful at securing mass voter support.

The stability of the Congress system broke down beginning in the late 1960s, however, as opposition parties began to defeat the Congress party in state assembly elections across India.

General elections in 1977, following a two-year period of emergency rule by executive decree – declared by Congress prime minister Indira Gandhi in the face of growing political challenges

– saw the election of the Janata party, the first non-Congress national government in India’s history. A merger of opposition parties which initially came into force in state-level legislatures, the Janata government soon collapsed due to internal divisions. However, its successor parties, as well as a range of regional, communist, and Hindu nationalist opposition parties, continued to play an important role in Indian politics and by the 1980s competitive multi-party politics was firmly established in each of India’s major states (Yadav, 1996). By the 1990s, multi-party competition became an embedded feature of national-level politics as well.

Figure 1 below provides a plot for each of India’s major states of the timing of the historical decline of Congress party dominance, where each state-year observation represents whether the state was ruled by a Congress government or an opposition party. The estimated structural breaks represent roughly when each state made a lasting, as opposed to transitory, transition away from Congress party dominance. A key pattern to note is the considerable sub-national variation in the incidence and timing of dominant party decline, a feature that the empirical analysis will exploit (using even further dis-aggregated constituency level election data).

8 Figure 1: Decline of Congress Party Dominance Across Major States

ANDHRA PRADESH ASSAM BIHAR ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●● ●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●● ●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●● 1.0 0.8 0.6 0.4 0.2 Congress Government Congress Government Congress Government

●●●●●● ●●●●●●●●● ● ●● ●●●●● ●●●●● ● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●● 0.0 GUJARAT HARYANA KARNATAKA ●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●● ●● ●●●●● ●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●● ●●●●●●● ●●● 1.0 ANDHRA PRADESH ASSAM BIHAR 0.8 0.6 0.4 0.2 Congress Government Congress Government Congress Government

●●●●● ●●●●●●●●●●●●●●●●●●●●● ● ●●● ●●●●●●●●● ●●●●●●●●● ●● ●●●●●●● ●●●●● ●●●●●●● 0.0 KERALA MADHYA PRADESH MAHARASHTRA ●●●●●● ●●●● ●● ●●●●● ●●●●● ●●●●● ●●●● ●●●●●●●●●●●●●●●● ●●●●●●●●● ●●●●●●●●●● ●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●● ●●●●●●●●●●●●●● 1.0 GUJARAT HARYANA KARNATAKA 0.8 0.6 0.4 0.2 Congress Government Congress Government Congress Government

●●●●●● ●●●●●●●●●● ●●● ●●●● ●●●●● ●●●●●● ●● ●● ●●● ●●●●●●●●●●●● ●● ●●●●● ●● Congress In Power at State Level Congress In Power 0.0 ORISSA PUNJAB RAJASTHAN ●●●●●●●●●●●●●●●● ●●●●● ●●●●●●●●●● ●●●●● ●●●●●●●●●●●●●●●● ●●●●● ●●●●● ●●●●● ●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●●● ●●●●●● ●●●●● 1.0 KERALA MADHYA PRADESH MAHARASHTRA 0.8 0.6 0.4 0.2 Congress Government Congress Government Congress Government

●●●●● ●●● ●●●●● ●●●●●●●●●●●●●●●● ●●●●● ●●● ●●●●●●● ●●●●● ●●●●●●●●● ●●● ●●●●●●●● ●●●●● ●● 0.0 TAMIL NADU UTTAR PRADESH WEST BENGAL ●●●●●●●●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●● ●●●●●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●● 1.0 ORISSA PUNJAB RAJASTHAN 0.8 0.6 0.4 0.2 Congress Government Congress Government Congress Government

●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ● ●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● 0.0

1950 1970 1990 2010 1950 1970 1990 2010 1950 1970 1990 2010

TAMIL NADU UTTAR PRADESH WEST BENGAL Year

Notes: Points represent state-years between 1951 and 2015 across India’s major states. Each point indicates whether the state was controlled by a Congress state-level government. Vertical dashed lines represent esti- mated structural break points, and horizontal solid lines represent estimated average Congress government probability for a segment of time. Minimum segment size is fixed at 15 years and structural breaks are esti- mated via the method of Bai and Perron (2003). 9 The Role of Agrarian Political Mobilization

What explains the demise of single-party dominance in India? Revisiting the districts in which

Weiner (1967) originally documented remarkably effective Congress local political machines,

Kohli (1990) found that the Congress party had experienced a de-institutionalization of its local branches, contributing to its diminished success in elections. Others highlight the effects of national crises, such as Indira Gandhi’s declaration of rule by executive decree in 1975, which galvanized opposition parties and alienated voters (Weiner, 1982). Historians point to the death of Nehru and fading glow of India’s independence struggle, erstwhile sources of prestige and legitimacy for the Congress party. Finally, Chhibber and Kollman attribute the emergence of multi-party competition in India to the decentralization and liberalization of economic policy

(Chhibber and Kollman, 1998).

While each of these explanations is compelling, to the extent that these are national-level explanations they do not account for sub-national variation in the incidence and timing of de- cline in support for the Congress party. Later empirical analyses control for all of the national- level explanations described above by including in all regression specifications year fixed ef- fects, which adjust for national events, trends and shocks affecting election outcomes similarly across constituencies in India, such as economic liberalization – the bulk of which took place following the time period analyzed in the paper.4 This paper, by contrast, focuses on the major role that agrarian political mobilization played in Congress’s long-run decline, and on the role of the green revolution in driving this mobilization.

The success of opposition parties from the late 1960s onward was often built upon appeals to and incorporation of agricultural producers, a group that while numerically large was his- torically under-represented in the leadership of the urban-biased Congress party (Lipton, 1977;

4Later empirical analyses also seek to control for sub-national variation in explanations for Congress party de- cline suggested by Kohli (1990) (weakening of local party organizations) and Ziegfeld and Tudor (2013) (opposition coordination). It also provides evidence that the green revolution had no impact on the rise of lower or intermedi- ate caste groups in politics, the theory suggested by Jaffrelot (2003).

10 Rudolph and Rudolph, 1987).5 The growing political activism of agricultural producers was among the most prominent features of Indian politics in the 1970s and 1980s, a phenomenon described by Brass as “the politicization of the peasantry" and by Varshney as “the rise of agrar- ian power" (Brass, 1980; Varshney, 1998). The quintessential example of agrarian power was the

Bharatiya Kranti Dal (BKD), a party which under the leadership of iconic kisan (farmer) mobi- lizer Charan Singh defeated the Congress party in Uttar Pradesh on the basis of policy platforms oriented around increased subsidies for farmers. The BKD’s successor party, the Bharatiya Lok

Dal (BLD), served as one of the major constituents of the Janata party which defeated Congress nationally, an indication of the growing political clout of agricultural producers as a “demand group" in Indian politics (Rudolph and Rudolph, 1987). Apart from the BKD and its descen- dants, regional opposition parties which disrupted Congress dominance in other states, such as such as the Akali Dal in Punjab and the Telugu Desam Party in Andhra Pradesh, also drew much of their support and leadership from agricultural producers. More broadly, agrarian political mobilization increased across the political spectrum from the late 1960s onward, resulting in over 40 percent of national legislators coming from agricultural occupational backgrounds in

1989, compared to just 23 percent in 1952 (Varshney, 1998).

The present era of multi-party politics in India owes much to the emergence of opposition parties which drew political support from agricultural producers. As Rudolph and Rudolph

(1980, p.587) put it, “The emergence of agrarian interests on the national scene undid the polit- ical settlement of the Nehru era...” However, the root cause of the emergence of a viable agrarian opposition remains an open question – especially in light of the pre-existing long-run stability of the clientelistic Congress system. The following sections theoretically connect the decline of

Congress party dominance to India’s green revolution.

5Roughly 70 percent of the Indian population was employed in the agricultural sector in the time period under analysis.

11 The Green Revolution

The decline of the Congress party coincided with a dramatic increase in agricultural produc- tivity due to the introduction of high-yielding variety (HYV) crops to India in the late 1960s.

HYV crops, highly productive crop cultivars invented in the 1950s by Norman Borlaug and sci- entists working in Mexico, revolutionized agricultural production across the developing world over the course of the 1960s, 1970s and 1980s, a transformation commonly known as the “green revolution" (Evenson and Gollin, 2003).

India’s green revolution began in 1966-67, when HYV seeds, at first primarily of wheat and rice, were finally adapted to Indian conditions and saw widespread uptake. Public distribution of HYV seeds occurred initially, before country-wide roll-out, via the Intensive Agricultural Dis- tricts Program (IADP), which targeted the 16 ‘model’ districts viewed as most agriculturally ad- vanced and equipped with irrigation infrastructure to take advantage of the new seeds (Frankel,

1971; Mohan and Evenson, 1975). This targeting was motivated by the fact that for biological reasons HYV crop cultivation requires access to controlled irrigation (Evenson and Gollin, 2003;

Khush, 2001; Rawlins, 1977). While in the presence of controlled irrigation, HYV crops delivered yield increases of up to 200 percent, in the absence of irrigation they typically delivered lower yields than did traditional strains (Cleaver, 1972).

The government’s new agricultural strategy, formulated in the early 1960s under the prime ministerships of Jawahrlal Nehru and Lal Bahadur Shastri and carried forward under Indira

Gandi, was a response to domestic food shortages as well as a desire to reduce dependence on

US food aid amid the Cold War (Evenson and Rosegrant, 1998). HYV crop seeds and techniques were disseminated by the Indian bureaucracy, in part with the aid of international agencies such as the Ford and Rockefeller foundations and World Bank, in part with the support of major domestic agricultural research centers and universities (Ladejinsky, 1970). The spread of HYV crops resulted in a massive increase in agricultural output and transformed India from a net food importing to a net food surplus country by the mid-1970s. The diffusion of HYV crops

12 also fundamentally altered the structure of India’s agricultural sector. Because of the greatly increased productivity of HYV crops, the green revolution encouraged a transition from subsis- tence to commercialized agriculture and also resulted in a steady decline in the market prices of crops. Because HYV crop yields were highly responsive to the usage of inputs, the green revolu- tion also greatly intensified the usage of inputs such as fertilizer, pesticides, irrigation, tractors, and electricity for operating tube-wells (Evenson and Rosegrant, 1998; Frankel, 1971; Rud, 2012;

Walker and Ryan, 1990).

Notably, the intensity of the green revolution varied across regions of India (Cleaver, 1972).

The northwestern states, such as Punjab and Haryana, were particularly suitable for HYV crop adoption due to the presence of extensive groundwater aquifers, which made private tubewell irrigation possible, as well as colonial-era irrigation infrastructure (Stone, 2002). Even within states there was often considerable inter-regional disparity in the intensity of the green revo- lution. For instance, due to colonial legacies of canal irrigation the coastal region of Andhra

Pradesh and the western region of Uttar Pradesh experienced much higher levels of HYV crop adoption than other parts of those states. This paper exploits this sub-national variation to estimate the effects of the green revolution.

3 Theory

Historical research has suggested a connection between the agrarian political mobilization of the 1970s and the agricultural economic growth sparked by HYV crop adoption (Brass, 1980;

Hasan, 1989b; Jaffrelot, 2003; Rudolph and Rudolph, 1987). Brass (1980), analyzing the social base of support for the Bharatiya Kranti Dal (BKD) during the 1960s and 1970s, noted that

“...politicization, as measured in turnout rates, has been highest in the districts where agricul- ture is most advanced in terms of yields." Hasan (1989a) also notes the “politicization of back- ward and middle castes in the wake of the Green Revolution..." However, these authors stop

13 short of offering a systematic theory or empirical test of this relationship. As I will discuss, the green revolution plausibly weakened the Congress party through two distinct channels, each corresponding to a very different theory of how economic growth contributes to dominant party decline, with important comparative implications.

Income Effects

The first potential channel is an income effect, the channel typically posited by modernization theory. Rising incomes have been argued to contribute to democratization in a number of dif- ferent ways (Lipset, 1959). One pathway is via impacts on the efficacy of clientelism. Since single-party dominance, including the Congress party in India (Bailey, 1970; Weiner, 1967), is sustained largely on the basis of patronage, economic growth may contribute to democratiza- tion by raising voters incomes and making voters less dependent on the largesse of the domi- nant party, and therefore more willing to support opposition parties.

Research into the political economy of clientelism finds that clientelism is more effective as an electoral strategy in low income environments (Kitschelt and Wilkinson, 2007). Scott (1969, p. 1154) argues that clientelist machine politics are especially likely to prevail in poor countries because poor voters are more “easily swayed by concrete, material incentives". Dixit and Lon- dregan (1996) argue that poor voters are cheaper to buy. Similarly, Robinson and Verdier (2013, p. 263) argue that clientelism is more common in developing countries because “at low income levels, the political allegiance of clients is cheaper to buy with employment offers". By rais- ing the incomes of farmers, the green revolution may thus have made it fundamentally more difficult for the Congress party to purchase votes in rural areas. Magaloni (2006), for example, argues that rising incomes contributed to the decline of PRI dominance in Mexico.

Several authors have implicitly or explicitly advanced versions of the modernization argu- ment in the context of India. Guha (2007, p. 532), for example, rising incomes among farmers

14 “converted themselves into political ambition" (Guha, 2007, p. 532).6 Similarly, Jaffrelot (2003) argues that agricultural economic growth made lower- and intermediate-caste voters wealth- ier and therefore more politically “assertive", resulting in an expansion of lower-caste political representation in legislatures: “The Jats of western Uttar Pradesh and Haryana grew wealthy, notably thanks to the increase in sugar cane production resulting from extensive irrigation programmes in the framework of the Green Revolution. The assertion of these middle farm- ers...largely explains the growing success of Charan Singh’s kisan [farmer] politics in the 1960s."

The argument resembles Deutsch’s (1961) theory that economic growth encourages political participation by raising incomes and eroding traditional social hierarchies.

However, rising rural incomes alone are unsatisfactory as an explanation for dominant party decline in India, for multiple reasons that characterize weaknesses in modernization theory more broadly. First, despite rapid agricultural economic growth, India remained among the poorest countries in the world, with a per capita GDP of roughly $900 (base year 1990) in 1975.

Second, even if increased incomes did reduce the effectiveness of clientelism, this channel does not on its own explain why farmers who benefited from the green revolution supported op- position parties instead of being incorporated directly into the Congress party. Third, it runs counter to a large literature in political science on retrospective voting which suggests that vot- ers in fact reward political incumbents for strong economic performance (Healy and Malhotra,

2013). Huntington (1991) and Reuter and Gandhi (2011) argue that economic performance is a source of legitimacy and strength for authoritarian regimes and dominant parties.

Mobilization Effects

Alternatively, this paper contends, a more plausible theory is that the green revolution may have impacted support for the Congress party via a mobilization effect. HYV crop adoption and

6Writes Guha, “The commercialization of agriculture and milk production had benefited a significant section of farmers in rural India. Crucially, economic gains had converted themselves into political ambition. In the 1960s it was these rising rural castes who came to dominate governments in northern India."

15 the commercialization of agriculture it brought about greatly increased the value to farmers of agricultural subsidies controlled by the state, which agricultural producers sought to capture by supporting agrarian opposition parties because they were under-represented in the urban- biased Congress party (Rudolph and Rudolph, 1987).

As discussed, a central feature of HYV crops is that they were highly responsive in terms of yield to the usage of inputs such fertilizer, pesticide, irrigation and mechanization (Evenson and Gollin, 2003). The adoption of HYV crops therefore greatly increased the value to farmers of agricultural subsidies controlled by the government, such as subsidies for seeds, fertilizer and pesticide; subsidized electricity for operating tube-wells; and low-interest loans and waivers for purchasing farm equipment. The increase in output resulting from HYV crop adoption placed downward pressure on crop market prices. This made the level at which the Indian government set “minimum support prices", a long-standing policy which guaranteed agricultural producers a minimum crop procurement price, an increasingly important issue for agricultural producers.

These changes in the agricultural sector plausibly created economic incentives for agricultural producers to mobilize politically in order to obtain increased levels of subsidies.

A large political economy literature highlights the value to firms of influence over govern- ment (Krueger, 1974). In the presence of complementarities between production and control of government, increases in productivity may increase the returns to political control, gener- ating incentives for political mobilization by economic interest groups under-represented in the incumbent regime and thereby contribute to democratization. This is consistent with a class of formal models of contestation in which an increase in the value of the prize can elicit increased investment in competition over those resources (Bates, 1983; Skaperdas, 1992). In the appendix, I provide a simple game theory model that formalizes this intuition. The theory relates to Ansell and Samuels (2014), who argue that democratization occurs when rising eco- nomic interests demand political influence because they fear expropriation. It differs, however, in highlighting the prospect of economic gain from capturing the state as a powerful motive for

16 opposition mobilization and democratization.

The mobilization theory is consistent with the politicization of agricultural policy beginning in the 1970s. Hasan (1989a) writes that “What helped [the Bharatiya Kranti Dal] most was its image as the voice of rural India speaking on behalf of the agricultural sector, arrayed against the urban industrial interests which dominated the Congress." Rudolph and Rudolph (1987) and

Varshney (1998) document the major role played by price supports in the demands of agrarian opposition parties. The mobilization theory is also consistent with the expanded subsidies for agriculture that agrarian opposition parties implemented when elected to office, as well as the overall rise in spending on agricultural subsidies coinciding with India’s transition to multi- party competition. Between 1980 and 2000, government spending on agricultural subsidies as a share of GDP nearly tripled, rising from 1.2% to over 3% (Gaiha and Kulkarni, 2005).

The mobilization effect channel differs sharply from a theory based upon rising incomes.

The mobilization effect channel provides a strategic account of what agricultural producers stood to gain – subsidies made more valuable by the green revolution – by mobilizing. This channel also provides a theoretical rationale for why agricultural producers supported opposi- tion parties instead of joining the Congress party directly. The urban bias of the Congress party made it an implausible vehicle for the demands of agricultural producers. Finally, the mobiliza- tion theory of how economic growth contributes to dominant party decline also makes sharply different empirical predictions than does modernization theory. It suggests that rising incomes on their should not weaken single-party dominance. In the context of India, it also suggests that the green revolution ought to have caused no significant change in the presence of lower-caste or other traditionally disadvantaged groups in legislatures, as has been argued by proponents of the income theory of dominant party decline. Instead, the primary prediction of the mobiliza- tion theory in the context of India is that the adoption of HYV crops should have led to the rise of opposition parties representing the interests of agricultural producers.

17 4 Data

Explanatory Variable

To estimate the effects of the green revolution, this paper utilizes annual district-level data on

HYV crop adoption. Annual district-level data on the share of agricultural land under HYV crop cultivation is available in a dataset compiled by Sanghi et al. (1998) from annual Indian agri- cultural surveys.7 The panel dataset covers 270 districts between 1957 and 1987. The districts covered span all of India’s major states except for Assam and Kerala, and account for over 85 percent of India’s land area. To account for administrative splits of districts over time, all data are aggregated to the level of 1961 district boundaries. Figure 2 displays a map of Indian districts shaded according to mean HYV crop adoption between 1967 and 1987.

To instrument for HYV crop adoption, this paper utilizes a time indicator of the introduc- tion of HYV crops to India, which is coded as occurring in 1967, when HYV crops first diffused widely, in interaction with a cross-sectional measure of suitability for HYV crops based on pre- existing irrigation coverage. This is defined as the share of agricultural land in a district with access to controlled irrigation in 1966, just prior to the widespread introduction of HYV crops.

This includes irrigation of any kind, ranging from canals to tube-wells. As a robustness test, some specifications replace pre-existing irrigation with a binary indicator of the presence of a naturally occurring thick aquifer as the suitability measure.8

7Data on crop coverage come random samples of areas within districts by state-level officials, who report the data to the Ministry of Agriculture for centralized compilation in annual publications. 8This comes from the National Atlas of India, which contains hydrological maps of the presence of three cate- gories of aquifer: (1) fairly extensive thick aquifers occurring beyond 150 meters, (2) aquifers with limited extent occurring between 100 and 150 meters, and (3) aquifers with restricted extent occurring up to 100 meters. Follow- ing Rud (2012), I utilize an indicator of the presence of the deepest category of aquifer as the measure of suitability, since these areas were especially suitable for HYV crop cultivation based on tube-well irrigation. I differ in using a district-level indicator of thick aquifers as opposed to the state-level measure in Rud (2012).

18 Figure 2: Mean HYV Crop Adoption Across Districts, 1967-1987

Notes: Districts shaded according to quintile of mean share of agricultural land under HYV crop cultivation between 1967-1987. HYV crops were first widely adopted in 1967. Darker shades indicate greater HYV crop adoption. Districts defined by 1961 boundaries. Un-shaded districts are missing data.

19 Outcome Variable

The main outcome variable is the election performance of the Congress party. This paper an- alyzes a dataset containing constituency-level information on state assembly election races between 1957 and 1987 and parliamentary races between 1962 and 1985 across India’s ma- jor states. Table A1 in the appendix reports the states and elections in the analysis. The data are based on PDF election reports produced by the Election Commission of India, which were scraped by Jensenius (2013) for elections from 1961 onward. This paper extends this election dataset back in time by scraping the data for state assembly elections held in 1957.9

The main outcome variable is either binary, indicating whether a Congress party legislator was elected, or continuous, defined as the Congress party’s share of votes cast.10 These vari- ables are multiplied by 100 so that all reported regression coefficients have a percentage point interpretation. Figure 3 provides a map of districts shaded by over-time change in the Congress party’s election performance, computed by subtracting the percentage of state assembly seats won in elections after the introduction of HYV crops in 1967 (inclusive) from the percentage of seats won before 1967.

To connect the political data to HYV crop adoption data, this paper geo-codes assembly con- stituencies and parliamentary constituencies to 1961 district boundaries. During the period under analysis, India experienced three rounds of re-districting of constituency boundaries: in

1961, 1967 and 1976. This paper therefore utilizes the 1956, 1961, 1967, and 1976 reports of the

Delimitation Commission to name match constituencies to contemporaneous districts. It then utilizes the Administrative Atlas of India to match contemporaneous districts to 1961 district

9I do not extent the dataset to 1957 in the case of parliamentary elections, however, because key information is unavailable for coding these results. For multi-member constituencies, only aggregate candidate-wise data re- ported and it is not possible to assign candidates to the specific seat for which they are contesting. Detailed geo- graphical data on which districts and assembly constituencies each parliamentary constituency intersects with is also missing. 10A small complication arises in defining these variables for multi-member constituencies, which existed in 1957. These constituencies elected multiple candidates, with typically one seat reserved for competition between can- didates from under-represented minority groups (scheduled caste or scheduled tribe). To deal with this, I simply treat the “reserved” and “open" seat as two separate races.

20 boundaries. Assembly constituencies are smaller than and fit contiguously within districts, so each assembly constituency is matched to a single 1961 district. Parliamentary constituencies, however, occasionally cut across multiple districts; in these cases, I matched parliamentary constituencies to a single “greatest overlap" 1961 district.11

After accounting for assembly constituencies falling in districts not covered in the agricul- tural panel dataset, this procedure yields a panel dataset linking district-level HYV crop adop- tion data to constituency-level political data for nearly 22,000 state assembly races between

1957 and 1987 and 3,000 parliamentary races between 1962 and 1985.

Other Political Variables

Some analyses dis-aggregate the political effects of HYV crop adoption by opposition party type.

This is measured by an indicator of electing a legislator from and vote share accruing to agrarian and non-agrarian opposition parties, respectively. The top twenty opposition parties, based on total seats won in sample, are coded on the basis of qualitative analysis of the secondary litera- ture as “agrarian" if they received significant political support from and/or advocated programs oriented around the interests of agricultural producers.12

The BKD family of parties, including the BLD and Janata parties, are coded as agrarian op- position parties because of their well-documented support from farmers and advocacy of agri- cultural subsidy policies. Regional parties, including the DMK and AIADMK in Tamil Nadu, the

Akali Dal in Punjab, and the Telugu Desam party in Andhra Pradesh, which are widely recog- nized for drawing support from farming caste groups and advocating for agricultural subsidies, are also coded as agrarian.13 By contrast, the Hindu nationalist Bharatiya Jana Sangh, and its

11“Greatest overlap" is identified by selecting the district in which fall the greatest number of assembly con- stituencies comprising a parliamentary constituency. All of the reported results are substantively identical to using alternative coding solutions, including aggregating over all intersected districts. See Figure A3 in the Appendix for an example of how assembly and parliamentary constituencies were matched to 1961 districts. 12However, in coding I do not require that these parties appeal exclusively to agricultural producers. 13An ambiguous case in the DMK. Relative to the AIADMK, the DMK has historically drawn more support in elections from urban areas in Tamil Nadu. However, the populist platforms of both parties have included consid-

21 Figure 3: Change in Congress State Assembly Seat Share by District, Pre-Post 1967

Notes: Districts shaded according to quintile of percentage point difference in share of assembly constituencies with a Congress legislator, subtracting the the pre-1967 period average from the post-1967 (inclusive) period average. HYV crops were first widely adopted in 1967. Darker shades indicate a larger loss of Congress seat share. Districts defined by 1961 boundaries. Un-shaded districts are missing data.

erable subsidies, notably free electricity, for farmers. I code the DMK as well as the AIADMK as agrarian, consistent with Wyatt’s (2009, p.53) observation that “both parties have shown willingness to accomodate rural interests. In contrast the Dravidian parties have been criticized for neglecting22 urban development." Omitting this party from the list of the top 20 opposition parties or altering this coding does not significantly change the results. descendent, the BJP, are coded as non-agrarian, as are communist opposition parties, such as the CPI and CPI(M). Opposition parties that drew support from rural areas but did not advo- cate for the interests of agricultural producers, such as the left-wing Peasants and Workers Party of India (pro-rural poor) or the conservative Swatantra party (pro-landlord), are coded as non- agrarian as defined in this paper. Table A2 in the appendix lists the coding of the top twenty opposition parties along with citations of secondary literature.

To assess the impact of HYV crops on traditional social hierarchies in political representa- tion, this paper utilizes two datasets. One, assembled by Jaffrelot (2003), collects data on the caste of elected MPs in India’s Hindi speaking states based on surname analysis, legislator bi- ographies and key informant interviews. This paper assembles a second dataset by scraping the official self-reported ‘Who’s Who’ biographies of Indian MPs across all states from the In- dian parliament’s website, and utilizing natural language processing to code the educational level and occupation of MPs.14

Other Economic Variables

To measure the economic impact of HYV crops, this paper utilizes several additional district- level economic variables from the Sanghi et al. (1998) dataset. Wages are measured in terms of the daily wages of male agricultural laborers (rupees/day).15 Output is measured in terms of annual yield (rupees/hectare), computed by multiplying the area normalized output of the six major crops – rice, wheat, sorghum, millet, maize, sugarcane – by the market price of each of the crops. All prices are deflated to 1973 levels based on the state-specific annual consumer price index in Ozler and Ravllion (1996). This paper measures usage of inputs in terms of fertilizer usage (kilograms/hectare) and tractor usage (tractors/1000 hectares). To assess average crop

14Occupational information is self-reported, and thus partly shaped by a legislator’s subjective self- representation. For instance, many MPs describe themselves as “agriculturists" despite clearly never having worked as a farmer. This is taken into account in interpretation of the results. See Figure A4 in the Appendix for an example of how digitized legislator biographies were coded. 15This data comes from the Agricultural Wages of India publication series.

23 market prices, this paper constructs a crop price index (rupees/quintal), computed by taking a revenue weighted average of the market price of the six major production crops.

Finally, some specifications utilize rainfall shocks as a source of exogenous variation in agri- cultural productivity. Rainfall is measured in terms of the natural log of annual rainfall (ln mm).

Because all analyses utilize district fixed effects, they implicitly identify the effects of within- district rainfall variation over time, i.e. shocks. Data on annual rainfall at the 1961 district boundary level comes from Jayachandran (2006), who aggregates 0.5◦ 0.5◦ gridded annual × rainfall maps from the Center for Climatic Research at the University of Delaware to 1961 dis- trict boundaries. Table 1 reports descriptive statistics for major variables in the analysis.

24 Table 1: Descriptive Statistics for Major Variables in Analysis

Variable 1957-66 Mean 1967-76 Mean 1977-87 Mean SD Within-SD Panel A: State Assembly Constituency-level Congress Win 63.77 51.77 42.77 49.99 47.85 Congress Vote % 43.30 37.54 37.45 16.26 14.70 Agrarian Win 0.87 7.70 32.53 36.88 33.99 Agrarian Vote % 1.00 5.59 26.80 21.68 18.96 Non-Agrarian Win 27.31 34.48 22.04 44.72 42.60 Non-Agrarian Vote % 42.62 44.30 29.44 22.59 21.09 Panel B: Parliamentary Constituency-level Congress Win 74.89 61.70 58.16 48.58 42.01 Congress Vote % 45.82 44.26 43.99 16.48 13.59 Agrarian Win 1.36 4.41 31.68 38.39 35.16 Agrarian Vote % 1.72 2.97 34.15 26.39 23.53 Non-Agrarian Win 17.87 30.39 8.62 37.84 31.17 Non-Agrarian Vote % 40.80 46.54 17.53 24.32 21.10 Lower caste 0.29 0.31 0.35 0.47 0.37 OBC caste 0.10 0.13 0.21 0.37 0.31

25 Upper caste 0.61 0.57 0.44 0.50 0.41 Agriculturist 0.28 0.34 0.49 0.49 0.42 Professional 0.36 0.38 0.39 0.48 0.43 University 0.70 0.72 0.74 0.44 0.40 Panel C: District-level HYV 0.00 0.14 0.33 0.20 0.17 Rainfall 6.79 6.82 6.86 0.52 0.29 Yield 1060.39 1320.37 1347.93 777.00 367.23 Wage 4.04 4.09 5.11 1.90 1.13 Crop Price 121.20 127.67 101.08 31.49 24.10 Fertilizer 1.92 10.17 28.23 21.35 17.14 Tractors 0.22 0.90 3.15 3.65 2.92 1966 Irrigation 0.21 0.21 0.21 0.20 0.00 Aquifer 0.08 0.08 0.08 0.27 0.00

Notes: Within-SD is within-district standard deviation after partialing out district fixed effects. Panel A and B: Win is a binary indicator of winning the seat, multiplied by 100. Vote % is vote share, multiplied by 100. See text and appendix for coding of agrarian vs. non-agrarian opposition parties and MP biographies. Panel C: HYV is share of agricultural land under HYV crops. Yield is agricultural productivity measured in terms of Rs./Ha for the top 6 major crops. Wage is the daily wage of a male agricultural laborer in Rs./day. Crop Price is a revenue weighted average of the price in Rs./quintal of the top 6 crops. All monetary amounts are inflation-adjusted, base year 1973. Fertilizer usage is measured in terms of Kg./Ha. Tractors is number of tractors per 1000 hectares of agricultural land. Rainfall is measured in ln(annual mm). 1966 Irrigation is share of agricultural land with irrigation access in 1966. Aquifer is an indicator for the presence of a naturally occurring thick aquifer. 5 OLS Empirical Strategy and Results

To estimate the political effects of HYV crops, in the basic specification this paper estimates an

OLS regression of the form,

Y γ τ αHYV X0 γ ε (1) di t = d + t + dt + dit + di t

where Ydi t is a measure of Congress election performance, in constituency i in district d and year t. The variable HYVdt is a district-level measure of HYV crop adoption, defined as the share of agricultural land planted with HYV crops. The main coefficient of interest is α, the marginal effect of HYV crop adoption on the Congress party’s election performance. All speci-

fications include district, γd , and year, τt , fixed effects. The term Xdit0 is a vector of covariates, which will be discussed along with the results. Because the explanatory variable, HYV crop adoption, is measured at the district level while election outcomes are observed at the assem- bly or parliamentary constituency level, all analyses report robust standard errors adjusted for clustering by district, which accounts for correlation of errors within districts as well as over- time correlation of errors within districts.

A fixed effects regression of this form estimates the effect of changes within districts over time in HYV crop adoption on the Congress party’s performance in elections, while also partial- ing out any time-based shocks and trends common to all constituencies. A simplified version of the analysis is displayed in Figure 4 in the form of a district-level scatter plot of change in

Congress party seat share from the pre- to the post-1967 HYV crop introduction period against mean post-1967 HYV crop adoption. A large, negative relationship is apparent, providing pre- liminary evidence that districts which experienced the green revolution with greater intensity experienced a larger decline in Congress party election performance.

26 Figure 4: HYV Crop Adoption vs. Change in Congress Seat Share by District ● ● ● State Assembly Elections Parliamentary Elections● ● ● ●

60 60 ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●● ● ● ● ● ●●

40 ● 40 ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● 20 ● 20 ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● 0 0 ● ●●●● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● −20 ● ● ● ● −20 27 ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● −40 ● ● ● ● ● −40 ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● Percentage Point Change in Seat Share Change in Seat Point Percentage ● ● ● Change in Seat Share Point Percentage ● ● ● ● ● ● ● ● ● ●●●●● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −60 −60 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −80 −80 ●● ● ● ● ● ● ● 0.0 0.2 0.4 0.6 0.0 0.2 ● 0.4 0.6

Mean HYV Adoption, 1967−1987 Mean HYV Adoption, 1967−1987

Notes: Points represent districts. Horizontal axis measures mean share of agricultural land under HYV crop cultivation between 1967-1987. Vertical axis measures percentage point difference in share of assembly constituencies with a Congress legislator, subtracting the the pre-1967 period average from the post-1967 (inclusive) period average. HYV crops were first widely adopted in 1967. Line represents fitted bivariate OLS regression line. For state assembly elections, slope of line is -50.3 (standard error: 9.8). For parliamentary elections, slope of line is -28.5 (standard error: 19.4). Table 2 reports the main OLS regression estimates of the effect of HYV crop adoption on the probability of electing a Congress legislator and Congress vote share in state assembly elections based on constituency level panel data. The estimates provide evidence that the green revolu- tion had a large negative effect on the Congress party’s performance in elections. The results of the baseline specification reported in columns (1) and (4) imply that a 0.10 increase in the share of agricultural land under HYV crops reduced the probability of electing a Congress legislator by 3.2 percentage points and Congress vote share by 1.2 percentage points.

The magnitude of the estimates suggest that the green revolution played a pivotal role in the decline of Congress party dominance. The average share of agricultural land under HYV crops across all district-years following the introduction of HYV crops to India was 0.24. This implies that, if we multiply this number by the estimated impact of HYV crops, the green revolution was overall responsible for a 7.8 percentage point reduction in Congress party seat share and a 2.8 percentage point reduction in Congress party vote share. These are very large estimates, accounting for nearly half of the Congress party’s observed decline of 16.8 percentage points in seat share and 6.4 percentage points in vote share from the pre- to the post-HYV crop introduc- tion period in the sample under analysis.

The estimates are robust to a range of additional specifications. The specifications reported in columns (2) and (5) additionally control for year fixed effects interacted with a dummy vari- able indicating location in a Hindi-speaking state and with a dummy variable indicating loca- tion in a region historically under direct British colonial rule.16 The analysis controls for time trends specific to the Hindi-speaking states because much of the mobilization in opposition to the Congress party during the 1970s, such as the mass protests organized by Jayprakash

Narayan, originated in India’s ‘Hindi belt’ (Jaffrelot, 2003). It also controls for time trends spe-

16The Hindi-speaking states are defined as Bihar, Haryana, Madhya Pradesh, Rajasthan and Uttar Pradesh. District-level data on direct colonial rule come from Iyer (2010). Interacting these variables with year dummy variables non-parametrically absorbs any time-based trends or shocks specific to these areas. These variables are interacted with year dummy variables, not included as lower order terms, because district fixed effects absorb any time-invariant variables.

28 Table 2: OLS Estimates: Effects of HYV Crop Adoption on Congress Performance in State Elections

Dependent variable: Congress Win Congress Vote %

(1) (2) (3) (4) (5) (6)

HYV 32.334∗∗∗ 26.848∗∗∗ 19.192∗∗∗ 11.654∗∗∗ 9.691∗∗∗ 5.854∗∗∗ − − − − − − (6.124) (5.854) (5.039) (2.298) (2.439) (1.818)

Opposition Fractionalization 39.927∗∗∗ 13.058∗∗∗ − (2.040) (0.917)

INC Presence 49.858∗∗∗ 36.500∗∗∗ (1.458) (0.550) 29 Hindi Belt Year FE Y Y Y Y × Direct Rule Year FE Y Y Y Y × Year FE Y Y Y Y Y Y District FE Y Y Y Y Y Y Observations 21,907 21,907 21,906 21,907 21,907 21,906 Adjusted R2 0.18 0.19 0.24 0.32 0.33 0.50 Clusters 270 270 270 270 270 270

Notes: Unit of observation assembly constituency-year for election years between 1957-1987. Outcomes: Congress Win indicates whether Congress won seat and Congress Vote % indicates Congress vote share, both multiplied by 100 so that coefficients have a percentage point interpretation. Explanatory variables: HYV crop adoption is district-level share of agricultural land with HYV crops (0-1). Opposition Fractionalization is one minus the Herfindahl index of opposition party vote shares (0-1). INC Presence is an indicator variable for the presence of a Congress party candidate. Hindi Belt is an indicator variable for location in a Hindi-speaking state. Direct Rule is an indicator variable for historical exposure to direct British

colonial rule. Analysis estimated by OLS. Standard errors adjusted for clustering within districts. ∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01 < < < cific to directly ruled areas because the Congress party was historically banned in areas of in- direct colonial rule, raising the possibility of different trajectories of Congress performance in places exposed historically to direct versus indirect colonial rule. Reassuringly, time trends spe- cific to the Hindi belt and to directly ruled areas do not drive the results.

Columns (3) and (6) report estimates from specifications which additionally control for two major alternative explanations for Congress’s decline. As discussed, district fixed effects and year fixed effects absorb any time-invariant or national-level alternative explanations for dom- inant party decline in India. Ziegfeld and Tudor (2013) argue that the absence of opposition coordination was an important determinant of the Congress party’s success in elections. To operationalize this explanation, the analysis includes a variable representing constituency level fractionalization of opposition party vote shares.17 Kohli (1990) argues that the institutional de- terioration of local party branches drove the decline of Congress. To operationalize this variable, the analysis controls for whether Congress fielded a candidate, intended to serve as a proxy for the bare minimum presence of local party organization. Including these covariates runs the risk of conditioning on post-treatment variables (Rosenbaum, 1984), which could plausibly bias the estimates of the effect of HYV crop adoption downward if these variables were themselves im- pacted by the green revolution. For instance these variables may themselves be outcomes of

Congress weakness, or perhaps channels through which HYV crop adoption reduced support for Congress. For this reason, the estimates from this specification are not preferred and likely to be conservative. Nonetheless, HYV crop adoption is estimated to reduce Congress perfor- mance in elections even when these variables are controlled for.

Table 3 reports a parallel set of results using national parliamentary election results instead of state assembly election results as the outcome variable. Again, the estimates suggest that across specifications HYV crop adoption caused a large decline in Congress party seat share

17This is constructed by computing the Herfindahl index of non-Congress party vote shares and subtracting this from 1. This variable ranges from 0-1, with a higher value indicating greater fractionalization.

30 Table 3: OLS Estimates: Effects of HYV Crop Adoption on Congress Performance in Parliamentary Elections

Dependent variable: Congress Win Congress Vote %

(1) (2) (3) (4) (5) (6)

HYV 37.145∗∗∗ 35.532∗∗∗ 20.829∗ 15.176∗∗∗ 14.957∗∗∗ 6.847∗∗ − − − − − − (12.385) (12.333) (10.991) (4.036) (3.891) (2.683)

Opposition Fractionalization 70.014∗∗∗ 3.990∗∗ − (5.292) (1.658)

INC Presence 84.738∗∗∗ 49.199∗∗∗ (4.712) (1.435) 31 Hindi Belt Year FE Y Y Y Y × Direct Rule Year FE Y Y Y Y × Year FE Y Y Y Y Y Y District FE Y Y Y Y Y Y Observations 2,953 2,953 2,953 2,953 2,953 2,953 Adjusted R2 0.26 0.33 0.44 0.32 0.37 0.70 Clusters 260 260 260 260 260 260

Notes: Unit of observation parliamentary constituency-year for election years between 1962-1985. Outcomes: Congress Win indicates whether Congress won seat and Congress Vote % indicates Congress vote share, both multiplied by 100 so that coefficients have a percentage point inter- pretation. Explanatory variables: HYV crop adoption is district-level share of agricultural land with HYV crops (0-1). Opposition Fractionalization is one minus the Herfindahl index of opposition party vote shares (0-1). INC Presence is an indicator variable for the presence of a Congress party candidate. Hindi Belt is an indicator variable for location in a Hindi-speaking state. Direct Rule is an indicator variable for historical exposure to

direct British colonial rule. Analysis estimated by OLS. Standard errors adjusted for clustering within districts. ∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01 < < < and vote share. The estimates suggest that 0.10 increase in the share of agricultural land under

HYV crops caused a 3.7 percentage point reduction in the probability of electing a Congress MP and a 1.5 percentage point reduction in Congress vote share. These estimates imply, based on average HYV crop adoption in the post-green revolution period (0.24), that the green revolution was overall responsible for a 8.9 percentage point decline in Congress seat share and a 3.6 per- centage point decline in Congress vote share from the pre to post-green revolution period in parliamentary elections. This suggests that the green revolution accounted for roughly half of the half of the Congress party’s overall decline of 15.2 percentage points in seat share and essen- tially all of the Congress party’s decline of 2.5 percentage points in vote share in parliamentary elections (indeed, the estimates suggest that counter-factually absent the green revolution the

Congress party might have overall gained in vote share in parliamentary elections).

Determinants of HYV Crop Adoption

In spite of controlling for district and year fixed effects and a range of potential confounding variables, the estimates reported thus far may be biased if the diffusion of HYV crops was related to the dynamics of the Congress party’s election performance in ways not controlled for. In particular, the Congress party may have strategically manipulated the targeting of HYV crops; the direction of the bias implied by such targeting is unpredictable, depending upon whether the Congress party targeted supporters, opposition areas, or swing voters (Dixit and Londregan,

1996). It is therefore important to understand the determinants of HYV crop adoption.

Qualitative information suggests that HYV crop adoption was not driven by political target- ing. The IADP and other programs used to propagate HYV crops in the early stages were formu- lated in the context of food shortages and a pressing need to increase food production regard- less of the long-run political consequences. Additionally, the technical management of these programs was often delegated to domestic and international technocratic agencies (Evenson and Rosegrant, 1998). Finally, the government’s new agricultural strategy was formulated in the

32 early- to mid-1960s, well before the emergence of agrarian political mobilization and threats to

Congress party dominance. For all of these reasons, there appears to have been little political targeting in the distribution of HYV crop seeds.

This view finds support in the data. Table 4 reports estimates from OLS regressions of two district-level indicators of the intensity of the green revolution on pre-treatment predictors. The measures of the intensity of the green revolution are mean HYV crop adoption, 1967-1987, and whether or not a district was selected for the IADP program. The regression estimates show that pre-treatment Congress seat share is completely uncorrelated with the diffusion of HYV crops or selection for the IADP program. However, pre-treatment irrigation levels strongly predict

HYV crop adoption and selection for the IADP program, consistent with qualitative information about the dependence of HYV crop cultivation on access to controlled irrigation. The presence of a thick aquifer, a source of variation in capacity for tubewell irrigation, upon which much

HYV crop cultivation was based, strongly predicts the diffusion of HYV crops but not selection for the IADP program; this is also consistent with qualitative information, as the IADP program was targeted at districts with pre-existing irrigation infrastructure, whereas tube-well irrigation expanded rapidly following the introduction of HYV crops (Rud, 2012).

These results give additional credibility to the OLS estimates reported so far by mitigating concerns about selection bias arising from electoral targeting of the diffusion of HYV crops.

The results are also of theoretical value. Previous research highlights the role that patronage and political manipulation of public resources play in sustaining dominant party rule (Greene,

2007). The apparent failure or inability of the Congress party to politicize the distribution of

HYV crops helps to explain why the green revolution did not reinforce the Congress party’s clien- telistic dominance in rural elections.

33 Table 4: OLS Estimates: Pre-treatment Determinants of HYV Crop Adoption

Dependent variable: HYV IADP (1) (2) (3) (4) (5) (6) (7) (8) State Assembly INC Seat Share (%) 0.001 0.00003 (0.0004) (0.001)

Parliamentary INC Seat Share (%) 0.0001 0.00000 (0.0002) (0.0004)

1966 Irrigation 0.516∗∗∗ 0.238∗∗∗ (0.032) (0.065) 34

Aquifer 0.195∗∗∗ 0.046 (0.030) (0.048)

Constant 0.203∗∗∗ 0.237∗∗∗ 0.129∗∗∗ 0.223∗∗∗ 0.046 0.052∗ 0.003 0.045∗∗∗ − (0.030) (0.018) (0.009) (0.009) (0.044) (0.029) (0.019) (0.014)

Observations 270 247 270 269 270 247 270 269 Adjusted R2 0.002 0.002 0.492 0.133 0.004 0.004 0.044 0.0002 − − − − Notes: Unit of observation is district. Outcomes: HYV Intensity is mean district-level HYV crop adoption, 1967-1987 (0-1). IADP is an indicator for whether a district was selected for the Intensive Agricultural Districts Program. Explanatory variables: INC Seat % is percentage point Congress seat share before 1967, for state assembly and parliamentary elections, respectively. 1966 irrigation is pre-treatment share of agricultural land with

access to irrigation in 1966 (0-1). Aquifer is an indicator for the presence of a naturally occurring thick aquifer. Analysis estimated by OLS. ∗p 0.1; < ∗∗p 0.05; ∗∗∗p 0.01 < < 6 IV Empirical Strategy and Results

An alternative strategy for dealing with possible endogeneity is to utilize an instrumental vari- ables empirical strategy. To construct an instrument for HYV crop adoption, this paper utilizes the fact that HYV crops, which require steady dosages of moisture, deliver higher yields only in areas with access to controlled irrigation. It is widely recognized that for this reason HYV crops tended to diffuse to areas with pre-existing irrigation infrastructure (Cleaver, 1972; Even- son and Gollin, 2003; Khush, 2001; Rawlins, 1977), as has also been shown empirically. Based on this information, this paper constructs a difference-in-differences style instrument for HYV crop adoption. The first stage regression takes the form,

1966 1967 HYV γ τ βIr r i g ation Post X0 κ ν , (2) dt = d + t + d × t + dit + di t where Ir r i g ation1966 Post 1967 is the excluded instrument, the interaction of a cross-sectional d × t measure of suitability for HYV crops based on irrigation coverage in 1966 with a dummy vari- able which ‘switches on’ with the introduction of HYV crops to India, from 1967 onward. Lower order terms are accounted for in the equation via district, γd , and year, τt , fixed effects, which also absorb any time-invariant omitted variables as well as time-based shocks and trends com- mon to all constituencies. The second stage regression is the main OLS specification (1). This identification strategy utilizes the combination of time variation arising from the introduction of HYV crops in 1967 together with cross-sectional variation in suitability for HYV crop cultiva- tion to estimate the impact of HYV crop adoption on Congress election performance.

Table 5 reports instrumental variable regression estimates of the effect of HYV crop adoption on the probability of electing a Congress legislator and on Congress party vote share in state assembly elections. The baseline IV estimate suggest that a 0.10 increase in the share of agri- cultural land under HYV crops resulted in a 7.6 percentage point decrease in the probability of electing a Congress legislator and a 2.4 percentage point reduction in Congress party vote share.

35 Table 5: IV Estimates: Effects of HYV Crop Adoption on Congress Performance in State Elections

Dependent variable: Congress Win Congress Vote %

(1) (2) (3) (4) (5) (6)

HYV 76.174∗∗∗ 77.589∗∗∗ 49.679∗∗∗ 23.272∗∗∗ 21.982∗∗∗ 11.052∗∗ − − − − − − (16.629) (19.831) (17.791) (5.650) (6.423) (5.151)

Opposition Fractionalization 39.604∗∗∗ 13.113∗∗∗ − (2.030) (0.919)

INC Presence 48.852∗∗∗ 36.329∗∗∗ (1.531) (0.570)

Hindi Belt Year FE Y Y Y Y × Direct Rule Year FE Y Y Y Y ×

36 Year FE Y Y Y Y Y Y District FE Y Y Y Y Y Y First-stage 1966 Irrigation Post 0.433∗∗∗ 0.402∗∗∗ 0.400∗∗∗ 0.433∗∗∗ 0.402∗∗∗ 0.400∗∗∗ × (0.051) (0.051) (0.051) (0.051) (0.051) (0.051) F-statistic 71.6 61.7 61.5 71.6 61.7 61.5 Observations 21,907 21,907 21,906 21,907 21,907 21,906 Adjusted R2 0.17 0.18 0.24 0.31 0.33 0.50 Clusters 270 270 270 270 270 270

Notes: Unit of observation assembly constituency-year for election years between 1957-1987. Outcomes: Congress Win indicates whether Congress won seat and Congress Vote % indicates Congress vote share, both multiplied by 100 so that coefficients have a percentage point interpretation. Explanatory variables: HYV crop adoption is district-level share of agricultural land with HYV crops (0-1). Instrument for HYV crop adoption is an interaction term between a time variable indicating the introduction of HYV crops in 1967 and a cross-sectional variable indicating the share of agricultural land with access to irrigation in 1966. Opposition Fractionalization is one minus the Herfindahl index of opposition party vote shares (0-1). INC Presence is an indicator variable for the presence of a Congress party candidate. Hindi Belt is an indicator variable for location in Hindi- speaking state. Direct Rule is an indicator variable for historical exposure to direct British colonial rule. Analysis estimated by 2SLS. Standard errors

adjusted for clustering within districts.∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01 < < < The sign and magnitude of the estimates are robust to the inclusion of a large set of covariates, already discussed. The first stage estimates also show an extremely strong relationship between the instrument and HYV crop adoption across specifications.

Table 6 reports a parallel set of results for the effect of HYV crop adoption on Congress elec- tion performance in parliamentary elections. The IV estimates suggest that a 0.10 increase in the share of agricultural land under HYV crops reduced Congress vote share by 8.0 percent- age points and Congress seat share by 2.7 percentage points. As before, the estimates are stable across specifications controlling for potential confounders and alternative explanations, except in the case of specifications which include “bad" post-treatment control variables.18

A discernible pattern is that the IV estimates are roughly double the magnitude of the OLS estimates. The larger IV estimates may be due in part to endogeneity and measurement error in attenuating the OLS estimates. Part of the explanation also lies in the fact that IV regressions es- timate a local effect of HYV crop adoption in a sub-population, better irrigated areas, in which

HYV crops were particularly productive. Therefore, while the IV estimates provide additional evidence for a large, causal effect of HYV crop adoption on the decline of single-party domi- nance in India, the estimates may be larger than the average effect of HYV crop adoption across all districts.

Robustness

The validity of the IV estimates rests on several important assumptions (Angrist, Imbens and

Rubin, 1996), perhaps most importantly the exclusion restriction. In this case, the exclusion restriction requires that the interaction of interaction of a time indicator representing the in- troduction of HYV crops in 1967 and a group indicator of pre-existing suitability for HYV crops based on irrigation coverage in 1966 is, after conditioning on covariates, correlated with the

Congress party’s election performance only via its impact on HYV crop adoption.

18Which could bias and attenuate the estimated impact of HYV crop adoption.

37 Table 6: IV Estimates: Effects of HYV Crop Adoption on Congress Performance in Parliamentary Elections

Dependent variable: Congress Win Congress Vote %

(1) (2) (3) (4) (5) (6)

HYV 80.733∗∗ 95.825∗∗∗ 35.216 26.846∗∗∗ 27.745∗∗∗ 1.416 − − − − − (31.551) (36.024) (33.339) (9.252) (10.760) (7.700)

Opposition Fractionalization 69.753∗∗∗ 3.840∗∗ − (5.197) (1.659)

INC Presence 84.156∗∗∗ 49.533∗∗∗ (4.805) (1.489)

Hindi Belt Year FE Y Y Y Y × Direct Rule Year FE Y Y Y Y

38 × Year FE Y Y Y Y Y Y District FE Y Y Y Y Y Y First-stage 1966 Irrigation Post 0.412∗∗∗ 0.380∗∗∗ 0.375∗∗∗ 0.412∗∗∗ 0.380∗∗∗ 0.375∗∗∗ × (0.055) (0.057) (0.056) (0.055) (0.057) (0.056) F-statistic 55.2 44.5 44.8 55.2 44.5 44.8 Observations 2,953 2,953 2,953 2,953 2,953 2,953 Adjusted R2 0.25 0.32 0.44 0.32 0.37 0.70 Clusters 260 260 260 260 260 260

Notes: Unit of observation parliamentary constituency-year for election years between 1962-1985. Outcomes: Congress Win indicates whether Congress won seat and Congress Vote % indicates Congress vote share, both multiplied by 100 so that coefficients have a percentage point interpre- tation. Explanatory variables: HYV crop adoption is district-level share of agricultural land with HYV crops (0-1). Instrument for HYV crop adoption is an interaction term between a time variable indicating the introduction of HYV crops in 1967 and a cross-sectional variable indicating the share of agricultural land with access to irrigation in 1966. Opposition Fractionalization is one minus the Herfindahl index of opposition party vote shares (0-1). INC Presence is an indicator variable for the presence of a Congress party candidate. Hindi Belt is an indicator variable for location in Hindi- speaking state. Direct Rule is an indicator variable for historical exposure to direct British colonial rule. Analysis estimated by 2SLS. Standard errors

adjusted for clustering within districts.∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01 < < < The difference-in-differences setup of the IV analysis, including both district and year fixed effects, rules out any time-invariant variables or common trend threats to the exclusion restric- tion. Concerns about the exclusion restriction therefore take the form of the possibility that counter-factually absent the introduction of HYV crops in 1967 areas with greater pre-existing irrigation coverage would have experienced a larger decline in Congress election performance than areas with less pre-existing irrigation coverage. There is no obvious evidence to support such a concern. There were no major innovations during this period that disproportionately impacted better irrigated areas besides HYV crops. Irrigation infrastructure has historically been accumulated slowly in India, with much of it dating to the colonial era (Stone, 2002). This makes strategic allocation of irrigation infrastructure, in ways that might be related to the dy- namics of the Congress party’s election performance around the time HYV crops were intro- duced to India, unlikely.

Though the exclusion restriction cannot be tested directly, one way to assess its plausibility is to identify the timing of the emergence of a negative reduced-form relationship between pre-existing irrigation coverage and the Congress party’s election performance. To do this, it is possible to estimate a regression of the form,

6 X 1966 k Ydi t τt θk Ir r i g ationd It udi t , (3) = + k 1 × + =

k where Ydi t is a measure of Congress party election performance, It is a dummy variable repre- 1966 senting a half decade k (in the six half decades between 1957 and 1987), and Ir r i g ationd is the cross-sectional suitability measure of pre-existing irrigation coverage in 1966. The analysis also controls for year fixed effects, τt . Figure 5 displays the coefficients θk , representing the cor- relation between 1966 irrigation coverage and Congress election performance in state assembly elections in the six half decades between 1957 and 1987.19

19A similar test for pre-trends is not available in the case of parliamentary elections, since election data before 1962 is missing important information. However, parliamentary and state assembly elections are highly correlated.

39 Figure 5: Correlation between Pre-existing Irrigation and Congress Performance Over Time

Outcome: Congress Win 40 20 ●

0 ●

● ● ● −20

● −40 Coefficient on Pre−existing Irrigation Coefficient on Pre−existing

T−2: 1957−1961 T−1: 1962−1966 T: 1967−1971 T+1: 1972−1976 T+2: 1977−1981 T+3: 1982−1987

Time Period

Outcome: Congress Vote Share 15 10 5 0

● ● −5

● ● ● −10 ● Coefficient on Pre−existing Irrigation Coefficient on Pre−existing −15

T−2: 1957−1961 T−1: 1962−1966 T: 1967−1971 T+1: 1972−1976 T+2: 1977−1981 T+3: 1982−1987

Time Period

Notes: Unit of analysis assembly constituency-year for election years between 1957-1987. Outcome is either Congress win or Congress vote share in percentage points. Vertical axis range represents zero plus and minus an in-sample standard deviation of the outcome variable. Points represent estimates of coefficient on time period dummy variables interacted with the cross sectional measure of suitability for HYV crop adoption based on pre-existing irrigation coverage in 1966. HYV crops first widely adopted in 1967. Regression also controls for year fixed effects. Analysis estimated by OLS. Standard errors adjusted for clustering within districts. Dashed bands represent 95% confidence interval.

Additionally, the analysis of pre-treatment determinants of HYV crop adoption show that Congress seat share in the 1962 parliamentary elections at the district level does40 not predict HYV crop adoption. The estimates suggest that pre-existing trends unrelated to HYV crop adoption do not drive the results. The point estimates of the correlation between 1966 irrigation coverage and Congress election performance are statistically indistinguishable from zero in the two half-decades be- fore the green revolution. By contrast, a large, negative and statistically significant reduced form relationship between 1966 irrigation coverage and Congress election performance emerges pre- cisely following the introduction of HYV crops to India in 1967.

As a final robustness test, it is possible to use the presence of a naturally occurring thick aquifer for the cross-sectional suitability indicator for HYV crops in instrumental variable re- gressions. Areas with abundant groundwater were especially suitable for tube-well irrigation, upon which HYV crop cultivation was widely based (Rud, 2012). A disadvantage of this indica- tor is that just 8 percent of districts in sample possessed a thick aquifer, limiting the statistical power and external validity of estimates based on this instrument. An advantage, however, is that aquifers are naturally occurring and therefore more clearly exogenous than pre-existing ir- rigation coverage. Tables A5 and A6 in the appendix reports IV regression estimates using an aquifer-based instrument for HYV crop adoption. The coefficient estimates are broadly consis- tent with the main results already reported, though the standard errors are somewhat larger.

7 Testing Competing Theories

Agrarian Opposition Parties

How did economic growth contribute to democratization in the case of India’s green revolution?

To reiterate, there are two plausible theories. The income effect channel, favored by modern- ization theory, suggests that rising rural incomes reduced the effectiveness of clientelism. The mobilization effect channel, by contrast, suggests that HYV crops provided economic incentives for political mobilization by opposition parties supported by agricultural producers.

To distinguish between the competing theories, this paper first dis-aggregates the political

41 effects of HYV crop adoption in order to investigate which type of opposition parties, agrarian versus non-agrarian, benefited in elections from HYV crop adoption.20 A mobilization theory predicts that specifically opposition parties with support from agricultural producers ought to have benefited in elections from HYV crop adoption. By contrast, if rising incomes reduced the effectiveness of Congress party clientelism, there is little theoretical rationale for why one type of opposition party, among the many forms of opposition to Congress available, should have benefited in elections from the green revolution. Table 7 reports the OLS and IV estimates for both state assembly and parliamentary elections.

Strikingly, both the OLS and IV results show that specifically opposition parties oriented around increased subsidies for the agricultural sector and supported by agricultural produc- ers, such as the BKD, Lok Dal and Janata parties, benefited in terms of vote share in state and parliamentary elections. By contrast, other types of opposition parties did not. The estimates reported in columns (2) and (6) are similar in magnitude and opposite in sign relative to esti- mates of the impact of HYV crops on the Congress party’s election performance, suggesting that nearly all of the green revolution’s negative impact on Congress party vote share was due to a rise in the vote share of agrarian opposition parties.

In terms of seat share, HYV crops only benefited agrarian opposition parties in state assem- bly elections and not other types of opposition parties, while in parliamentary elections other types of opposition parties did benefit slightly, but to a lesser extent than did agrarian oppo- sition parties. The spillover benefits in terms of seat share to non-agrarian opposition parties, despite no increase in their vote share, is attributable to a first-past-the-post electoral system which permitted vote share gains by agrarian opposition opposition parties at the Congress party’s expense to benefit other types of opposition parties as well.

20I take the top 20 opposition parties in sample by total seats won and code each as agrarian or non-agrarian, as discussed in greater detail the data section (see Table A2 for a list of coded parties). Parties are coded on the basis of historical research as agrarian if they recieved significant support from and/or advocated programs oriented around the interests of agricultural producers.

42 Table 7: OLS and IV Estimates: Effects of HYV Crop Adoption by Opposition Party Type

Dependent variable: State Assembly Elections Parliamentary Elections Agrarian Opposition Parties Non-Agrarian Opposition Agrarian Opposition Parties Non-Agrarian Opposition

Win Vote % Win Vote % Win Vote % Win Vote % (1) (2) (3) (4) (5) (6) (7) (8) Panel A: OLS HYV 30.203∗∗∗ 19.355∗∗∗ 0.729 5.509∗ 24.904∗∗∗ 11.699∗∗ 16.173∗ 2.842 − (5.801) (3.425) (5.048) (3.306) (9.182) (4.963) (8.969) (6.013)

43 Panel B: 2SLS HYV 73.479∗∗∗ 53.486∗∗∗ 2.844 24.963∗∗∗ 50.213∗∗∗ 31.596∗∗∗ 32.948 12.803 − − (12.361) (8.288) (14.306) (7.834) (12.993) (7.711) (25.761) (14.803)

Year FE Y Y Y Y Y Y Y Y District FE Y Y Y Y Y Y Y Y

Observations 21,907 21,907 21,907 21,907 2,953 2,953 2,953 2,953 Clusters 279 270 270 270 260 260 260 260

Notes: Panel A is OLS specification. Panel B instruments for HYV crop adoption with interaction term between a time dummy variable indicating the introduction of HYV crops in 1967 and measure of pre-existing irrigation coverage in 1966. See text and appendix for coding of the top 20 opposition parties as agrarian or non-agrarian. Win is binary indicator of winning the seat and Vote % is vote share, both multiplied by 100 so that coefficients have a percentage point interpretation. HYV crop adoption is district-level share of agricultural land planted with HYV crops (0-1).

Standard errors adjusted for clustering within districts. ∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01. < < < Economic Impacts of HYV Crops

Next, this paper estimates the economic effects of HYV crops using detailed district-level agri- cultural data. The central predication of an income effect theory is that HYV crop adoption increased rural incomes, while the central prediction of an incentive effect theory is that HYV crop adoption increased the usage of inputs and decreased the market price of crops, thereby increasing the value to agricultural producers of state-controlled input subsidies and crop price supports. The results of these analyses are reported in Table 8.

The economic impacts of the green revolution are consistent with an income effect chan- nel. Columns (1) and (2) show that HYV crop adoption increased agricultural productivity and wages, increasing rural incomes. This is in contrast to some early historical field reports that

HYV crops might have reduced real wages (Frankel, 1971). The OLS estimates reported in panel

A imply, based average post-1967 HYV crop adoption (0.24), that the green revolution was re- sponsible for a 103.5 Rs/Ha or 0.28 within-district standard deviation increase in agricultural yield and a 0.2 Rs/Day or 0.19 within-district standard deviation increase in agricultural wages.

These are moderate but not extremely large estimates. Consistent with qualitative information that HYV crops were most productive in better-irrigated areas – the sub-population to which the IV estimates apply – the IV estimates reported in panel B are larger in magnitude.

The economic impacts of the green revolution are also consistent with an incentive effect channel. Columns (3), (4) and (5) provide evidence that the green revolution caused a massive decrease in crop prices and increase in the usage of inputs. The OLS estimates reported in panel A imply, based on average post-1967 HYV crop adoption (0.24), that the green revolution was responsible for a 7.7 Rs/quintal or 0.32 within-district standard deviation decrease in crop prices, a 15.8 Kg/Ha or 0.92 within-district standard deviation increase in fertilizer usage, and a

1.4 tractors/1000 Ha or 0.47 within-district standard deviation increase in tractor usage. These estimates provide evidence that the green revolution increased the usage of inputs and placed downward pressure on crop market prices, providing incentives for agricultural producers to

44 Table 8: OLS and IV Estimates: Economic Effects of HYV Crop Adoption

Dependent variable: Yield Wage Crop Price Fertilizer Tractors (1) (2) (3) (4) (5) Panel A: OLS

HYV 431.145∗∗∗ 0.886∗∗∗ 31.931∗∗∗ 65.732∗∗∗ 5.659∗∗∗ − (109.941) (0.288) (5.210) (5.444) (1.566)

Panel B: 2SLS

HYV 1,288.654∗∗∗ 3.666∗∗∗ 26.655∗∗∗ 115.612∗∗∗ 11.738∗∗∗ − (200.820) (0.655) (9.897) (6.785) (2.108) 45

Year FE Y Y Y Y Y District FE Y Y Y Y Y

Observations 8,364 8,370 8,364 8,370 8,370 Clusters 270 270 270 270 270

Notes: Unit of observation district-year for years 1957-1987. Panel A is OLS specification. Panel B instruments for HYV crop adoption with inter- action term between a time dummy variable indicating the introduction of HYV crops in 1967 and measure of pre-existing irrigation coverage in 1966. Outcomes: Yield is measured in terms of rs./ha for the top 6 major crops. Wage is the daily wage of a male agricultural laborer in rs./day. Crop Price Index is a revenue weighted average of the market price in rs./quintal of the top 6 crops. Fertilizer usage is measured in terms of kg/ha. Tractor usage is measured in terms of tractors/1000 ha. All monetary amounts are deflated by state-specific consumer price indices to obtain real prices, base year 1973. Explanatory variable: HYV crop adoption is district-level share of agricultural land planted with HYV crops (0-1). Standard

errors adjusted for clustering within districts. ∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01. < < < mobilize politically in order to obtain higher levels of agricultural subsidies in response to these changes in the agricultural sector. The IV estimates reported in panel B are, again, consistent in sign and larger in magnitude.

Rainfall Shocks and Incomes

How to separate the effects of rising incomes from increased incentives for political mobiliza- tion? A central challenge is that both resulted from HYV crop adoption.21 To isolate the effects of of increased incomes alone on Congress party dominance, this paper utilizes rainfall shocks as a placebo test, a source of exogenous variation in rural incomes unconnected to other con- founding variables (Miguel, Satyanath and Sergenti, 2004), in particular the structure of the agricultural sector. Table 9 reports estimates from OLS regressions of Congress and opposition party performance on rainfall.

The results provide evidence that increased incomes did not contribute to the decline of the

Congress dominance. The inclusion of district fixed effects means that this analysis estimates the effects of within-district deviations over time, i.e. rainfall shocks. Positive rainfall shocks improved the Congress party’s election performance, consistent with argument highlighting the role of rising incomes in enhancing the performance-based legitimacy of the incumbent regime

(Cole, Healy and Werker, 2012; Huntington, 1991; Reuter and Gandhi, 2011), and in sharp con- trast to the argument that rising incomes per se contribute to democratization. A one within- district standard deviation (0.29) improvement in rainfall increased the percentage probability of electing a Congress party state legislator by 3.3 percentage points and Congress MP by 5.5 percentage points. By contrast, positive rainfall shocks decreased the probability of electing an agrarian opposition party legislator and had no statistically significant impact on electing a non-agrarian opposition party legislator, in both state assembly and parliamentary elections.

21Identifying the relative salience of competing potential mechanisms when both are a result of treatment is a notorious methodological problem. See Green, Ha, and Bullock (2010).

46 Table 9: OLS Estimates: Effects of Rainfall Shocks on Congress and Opposition Party Seat Shares

Dependent variable: State Assembly Win Parliamentary Win

Congress Agrarian Opposition Non-Agrarian Congress Agrarian Opposition Non-Agrarian (1) (2) (3) (4) (5) (6)

Rainfall (ln mm) 11.516∗∗∗ 15.351∗∗∗ 2.796 21.596∗∗∗ 16.685∗∗∗ 0.707 − − − (2.506) (2.208) (1.845) (4.456) (3.045) (3.588) 47 Year FE Y Y Y Y Y Y District FE Y Y Y Y Y Y Observations 21,907 21,907 21,907 2,953 2,953 2,953 Clusters 270 270 270 260 260 260

Notes: Unit of observation constituency-year for state and parliamentary elections, respectively. Outcomes: Vote % indicates vote share and Win indicates whether or not party won the seat, both multiplied by 100 so that coefficients have a percentage point interpretation. See text and table A2 for coding of opposition parties as agrarian or non-agrarian. Explanatory variable: Rainfall is measured in ln(annual mm). The inclusion of district fixed effects means that the effects of rainfall are identified from within-district deviations from long-run mean i.e. rainfall shocks. Analysis

estimated by OLS. Standard errors adjusted for clustering within districts. ∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01. < < < Table 10 examines the economic impacts of rainfall shocks. These estimates rationalize the political effects of rainfall shocks. Rainfall improvements increased rural incomes but had little impact on the underlying structure of the agricultural sector. Positive rainfall shocks increased agricultural productivity and wages. By contrast, positive rainfall shocks did not have a statisti- cally or economically significant impact on the intensity of input usage as measured by fertilizer and tractor usage or on crop prices. These estimates therefore confirm that increased incomes on their own, unconnected to deeper economic incentives for political mobilization by agricul- tural producers, did not contribute to the decline of the Congress party but instead enhanced its performance legitimacy.

Erosion of Traditional Social Hierarchies

Arguably, rising incomes generated by transitory rainfall shocks do not simulate the sociolog- ical changes generated by the lasting rise in rural incomes brought about by the green revolu- tion. As discussed, some scholarship suggests that rising incomes created a class of upwardly mobile lower-caste farmers, disrupting patron-client ties of dependence upon the local upper caste elites who served as brokers for the dominant Congress party (Jaffrelot, 2003; Rudolph and Rudolph, 1987). If HYV crops contributed to democratization by eroding traditional social hierarchies and enabling the entry of new social groups into politics, as modernization the- ory suggests (Deutsch, 1961), this ought to be observable in the sociological profile of elected politicians. To test this hypothesis, I estimate the impact of HYV crop adoption on the caste, education, and reported occupation of elected MPs. The results are reported in Table 11.

The income effect channel and modernization theory again find little support in the data, with HYV crop adoption causing no increase in the probability of electing a lower caste, less educated, or agriculturist MP.The only marginally statistically significant estimate (at the 10% level), the OLS estimate in column (2), implies that HYV crop adoption caused a decrease in the likelihood of electing an other backwards class (OBC)/intermediate caste MP; however, the sign

48 Table 10: OLS Estimates: Economic Effects of Rainfall Shocks

Dependent variable: Yield Wage Crop Price Fertilizer Tractors (1) (2) (3) (4) (5)

Rainfall 96.528∗∗∗ 0.221∗∗∗ 1.896 0.771 0.227 − (24.678) (0.062) (1.258) (0.795) (0.371)

Year FE Y Y Y Y Y

49 District FE Y Y Y Y Y Observations 8,332 8,337 8,332 8,337 8,337 Adjusted R2 0.81 0.77 0.58 0.66 0.50 Clusters 269 269 269 269 269

Notes: Unit of observation district-year for years 1957-1987. Outcomes: Yield is measured in terms of rs./ha for the top 6 major crops. Wage is the daily wage of a male agricultural laborer in rs./day. Crop Price Index is a revenue weighted average of the price in rs./quintal of the top 6 crops. Fertilizer usage is measured in terms of kg/ha. Tractor usage is measured in terms of tractors/1000 ha. All monetary amounts are deflated by state- specific consumer price indices to obtain real prices, base year 1973. Explanatory variable: Rainfall is measured in ln(annual mm). The inclusion of district fixed effects means that the effects of rainfall are identified from within-district deviations from long-run mean i.e. rainfall shocks. Analysis

estimated by OLS. Standard errors adjusted for clustering within districts. ∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01. < < < Table 11: OLS and IV Estimates: Impact of HYV Crops on MP Caste, Education, and Occupation

Dependent variable: Caste Socioeconomic

Lower caste OBC caste Upper caste University Agriculturist Professional

(1) (2) (3) (4) (5) (6) Panel A: OLS

HYV 0.124 0.257∗ 0.133 0.116 0.062 0.135 − − − (0.135) (0.134) (0.150) (0.115) (0.118) (0.119)

Panel B: 2SLS

50 HYV 0.400 0.548 0.148 0.224 0.067 0.073 − − − − − (0.801) (0.569) (0.955) (0.270) (0.311) (0.349)

Year FE Y Y Y Y Y Y District FE Y Y Y Y Y Y Observations 1,330 1,330 1,330 2,384 2,2384 2,2384 Clusters 133 133 133 245 245 245

Notes: Unit of observation parliamentary constituency-year for years 1962-1985. Outcomes: Indicator for upper, intermediate or other backward class caste, and for lower caste and/or religious/ethnic minority. Caste data from Jaffrelot (2003) dataset for Hindi-speaking states only. University indicates whether MP self-reports a university degree. Agriculturist indicates whether MP self-reports as an "agriculturist". Professional indicates whether MP self-reports as a doctor, lawyer, journalist, writer, or teacher. Socieconomic variables coded through natural language processing of digitized ‘Who’s Who’ MP biographies. Explanatory variable: HYV crop adoption is district-level share of agricultural land planted with HYV crops

(0-1). Analysis estimated by OLS. Standard errors adjusted for clustering within districts. ∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01. < < < on the coefficient switches in the IV specification, suggesting that this result may be spurious and not particularly robust. Overall, the evidence suggests that the erosion of traditional so- cial hierarchies in political representation was not a salient channel for the green revolution’s democratizing effects.

It is informative that HYV crop adoption did not cause an increase in the likelihood that an

MP self-describes as an “agriculturist", despite causing large improvements in the election per- formance of opposition parties representing the interests of agricultural producers. This may be due in part to the fact that many MPs simply adopted the label of “agriculturist" as a way of signifying their grassroots credentials, with the term amounting essentially to “cheap talk".

More importantly, the results provide evidence that the green revolution did not cause a sig- nificant shift in the sociological composition of elected politicians at the parliamentary level.

The evidence instead suggest that the green revolution created a structural opportunity for the rise of opposition parties running on platforms geared toward agricultural producers, an inter- est group neglected by the dominant Congress party and with growing economic incentives to seek political influence through opposition parties seeking their votes.

8 Conclusion

This paper has shown, exploiting sub-national variation in the local adoption of HYV crops, that agricultural economic growth due to the green revolution played a pivotal role in the historical decline of single-party dominance in India. The democratizing effect of the green revolution was due not to rising rural incomes nor to the erosion of traditional social hierarchies in political representation, but to the rise of opposition parties supported by agricultural producers and motivated by the growing value of state-controlled agricultural subsidies.

The findings hold lessons for debates about the impact of economic growth on dominant parties and democratization more generally (Acemoglu et al., 2008; Boix and Stokes, 2003; Greene,

51 2010; Limongi and Przeworski, 1997). The theoretical lesson with comparative relevance is that economic growth can contribute to democratization, but in a way that the existing literature has tended to overlook. Contrary to modernization theory (Lipset, 1959), rising incomes on their own reinforce and legitimize the incumbent regime and are unlikely to lead to democra- tization. However, economic growth that increases the value of political control and provides economic incentives for regime outsiders to mobilize politically in order to capture increasingly valuable state resources can contribute to democratization.

This pathway to democratization is clearly relevant beyond the Indian case. Similar dynam- ics have been highlighted in other historical episodes of democratization. For example, Ace- moglu and Robinson (2005) and Jha (2015) highlight the the growing value of Atlantic trade as a motive for mercantile interests to lobby for executive constraints and early democratic reforms in England.22 Similarly, economic liberalization and urbanization have been argued to have contributed to the demise of PRI dominance in Mexico (Greene, 2007; Magaloni, 2006). Rather than rising incomes per se, part of the mechanism at work could have been the rise of urban and trade oriented business interests, which have served as major source of support for the ma- jor opposition party, the PAN. The mechanism highlighted in this paper is also related to Ansell and Samuels (2014), who argue that democratization occurs when rising economic interests demand political influence because they fear expropriation. The findings of this paper suggest that the prospect of economic gain from capturing the state is a flip side of the same coin and an equally powerful motive for democratization.

An important scope condition may be the capacity of the incumbent regime to politicize productivity gains. In the case of India, this paper has provided evidence that the dominant

Congress party failed to strategically manipulate the diffusion of HYV crops. In the case of other technologies or institutional contexts, such manipulation is more likely, and may mitigate the democratizing effects of such innovations. Studying how different types of productivity shocks

22North and Weingast (1989) attribute the Glorious Revolution to creditors as an interest group, by contrast.

52 impact dominant parties and democratization more generally across different settings is a valu- able topic for future research. The findings of this paper more broadly highlight the importance of studying the heterogeneous effects of different types of economic growth, as opposed to ris- ing incomes per se.

References

Acemoglu, Daron, Simon Johnson, James A Robinson and Pierre Yared. 2008. “Income and Democracy.” American Economic Review 98(3):808–842.

Acemoglu, Daron, Simon Johnson and James Robinson. 2005. “The rise of Europe: Atlantic trade, institutional change, and economic growth.” American economic review pp. 546–579.

Angrist, Joshua, Guido Imbens and Donald Rubin. 1996. “Identification of Causal Effects using Instrumental Variables.” Journal of the American Statistical Association 91(434):444–455.

Ansell, Ben W and David J Samuels. 2014. Inequality and Democratization: An Elite-competition Approach. Cambridge University Press.

Bai, Jushan and Pierre Perron. 2003. “Computation and Analysis of Multiple Structural Change Models.” Journal of Applied Econometrics 18(1):1–22.

Bailey, Frederick G. 1959. “The Ganatantra Parishad.” Economic Weekly 24:1469–1476.

Bailey, Frederick George. 1970. Politics and Social Change: Orissa in 1959. University of Califor- nia Press.

Bates, Robert. 1983. “Modernization, Ethnic Competition, and the Rationality of Politics in Contemporary Africa.” State versus Ethnic claims: African Policy Dilemmas 152:171.

Boix, Carles and Susan Carol Stokes. 2003. “Endogenous Democratization.” World Politics 55(4):517–549.

Brass, Paul R. 1980. “The Politicization of the Peasantry in a North Indian State: I.” The Journal of Peasant Studies 7(4):395–426.

Brass, Paul R. 1981. “Congress, the Lok Dal, and the Middle-peasant Castes: An Analysis of the 1977 and 1980 Parliamentary Elections in Uttar Pradesh.” Pacific Affairs pp. 5–41.

Brownlee, Jason. 2007. Authoritarianism in an Age of Democratization. Cambridge University Press.

Chhibber, Pradeep and Ken Kollman. 1998. “Party Aggregation and the Number of Parties in India and the United States.” American Political Science Review pp. 329–342.

53 Chhibber, Pradeep and Ken Kollman. 2009. The formation of national party systems: Federal- ism and party competition in Canada, Great Britain, India, and the United States. Princeton University Press.

Cleaver, Harry M. 1972. “The Contradictions of the Green Revolution.” The American Economic Review 62(1/2):177–186.

Cole, Shawn, Andrew Healy and Eric Werker. 2012. “Do Voters Demand Responsive Govern- ments? Evidence from Indian Disaster Relief.” Journal of Development Economics 97(2):167– 181.

Dasgupta, Aditya. 2016. How Dominant Parties Survive, Decline, and Shape Democracy: Lessons from India PhD thesis Harvard University.

Deutsch, Karl W. 1961. “Social mobilization and political development.” American Political Sci- ence Review 55(03):493–514.

Dixit, Avinash and John Londregan. 1996. “The Determinants of Success of Special Interests in Redistributive Politics.” Journal of Politics 58(04):1132–1155.

Duncan, Ian. 1988. “Party Politics and the North Indian Peasantry: The Rise of the Bharatiya Kranti Dal in Uttar Pradesh.” The Journal of Peasant Studies 16(1):40–76.

Erdman, Howard L. 2007. The Swatantra Party and Indian Conservatism. Vol. 5 Cambridge University Press.

Evenson, Robert E and Douglas Gollin. 2003. “Assessing the impact of the Green Revolution, 1960 to 2000.” Science 300(5620):758–762.

Evenson, Robert Eugene and Mark W Rosegrant. 1998. Agricultural Research and Productivity Growth in India. Vol. 109 International Food Policy Research Institute.

Fickett, Lewis P.1973. “The Praja Socialist Party of India–1952-1972: A Final Assessment.” Asian Survey pp. 826–832.

Foster, Andrew D and Mark R Rosenzweig. 1996. “Technical Change and Human-Capital Re- turns and Investments: Evidence from the Green Revolution.” The American Economic Re- view pp. 931–953.

Frankel, Francine R. 1971. India’s Green Revolution. Economic Gains and Political Costs. Prince- ton, NJ: Princeton University Press.

Gaiha, Raghav and Vani S Kulkarni. 2005. “Foodgrain Surpluses, Yields and Prices in India.”.

Green, Donald P, Shang E Ha and John G Bullock. 2010. “Enough Already about “Black Box" Experiments: Studying Mediation is More Difficult than most Scholars Suppose.” The Annals of the American Academy of Political and Social Science 628(1):200–208.

54 Greene, Kenneth F.2007. Why Dominant Parties Lose: Mexico’s Democratization in Comparative Perspective. Cambridge University Press.

Greene, Kenneth F. 2010. “The Political Economy of Authoritarian Single-party Dominance.” Comparative Political Studies 43(7):807–834.

Guha, Ramachandra. 2007. India after Gandhi: The History of the World’s Largest Democracy. Pan.

Hardgrave, Robert L. 1970. “The Congress in India–Crisis and Split.” Asian Survey pp. 256–262.

Hasan, Zoya. 1989a. Dominance and Mobilisation: Rural Politics in Western Uttar Pradesh, 1930-1980. Sage New Delhi.

Hasan, Zoya. 1989b. Power and Mobilization: Patterns of Resilience and Change in Uttar Pradesh Politics. Vol. 1 Oxford University Press Delhi pp. 133–203.

Healy, Andrew and Neil Malhotra. 2013. “Retrospective Voting Reconsidered.” Annual Review of Political Science 16:285–306.

Huntington, Samuel P.1991. “Democracy’s Third Wave.” Journal of Democracy 2(2):12–34.

Iyer, Lakshmi. 2010. “Direct versus Indirect Colonial Rule in India: Long-term Consequences.” The Review of Economics and Statistics 92(4):693–713.

Jaffrelot, Christophe. 1999. The Hindu Nationalist Movement and Indian Politics, 1925 to the 1990s: Strategies of Identity-building, Implantation and Mobilisation. Penguin Books India.

Jaffrelot, Christophe. 2003. India’s silent revolution: the rise of the lower castes in North India. Orient Blackswan.

Jayachandran, Seema. 2006. “Selling Labor Low: Wage Responses to Productivity Shocks in Developing Countries.” Journal of Political Economy 114(3):538–575.

Jensenius, Francesca. 2013. Power, Performance and Bias: Evaluating the Electoral Quotas for Scheduled Castes in India. Dissertation (UC-Berkeley).

Jha, Saumitra. 2015. “Financial Asset Holdings and Political Attitudes: Evidence from Revolu- tionary England.” Quarterly Journal of Economics 103.

Khush, Gurdev S. 2001. “Green Revolution: the Way Forward.” Nature Reviews Genetics 2(10):815–822.

Kitschelt, Herbert and Steven I Wilkinson. 2007. Patrons, Clients and Policies: Patterns of Demo- cratic Accountability and Political Competition. Cambridge University Press.

Kohli, Atul. 1983. “Parliamentary Communism and Agrarian Reform: The Evidence from India’s Bengal.” Asian Survey pp. 783–809.

55 Kohli, Atul. 1988. “The NTR phenomenon in Andhra Pradesh: Political change in a South Indian state.” Asian Survey pp. 991–1017.

Kohli, Atul. 1990. Democracy and discontent: India’s Growing Crisis of Governability. Cambridge University Press.

Kothari, Rajni. 1964. “The Congress System in India.” Asian Survey pp. 1161–1173.

Krueger, Anne O. 1974. “The Political Economy of the Rent-seeking Society.” The American Economic Review 64(3):291–303.

Kumar, Ashutosh. 2004. “Electoral Politics in Punjab: Study of Akali Dal.” Economic and Politi- cal Weekly pp. 1515–1520.

Ladejinsky, Wolf. 1970. “Ironies of India’s Green Revolution.” Foreign Affairs pp. 758–768.

Levitsky, Steven and Lucan A Way. 2010. Competitive authoritarianism: Hybrid regimes after the cold war. Cambridge University Press.

Limongi, Fernando and Adam Przeworski. 1997. “Modernization: Theories and Facts.” World Politics 49(2):155–183.

Lipset, Seymour Martin. 1959. “Some Social Requisites of Democracy: Economic Development and Political Legitimacy.” American Political Science Review 53(01):69–105.

Lipton, Michael. 1977. Why Poor People Stay Poor: a Study of Urban Bias in World Development. Temple Smith London.

Magaloni, Beatriz. 2006. Voting for Autocracy: Hegemonic Party Survival and its Demise in Mex- ico. Cambridge University Press New York.

Magaloni, Beatriz and Ruth Kricheli. 2010. “Political Order and One-Party Rule.” Annual Review of Political Science 13:123–143.

Manor, James. 1980. “Pragmatic Progressives in Regional Politics: The Case of Devaraj Urs.” Economic and Political Weekly pp. 201–213.

Miguel, Edward, Shanker Satyanath and Ernest Sergenti. 2004. “Economic Shocks and Civil Conflict: An Instrumental Variables Approach.” Journal of Political Economy 112(4):725–753.

Mohan, Rakesh and Robert E Evenson. 1975. “The Intensive Agricultural Districts Programme in India: A New Evaluation.” The Journal of Development Studies 11(3):135–154.

North, Douglass C and Barry R Weingast. 1989. “Constitutions and Commitment: the Evolu- tion of Institutions Governing Public choice in Seventeenth-century England.” The Journal of Economic History 49(04):803–832.

56 Özler, Berk, Gaurav Datt and Martin Ravallion. 1996. “A Database on Poverty and Growth in India.” Policy Research Department, World Bank .

Pempel, T John. 1990. Uncommon democracies: The one-party dominant regimes. Cornell Univ Pr.

Rawlins, SL. 1977. “High-Frequency irrigation and Green Revolution Food Production.” Annals of the New York Academy of Sciences 300(1):121–128.

Reuter, Ora John and Jennifer Gandhi. 2011. “Economic Performance and Elite Defection from Hegemonic Parties.” British Journal of Political Science 41(1):83–110.

Robinson, James A and Thierry Verdier. 2013. “The Political Economy of Clientelism*.” The Scandinavian Journal of Economics 115(2):260–291.

Rosenbaum, Paul R. 1984. “The Consquences of Adjustment for a Concomitant Variable that has been Affected by the Treatment.” Journal of the Royal Statistical Society. Series A pp. 656– 666.

Rud, Juan Pablo. 2012. “Electricity Provision and Industrial Development: Evidence from In- dia.” Journal of Development Economics 97(2):352–367.

Rudolph, Lloyd I and Susanne Hoeber Rudolph. 1984. Determinants and Varieties of Agrarian Mobilization. Oxford University Press Delhi.

Rudolph, Lloyd and Susan Hoeber Rudolph. 1987. In Pursuit of Lakshmi: The Political Economy of the Indian State. University of Chicago Press.

Rudolph, Susanne Hoeber and Lloyd I Rudolph. 1980. “The centrist future of Indian politics.” Asian Survey pp. 575–594.

Sanghi, Apurva, Kavi Kumar, Robert Evenson and James McKinsey. 1998. “India Agriculture and Climate Dataset.”.

Schedler, Andreas. 2006. “Electoral authoritarianism: The dynamics of unfree competition.”.

Scott, James C. 1969. “Corruption, Machine Politics, and Political Change.” The American Polit- ical Science Review 63(4):1142–1158.

Sinha, Lalan Prasad. 1965. The Left-Wing in India, 1919-47. Vol. 1 New Publishers.

Skaperdas, Stergios. 1992. “Cooperation, Conflict, and Power in the Absence of Property Rights.” American Economic Review 82(4):720–739.

Srinivas, Mysore Narasimhachar. 1957. “Caste in Modern India.” The Journal of Asian Studies pp. 529–548.

57 Stone, Ian. 2002. Canal Irrigation in British India: Perspectives on Technological Change in a Peasant Economy. Vol. 29 Cambridge University Press.

Varshney, Ashutosh. 1998. Democracy, Development, and the Countryside: Urban-Rural Strug- gles in India. Cambridge University Press.

Walker, Thomas S and James G Ryan. 1990. Village and Household Economics in India’s Semi- arid Tropics. Johns Hopkins University Press.

Weiner, Myron. 1967. Party Building in a New Nation: The Indian National Congress. University of Chicago Press Chicago.

Weiner, Myron. 1971. “The 1971 Elections and the Indian Party System.” Asian Survey pp. 1153– 1166.

Weiner, Myron. 1982. “Congress Restored: Continuities and Discontinuities in Indian Politics.” Asian Survey 22(4):339–355.

Wyatt, Andrew. 2009. Party System Change in South India: Political Entrepreneurs, Patterns and Processes. Routledge.

Yadav, Yogendra. 1996. “Reconfiguration in Indian politics: State assembly elections, 1993-95.” Economic and Political Weekly pp. 95–104.

Ziegfeld, Adam and Maya Tudor. 2013. “Bringing Down Dominant Parties: Evidence from In- dian States.” Mimeo.

58 Appendix

A Simple Model of Dominant Party Survival and Decline

A democracy contains three voters, two of them citizens M and one of them an elite E. Elections are held according to majority rule. Two parties compete for votes. The dominant party, D, is controlled by elites and benefits only from elite expenditures, while the challenger party, C, rep- resents the citizens and benefits only from expenditures on citizens. The incumbent dominant party possesses control of state resources worth 1, which it can allocate either in investment in

M M citizen favoring policies G , direct transfers to each citizen Ti , or direct transfers to the elite T E . The dominant party can credibly promise transfers to individual voters by virtue of patron- age networks. The challenger party, however, cannot credibly promise transfers since it is not currently in power and does not possess patronage networks, but it is rationally expected to use state resources to implement citizen favoring policies if elected to office.

In this model, the game takes place as follows. Each party chooses its platform, subject to the budget constraint:

G M T M T M T E 1 (4) + 1 + 2 + =

Then, each voter strategically casts his vote in order maximize his own utility. The utility func- tions of each of the three voters are below:

πE T E (5) = πM αG M T M (6) 1 = + 1 πM αG M T M (7) 2 = + 2

Note that both citizen 1 and citizen 2 derive linear utility from the same unit expenditure G M on citizen favoring policies, which may be conceptualized as spending on the agricultural sector,

59 intended the capture efficiencies of scale on spending on policies with diffuse sectoral benefits relative to private transfers, the extent of which is modelled by α 1 . Finally, the party winning > 2 the most votes is elected to office and its distributive platform is implemented.

In this simplified game, the challenger’splatform is fixed to benefit citizens through spending all 1 of the government’s resources on citizen favoring policies, since it cannot credibly offer targeted transfers. The dominant party’s optimization problem is to offer the cheapest targeted transfer to one citizen that exceeds the benefits that the citizen would receive from the election of the challenger party (α), while preserving the rest for elite spending.

If α 0.6, in equilibrium the incumbent party spends just more than 0.6 on buying one cit- = izen’s vote, just less than 0.4 on elite expenditures, and nothing on citizen favoring policies, winning the election with the vote of the elite and at least one citizen (the citizen who does not receive a targeted transfer is non-pivotal). Note that the total payoff the population is 1 (0.6 to citizen 1, 0 to citizen 2, and 0.4 to the elites), even though the platform promised by the chal- lenger party would have yielded a total payoff of 1.2 (0.6 to citizen 1, 0.6 to citizen 2, and 0 to the elite), benefiting in particular the citizenry as a whole relative to the elite. This shows how an an elite-controlled dominant party can dominate elections without catering to mass policy preferences by utilizing patronage networks.

However, single-party dominance can decline when changes in the economy increase the value to regime outsiders of favorable policy provision, making it too expensive for the domi- nant party to continue to purchase the votes of a strategic subset of the citizenry. In this simple model provided earlier, if the term α, representing the value of public policies with diffuse ben- efits, exceeds a certain threshold (in the specific model above, 1), then the incumbent party’s resources are no longer sufficient to dissuade citizens from from voting for the challenger party.

HYV crops can be thought to have shifted the α parameter, and contributed to support for agrar- ian opposition parties instead of the elite-dominated Congress party .

60 Table A1: States and Election Years in Analysis State Election Year Panel A: State Assembly Elections Andhra Pradesh 1957, 1962, 1967, 1972, 1978, 1983, 1985 Bihar 1957, 1962, 1967, 1969, 1972, 1977, 1980, 1985 Gujarat 1957, 1962, 1967, 1972, 1975, 1980, 1985 Haryana 1957, 1962, 1967, 1968, 1972, 1977, 1982, 1987 Karnataka 1957, 1962, 1967, 1972, 1978, 1983, 1985 Madhya Pradesh 1957, 1962, 1967, 1972, 1977, 1980, 1985 Maharashtra 1957, 1962, 1967, 1972, 1978, 1980, 1985 Orissa 1957, 1961, 1967, 1971, 1974, 1977, 1980, 1985 Punjab 1957, 1962, 1967, 1969, 1972, 1977, 1980, 1985 Rajasthan 1957, 1962, 1967, 1972, 1977, 1980, 1985 Tamil Nadu 1957, 1962, 1967, 1971, 1977, 1980, 1984 Uttar Pradesh 1957, 1962, 1967, 1969, 1974, 1977, 1980, 1985 West Bengal 1957, 1962, 1967, 1969, 1971, 1972, 1977, 1982, 1987 Panel B: Parliamentary Elections All States Above 1962, 1967, 1971, 1977, 1980, 1984 (Punjab in 1985)

Notes: Karnataka and Tamil Nadu are the states of Mysore and Madras before 1967. The districts that comprise Haryana were a part of Punjab before the 1967 elections. In the 1957 elections, the districts that comprise the states of Maharashtra and Gujarat were part of Bombay state. Districts in all major states except Assam and Kerala are included in the analysis, covering 85% of India’s land area.

61 Table A2: Coding of Top 20 Opposition Parties

AC/PC Agrarian Notes Sources Party Seats TPS 0/12 No Telangana Praja Samiti: Telangana separatist party. Weiner, 1971 RSP 77/12 No Revolutionary Socialist Party: socialist party. Sinha, 1965 NCU 82/7 No Congress (Urs): split from Congress in 1978. Manor, 1980 GP 87/4 No Ganatrantra Parishad: conservative party supported by Bailey, 1959 landlords. Joined SWP in 1962. PWP 109/8 No Peasants and Workers Party of India: left-wing party formed Srinivas, 1957 in 1947. FBL 127/12 No Forward Block: left-wing nationalist party, formed in 1939 by Sinha, 1965 Subhas Chandra Bose. NCO 166/21 No Congress (Organisation): conservative wing which split from Hardgrave, 1970 the Indira-Gandhi led faction of the Congress party in 1969. TDP 203/31 Yes Telugu Desam Party: regional party founded in 1982. Lead- Kohli, 1988 ership and support drawn considerably from the Kammas, a prosperous farming caste. BKD 212/1 Yes Bharatiya Kranti Dal: pro-rural, anti-urban party founded in Duncan, 1988 1967 by INC dissident Charan Singh SAD 235/19 Yes Shiromani Akali Dal: Sikh-Punjabi sub-nationalist party Kumar, 2004 founded in 1920. Leadership drawn significantly from pros- perous farming communities. LKD 247/289 Yes Bharatiya Lok Dal: founded in 1974, descended from BKD. Brass, 1981 SSP 249/26 No Samyukta Socialist Party: split from PSP in 1964. Fickett, 1973 BJP 357/2 No Bharatiya Janata Party: Hindu nationalist party founded in Jaffrelot, 1999 1980, descended from BJS. ADK 385/35 Yes All India Anna Dravida Munnetra Kazhagam: split from DMK Wyatt, 2009 in 1972. DMK 400/66 Yes Dravida Munnetra Kazhagam: Dravidian sub-nationalist Wyatt, 2009 party. Populist platforms consisted significantly of subsidies for farmers. PSP 460/24 No Praja Socialist Party: socialist party formed in 1952. Fickett, 1973 SWA 505/78 No Swatantra Party: conservative party founded in 1959. Erdman, 2007 BJS 569/70 No Bharatiya Jana Sangh: Hindu nationalist party founded in Jaffrelot, 1999 1951 by Syama Prasad Mookherjee. CPI 691/75 No Communist Party of India: India’s first communist party, founded in 1920. CPM 885/103 No Communist Party of India (Marxist): split from CPI in 1964. Kohli, 1983 JNP 1716/79 Yes Janata Party: merger of several opposition parties in 1977, Rudolph and Rudolph, 1980 with large support base amongst farmers. IND 1780/83 No Independent: independent candidates were typically local Chhibber and Kollman, 2009 notables without organized party ties.

Notes: Table contains top 20 opposition parties by seats won. Party is classified as agrarian if it received signif- icant political support from and/or advocated programs oriented around interests of agricultural producers.

62 Figure A3: Matching Assembly and Parliamentary Constituencies to 1961 Districts

Panel A: Split of 1961 Shahabad District Panel B: Assembly Constituencies Panel C: Parliamentary Constituencies

BRAHMPUR SHAHPUR BARHARA ARRAH BUXAR BHOJPUR BUXAR JAGDISHPUR DUMRAON SANDESH ARRAH

RAJPUR PIRO RAMGARH SAHAR BIKRAMGANJ DINARA MOHANIA BIKRAMGANJ KARAKAT

CHAINPUR NOKHA

ROHTAS DEHRI CHENARI SASARAM

BHABHUA 63

SASARAM

Notes: Panel A: The 1961 District Shahabad subsequently split into two constituent districts, Rohtas and Bhojpur. Panel B: Post-1976 delimitation assembly constituencies comprising Rohtas and Bhojpur. Assembly constituencies fit contiguously within district boundaries. Panel C: Post-1976 delimitation parliamentary constituencies matched to Rohtas and Bhojpur districts. Note that Bikramganj was matched to Rohtas district and Arrah to Bhojpur district, despite partially intersecting with adjoining districts, because the majority of their constituent assembly constituencies fell in those districts. All matching of 1961 districts to assembly and parliamentary constituencies was done on the basis of tables in the Administrative Atlas of India and the 1956, 1961, 1967, and 1976 Delimitation Commission Reports. Figure A4: Example of Automated Coding of Digitized Legislator Biography 64

Notes: Who’s Who biography of Chaudhury Charan Singh – Member of Parliament, kisan (farmer) mobilizer, and leader of the Janata party – scraped from http://164.100.47.132/LssNew/biodata_1_12/2336.htm. Occupation coded as Professional and Career Politician based on detected keywords “lawyer" and "political worker". Coded as possessing a university education based on keywords "b.sc.", "m.a.", and "ll.b.". Caste coded as OBC/Intermediate based on Jaffrelot (2003) dataset. Table A5: IV Estimates: Effects of HYV Crop Adoption on Congress Performance in State Assembly Elections using Aquifer Instrument

Dependent variable: Congress Win Congress Vote %

(1) (2) (3) (4) (5) (6)

HYV 86.200∗∗ 89.187∗ 65.074 11.730 8.580 17.046 − − − − − − (37.283) (47.444) (43.745) (10.037) (12.141) (11.828)

Opposition Fractionalization 39.560∗∗∗ 13.133∗∗∗ − (2.114) (0.929)

INC Presence 48.344∗∗∗ 36.131∗∗∗ (2.145) (0.724)

Hindi Belt Year FE Y Y Y Y × Direct Rule Year FE Y Y Y Y

65 × Year FE Y Y Y Y Y Y District FE Y Y Y Y Y Y First-stage Aquifer Post 0.133∗∗∗ 0.113∗∗∗ 0.113∗∗∗ 0.133∗∗∗ 0.113∗∗∗ 0.113∗∗∗ × (0.034) (0.034) (0.034) (0.034) (0.034) (0.034) F-statistic 14.9 11.0 11.0 14.8 11.0 11.0 Observations 21,843 21,843 21,842 21,843 21,843 21,842 Adjusted R2 0.17 0.18 0.23 0.32 0.33 0.50 Clusters 268 268 268 268 268 268

Notes: Unit of observation assembly constituency-year for election years between 1957-1987. Outcomes: Win indicates whether Congress won seat and Vote % indicates Congress vote share, both multiplied by 100 so that coefficients have a percentage point interpretation. Explanatory variables: HYV crop adoption is district-level share of agricultural land with HYV crops (0-1). Instrument for HYV crop adoption is an interaction term between a time dummy variable indicating the introduction of HYV crops in 1967 and a cross-sectional variable indicating presence of a naturally occurring thick aquifer. Opposition Fractionalization is one minus the Herfindahl index of opposition party vote shares (0-1). INC Presence is an indicator variable for the presence of a Congress party candidate. Hindi Belt is an indicator variable for location in a Hindi-speaking state. Direct Rule is an indicator variable for historical exposure to direct British colonial rule. Analysis estimated by 2SLS. Standard errors adjusted for clustering within

districts. ∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01 < < < Table A6: IV Estimates: Effects of HYV Crop Adoption on Congress Performance in Parliamentary Elections using Aquifer Instrument

Dependent variable: Congress Win Congress Vote %

(1) (2) (3) (4) (5) (6)

HYV 85.457∗∗ 91.420∗ 64.047 15.031 12.655 21.120∗ − − − − − − (38.147) (48.588) (44.619) (10.016) (11.729) (11.568)

Opposition Fractionalization 41.561∗∗∗ 16.664∗∗∗ − (2.185) (0.704)

INC Presence 42.887∗∗∗ 34.523∗∗∗ (2.458) (0.669)

Hindi Belt Year FE Y Y Y Y × Direct Rule Year FE Y Y Y Y

66 × Year FE Y Y Y Y Y Y District FE Y Y Y Y Y Y First-stage Aquifer Post 0.106∗∗∗ 0.091∗∗∗ 0.090∗∗∗ 0.106∗∗∗ 0.091∗∗∗ 0.090∗∗∗ × (0.029) (0.030) (0.030) (0.029) (0.030) (0.030) F-statistic 13.3 9.0 8.9 13.3 9.0 8.9 Observations 2,940 2,940 2,940 2,940 2,940 2,940 Adjusted R2 0.26 0.33 0.44 0.32 0.37 0.70 Clusters 259 259 259 259 259 259

Notes: Unit of observation parliamentary constituency-year for election years between 1962-1985. Outcomes: Win indicates whether Congress won seat and Vote % indicates Congress vote share, both multiplied by 100 so that coefficients have a percentage point interpretation. Explanatory variables: HYV crop adoption is district-level share of agricultural land with HYV crops (0-1). Instrument for HYV crop adoption is an interaction term between a time dummy variable indicating the introduction of HYV crops in 1967 and a cross-sectional variable indicating presence of a nat- urally occurring thick aquifer. Opposition Fractionalization is one minus the Herfindahl index of opposition party vote shares (0-1). INC Presence is an indicator variable for the presence of a Congress party candidate. Hindi Belt is an indicator variable for location in a Hindi-speaking state. Direct Rule is an indicator variable for historical exposure to direct British colonial rule. Analysis estimated by 2SLS. Standard errors adjusted for

clustering within districts. ∗p 0.1; ∗∗p 0.05; ∗∗∗p 0.01 < < <