WORKING PAPER NO. 2017-03

Traditional Elites: Political Economy of

and Tenancy

by

Sabrin Beg

WORKING PAPER SERIES Traditional Elites: Political Economy of Agricultural Technology and Tenancy

Sabrin Beg1 July 2017

Abstract

Traditional elites can perpetuate their political influence through agricultural relation- ships. I show that landlords in can make cost-effective transfers to sharecropper- tenants, thereby gaining tenants’ electoral support and controlling policy. Technological change in makes less optimal, attenuating landlords’ electoral advantage. Exogenous productivity change lowers the rate of sharecropping and low- ers the likelihood of election of landlords in landlord-dominated areas; in turn electoral competition improves and the composition of public goods shifts. While demonstrating clientelism in rural agrarian societies through sharecropping contracts, I also highlight how changes in agricultural technology affect it.

Keywords: Land Inequality; Clientelism; Public Goods; Colonial Institutions; Electoral Competition, Traditional Chiefs; Political Economy; Elite Capture; Agricultural Produc- tivity.

1University of Delaware (e-mail: [email protected]) I would like to thank Shameel Ahmad, David Atkin, Kate Balwin, Ali Cheema, Alexander Debs, Rahul Deb, Yingni Guo, Naved Hamid, Susan Hyde, Dean Karlan, Asim Khwaja, Thomas Kirk, Dan Keniston, Phillip Keefer, Adrienne Lucas, Nancy Qian, Mark Rosenzweig, Laura Schechter, Raul Sanchez de la Sierra, Christopher Udry, Eric Weese and Matthew White for comments, advice and support. I acknowledge feedback received from participants at the Yale Development Seminar, Leitner Political Economy seminar, Princeton, Boston University, World Bank, Texas A&M University and University of Delaware. I also thank Jacob Shapiro and Abdul Ghaffar for useful data, without which this project would not have been possible. I acknowledge financial support from the Sylff Foundation and Georg W. Leitner Program in International and Comparative Political Economy. All errors are mine. 1 Introduction

Land ownership, power and paternalism have been central to historical development— landlords were politically and economically dominant in the Roman Empire, medieval Europe, the traditional societies in Latin America and East Asia, and the pre-industrial United States South. These landlords used coercion and other oppressive tactics to con- trol tenants, but they also provided patronage and public goods.2 As patrons, landlords precluded the provision of broad welfare by the state, impeding the effectiveness of democ- racy and the state. The de facto power of traditional elites, such as landlords, and the clientelistic equilibrium were remarkably persistent, and eventually broken via the entry of new political forces (Shefter 1977), democracy (Lizzeri and Persico 2004), or economic opportunities (Chubb 1982). The decline of landlords’ power led to the formation of national welfare states by weakening the paternalistic ties between landlords and their rural clients. Indeed, rapid growth in East Asia has been attributed in part to extensive land reforms, which facilitated elimination of the landlord class and thus helped establish an equitable distribution of the benefits of economic growth (Grabowski 2002; Perkins 2013; Kitamura 2016). Paternalism continues to prevail in rural economies despite democratization, par- ticularly in Pakistan, the context for this study. Sixty-one percent of the population in Pakistan is rural, and in some cases residing in areas characterized by high land inequal- ity, pervasive landlessness, and few large landowning families with economic and political influence (World Bank 2015; Baker 2008).3 Sharecropping tenancy, which is historically common in agricultural societies but dramatically reduced in most of the world due to agricultural modernization, is persistently high in Pakistan: of the 30% of agricultural land farmed by tenants, 70% is sharecropped). Landlords in Pakistan appear to obtain political offices via the support of their sharecropping tenants (Baker 2008). In this paper, I empirically identify a microfounded mechanism that defines the patron-client

2For more information about landlords in Chile, see Baland and Robinson (2012); in pre-industrial Europe, see Brenner (1976); in the US South, see Alston and Ferrie (1999); and in Peru, see Dell (2012). 3About 70% of members elected to the provincial assembly are agricultural landowners (see section 2 for details).

1 relationship between the traditionally powerful landlord and sharecroppers. I further highlight the political economy of technological change in agriculture, which can affect landlord-tenant contracts and consequently political outcomes. I pose two major questions. First, how do historic landowning elites acquire and maintain political power through clientelistic ties with their tenants? Second, does a landlord’s ability to acquire political support shift with technological change in agricul- ture, and if so, how does this affect electoral and public goods outcomes? These questions are vital in the study of development economics because of the prevalence of paternalism and clientelism, especially in pre-industrialized and poor countries. While paternalism has been studied descriptively in a number of contexts, I use well-identified empirical analysis to understand the mechanisms through which paternalistic systems persist and what factors contribute to the eventual breakdown of these systems. I develop a model by merging insights from the literature on contractual arrange- ments in agriculture (Braverman and Stiglitz 1982; Stiglitz 1974) and on political clien- telism (Dixit and Londregan 1996; Robinson and Verdier 2002) to study an environment in which landowners and sharecropping tenants participate in traditional relationships of reciprocity. The landlord and tenant share input costs in a sharecropping arrangement; thus the landlord can raise the tenant’s income share by paying a higher share of the in- put cost. The increase in the tenant’s effective share induces higher effort, which results in higher overall output that accrues to the landlord through his share (Burchardi et al. 2017).4 Thus by paying more for inputs on sharecropped plots, landlords bear a lower cost of income transfers to their tenants relative to a lump sum transfer. This interlinkage of input and output markets in sharecropping agriculture renders an electoral advantage for dominant landowners through their ability to capture votes by making ‘cheap’ transfers to landless or smallholder tenants. I incorporate this electoral advantage into an election framework in which candi- dates offer both private transfers and public goods to voters. Landlord politicians win a higher vote share as a result of this ability to make cheap transfers, and conditional

4Output sharing generally lowers the agent’s (tenant’s) incentives resulting in a classic moral hazard problem; increasing the tenant’s income share improves his incentives.

2 on winning spend more on types of public goods that benefit landowners. Technological change in agriculture increases productivity and decreases sharecropping under certain conditions, as noted by Eswaran and Kotwal (1982) and Stiglitz (1974). Two mecha- nisms affect landlord politicians’ election outcomes in opposite ways. Firstly, technolog- ical change improves landowners’ wealth and thus their ability to make transfers to all voters, increasing their votes and likelihood of winning elections. The second mechanism operates specifically for political landowners who engage tenants for electoral support—if technological change lowers sharecropping, the landlords’ electoral advantage in securing tenants’ votes cheaply goes down. Thus, their probability of winning elections decreases. Thus the model predicts that: (1) Landlord politicians are more likely to hire sharecroppers when electoral incentives are high and offer better income to them (by paying a higher share of input costs). Technological change in agriculture makes landlords wealthier everywhere, which improves their chances of electoral success. At the same time, technological change affects optimal tenancy contracts. This results in prediction (2): if technological change makes sharecropping costly, then previously landlord dominated areas experience (a) lower likelihood of landlords’ election (b) better electoral competition and (c) shift of policy away from landlord-preferred, relative to other (non-landlord- dominated) areas. I use household-, constituency-, and district-level data from Pakistan to test these mechanisms. Pakistan is well suited to this analysis, because it has a history of landown- ing elites dating to the colonial and precolonial eras. Landlords continue to exert agri- cultural and political influence, and sharecropping remains high. I conduct two sets of empirical analyses for each of the two main predictions. In the first analysis, I examine tenancy contracts offered by political office holding landowners— I isolate the effect of landowner’s political status using tenant-level panel data and ex- ploiting the introduction of elections in 2002, following a military regime, as a natural experiment.5 I compare the agricultural contractual terms for plots owned by landlord

5Pakistan is a federal republic, but historically it has alternated between democratic and military regimes. The 2002 election that followed military rule provides a natural experiment to study landlord politicians who have electoral incentives.

3 politicians post-election, relative to other contracts of the same tenant pre-election. I find that a tenant household is more likely to be engaged in sharecropping on plots that are leased by a politician landlord, specifically after the 2002 election. Further, tenants of winning politicians pay a lower share of input costs and are more likely to report access to after the election. Politician landlords also engage in more direct supervision of their sharecropping tenants after the election. This classic form of clientelism in the case of landlord politicians and tenants constitutes electoral transfers in the form of more favorable tenancy contracts. In the second empirical analysis, I study the effect of technological change on the electoral success of traditionally influential landlords and the allocation of public goods using constituency- and district-level data. In theory agricultural technology that lowers sharecropping rates should affect political outcomes in initially landlord dominated areas. To proxy for landlords’ historical dominance in any area, I classify areas of the country as ‘landlord-dominated’ or not, based on the existence of prominent landlords during the colonial period. I also construct a measure of exogenous technology change using high yielding variety (HYV) technology—I compare areas with high HYV suitability due to area-specific, exogenous geological factors to areas with low HYV suitability, as these seeds become available. Specifically, I use the interaction of area-specific suitability for HYV seeds with time-varying, aggregate availability of the HYV seeds to get a proxy for technological change by geographic region and year.6 I find my measure of technological change to be positively correlated with higher agricultural productivity and lower rates of sharecropping. Next, I examine the political effect of the lower sharecropping due to agricultural technology change in areas where landlords were previously dominant using constituency level data from 2002–2013. I find that technological advancement due to HYV seeds shifts power away from landlords in areas where they have been traditionally dominant. A technological improvement that raises wheat yields by 0.25 ton/ha reduces landowners’ election probability by one-half.7

6Bustos, Caprettini, and Ponticelli (2013) use a similar construction to study the effect of agricultural pro- ductivity on structural transformation in Brazil. 7Wheat is the most widely grown in Pakistan, with 84% of private reporting cultivating wheat (Agricultural Statistics of Pakistan 2011–12).

4 I also find that technological change in agriculture improves electoral competition and shifts the composition of public goods in initially landlord dominated areas; public goods prioritized by landowners decrease relative to other public goods. Through the political economy of technological change, the shift in the electoral success of landlords causes shifts in the composition and spatial allocation of public goods. The results make three key contributions to the literature. First, the analysis em- pirically demonstrates a microfounded mechanism for landlord-tenant reciprocity in rural contexts (see Banfield 1967; Brenner 1976, Popkin 1979; Powell 1970; Scott 1972, 1976; Finan and Schechter (2012); Piliavsky (2014)) by expanding on the literature on in- terlinked agrarian markets (Bell and Srinivasan 1989; Braverman and Srinivasan 1981; Braverman and Stiglitz 1982). Baland and Robinson (2008, 2012) studied landlord-tenant relations in Chile in 1950s, where landlords conceded rents to workers to induce effort and threatened to withdraw those rents to control their workers’ voting behavior. Acemoglu and Robinson (2001) and Alston and Ferrie (1999) similarly document labor repression by landlords in the of the antebellum U.S. South, in which landlords exchanged paternalistic favors (e.g., economic and social protection) for loyalty and hard work, which saved the landowners the costs of labor turnover and monitoring. This paper studies paternalistic relations between landlords and tenants using rich data and context, showing that tenancy contracts allow landlords in Pakistan to gain political support from tenants. Transfers to sharecroppers by landlords demonstrate a specific example of clien- telism in which landlord politicians offer favorable contracts and better input access as electoral transfers to loyal tenants. Thus, I also contribute in a novel way to the litera- ture on political clientelism (Diaz-Cayeros and Magaloni 2003; Keefer 2007; Keefer and Valaicu 2008; Robinson and Verdier 2002; Torvik 2005; Vicente and Wantchekon 2009; Anderson, Francois and Kotwal 2015). The literature has noted mechanisms that lead to the breakdown of landlord dom- inated systems, like introduction of the secret ballot in Chile or economic forces like migration and mechanization of farming as in the U.S. South (Alston and Ferrie 1999). The second contribution of this paper is in understanding how agricultural transformation

5 affects landlord-tenant relations, making it costlier for landlords to control tenant votes, eroding the influence of traditionally dominating landlords and potentially inducing the transition of political power. I document this transition in the context of Pakistan, using well-identified empirical analysis. Sharecropping, which had allowed land-owning elites to maintain political control, declined due to technological changes in agriculture. These changes resulted in a shift away from rural paternalism. Thus far, this mechanism of transition in rural societies has not been documented with rigorous empirical analysis. Increased agricultural productivity is a well-documented precedent for industrialization and development (Nurkse 1953; Rostow 1960). Highlighting the political implications of agricultural technology offers a key mechanism that links agricultural productivity to development, and is a novel aspect and important contribution of the paper. Lastly, by presenting new microfounded mechanisms to understand the impact of historic land institutions on development outcomes, the results corroborate the re- search documenting the persistence of institutions, particularly historic elites (Nunn 2009, Glaeser and Shleifer 2002, Acemoglu, Johnson, and Robinson 2001,2002, Galor, Moav and Vollrath 2003, 2009, Engerman and Sokoloff 2005, 2007, Banerjee and Iyer 2001). Acemoglu, Reed, and Robinson (2014) document that traditional elites or ‘Chiefs’ were endowed with institutional and political powers by colonial governments.8 The existing literature does little to reconcile how the pre-existing elites are able to maintain political influence even after the end of colonial era and subsequent democratization (Logan 2011). I identify a microfounded mechanism through which we can trace the impact of historic land institutions on the existence of landed elites and eventually the provision of public goods. Economic growth, in turn, can reinforce or undermine the power of these elites. I therefore examine the impact of agricultural development on the political dominance of landowning elites and the consequent implications for public goods provision.

8There is also work by Boone (1994), Chanock (1985), Mamdani (1996), Merry (1991), Migdal 1988, Roberts and Mann (1991).

6 2 Institutional and Historical Background

Land and power were undoubtedly linked during the precolonial and colonial eras in the Indian subcontinent (comprising present day Pakistan, India and Bangladesh). Feuda- tories or jagirdars (landholders under the feudal tenure system) were offered large land assignments as gifts from the State for allegiance or military services, rendering consid- erable local influence for them.9 The initial ruling dynasties (e.g. Talpurs, Kalhoras, Sikhs, Mughals) gave jagirdars temporary authority to collect revenues from their lands (Hussain 1979). The British government then made these assignments permanent to help establish a social class that was loyal to the empire. The British initiated an institution- alization of aristocracy, maintaining detailed records of the landed aristocrats, e.g. in the publication titled ‘The Chiefs of the Punjab’ (Massy 1940). The jagirdars were also the zamindar in the zamindari system of revenue collection, as described in Banerjee and Iyer (2002, 2008), and had the status of local chiefs, as described in Acemoglu et al. (2014). The political and economic influence enjoyed by jagirdars persisted post-independence. Attempts to institute land reforms after independence have had limited success in Pak- istan (Gazdar 2009; Joshi 1970; Rashid 1985). Land concentration continues to be high, especially in the jagirdar-dominated areas. The top percentile of landowning households own over 150 acres and up to several thousand acres each, therefore, control a consid- erable fraction of agricultural land; whereas, the average landholding of the remaining households is only 3.5 acres (Pakistan Agricultural Census 2010). Within a village, which reasonably represents an independent land market, there are typically three or fewer large landowners while 75% of households own 0–5 acres (Pakistan Rural Household Survey 2000). Thus these large landlords are considered monopsonist in the village land mar- ket. On the aggregate level, the large landholders constitute an oligarchy (Powell 1970). Given the high percentage of landless and small-landholding households, tenancy (par- ticularly sharecropping) is prevalent: 30% of land is farmed by tenants, and more than

9The land assignments, also know as jagirs could be a village, a group of villages, or even as large as an entire sub-district.

7 70% of leased farms are sharecropped.10 Therefore, the landholding oligarchy may act in coordination to control voting behavior of tenants who depend on them for livelihood. For instance, the prominent family achieves its political agenda through support from thousands of sharecroppers who work on over 10,000 acres of land that the family has possessed for 500 years (Baker 2008).

Sharecroppers till the lands, exchanging half they produce—, wheat and sugarcane—for a place to live, seeds and . And patronage. ‘If my tenants are happy with me, they work more efficiently on the lands,’ says Mumtaz Bhutto. ‘You help the people and they will help you... The tenants support any candidate their landlords put up.’ (Baker 2008)

Historically and in other contexts, landowners have procured political support by threatening eviction or by coercion (Baland and Robinson 2008, 2012; Powell 1970; Ri- cardo 1824; Scott and Kerkvliet 1976). However, little evidence of such threats exist in survey data from rural Pakistan. Instead, landlords supply several favors to tenants, including consumption loans, agricultural credit, and key agricultural resources like ir- rigation access. These inter-linkages of the land, inputs, credit, and electoral markets guarantee a large rural electoral base for landowning families. Given the number of ten- ants working for large landlords, I estimate the large landowners can acquire 40% of the votes needed to win from tenants alone (see the appendix).11 Recent anthropological literature identifies Pakistani politics as constituting national and local ‘political settle- ments’ or political coalitions, which legitimize their domination by extending services to their clients (Kaplan 2013; Zaidi 2014). The oligarchy of landholding elites is such a coalition, strengthening itself thorough alliances among their class. Seventy percent of Provincial Assembly members declare agricultural land owner- ship (Election Commission of Pakistan). The Provincial Assembly is the lowest level of government—like the state legislatures in the United States, officials are elected by ma- jority voting. The landowners’ overwhelming representation in government allows them to stall land reforms and keep agricultural taxes low. Thus, Pakistan’s government rev-

10In contrast, the tenancy rate in India is less than 1%, according to the latest census in 2010. 11In my analysis, I test one specific mechanism through which landlords’ can acquire tenants electoral support, but landlords are influential for many reasons. Landlords’ knowledge about voters’ preferences, higher credibility as politicians, and ability to monitor and enforce voters can be other mechanisms through which landlords secure their tenants’ votes. 8 enue was 15% of GDP in 2015, compared with 28% for emerging and middle-income economies as a group (IMF 2015). More recently, politics has experienced a shift. As farming becomes capital-intensive, the landowners have fewer tenants and thus fewer votes—“as labor-intensive plantations decline, ... farming modernizes ... old families lose clout” (The Economist 2013). Industrialists and urban professionals are the new alternate political classes (Shafqat 1998).

3 Theoretical Framework and Testable Predictions

In the appendix, I outline a basic model to illustrate how the electoral and land markets operate simultaneously when landlords are politically involved. The model is guided by the context described above and aims to: (a) explain how landowners can gain votes from sharecropping tenants, (b) illustrate the implication of the electoral advantage of landlord politicians for electoral competition and policy, and (c) illustrate the effects of a shift in agricultural technology on landlords’ political participation and public goods provided. I build on two canonical models: an election model of redistributive politics (Persson and Tabellini 2000) and a model of tenure choice in agriculture (Braverman and Stiglitz 1982; Stiglitz 1974). I show that a landlord can transfer utility to the sharecropper cheaply (i.e., the landlord’s cost of raising the tenant’s income is less than the cost of offering an equivalent lump sum transfer) by offering a higher net income in the sharecropping contract. Specifically, landlords can increase tenant payoff in a sharecropping contract by paying a higher share of costs—by doing so they incentivize the tenant to supply higher inputs and effort, alleviating the moral hazard problem, which results in higher overall output.12 The landlord gets a share of the higher output, making the net cost of the income transfer lower. I consider the large landowners to constitute an oligarchy acting in coordination

12This prediction is supported by evidence from Burchardi et al. 2017, where output share of sharecropping tenants is experimentally varied. Higher tenant share incentivizes him to apply higher inputs and change crop choice, resulting in higher output. Since the output share is predominantly fixed at 50% in the data and input shares are variable, I allow the input share to vary in the sharecropping contract. Both higher output share to tenant or higher input share paid by landlord effectively improve tenants income share.

9 to gain votes. A landlord candidate in an election has an advantage over a non-landed competitor because landlords can promise higher transfers to tenants. I incorporate this landlord advantage into an election model where candidates offer both private transfers and public goods. I derive the electoral equilibrium and predict the agricultural tenure choices, electoral competition, and electoral platforms when productivity shifts due to technological changes in agriculture. The model yields the following main predictions. Prediction 1: Landlord politicians with incentives to offer private transfers are more likely to hire sharecroppers and offer better income to them (by paying a higher share of input costs). Technological change in agriculture increases the profits of landlords and makes sharecropping less optimal if the cost of monitoring labor remains stable (Eswaran and Kotwal 1982; Stiglitz 1974). Higher agricultural profits should increase political par- ticipation by landowners in all regions. On the other hand, changes in incentives for sharecropping can impact the ability of dominant landlords to offer patronage to tenants and consequently affect political outcomes specifically in previously landlord dominated regions. High efficiency cost of sharecropping lowers landlords’ ability to gain ‘cheap votes’, resulting in decreased votes and likelihood of winning among landlord politicians. Prediction 2: If technological change lowers sharecropping, then regions with initial landlord dominance experience (a) lower chances of election wins for landlords (b) better electoral competition, and (c) lower provision of landlord-prioritized public goods, relative to non-landlord-dominated regions. For ease of narration, I describe the empirical strategy, data and results for predic- tion 1 first, and then go on to do the same for prediction 2.

4 Empirical Strategy, Data and Results

4.1 Landlord Politicians and Agricultural Tenancy Contracts—Prediction 1

Testable prediction 1 states that a landlord with political incentives will be more likely to have sharecropping tenants. Political status is not randomly assigned, and observable and unobservable characteristics of landowners and tenants can simultaneously drive

10 contractual terms and the political status of landlords, making it difficult to identify the effect of political incentives. To disentangle the effect of landlords’ political incentives, I would need to compare the contracts of the same landlord with and without political incentives. I examine the effect of electoral incentives on agricultural contracts using a panel of tenant households and exploiting an introduction of an election after a military regime as a natural experiment. The tenant-level panel consists of households surveyed in the agricultural season of 2000–01 when the country was under a military regime after a coup in 1999, and again in 2004–05, after a general election was held in late 2002.13 Thus, in the pre-election round, there are no active and directly elected politicians, whereas the post-election data indicates plots owned and rented out by an office-holder’s household.14 The survey data does not follow the same landlord over time, but facilitates a tenant level panel, allowing me to control for tenant-fixed effects when a particular household is a tenant in both survey rounds. Under the assumption that tenants do not switch landlords between the two rounds, I can compare contractual terms for a plot with a politician landlord relative to contractual terms on other plots leased in by the same tenant in the pre-election round. Consider the following plot-level specification that classifies plots rented out by politically involved landlords:

yi,j,p,t = γ0 + γ1P oliticianLL P ostElectioni,j,p,t + γ2P ostElectiont

+ γ3ηp + γ4σj,t + γ5ςi,t + κi + εi,j,p,t, (1)

where yi,j,p,t is the contractual arrangement for tenant i, landlord j, and plot p in

13The previous general election was held in 1997; however, the government was dissolved and replaced by a military government during a Musharraf-led coup in 1999. The Supreme court had ordered Musharraf to hold general elections in 2002 and he had agreed (Associated Press 2002). The 2002 election was thus not a surprise; even if Musharraf’s compliance was not credible, the general election was likely to be anticipated as local government elections were already being conducted during 2001 (Mezzera, Aftab and Yusuf 2010). Moreover, the alliance between Pakistan and the U.S. on the ”war on terror” likely pressured Pakistan to demonstrate a shift toward more democratic institutions. 14We observe if the landlord of a certain plot is from the household of a politician/officeholder, but not if he is an officeholder himself. As argued in Section 2 tenants associate themselves with the families of their landlord rather than the individual landlord and politically involved families act in coordination to influence voters. Thus, the effect of an individual landlords’ political status is likely exhibited on all plots rented by the politician’s household. Plots rented out by members of a political households are thus classified as “treatment” plots, i.e. plots leased by politician landlords.

11 year t, and P oliticianLL P ostElectioni,j,p,t equals 1 for leased plots in the P ostElection period where the landlord comes from a officeholding household. κi is a tenant fixed effect.

Controls include plot characteristics, ηp, landlord characteristics σj,t, and time-varying tenant characteristics ςi,t. The P ostElection dummy is a fixed effect for agricultural season 2004-5, i.e. after the election and accounts for the fact that election years are systematically different from other years (Khemani 2004). Under the assumption that the households rent from one landlord do not change their landlord pre- and post-election, the tenant fixed effect also accounts for the landlords’ identity. In this case, this specification allows me to compare a leased plot where the landlord is a politician post-election to the leased-in plot of the same tenant (and same landlord) from the pre-election period. The data shows that two-thirds of the farmers rent in one plot and the remainder rent two or more plots from different landlords. Additionally, the average length of landlord-tenant relationships is over 8 years—landlords tend to stay same across the pre- and post-election survey rounds.15 For tenants who rent from different landlords in the same period or across the two survey rounds, the above specification compares a leased plot where the landlord is a politician post-election to the average leased in plot of the same tenant from the pre-election period. The leased in plots from the pre-election period may be owned by different landlords. As mentioned above, in this case the main challenge to identification is unobserved landlord characteristics that are associated with their political status and simultaneously affect their contract terms. For instance, if some landowners have more wealth, charisma or ability they will be more likely to win political office and may also engage in agricultural contracts that are more generous for their tenants. Thus, even for the same tenant, politician landlords in the post election period may be systematically different from landlords from the pre-election period. To account for these concerns, I additionally control for politician landlords in the pre election period. Tenants report if their landlord is a politician in the pre-election round, which implies that the landlord was elected in the previous election but is not currently in office due to the coup. If these

15Sharecroppers report to have been with the same landlord for 10 years on average, and tenants of politicians have been with their landlords for over 12 years on average.

12 politicians from the last election do not have electoral incentives, then I can identify the effect of political incentives by comparing plots leased by politician landlords in the pre- election period to plots leased by politicians in the post-election period. The specification is as follows:

yi,j,p,t = β0 + β1P oliticianLL P ostElectioni,j,p,t + β2P oliticianLL P reElectioni,j,p,t

+ β3P ostElectiont + β4ηp + β5σj + β6ςi,j + κi + εi,j,p,t (2)

P oliticianLL P ostElectioni,j,p,t, same as previously defined, is the explanatory

variable of interest. P oliticianLL P reElectioni,j,p,t is an indicator for a leased plot during

the pre-election period where the landlord is a politician. P oliticianLL P reElectioni,j,p,t controls for characteristics specific to politically involved landlords and provides an appro-

16 priate reference group to compare the effect of P oliticianLL P ostElection. β1−β2 mea- sures the difference in the outcome variable for political landlords with electoral incentives relative to those without. P oliticianLL P ostElection and P oliticianLL P reElection may represent the same landlord or different landlords. The key idea is that I am com- paring plots leased by the same tenant from similar (or the same) landlord when electoral incentives are present to when they are not.

I additionally control for the case where P oliticianLL P ostElectioni,j,p,t represents a new landlord, who recently entered into a contract with the tenant and was thus not included in the pre-election landlord composition of the tenant. Households may enter into contracts with new landlords between the two survey rounds, specifically they may rent land from landlords striving to enter office.17 Using tenants’ responses about the length of time they have been renting from a certain landlord, I construct NewLL, which is 1 if the tenant rents land from a landlord for 3 or fewer years, i.e. since the last time the household was surveyed. Controlling for NewLL × P oliticianLL P ostElection enables me to specifically examine differences in contractual terms before and after the election when the tenant has had the same landlord over this time; thus I can isolate the effect of electoral incentives holding landlord identity constant.

16See data appendix for coding of politician landlords across the two survey rounds. 17About 30% of the post-election plots are reported to have been leased from a new landlord in the last 3 years, i.e. since after the last survey (though, only 4% of these plots are leased by politician landlords). 13 The statistic of interest in the above specification is β1 − β2; this is similar to a difference in difference framework. Noting the caveat previously mentioned with respect to the slight difference in how the political status of landlords is solicited across the survey rounds, I assign variable P oliticianLL to indicate a plot where the landlord is politician either pre- or post-election and present results from a difference in difference specification as follows (subscripts omitted):

y = α0 + α1P oliticianLL × P ostElection + α2P oliticianLL + α3P ostElection

+ α4η + α5σ + α6ς + κ + ε (3)

The coefficient of interest is α1, and is the same as β1 − β2 from the previous specifica- tion. The outcome variables include the contract type (fixed rent versus sharecropping), access to irrigation, landlord’s input cost-shares for various inputs and supervision on sharecropped plots. Standard errors are clustered at the tenant level. In a different specification, I analyse the same outcomes at the household level. I create a dummy, T enantofP olitician, which equals 1 for a household that rents land from a politician in the post-election round. I run the following tenant level difference- in-difference regression, which identifies the change in outcome for tenants of politicians after the election relative to before.

yi,t = λ0 + λ1T enantofP olitician × P ostElectioni,t + λ2P ostt + κi + εi,t (4)

The specification demonstrates if tenants of politicians experience significantly different contracts on average after the election relative to before. The household level regression alleviates the problem of differences across plots and landlords within a tenant household, and identifies the change in outcome when one of the landlords of a household enters politics, or when a household contracts with a politician landlord as elections approach.18 Similar to above, I include an interaction with NewLL, which in this case identifies a tenant who rents land from a new landlord after the last survey round. This allows me to look at the post-election change for tenants who did not change their landlord composition since the pre-election survey round.

18I thank two anonymous referees for this suggestion.

14 If elections are expected then electoral incentives may still lead landlords to change their behavior in the pre-election period. However, this should weaken the difference between post- and pre-election. As I show later, I find an effect after the election relative to before, thus demonstrating the higher electoral incentives during the post-election survey round. Data for Prediction 1 I use data from two rounds of the Pakistan Rural Household Survey (PRHS) in 2000 and 2004. Round I (PRHS 2000) provides data for agricultural seasons Kharif (March-October) 2000 and Rabi (November-February) 2001. Round II (PRHS 2004) covers Kharif 2004 and Rabi 2005.19 For each leased-in plot the tenants are asked if their landlord is a politician in the 2000 survey round, and they are asked if their landlord is from the household of an office-holder during the 2004 survey round. Additionally, I use information at the plot level, including plot characteristics, leasing status (self-cultivation versus sharecropping contract or fixed-rent contract), contract terms, input and outputs, and characteristics of the cultivating households. Summary statistics are provided in Table 1b. The data comprises a panel of 484 households for two periods, with a total of 1586 plots of which 1005 are leased in. The regression analysis focuses mostly on the leased plots; 75% of the leased plots are sharecropped. Apart from sharing the output in share- cropping contracts, landlords also pay some fraction of input costs. Landlords provide about 50% of fertilizer costs and 26% of the cost of activities (including labor and the use of thresher). Landlord supervision of sharecropping tenants is almost universal in the data, either directly through meetings or through labor managers employed by the landlord. Landlords meet their sharecroppers over 20 times on average during the season, while labor managers, when employed, have an average of 30 meetings with the tenant over the course of the season. On average, tenants interact with the landlord or his manager almost every week during the agricultural season. Results for Prediction 1

19I adopt the names provided by the survey administrating organization, Pakistan Institute of Development Economics.

15 Table 1b shows that politicians are more likely to lease out land on sharecropping contracts over the two survey rounds. Column 1 of Table 2 shows that a given tenant is 10 percentage points more likely to have a sharecropping contract on the plot leased out by a politician relative to the average leased plot of the tenant during a nondemocratic period. This difference may arise due to election incentives or due to other characteristics specific to politicians; the average leased plot of the tenant may not be a comparable control group as plot and landlord characteristics may differ across politician and non-politician landowners. The regression in Column (2) compares politician landlords’ contracts post- election to contracts of similar or the same landlord before the election. Column (2) shows that the rate of sharecropping is 15 percentage points higher on politician owned plots after the election relative to politician owned plots before the election. This is 20% higher than the average pre-election rate of sharecropping in the sample. The higher sharecropping rate may occur as landlords lease out more land to existing or new tenants when electoral incentives are high, or they may change the contractual arrangement on existing plots of existing tenants. The latter is unlikely as the data does not suggest that the type of tenancy arrangement (sharecropping or fixed rent) between a particular landlord and tenant on a given plot changes from year to year. The interaction between P olitician × P ostElection and NewLL, a dummy for plots leased from a new landlord in the last 3 years, is intended to see if electoral incentives induce politically inclined landowners to lease out plots to new sharecropping tenants. The effect is insignificant, but large—plots that are leased out by politician landlords close to election time comprise a small fraction of the plots, but are 31 percentage points more likely to be sharecropped than plots leased out by nonpolitical landowners pre-election. The effect on sharecropping rate for longstanding tenants of politician landlords is still positive, though insignificant, implying that landlords may be renting plots to both new and previous tenants as elections approach. One concern is that the recently leased plots or plots specifically leased by politicians after the election are significantly different. Plot controls account for plot specific differences in all regressions; additionally, appendix table A1 ascertains that NewLL = 1 or LL P olitician do not indicate significantly

16 different plots in terms of observable characteristics like plot area and and terrain characteristics etc. To augment the findings from the above regression and alleviate the confounding variation due to differences across plots, I run tenant-level regressions in a difference-in- difference framework examining the change in the rate of sharecropping for tenants of politicians between the post- and pre-election periods. Table 3 shows regressions at the tenant level including all households that report being a tenant in at least one survey round. Tenants of politicians are more likely to sharecrop after the election relative to before, as measured by a dummy for sharecropping and also by the total number of sharecropped plots. The rate of sharecropping measured by percentage of cultivated plots of a household that are sharecropped increases by 7.1 percentage points (13%) after the election if the household is a tenant of a politician. Further, I specify households who have entered a contract with the politician landlord after the last survey round, i.e. in the last 3 years. These households comprise about 15% of all households who are tenants of politicians, and because of the small number of households in this category the coefficients are not precise. However, these households are indeed more likely to sharecrop after the election relative to before. NewLL = 0 comprise the tenants who were renting from the politician landlord over the entire span of the data; these tenants also demonstrate higher likelihood of sharecropping, confirming that electoral incentives lead politicians to engage new sharecroppers as well as offer more land to existing tenants. These findings corroborate the evidence from Table 2. Sharecropping allows landlords to offer cheap patronage through provision of inputs as argued in section 3; moreover clientelism through brokered relations is generally more efficient due to ease of monitoring (Larreguy 2013; Stokes 2013), which is also true for sharecropping arrangements as they allow landlords to easily supervise their voters. In the next set of regressions I examine supervision and input provision for sharecropped plots by electorally motivated landlords. Table 4 shows that politician landlords pay a higher share of fertilizer and har- vesting costs for their sharecropped plots after the election relative to politicians before

17 the election. Additionally, politicians increase direct meetings with their sharecroppers and lower monitoring through managers after the election. This may indicate more supervision by the political landlords, but also more importantly, greater paternalistic engagement between politicians and their tenants. Again, I verify that the higher avail- ability of inputs and landlord interaction is not because the ‘treatment’ plots (i.e. plots with P olitician×P ostElection = 1) are those leased from new and significantly different landlords after the election. Plots where the landlord contracted with the tenant since the pre-election period receive better input access post election. The output share of the politician landlord is higher after the election (although this is small relative to the input costs paid by him) on plots leased to long term tenants. At the same time, politicians also pay higher input costs when they lease out plots to new tenants close to the election, while demanding a lower share of the output. These predictions are supported by the tenant-level analysis (Table 5) where out- comes are averages of landlord input cost shares across sharecropped plots of a tenant— long-term tenants of politicians tend to receive higher inputs and more direct interaction with their landlord after the election. Recent tenants contracted by politicians near elec- tion time similarly receive better access to some inputs. I also examine access to canal water, which is a strategic input and appears to be central to village-level conflicts in the survey data. Table 6 shows that, specifically on sharecropped plots, tenants of politicians are more likely to have access to canal irrigation. While the availability of irrigation is not significantly different for politician-owned plots, usage of canal water, i.e. receiv- ing a turn to access water to irrigate ones field is higher for sharecroppers of politicians post-election. Overall these results suggest politically motivated landowners seem to rely on their access to land and agricultural inputs to fulfill clientelistic endeavors towards tenants and landless farmers in their villages. The P ostElection dummy, indicating the survey round after the election, is signif- icant in most regressions. This coefficient captures how contractual terms change after the election when the tenant does not have a politician landlord. The significance may be caused by the political effect of a transition from a military regime, which is indepen-

18 dent of the landlord’s political status, or as a result of periodic adjustment of contracts. Survey data indicate that contracts terms, especially the type of tenancy, are set between the landlord and tenant and are relatively stable until the tenancy ends. In section 5, I examine pre-trends in contract terms from a different survey (and in a different time- frame); there seem to be no significant changes in average contractual terms from year to year, except in an election year after a military regime. Thus, P ostElection is likely capturing the election effect or a one-time significant weather or agricultural shock. Even though P ostElection has an effect on contractual terms, P oliticianLL × P ostElection has its own independent effect that is different from the P ostElection coefficients. The finding that winning politicians are more likely to hire sharecroppers does not specify if politicians make these offers after or just before the election. Moreover, the findings do not preclude the fact that all politically motivated landlords may prefer sharecropping, regardless of winning an election. The data set does not specify landlord politicians in 2004 who contested in the election but did not enter office, thus I cannot estimate the propensity to hire sharecroppers for such landowners. I note that during 2001, when the pre-election data is collected, an election was expected and any electoral transfers offered by politician landlords in preparation for the election should already be reflected in the pre-election data. I do not find a positive coefficient for politician landlords in the pre-election round. The significantly different contract post-election indicate these may be post-electoral transfers. The findings do establish that office-seeking landlords offer more generous agricul- tural contracts when they face higher electoral incentives than when they don’t, demon- strating clientelism through the use of tenancy contracts. Electoral incentives are known to cause cyclicality in policies and outcomes, referred to as ‘political business cycles’. The results indicate the presence of political business cycles in agricultural contracts in this rural agrarian setting. Nichter (2011) has shown that political business cycles occur in sterilization surgeries in Brazil, and Labonne (2016) has shown that they occur in employment in the Philippines. Moreover, I note that the clientelistic nature of rural politics in Pakistan entails more than coercion or threats, as tenants seem to receive high

19 transfers. As Bursztyn (2015) suggests, rural voters actually value private transfers more than public goods, like education. One competing argument may be that winners offer better contracts not as electoral transfers but because of other office-related incentives. For instance, politicians may fear scandals and hence are incentivized to treat tenants well, or they may be able to appro- priate public funds that could relax cash constraints and allow them to invest in their farms. These plausible and valid explanations for the effect that I find cannot be explic- itly ruled out in some cases. However, I argue that my story about the ability of landlords to make transfers to sharecroppers is adequately supported by the results. Firstly, other office-related incentives would induce landlords to favor all tenants, and I find that only sharecropping tenants benefit, specifically, in access to irrigation. Moreover, the empirical findings are in line with the descriptive evidence in section 2 demonstrating that landlord politicians’ electoral incentives drive transfers to tenants. My argument is also supported by the results in the next section, which indicate that the decline of sharecropping is as- sociated with a shift in landlords’ electoral successes, further illustrating the interlinkage of agricultural contracts and electoral transfers.

4.2 Technological Change, Tenancy, and Electoral Outcomes—Predictions 2(a)–(c)

Prediction 2(a)–(c) state that shifts in technology should shift tenure contracts; this in turn differentially affects electoral and policy outcomes in areas where landlords are initially dominant. I test the effect of technological change using general elections data between 2002 and 2013 on voting outcomes and politicians’ assets, and data on public goods allocation.20 I estimate a regression of the form:

yjt = β1 + β2AgT echnologyjt + β3AgT echnologyjt × LL Dominatedj + νj + µpt + jt (5)

20The Pakistan Rural Household Survey data set does not cover all districts of the country. Thus I use national datasets for these regressions.

20 where yjt is the outcome of interest in area j (constituency or district) and year t;I include constituency (or district) as well as province-year fixed effects to account for time-invariant geographic heterogeneity and province-specific time trends, respectively. Below I describe the construction for the measures of agricultural technology and initial landlord dominance. Constructing a Measure of Exogenous Technological Change Agricultural technology and electoral outcomes are likely to affect each other and to be affected by underlying human capital and institutional characteristics of a region. I construct a measure of exogenous technological change using variations in suitability and availability of high-yielding-variety (HYV) seeds. An important feature of produc- tivity gains from HYV seeds allows me to construct a plausibly exogenous measure for agricultural productivity due to this technology. Foster and Rosenzweig 1996, among others, argue that the profitability of these seeds is heterogeneous across space because of (exogenous) differentials in local soil and climate conditions. I build on this aspect of productivity gains from HYVs and exploit spatial variation in HYV suitability in combination with time variation in the aggregate availability of HYV seeds to construct a local measure of productivity shock due to HYV seeds in area j and year t, for crop c.

SuitHjc is suitability for growing crop c in area j under hypothetical high-technology conditions (i.e., mechanized inputs, fertilizer, and irrigation), and SuitLjc is the same under hypothetically low-technology conditions (i.e., traditional inputs and rainfed).21 These suitability indices are constructed using a crop-growth model based on soil mois- ture, terrain, and climate characteristics for a geographic location and for each hypothet- ical technology level. Thus, the indices vary at geographic level, are time-invariant and capture the underlying geological conditions that determine a region’s suitability for cul- tivating a crop. These indices are at coordinate-level and aggregated to the constituency- or district-level depending on the unit of analysis in the regressions. HYV seeds are most profitable when mechanized inputs and irrigation are used (Shiva 1991); thus, the differ- ence between the two suitability indices, SuitDiffjc, captures the potential gains in area

21These data from FAO-GAEZ are explained in data description below.

21 j from using HYV seeds due to natural geological characteristics of area j.

The suitability index SuitDiffjc varies across space and higher suitability in any area should result in higher agricultural productivity as more HYV seeds become avail-

able. To measure availability of HYVs, I construct HYVcpt, which is the total volume of improved seeds for a crop distributed or supplied in a given province in any year. Demand and supply factors may drive the total volume of high yielding seeds for a certain province, to control for which I include a full set of province-year interactions and only use within province variation in geological factors affecting suitability for these seeds. This gives me a within province (district- or constituency-level) proxy for agricultural productivity that varies over time as the overall level of improved seeds changes.22 The constructed measure is an interaction between area-specific suitability for the high- technology and over- time availability of the technology at the overall province level: SuitDiffcj × HYVcpt gives a time- and area-varying measure of shock to agricultural productivity (see Figures 1–2). The estimation strategy compares districts or constituencies within provinces, and all regressions have area fixed effects for j and province-year fixed effects for each p, t. I show in table 7 that the constructed measure is significantly correlated with the actual yields by crop, allowing me to use it as a proxy for shifts in agricultural technology resulting from innovations in seeds. For any district, I use SuitDiffcj × HYVcpt, where c is the most widely grown crop in that district in 1980, to measure technological change at the region-time level.23 Because electoral outcomes and the productivity instrument are at the constituency level but actual yields are not available at constituency level, I run reduced-form regressions instead of 2SLS.24 Constructing a Measure of Landlord Dominance

22 Variation in HYVcpt is partly driven by the availability of these seeds from foreign producers in any year, making it less likely to be endogenous to any area-specific characteristics. I show in the appendix that demand factors as measured by annual rainfall do not have a strong correlation with HYVcpt (Table A2). Even if total HYVcpt is driven by political or economic determinants at the national or provincial level, it remains exogenous to local area-specific characteristics after controlling for province-year fixed effects. Adoption and usage of seeds at the local level is undoubtedly endogenous, which is why I use the total amount of the seeds at the aggregate, province level as a measure of overall availability of the technology, which is unrelated to local characteristics of a constituency or district. I always include province-year fixed effects in the regression analysis to account for any province-time variation that is correlated with total seeds available in any province in any year. 23By using the crop choice from an initial period (1980), I avoid confounding factors due to an endogenous crop choice. 24The smallest level for which I have actual yields is district; I estimate and report the 2SLS regressions using predicted district-level yields for the outcome available at district-level.

22 Agricultural technology inarguably affects politico-economic outcomes through a number of channels; the channel I intend to identify affects these outcomes differentially in regions where landlords are politically active, as I have shown, through patronage to tenants. Landlords’ political dominance as described in section 2 is a legacy of colonial times, thus I use historic data to assign areas that were initially landlord dominated. I use colonial district gazetteers, land settlement reports and documentation of chiefs and landed families to construct a district-level dummy variable LL Dominated, which captures a historical presence of prominent ‘jagirdars’ or landowners. Figure 5a in the appendix shows an illustration from one of the colonial gazetteers listing all prominent ‘jagirdars’ by subdistrict. Similarly, figure 5b shows an excerpt from ‘Chiefs of Punjab’ outlining the ‘jagirs’ held by one prominent chief. LL Dominated is assigned value 1 if a prominent ‘jagirdar’ is featured in either the district gazetteer or the compilation of chiefs. LL Dominated is assigned value 0 when the district gazetteer indicates no jargirdars in a district or when there are numerous small jagir holders in a district, signifying a lack of any prominent or influential landowner in the district. A comprehensive and detailed list of jagirs is not available for some districts—thus to maximize the data points and use all the districts in the analysis, I use the indicator variable instead of a continuous measure. Since the political and economic dominance of landlords was originally established during colonial times and is not a phenomenon caused by post-colonial policies, I clas- sify landlord dominated districts based on historical status.25 The prominent historical ‘jagirdars’ manifest the dominance that allows landowners to gain electoral support as captured in the theoretical framework and that I aim to quantify for the empirical results.

A quick analysis of the land census data of 2000 lends assurance that LLDominated is indeed correlated with presence of influential landlords in more recent times. Table 8 shows that the largest landlords (owning 150 acres or more) control a significantly higher share of agricultural land in historically landlord dominated districts and landlords are significantly more likely to win elections in these districts. At the same time percentage of landless agricultural households and rate of sharecropping is higher in these districts.

25Additionally, the historic data on landlords is surprisingly more comprehensive than any other recent data sources.

23 Thus, the historically LL Dominated districts, as classified by my original construction, have politically influential landlords in post-colonial times, causing these regions to be differentially affected by agricultural technology change that affects landlord-tenant rela- tions. The historical accounts suggest that land assignments or jagirs are likely not driven by any specific agenda other than to reward loyal locals in different parts of the empire. I confirm that landlord-dominated and other areas are balanced along observable historic characteristics like population density, land revenue per acre, and religious composition of

population (see Table A3 in the appendix). Furthermore, SuitDiffj is not systematically different across regions with and without dominant landlords; thus, landlord-dominated areas are not significantly different agriculturally (see Table 8). As expected, the landlord- dominated areas are more likely to have the zamindari (or landlord) system of tax collec- tion, as outlined in Banerjee and Iyer (2002, 2008). Area fixed effects in the regressions account for any systematic, time-invarying differences between LL Dominated and other areas.26

I first investigate the effect of exogenous technology change as measured by SuitDiffjt×

HYVpt on agricultural productivity and the rate of sharecropping tenancy. Next, I iden- tify the effect of exogenous technology change on landlords’ political success and elec- toral and public goods outcomes specifically in the historically landlord dominated areas. Agricultural productivity is correlated with changes in income and land distribution and structural transformation, which are likely to affect electoral outcomes. The interac- tion with LL Dominated thus helps to identify specifically the mechanism that operates through landlords’ ability to offer clientelism through tenancy relationships when agri- cultural technology changes. Thus, I estimate the following specification:

yjt = β1 + β2SuitDiffj × HYVpt + β3SuitDiffj × HYVpt × LL Dominatedj

+ νj + µpt + jt (6)

26Later I also include LL Dominated × Y ear fixed effects to account for differential trends in these areas.

24 The coefficient of interest is β3; if the agricultural technology lowers landlords’ political participation, β3 should be negative when the dependent variable is an indicator for a landowning politician or if the dependent variable is an indicator for low electoral competition. Similarly, lower political dominance among landlords also implies that β3 will be negative for measures of landlord-preferred public goods. I run robustness specifications with additional LL Dominated×Y ear and District× Y ear fixed effects to account for differential trends in agricultural tenure and electoral outcomes that were specific to historically landlord-dominated areas. Additionally, to account for the income effect of increased productivity, I run this specification replacing

SuitDiffjt × HYVpt with rainfall in j, t as a proxy for income. SuitDiffjt × HYVpt is a measure of a permanent improvement in land productivity, which can shift incentives for sharecropping. Rainfall measures, however, affect agricultural output in the short run, and although they capture variations in incomes of landlords and rural voters, they do not affect the incentives for sharecropping, which is the mechanism for my story. Data for Prediction 2 The data for agricultural tenancy and agricultural yields (district-level) and HYV seeds (province-level) covers the over 100 present-day districts and four provinces of the country and are obtained from issues of the annual Agricultural Statistics of Pakistan (1970–2013) and the decennial Agricultural Census (1970–2010). The data for politicians’ assets and voting outcomes at the electoral constituency-level comes from the Election Commission of Pakistan (a 2002 bill requires that all office holders declare all assets and liabilities). Election data cover the 2002–2013 period, during which three direct elec- tions have elected three sets of office holders. I use only politicians who were elected to the Provincial Assembly, the lowest level at which voters can elect their representatives. Constituencies are contained within districts; each district has several constituencies av- eraging 120,000 voters per constituency. For constructing the measure of technological change as described above, I use the Global Agro-Ecological Zones database of the Food and Agricultural Organization (FAO-GAEZ), which provides suitability indices by hypothetical technology level for all

25 , for a worldwide grid at a resolution of 9.25 x 9.25 km. I chose two extreme cases of the hypothetical technology levels—high (mechanized inputs and irrigation) and low (traditional inputs and rainfed). To match the FAO crop suitability data with electoral outcomes, I superimpose each suitability map (for both high- and low-technology) with political maps of Pakistan reporting the constituency and district boundaries. Next, I compute the average suitability values across all cells within the boundaries of every con- stituency or district, to give me constituency- or district-level measures of crop suitability. The historical data for classifying landlord-dominated districts comes from province and district gazetteers and land settlement reports compiled by the British government from the late 19th to the early 20th centuries. These sources include information on zamindars and jagirdars and land assignments made to them. In addition to these publications, colonial authorities also compiled an extensive list of notable, district-level jagirdars in the publication titled ‘The Punjab Chiefs’ (Griffin and Craik 1865). These sources are used in combination to obtain a district-level indicator for landlord dominance. The data on public goods are from the Pakistan Living Standards Measurement Surveys in 2006, 2008, 2010. Two separate governments were elected during the time frame covered by these surveys. This district-level survey asks respondents about their use of, availability of, and quality of various public services. Results for Prediction 2 To test prediction 2(a)–(c), I use specification (6). As mentioned before, the con- structed instrument for agricultural technology is correlated with significantly higher yield for different crops. If the technological change due to HYV seeds increases per acre yield relative to monitoring costs, we expect landowners to shift away from sharecrop- ping contracts towards self cultivation using hired wage labor (Eswaran and Kotwal 1982; Stiglitz 1974). Figure 3 shows a drastic fall in the rate of sharecropping tenancy over five decades. The regression results in Table 9 confirm that the technology shift due to HYV seeds indeed caused the rate of sharecropping to go down and the rate of self-cultivation to go up. The 2SLS estimates using predicted yield (instrumented by the constructed measure SuitDiff × HYV ) show that 0.25 ton/ha increase in yield lowers the rate of

26 sharecropping by 27%; these effects are large and significant.27 The next set of regressions (Table 10) looks at the landholdings of the elected members of the Provincial Assemblies and the degree of electoral competition in these elections. Since agricultural yield is not available at constituency level, the regressions show the reduced form effect of SuitDiff × HYV on the constituency level outcomes. There in no significant overall effect of technological change on electoral outcomes (Panel A of Table 10). However, technological change specifically within a landlord-dominated district leads to a lower likelihood that the winning politician is an agricultural landowner. Columns (2)-(6) of Table 10 (Panel B) shows that a lower probability of a landlord winning an election corresponds to improved electoral competition. This is shown using two alternate measures for electoral competition—a Herfindahl-Hirschman index for win- margin and alternately Lowcomp, a categorical variable for low competition, which is 1 when the winning margin is above the 20th percentile.28 Using the actual win-margin as the dependent variable or shifting the cut-off to classify Lowcomp yields similar results (see appendix table A4). The sum of the two coefficients shows that an increase in productivity corresponding to a 1 SD increase in SuitDiff × HYV lowers the winning probability of a landowner in historically landlord-dominated areas by 16 percentage points or by 23%. and lowers the likelihood of low competition by 33% in areas with historic landlord dominance.29 In more stringent specifications, I add LL Dominated × Y ear and District × Y ear fixed effects. These results, explained in section 5, ensure that the findings are not driven by differential trends in historically landlord-dominated areas or by other alternate mechanisms in the district, like growth in the non-agricultural sector or urbanization. Similarly, using rainfall as a proxy for income, I verify that the mechanism of rising incomes due to the productivity increase is not entirely responsible for the shift in political

27It could be changes in land ownership distribution due to increased productivity that are simultaneous with these trends in tenancy. However, the relatively stable distribution of land ownership over this time, as shown in figure 4, indicates that changes in the distribution of land ownership can only be partly responsible for shifts in the tenancy rates. 28The Herfindahl-Hirschman index is calculated by squaring the vote share share of each candidate competing in a constituency, and then summing the resulting numbers. Lower index corresponds to lower concentration in the electoral race, or lower competition. 29A 3 s.d increase in the measure of technological change corresponds to a 0.25ton/ha improvement in wheat yield. For reference, wheat yields increased from 0.84ton/ha in 1965 to over 2 ton/ha in 2000. 27 power and electoral competition. This further confirms that the alternate modernization channel cannot fully explain the effects I find. The outcome directly relevant to development is public goods; the agricultural technology shock affects the identity of winning politicians, who in turn allocate public goods. Thus, to estimate the impact of the exogenous technological shift on public goods, I use the SuitDiff × HYV measure from the year preceding the last election before the public goods data were collected. By doing so, I capture the impact of the agricultural technology shock through its effect on the identity of elected politicians who determine how public goods are allocated. The outcome I consider is the percentage of respondents from any district who report an improvement in the availability of a facility over the past year. First, I distinguish public goods categories by examining service use across land- owning statuses (Panel A in Table 11). To identify public goods that are specifically prioritized by large landowners, I regress a household’s self-reported frequent use of a facility on an indicator for whether the household’s land holdings are in the top 5th per- centile and an indicator for a landless or smallholder (owning less than 1 acre) household. Landowners are more likely to use roads, police stations and veterinary services and less likely to use a basic health unit or family planning services. I also distinguish publicly provided facilities in rural or landlord-preferred areas from those in urban areas, as the literature on favoritism and ‘core voter’ targeting (Cox and McCubbins 1986) suggests that landlords may target their voter bases by providing more facilities in rural areas. Panel B-D of Table 11 shows the regression results. Panel B suggests that ex- ogenous technological change causes improved basic health and worse family planning services. However, Panel C shows that specifically in landlord-dominated areas, health, family planning and drinking water services improve relative to other areas as agricultural technology changes. Of these services the first two are prioritized less by large landlords. The effect on other public goods is not significantly different in landlord dominated areas. Panel D helps unpack this finding by showing the effect of the agricultural technology shift and the resulting shift in political influence of landlords on public goods across rural

28 and urban areas. For almost all public goods, the coefficient for technology interacted with LL Dominated is significantly higher for urban areas. This implies an improvement in public goods in urban areas relative to rural areas. Basic health, family planning, and drinking water improve in landlord dominated areas, but are significantly better in urban areas relative to rural areas. On the other hand, the landlord-preferred public goods are overall lower in rural areas, and seem to be diverted toward urban areas. This is sugges- tive evidence that landlord politicians steer resources towards their preferred services and favor rural areas when providing all services. A shift in power away from the traditional rural elite results in a shift of resources towards urban areas.30 The results highlight that tenants who benefit from high transfers when traditional rural elites dominate politics stop receiving those benefits as clientelism plausibly shifts toward a different set of voters. Landless tenants seemingly fare poorly in the aftermath of technological change for several reasons, including low wages and rents, low electoral transfers, and diversion of public services away from rural areas. The results thus shed some light on the rural-urban inequality in developing economies.

5 Robustness

Regressions in section 4.1 use a difference-indifference strategy so I check for pretrends in tenancy contracts. I use an earlier panel data set by the International Food Policy Research Institute, which covers 12 rounds from April 1986 to September 1989 in a sub- sample of the villages in the Pakistan Rural Household Survey. This timeframe is useful, because it covers a non-democratic regime followed by a general election in 1988, however the data does not indicate the specific plots owned by politicians. In appendix figure A3, I plot the coefficient on each year dummy from a household-level regression of an indicator for sharecropping on year dummies, household (tenant) fixed effects, and survey round and season fixed effects. There are no specific trends in the rates of sharecropping or the inputs provided by the landlords during the non-democratic regime. However, as expected, after the election in 1988, the rate of sharecropping and the landlords’ payments

30Another data set for rural public goods confirms these results. Results are available on request.

29 for inputs increase. Since, I do not have information on the landlords’ political statuses in this dataset, I cannot verify which landlords cause the spike, as I do in the main regressions in section 4.1. I argue that the results in section 4.2 are not driven by differential growth paths in areas that are historically different, other than through the differential effects of variations in agricultural technology. I control for differential trends that are specific to areas with historic landlord dominance by adding LL Dominated × Y ear fixed effects. Results in columns 1–3 of Table A6 in the appendix show that the effect of technological change in landlord-dominated areas is robust to controlling for trends that are specific to the landlord-dominated areas that correlate with productivity improvements. Alternate trends in some areas could coincide with technology improvements. For example, changes in value of capital, rural-urban migration, or a shift in agricultural wages could lower the traditional landowner’s prominence relative to urban elites, thus decreasing their election wins and affecting electoral competition. For the previous re- sults to be biased, these trends must operate differently in landlord-dominated areas. The province-year fixed effects in the regression account for any province-specific trends in these factors, but I run a stringent specification by adding district-year fixed effects. The results in columns 4–6 of Table A6 in the appendix estimate the effect of differences in agricultural technology in constituencies within the same district and year. The co- efficients are lower due to the strict controls, but the effect is robust. These rigorous specifications confirm that the landowners’ weakened election success and the improved electoral competition cannot be entirely explained by alternate district-specific trends. Finally, the results may be driven by an ‘income effect’. As agricultural productivity improves, voters’ incomes improve and they change their voting behavior, leading to different electoral outcomes. To test for this effect, I run the same specification but replace the proxy for agricultural technology with rainfall as a proxy for income. Rainfall captures shocks to income, which are unrelated to permanent improvement in agricultural productivity. Table A5 in the appendix shows the results and highlights that the income effect is not entirely responsible for changes in response to exogenous improvement in

30 agricultural technology. There is a differential income effect in landlord dominated areas on electoral competition measured by Herfendhal index of vote shares, but the overall effect in these areas is zero. Thus, the electoral effect of agricultural technology change, via the cost of sharecropping to large landowners, is robust and operates specifically in landlord-dominated areas.

6 Concluding Remarks

The results show one way in which traditional landowning elites can perpetuate their political influence—by gaining the loyalty of their sharecropping tenants. Agricultural technological change lowers sharecropping and effectively alters the nature of paternalistic institutions. These findings have implications for landlords’ success in politics and the resulting electoral competition and public goods allocation. I study the exchange between tenants and landlords who are bound in contracts that link land, factor and electoral markets. I argue that landlords can improve their tenants’ income efficiently, thus gaining their electoral support. I find that landed elites, when in power, provide facilities that benefit farmers and their voter base. The paternalistic incentives of landlords are strongest when land productivity is low and the efficiency cost of sharecropping is low. Changing agricultural technology in a way that lowers share- cropping, undermines the paternalistic ties with tenants and eventually affects electoral outcomes and public goods. This development can influence the democratic process by supporting the transition from clientelism to a more effective and competitive democratic regime. However, this transition may not imply an absolute improvement in rural clients’ welfare. Their transfers remain low, and whereas some public goods improve, others are re-directed away from rural areas to urban areas. The transition in identity of elites can modify the extent and nature of elite capture and help to redistribute resources across society. Plausibly, a new form of clientelism emerges, where resources are redirected toward non-rural areas. These results speak to rural-urban inequality and rural-urban migration, which are common features in developing economies.

31 References

Acemoglu, Daron, Simon Johnson, and James A Robinson. 2000. “The Colonial Origins of Comparative Development: An Empirical Investigation.” . 2001. “Reversal of Fortune: Geography and Institutions in the making of the Modern World Income Distribution.” Acemoglu, Daron, Simon Johnson, James A Robinson, and Pierre Yared. 2008. “Income and Democracy.” The American Economic Review 98 (3): 808–842. Acemoglu, Daron, Tristan Reed, and James A Robinson. 2014. “Chiefs: Economic Development and Elite Control of Civil Society in Sierra Leone.” Journal of Political Economy 122 (2): 319–368. Agricultural Census of Pakistan (various issues). Pakistan Bureau of Statistics. Agricultural Statistics of Pakistan (various issues). Pakistan Bureau of Statistics. Alston, Lee J, and Joseph P Ferrie. 1999. Southern Paternalism and the American Wel- fare State: Economics, Politics, and Institutions in the South, 1865-1965. JSTOR. Anderson, Siwan, Patrick Francois, and Ashok Kotwal. 2015. “Clientelism in Indian Villages.” The American Economic Review 105 (6): 1780–1816. Baker, Aryn. 2008. “Landowner Power in Pakistan Election.” Time, February 13, 2008. http://content.time.com/time/world/article/0,8599,1712917,00.html. (ac- cessed on January 10, 2017). Baland, Jean-Marie, and James A Robinson. 2008. “Land and Power: Theory and Evidence from Chile.” The American Economic Review 98 (5): 1737–1765. . 2012. “The Political Value of Land: Political Reform and Land Prices in Chile.” American Journal of Political Science 56 (3): 601–619. Banerjee, Abhijit, and Lakshmi Iyer. 2005. “History, Institutions, and Economic Per- formance: the Legacy of Colonial Land Tenure Systems in India.” The American Economic Review 95 (4): 1190–1213. Banerjee, Abhijit V, and Lakshmi Iyer. 2008. Colonial Land tenure, Electoral Compe- tition and Public Goods in India. Harvard Business School. Banfield, Edward C. 1967. The Moral Basis of a Backward Society. Free Press. Bell, Clive, and TN Srinivasan. 1989. “Interlinked Transactions in Rural Markets: An Empirical Study of Andhra Pradesh, Bihar and Punjab.” Oxford Bulletin of Economics and Statistics 51 (1): 73–83. Boone, Catherine. 1994. States and Ruling Classes in sub-Saharan Africa: the Enduring Contradictions of Power. Cambridge University Press. Braverman, Avishay, and Thirukodikaval N Srinivasan. 1981. “Credit and Sharecropping in Agrarian Societies.” Journal of Development Economics 9 (3): 289–312. Braverman, Avishay, and Joseph E Stiglitz. 1982. “Sharecropping and the Interlinking of Agrarian Markets.” The American Economic Review 72 (4): 695–715. . 1986. “Cost-sharing Arrangements under Sharecropping: Moral hazard, In- centive Flexibility, and Risk.” American Journal of Agricultural Economics 68 (3): 642–652.

32 Brenner, Robert. 1976. “Agrarian Class Structure and Economic Development in pre- industrial Europe.” Past & present, no. 70:30–75. Brockett, Charles D. 1992. “Measuring Political Violence and Land Inequality in Central America.” American Political Science Review 86 (01): 169–176. Burchardi, Konrad, Selim Gulesci, Benedetta Lerva, and Munshi Sulaiman. 2017. “Moral Hazard: Experimental Evidence from Sharecropping Contracts.” Working paper. Bursztyn, Leonardo. 2013. “Poverty and the Political Economy of Public Education Spending: Evidence from Brazil.” Unpublished Manuscript. Chanock, Martin. 1985. “Law, Custom and Social Order: The Colonial Experience in Malawi and Zambia.” Chubb, Judith. 1982. Patronage, Power and Poverty in Southern Italy: a Tale of Two Cities. Cambridge University Press. Cox, Gary W, and Mathew D McCubbins. 1986. “Electoral Politics as a Redistributive Game.” The Journal of Politics 48 (02): 370–389. Dell, Melissa. 2010. “The Persistent Effects of Peru’s Mining Mita.” Econometrica 78 (6): 1863–1903. Dixit, Avinash, and John Londregan. 1996. “The Determinants of Success of Special Interests in Redistributive Politics.” the Journal of Politics 58 (04): 1132–1155. Engerman, Stanley L, and Kenneth L Sokoloff. 2005. “Colonialism, Inequality, and long-run Paths of Development.” Eswaran, Mukesh, and Ashok Kotwal. 1985. “A Theory of Contractual Structure in Agriculture.” The American Economic Review 75 (3): 352–367. Galor, Oded, Omer Moav, and Dietrich Vollrath. 2002. “Land Inequality and the Origin of Divergence and Overtaking in the Growth Process: Theory and Evidence.” Hebrew Univ. Dept. of Econ. Working Paper, no. 2002-11. Gazdar, Haris. 2009. “The Fourth Round, and why they Fight on: An Essay on the History of Land and Reform in Pakistan.” PANOS South Asia, Collective for Social Science Research, Karachi. Glaeser, Edward L, and Andrei Shleifer. 2002. “Legal Origins.” Quarterly Journal of Economics, pp. 1193–1229. Grabowski, Richard. 2002. “East Asia, Land Reform and Economic Develop- ment.” Canadian Journal of Development Studies/Revue canadienne d’´etudesdu d´eveloppement 23 (1): 105–126. Hussain, Asaf. 1979. Elite politics in an Ideological State: the case of Pakistan. Dawson. IIASA/FAO. 2012. Global Agro-ecological Zones (GAEZ v3.0). IIASA, Laxenburg, Austria and FAO, Rome, Italy. IMF. 2015. Fiscal Monitor, April 2015. International Food Policy Research Institute (IFPRI). 2015. Pakistan Panel Survey, 1986-1991. http://hdl.handle.net/1902.1/11189. Joshi, PC. 1970. “Land Reform in India and Pakistan.” Economic and Political Weekly, pp. A145–A152. 33 Kaplan, Seth. 2013. “Power and Politics in Pakistan.” Norwegian Peacebuilding Resource Centre, April. Keefer, Philip, and Razvan Vlaicu. 2002. “Clientelism, Credibility and Democracy.” World Bank: Washington, DC Processed. Khan, M. 1995. “State Failure in Weak States: a critique of New Institutional Eco- nomics.” Harris, J. Khan, Mushtaq. 2010. “Political Settlements and the Governance of Growth-enhancing Institutions.” London: SOAS. Khemani, Stuti. 2004. “Political Cycles in a Developing Economy: Effect of Elections in the Indian states.” Journal of development Economics 73 (1): 125–154. Kitamura, Shuhei. 2013. “Loyalty and Treason: Theory and Evidence from Japan’s Land Reform.” Working paper. Labonne, Julien. 2016. “Local Political Business Cycles: Evidence from Philippine Municipalities.” Journal of Development Economics 121:56–62. La Porta, Rafael, Florencio Lopez-de Silanes, and Andrei Shleifer. 2008. “The Economic Consequences of Legal Origins.” Journal of economic literature 46 (2): 285–332. Larreguy, Horacio A. 2013. “Monitoring Political Brokers: Evidence from Clientelistic Networks in M´exico.”Working paper. Lizzeri, Alessandro, and Nicola Persico. 2004. “Why did the Elites Extend the Suffrage? Democracy and the Scope of Government, with an Application to Britain’s ‘Age of Reform’.” The Quarterly Journal of Economics, pp. 707–765. Logan, Carolyn. 2013. “The Roots of Resilience: Exploring Popular Support for African Traditional Authorities.” African Affairs 112 (448): 353–376. Mamdani, Mahmood. 1996. Citizen and Subject: Contemporary Africa and the Legacy of Late Colonialism. Princeton University Press. Martinez-Bravo, Monica, Nancy Qian, Yang Yao, et al. 2011. “Do Local Elections in non-Democracies Increase Accountability? Evidence from Rural China.” Technical Report, National Bureau of Economic Research. Mason, T David. 1986. “Land Reform and the Breakdown of Clientelist Politics in El Salvador.” Comparative Political Studies 18 (4): 487–516. Massy, Charles Francis. 1940. “Chiefs and Families of Note in the Punjab.” Government Printing Press, Lahore 19091:1739–68. Medina, Luis Fernando, and Susan Stokes. 2002. “Clientelism as Political Monopoly.” the 2002 Annual Meetings of the American Political Science Association Conference. Boston. August, Volume 29. Merry, Sally Engle, Martin Chanock, Sally Falk Moore, Joan Vincent, Peter James Nelligan, Robert J Gordon, Mervyn J Meggitt, Yash Ghai, Robin Luckham, Francis Snyder, et al. 1991. Law and Colonialism. Mezzera, Marco, Safiya Aftab, and Sairah Yusuf. 2010. “Devolution Row: An Assess- ment Of Pakistan’s 2001 Local Government Ordinance.” Migdal, Joel S. 1988. Strong Societies and Weak States: State-society Relations and State Capabilities in the Third World. Princeton University Press. 34 Migdal, Joel Samuel, Atul Kohli, and Vivienne Shue. 1994. State power and Social forces: Domination and Transformation in the Third World. Cambridge University Press. Nichter, Simeon. 2011. “Electoral Clientelism or Relational Clientelism? Healthcare and Sterilization in Brazil.” Available at SSRN: https://ssrn.com/abstract=1919567. Nunn, Nathan. 2009. “The Importance of History for Economic Development.” Annu. Rev. Econ. 1 (1): 65–92. Pakistan Living Standards Measurement Surveys (2006-7, 2008-9, 2010-11). Pakistan Bureau of Statistics. Pakistan Rural Household Survey (2000–1, 2003–4). Pakistan Institute of Development Economic (PIDE) and World Bank. Perkins, Dwight H. 2013. East Asian Development. Harvard University Press. Piliavsky, Anastasia. 2014. Patronage as Politics in South Asia. Cambridge University Press. Powell, John Duncan. 1970. “Peasant Society and Clientelist politics.” American Political Science Review 64 (02): 411–425. Rashid, Shaikh Muhammad. 1985. “Land Reforms in Pakistan.” Social Scientist, pp. 44–52. Roberts, Richard, and Kristin Mann. 1991. Law in Colonial Africa. Heinemann Portsmouth, NH. Robinson, JA, and T Verdier. 2003. “The Political Economy of Clientelism.” Harvard University. Shefter, Martin. 1994. “Party and Patronage: Germany, England, and Italy.” The state: Critical concepts 3:103–143. Stiglitz, Joseph E. 1974. “Incentives and Risk sharing in Sharecropping.” The Review of Economic Studies 41 (2): 219–255. Stokes, Susan C, Thad Dunning, Marcelo Nazareno, and Valeria Brusco. 2013. Brokers, Voters, and Clientelism: The Puzzle of Distributive Politics. Cambridge University Press. The Economist. 2013. “Pakistan’s Waning Feudalism: Gone with the wind.” May 18, 2013. http://www.economist.com/news/asia/ 21578104-wrestlers-son-overthrows-landed-gentry-gone-wind. Accessed January 10, 2017. Vicente, Pedro C, and Leonard Wantchekon. 2009. “Clientelism and Vote Buying: Lessons from Field Experiments in African Elections.” Oxford Review of Economic Policy 25 (2): 292–305. Willmott, C. J., and K Matsuura. 2001. Terrestrial Air Temperature and Precipitation: Monthly and Annual Time Series (1950-1999). Zaidi, S Akbar. 2014. “Rethinking Pakistan’s Political Economy.” Economic & Political Weekly 49 (5): 47.

35 7 Figures

Figure 1: FAO Suitability for Wheat: High Input Level and Irrigation (Left), Low Input Level and No Irrigation (Right)

Figure 2: HYV Availability across Years and Crops for Different Provinces

BALUCHISTAN N.W.F.P 2 15 1.5 10 1 5 .5 0 0

PUNJAB 2 15 1.5 10 1 5 .5 0 0

Improved VarietiesImproved hectar) 10,000 per (tonnes of Seeds 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 year Wheat Cotton Maize Rice

Graphs by province

36 Figure 3: Distribution of Sharecropping Rate (percent of area under sharecropping) across Districts .06 .04 Density .02 0 0 20 40 60 80 Perc. of Total Area under Sharecropping (%)

1960 1972 1980 1990 2000 2010

Source: Relevant Annual Census of Agriculture

Figure 4: Distribution of Land Concentration across Districts .04 .03 .02 Density .01 0 20 40 60 80 Land Concentration Index

1972 1980 2000 2010

Note: The Land Concentration index is calculated as in Brockett 1992: (% of land with largeholders + % of land with smallholders/average size with smallholder) Source: Relevant Annual Census of Agriculture

37 Figure 5: Historic Landlord Records

THE DERA ISMAIL KHAN DISTRICT. 591

SARDAR ALA WARDI KHAN OF HAZARA.

H^ji Khan.

I Khudadad Khan.

I Sher Mahomed Khan.

I

1 1 1 1 1 1 Haidar Ashak ALA WaRDI Yusaf Ali Alayar Abdula Sadat Ali Hasain Khan. Khan. Khan. Khan. Khai Khan. Khan.

i

I I I I I I All Mahomed Sadik Yar Ghulam Fida Hasain Akbar Ali Mahomed Muilza. Hasain. Khan. Khan. Khan. Khan.

Ala Wardi Khan's father, a Kazalbash, accompanied Shah Shujah to India. He himself obtained a command of sowars locally raised in the First Afghan War, and was after- wards made Rasaldar in the 17th Irregular Cavalry. His

38 regiment behaved well during the Mutiny, He led a bril- liant charge against the mutineers of the 9th Cavalry, and slew their leader Wazir Khan with his own hand. His services w^ere also conspicuous in the Mahsud Expedition of i860. He was given a jagir of Rs. 2,400 in 1862, of which one-fourth was in perpetuity in lieu of cash allowances, without prejudice to his military pension of Rs. 970 per annum.

The Sardar has acquired about one hundred and seventy acres by purchase in Mauzas Hazara and Bahal in the Bhak- kar Tahsil, and he has received a grant of three thousand five hundred acres at Pak Patan, Montgomery. He exercises magisterial powers within the limits of his jagir. He is a member of the District Board, and has lambardari rights in two villages, besides being Zaildar of the Bahal Ilaka. He is decorated with three War Medals, and he wears the Orders of Me:it and of British India.

His eldest son is the Rasaldar-Major of the 13th Bengal Lancers. Many of his relatives are serving in the Army. (a) Gazetteer of Sindh (b) Chiefs and Families of Note in Punjab (Massy 1940) 8 Tables

Table 1a: Summary Statistics: Data for Testing Prediction 1

Pre Election Post Election Total Number of Plots (Owned and Leased) 709 877 Number of plots leased in 461 544 % of total plots 0.65 0.62 Number of plots leased in on sharecropping contract 361 398 % of total plots 0.51 0.45 % of leasedin plots 0.78 0.73 Number of leased in plots with politician landlords 46 48 % of leasedin plots 0.10 0.09 Plots leased in where landlord and tenant contracted after preelection survey round na 176 % of leasedin plots 0.32 Number of tenant households 484 484 Number of households who sharecrop 327 298 % of total tenant households 0.68 0.62

Table 1b: Summary Statistics: Data for Testing Prediction 1

LL is not Politician LL is Politician Plot is sharecropped 0.606 0.904 (0.489) (0.296) Plot area (kanals) 48.75 28.89 (79.27) (20.31) Supervise (times/season) 19.52 9.524 (28.56) (16.58) Labor Manager (times/season) 7.505 25.14 (22.30) (29.85) Canal Irr Available 0.844 0.904 (0.363) (0.296) % area under wheat 0.459 0.502 (0.373) (0.394) % area under cotton 0.201 0.251 (0.319) (0.378) % area under rice 0.364 0.285 (0.389) (0.387) Number of tenants 14.50 63.85 (77.03) (143.7) For sharecropped plots, Landlord Share of: Output 53.66 50.95 (13.15) (6.696) Harvest/Thresher 39.25 33.94 (34.34) (29.19) Fertilizer 51.04 46.18 (18.59) (15.67) Seeds 38.56 31.18 (34.73) (28.84)

Number of plots 911 94 Note: Summary statistics for all leased in plots in both survey rounds.

39 Table 2: Landlord Politicians and Sharecropping Contracts: Plot Level Regression 1 2 3 Dependant variables is indicator for a sharecropped plot (1) (2) (3)

PoliticianLL_PostElection 0.103* (0.0618) Politician x Post 0.153** 0.108 (0.0694) (0.0717) Politician x Post x New LL 0.210 (0.215) Politician -0.0655 -0.0631 (0.0530) (0.0552) Post -0.0354 -0.0398 -0.00688 (0.0327) (0.0325) (0.0321) Constant 0.749*** 0.753*** 0.761*** (0.0552) (0.0547) (0.0539)

Observations 1,005 1,005 1,005 R-squared 0.049 0.052 0.065 Politician x Post + Politician 0.0870 0.0449 pvalue 0.190 0.505 Politician x Post + Politician x Post x newLL 0.318 p-value 0.126 Notes: Standard errors in parenthesis are clustered at household level. The sample consists of all rented in plots of tenant-households. All regressions include tenant fixed effects, plot level controls including plot soil, depth, area, crops grown, and seasons leased. Additional controls include landlord and tenant landholdings. NewLL indicates plots leased from new landlords, i.e. the tenant contracted with the landlord after the preelection survey round.

40 Table 3: Landlord Politicians and Sharecropping Contracts: Tenant Level Regression 1 2 3 4 (1) (2) (3) (4)

HH is a Sharecropped/ HH is a Sharecropped/ sharecropper Total no. of plots sharecropper Total no. of plots

Tenant of Politician x Post 0.170*** 0.0711** 0.150** 0.0699** (0.0597) (0.0352) (0.0593) (0.0317) Tenant of Politician x Post x New LL 0.120 0.00667 (0.187) (0.128) Post -0.0705*** -0.0766*** -0.0705*** -0.0766*** (0.0235) (0.0190) (0.0236) (0.0190) Constant 0.676*** 0.600*** 0.676*** 0.600*** (0.0112) (0.00894) (0.0112) (0.00894)

41 Observations 968 968 968 968 R-squared 0.021 0.035 0.022 0.035 Mean of Dependant Var Preelection 0.674 0.598 0.674 0.598 Tenant of Politician x Post + Tenant of Politician x Post x NewLL 0.270 0.0766 p-value 0.135 0.545 Notes: All regressions include household fixed effects, with standard errors clustered at households level. The sample consists of all household that are tenants in atleast one round of the survey. NewLL indicates households that contract with a politician landlord after the preelection survey round. Table 4: Landlord Politicians and Input Costs: Plot Level Regression

1 2 3 4 5 6 7 (1) (2) (3) (4) (5) (6) (7) LL share of cost of: LL share of Land Preparation Supervise Labor Manager Fertilizer Labor/Tractor Seeds Harvest/Thresher Output (times/season) (times/season)

Politician x Post 9.217*** -0.760 2.273 22.74** 5.571** 21.87*** -27.19*** (3.405) (7.950) (5.983) (8.925) (2.243) (7.734) (9.698) Politician x Post x NewLL 0.915 -5.748 1.451 -30.03 -22.06* -26.47** -17.37 (5.573) (12.97) (7.030) (21.94) (11.64) (11.32) (16.21) Politician -4.017 -3.809 -4.610 -8.440 -1.436 -10.22* 12.74 (3.371) (5.123) (5.488) (6.541) (1.373) (5.977) (8.187) Post -12.19*** -8.356* -8.167** -10.52** -4.542*** -22.13*** -6.412* (2.052) (4.581) (3.764) (4.800) (1.377) (4.628) (3.634)

Observations 684 684 684 684 684 683 642 R-squared 0.250 0.078 0.130 0.260 0.135 0.296 0.291

42 Mean of Dependant Var Preelection 52.03 25.59 37.82 43.75 54.66 29.67 15.48 Politician x Post + Politician 5.199 -4.569 -2.337 14.30 4.135 11.66 -14.45 p-value 0.113 0.530 0.634 0.0259 0.00709 0.0685 0.0464 Politician x Post + Politician x Post x NewLL 10.13 -6.509 3.724 -7.295 -16.49 -4.594 -44.56 p-value 0.0553 0.580 0.612 0.743 0.167 0.627 0.00969 Notes: Standard errors in parenthesis are clustered at household level. The sample consists of all sharecropped plots of tenant-households. All regressions include tenant fixed effects, plot level controls including plot soil, depth, area, crops grown, and seasons leased. Additional controls include landlord and tenant landholdings. NewLL indicates plots leased from new landlords, i.e. the tenant contracted with the landlord after the preelection survey round.

Table 5: Landlord Politicians and Input Costs: Tenant Level Regression 1 2 3 4 5 6 7 (1) (2) (3) (4) (5) (6) (7) LL share of cost of: LL share of Land Preparation Supervise Labor Manager Fertilizer Labor/Tractor Seeds Harvest/Thresher Output (times/season) (times/season)

Tenant of Politician x Post 5.305* 1.552 -2.757 16.11** 2.273*** 12.52** -21.91** (2.935) (6.759) (5.082) (7.750) (0.812) (5.009) (8.516) Tenant of Politician x Post x New LL 2.273 -11.79 10.23** -32.54** -15.95 -14.63 -11.84 (2.755) (14.86) (4.755) (15.53) (13.13) (11.80) (8.694) Post -7.577*** -6.097*** -7.470*** -16.90*** -2.318*** -18.50*** -8.502*** (1.012) (2.203) (1.795) (2.448) (0.811) (2.554) (2.011) Constant 53.76*** 22.01*** 34.38*** 43.41*** 54.88*** 28.15*** 17.45*** (0.471) (1.032) (0.832) (1.151) (0.373) (1.167) (0.985)

Observations 504 452 462 488 482 504 450 R-squared 0.191 0.039 0.086 0.175 0.057 0.189 0.165 43 Mean of Dependant Var Preelection 54.50 21.46 35.76 42.78 55 28.64 18.62 Tenant of Politician x Post + Tenant of Politician x Post x NewLL 7.577 -10.24 7.470 -16.43 -13.68 -2.109 -33.75 p-value 0 0.452 4.45e-05 0.238 0.299 0.852 0 Notes: All regressions include household fixed effects, with standard errors clustered at household level. The sample consists of all households with atleast one sharecropped plot in both survey rounds. The outcomes are averaged across all sharecropped plots of a household in any survey round. NewLL indicates households that contract with a politician landlord after the preelection survey round. Table 6: Landlord Politicians and Canal Irrigation Access 1 2 3 4 (1) (2) (3) (4) Canal Irr Dependant Variable: Available Canal Irr used Canal Irr used Canal Irr used

Politician x Post x Sharecropped 0.0119 0.824*** 0.857*** 0.840*** (0.0552) (0.0770) (0.0788) (0.0744) Politician x Post -0.0286 -0.768*** -0.781*** -0.797*** (0.0565) (0.0736) (0.0789) (0.0764) Politician x Sharecropped -0.124* -0.0212 -0.00213 0.0232 (0.0683) (0.0683) (0.0632) (0.0654) Politician 0.128** 0.0123 -0.0275 -0.0529 (0.0548) (0.0644) (0.0606) (0.0634) Post 0.0806* -0.156** -0.159** -0.141** (0.0485) (0.0639) (0.0674) (0.0667) Politician x Post x Sharecropped x New LL 0.256 (0.242) Observations 911 911 834 834 R-squared 0.113 0.195 0.181 0.191 All leased in All leased in All leased in All leased in plots with plots with Sample plots plots irrigation irrigation Mean of Dependant Var Preelection 0.849 0.855 0.855 0.855

Politician x Post x Sharecropped + Politician x Post -0.0167 0.0555 0.0762 0.0434 p-value 0.550 0.128 0.0270 0.198 Notes: Standard errors in parenthesis are clustered at household level. The sample consists of all leasedin plots of tenant-households. All regressions include tenant fixed effects, plot level controls including plot soil, depth, area, crops grown, and seasons leased. Additional controls include landlord and tenant landholdings. NewLL indicates plots leased from new landlords, i.e. the tenant contracted with the landlord after the preelection survey round.

44 Table 7: Actual Yields and Constructed Instrument

(1) (2) (3) (4) (5) (6) Dependent Variable is Actual Annual Yield of: Wheat Cotton Rice Maize Maxcrop Maxcrop

Suit Diff x HYV (by crop) 0.0514*** 0.0593*** 0.285*** 0.853*** 0.503*** 0.470*** (0.00997) (0.0152) (0.0260) (0.315) (0.0998) (0.116) Suit Diff x HYV x LL Dominated 0.197 (0.181)

Observations 1804 955 1110 1162 1804 1804 Notes: The regressions use FAO suitability indices for each crop. Suit Diff x HYV is the interaction of the difference in FAO crop suitability index under high and low technology with the province level availability of HYV in any year. The LHS variable is district-level annual yield from 1979 onward. Standard errors clustered at district level. All regression include district FE and province-year FE. Columns (5) and (6) use the yield and suitability of the most commonly grown crop in any district.

45 Table 8: Summary Statistics: Data for Testing Prediction 2(a)-(c)

LL Dominated=0 LL Dominated=1 Difference Unit Source Agriculture Perc. under sharecropping 23.2 31.0 7.8*** District Ag. Census (17.0) (18.5) Percentage of area under large landowners in 2000 12.8 25.0 12.2*** District Ag. Census (13.1) (19.6) Percentage of agricultual households with no land in 2000 57.1 60.8 3.7 District Ag. Census (17.8) (17.2) HYV Suitability 0.20 0.22 0.02 9.25 × 9.25 km FAO (0.23) (0.19) Wheat Yield in 1965 (ton/ha) 0.84 0.68 -0.16 District Ag. Census (0.27) (0.16) Wheat Yield in 1980 (ton/ha) 1.44 1.38 -0.06 District Ag. Census (0.39) (0.47) Wheat Yield in 2000 (ton/ha) 2.18 2.00 -0.18 District Ag. Census (0.73) (0.70)

Electoral Provincial Assembly member (MPA) is landowner 0.65 0.79 0.14*** Constituency Elec. Comm (0.477) (0.410) Agr. Land Declared by MPA (Acres) 97.2 329 231.8** Constituency Elec. Comm (331) (1018) Herfendahl Index of Vote Shares 0.36 0.38 0.02** Constituency Elec. Comm (0.137) (0.157) Win Margin 17.6 21.2 3.6** Constituency Elec. Comm (17.88) (20.22) Low Competition 0.20 0.27 0.07*** Constituency Elec. Comm (0.40) (0.44) Public Goods Police Station 86 79 -7*** District PSLM (19) (22) Vet 82 77 -5*** District PSLM (19) (22) Basic Health Unit 49 43 -6*** District PSLM (28) (28) Family Planning 69 62 -7*** District PSLM (21) (25)

46 Table 9: Land Tenure in response to Technology Shift

(1) (2) (3) Perc of Tot Area Perc of Tot Area Perc of Leased Area Dependant Variable: Sharecropped (SC) Self-cultivated (W) Sharecropped (SC) A: OLS with actual yield Yield -10.15*** 9.039*** -16.29*** (2.826) (2.913) (3.620)

B: OLS with Constructed Instrument Suit Diff x HYV -1.356*** 1.192*** -1.688*** (0.380) (0.388) (0.312)

C: 2SLS with predicted Yield Yield -24.80*** 21.98*** -29.93*** (6.565) (5.754) (10.14)

Observations 244 244 244 Mean of dependant variable 23.0 70.7 76.3 Notes: Standard errors clustered at district level. All regressions use district-level data from Agricultural Censuses of 1970 to 2010 and include district and province-year FE. Suit Diff x HYV is the interaction of the difference in FAO crop suitability index under high and low technology with the province level availability of HYV in any year.

47 Table 10: Electoral Outcomes in Landlord Dominated Areas 1 2 3 4 5 6 (1) (2) (3) (4) (5) (6) Winner owns Low Dependant Variable: AgLand Agland (Acres) H-dhal Index Winmargin (%) Competition # of Candidates Panel A: Suit Diff x HYV 0.001 10.410 -0.006* -0.259 -0.006 0.491*** (0.014) (21.196) (0.003) (0.464) (0.013) (0.147) Panel B: Suit Diff x HYV 0.003 11.993 -0.006* -0.223 -0.004 0.468*** (0.014) (21.140) (0.003) (0.465) (0.013) (0.148) Suit Diff x HYV x LL Dominated -0.110** -140.945* -0.030* -2.916 -0.150*** 1.880 (0.055) (77.458) (0.016) (2.046) (0.049) (1.143) 48 Suit Diff x HYV + Suit Diff x HYV x LL Dominated -0.107 -129 -0.0361 -3.139 -0.154 2.347 p-value 0.0537 0.110 0.0229 0.131 0.00224 0.0395

Mean of Dependant Variable 0.70 116.02 11.87 45.13 0.699 116 Observations 1,568 1,402 1,568 1,568 1,568 1,568 Notes: All regressions include constituency and province-year FE and standard errors are clustered at constituency level. Column 2 includes officeholders who declare exact acerage of agricultural land in their household's possession. Low Competition is a dummy indicating the winmargin in any election race is higher than 32% (which is the 80th percentile). Suit Diff x HYV is the interaction of the difference in FAO crop suitability index under high and low technology with the province level availability of HYV in any year. LL Dominated is a dummy indicating historic landlord dominance. Table 11: Public Goods usage in Landlord Dominated Areas 1 2 3 4 5 6 (1) (2) (3) (4) (5) (6) Family Basic Planning Drinking Health Unit Center Water Road Veterniary Police Panel A: Public goods useage Landless/Smallholder 0.00285 -0.00673** 0.000522 0.00571 -0.0405*** -0.0184*** (0.00707) (0.00309) (0.00216) (0.00509) (0.00518) (0.00434)

Largeholder -0.0523** -0.0127 0.00656 0.0322** 0.0506** 0.0505** (0.0253) (0.00823) (0.00487) (0.0140) (0.0206) (0.0197)

Observations 73944 73944 73944 73944 73944 73944 Public Goods Change in Quality Panel B: Suit Diff x HYV 6.713* -13.67*** -3.392 -7.342 -7.303 -9.500 (3.879) (5.056) (4.702) (4.983) (4.943) (6.848)

Panel C: Suit Diff x HYV 5.000 -17.77*** -5.331 -9.214* -8.881 -9.188 (4.026) (6.070) (5.092) (5.522) (5.401) (7.528)

Suit Diff x HYV x LL Dominated 11.78* 25.83** 13.28** 13.02 10.80 -1.291 (6.814) (11.81) (6.543) (9.184) (7.001) (9.843)

Panel D: Suit Diff x HYV x LL Dominated 9.451 23.40** 11.51* 8.211 2.934 -2.495 (6.363) (11.69) (6.736) (10.05) (5.745) (11.50)

Suit Diff x HYV x LL Dominated x Urban 2.974** 3.681 1.593* 4.986*** 8.306*** 1.811 (1.430) (2.215) (0.946) (1.252) (2.890) (1.621)

Observations 411 411 411 411 411 411

Overall effect in rural areas + + + - - - Overall effect in urban areas + + + + + - Notes: Panel A shows coefficients from individual level regressions where outcome is a dummy for frequent usage of each facility. Standard errors are clustered at district level. Largeholder is a dummy indicating the respondent is in the top 5th percentile of the landowners in the sample. Smallholders are households with less than 1 acre of land. Data is obtained from the Pakistan Living Standards Measurement Surveys of 2006. Panels C-D regressions are at district-region-year level (region corresponds to rural/urban) and control for district, rural status, province-year fixed effects. Standard errors are clustered at district level. Suit Diff x HYV is the interaction of the difference in FAO crop suitability index under high and low technology with the province level availability of HYV in any year. Suit Diff x HYV is taken for the year corresponding to the last election before the survey data was collected. LL Dominated is the historic measure of landlord dominance. Data is obtained from the Pakistan Living Standards Measurement Surveys between 2006 and 2010.

49 FOR ONLINE APPENDIX

A Figures and Tables

Figure A1: Area controlled by Large Landowners .08 .06 .04 Density .02 0 0 20 40 60 80 100 Percentage of Area Held by Top 1% of Landowners (%)

Notes: The charts plots the distribution of the percentage of total area in a subdistrict held by the top 1% of landowners as a whole. The top 1% comprises landowners who own more than the 99th percentile of the distribution of holdings of all landowners within a subdistrict.

Figure A2: Divide between Large and Small Holders 100 80 60 40 20 Smallholder or Landless Households (%) Households or Landless Smallholder 0 0 20 40 60 80 100 Percentage of Area Held by Top 1% of Landowners (%)

Notes: The charts plots the share of area held by the top 1% in a subdistrict, and the percentage of agricultural households who are landless or own 5 or less acres of land. The dotted lines mark the average value for the 2 axes across the subdistricts. The sub-districts in the top left will have a higher degree of land concentration.

50 Figure A3: Pre-trends in Rate of Sharecropping (Left) and Inputs provided by Landlord in Sharecrop- ping Contracts (Right)

Election Year Election Year .06 600 .04 400 .02 200 0 0 -.02 -.04 -200 1986 1987 1988 1989 1986 1987 1988 1989

Notes: The left panel shows the coefficient on each year dummy from a regression of tenant households, regressing whether the household is a sharecropper as a function of household, round, season and year fixed effects. The right panel shows the coefficient on year dummies from the same regression with the share of inputs provided by landlord as the explanatory variable.

Table A1: Plot Characteristics 1 2 (1) (2) Dependant variable is indicator for: NewLL LL Politician

LL Politician -0.184*** (0.0633) Plot Area -0.000235 -0.000342 (0.000328) (0.000262) Plot Slope is: Slight -0.105 -0.0198 (0.0902) (0.0525) Moderate -0.0663 0.0333 (0.0459) (0.0377) Steep -0.139 -0.0641 (0.120) (0.0717) Terraced -0.125* 0.0316 (0.0652) (0.0245) Plot Soil is: Clay -0.0176 0.0312 (0.0411) (0.0295) Sandy -0.0142 -0.0515 (0.126) (0.0587) Other -0.158 0.108** (0.116) (0.0494) Crop Composition: Wheat -0.0936* -0.0454 (0.0508) (0.0477) Rice -0.105 -0.0918 (0.0745) (0.0595) Maize 0.0403 -0.0166 (0.139) (0.0571) Cotton -0.119* -0.0430 (0.0612) (0.0537) Sugarcane -0.156*** -0.0414 (0.0432) (0.0367) Observations 1,005 1,005 R-squared 0.299 0.047 Notes: Observations are plot level for all leased in plots. NewLL indicates plots leased from new landlords, i.e. the tenant contracedt with the landlord after the preelection51 survey round. Regressions include tenantd fixed effects Table A2: HYV and provincial demand

(1) (2) (3) (4) (5) Improved Seeds (tonnes/ha) Wheat Cotton Paddy Maize Overall

Provincial Annual Rainfall -0.0406 0.000467 -0.00347 -0.000610 -0.00692 (0.0428) (0.00545) (0.00384) (0.00160) (0.00837)

Observations 164 164 164 164 1,136 Notes:Regressions show total volume of improved variety seeds supplied at province-year level and their correlation with average provincial level rainfall in any year. All regressions include fixed effects for province and year; Column (1) also includes fixed effects for crop.

Table A3: Historic Land Assignments and Observables

Dependant Variable is LL Dominated==1 (1) (2) (3) (4) (5)

Population Density -0.000223 0.000118 (0.000303) (0.000350) Land Revenue per Acre -0.000147 -0.000249 (0.000171) (0.000193) Percentage Muslim Population 0.00353 -0.00336 (0.00379) (0.00341) Share of Zamindari Villages 0.267 0.199 (0.149) (0.156) Observations 158 151 131 83 131 Notes: All regressions have province fixed effects. Regressions are at historic sub-district level. Share of zamindari villages are the share of villages in each subdistrict where the landlord tenure was implemented by the British (Banerjee and Iyer 2005).

52 Table A4: Electoral Competition in Landlord Dominated Areas (using alternate cut-offs for Low Competition)

Dependant Variable is a dummy for Low Competition using various cut-offs: (1) (2) (3) (4) (5) (6) (7) (8)

Suit Diff x HYV -0.002 -0.050 0.001 -0.001 -0.004 -0.029 -0.000 0.004 (0.016) (0.124) (0.014) (0.142) (0.013) (0.168) (0.010) (0.193) Suit Diff x HYV x LL Dominated -0.156** -0.858** -0.112 -0.696 -0.150*** -1.295*** -0.137*** -1.305*** (0.071) (0.422) (0.068) (0.443) (0.049) (0.460) (0.049) (0.471) p-value of sum of coefficients 0.0279 0.0348 0.107 0.121 0.00224 0.00472 0.00557 0.00703

Mean of Dependant Variable 0.267 0.267 0.236 0.236 0.201 0.201 0.176 0.176 Cut-off value for winmargin used to determine Low Competition (%) 23.85 23.85 25.00 25.00 27.78 27.78 30.00 30.00 Estimation OLS PROBIT OLS PROBIT OLS PROBIT OLS PROBIT Observations 1,568 1,568 1,568 1,568 1,568 1,568 1,568 1,568 Notes: All regressions include constituency and province-year FE and standard errors are clustered at constituency level. Suit Diff x HYV is the interaction of the difference in FAO crop suitability index under high and low technology with the province level availability of HYV in any year. LL Dominated is a dummy indicating historic landlord dominance. 53 Table A5: Electoral Outcomes in Landlord Dominated Areas (with Rainfall as proxy for Income) 1 2 3 4 5 (1) (2) (3) (4) (5) Dependant Variable: Agland=1 H-dhal Index Winmargin (%) Low Competition # of Candidates

Annual Rainfall 0.022 0.017*** 0.446 -0.000 -0.271 (0.016) (0.004) (0.696) (0.009) (0.173) Annual Rainfall x LL Dominated -0.019 -0.014** 0.978 -0.037 1.053*** (0.024) (0.006) (0.922) (0.032) (0.405)

p-value of sum of coefficients 0.908 0.616 0.156 0.883 0.0880

Observations 1,568 1,568 1,568 1,568 1,568 Notes: All regressions include constituency and province-year FE and standard errors are clustered at constituency level. Mean for annual rainfall is 3.87 and SD is 2.85. Low Competition is a dummy indicating the winmargin in any election race is higher than 32% ( 80th percentile). LL Dominated is a dummy indicating historic landlord dominance. Table A6: Electoral Outcomes in Landlord Dominated Areas (additional FE)

(1) (2) (3) (4) (5) (6) Dependant Variable: Agland=1 H-dhal Index Low Competition Agland=1 H-dhal Index Low Competition

Suit Diff x HYV 0.002 -0.006* -0.004 0.068* 0.006 -0.001 (0.014) (0.003) (0.013) (0.041) (0.006) (0.022) Suit Diff x HYV x LL Dominated -0.082 -0.032* -0.204*** -0.082 -0.029* -0.230*** (0.065) (0.019) (0.063) (0.052) (0.018) (0.082)

p-value of sum of coefficients 0.224 0.0467 0.00124 0.797 0.208 0.00581

LL Dominated x LL Dominated x LL Dominated x Additional Fixed Effects Year Year Year District x Year District x Year District x Year

Observations 1568 1568 1568 1568 1568 1568 54 Notes: All regressions include constituency and province-year FE and standard errors are clustered at constituency level. Low Competition is a dummy indicating the winmargin in any election race is higher than 32% (which is the 80th percentile). Suit Diff x HYV is the interaction of the difference in FAO crop suitability index under high and low technology with the province level availability of HYV in any year. LL Dominated is a dummy indicating historic landlord dominance. B Data Appendix

B.1 Pakistan Rural Household Survey—Identifying Politicians In the first round of the PRHS data, I use the following question asked to all tenants about each landlord they rent land from: ”Is your landlord also a politician?” I code the variableP lothasP oliticianLandlord as 1 if the response to this question is positive. During the time of the survey, there no elected officials in office due to the 1999 coup. Based on conversations with the organization responsible for conducting these surveys, this question identifies landlords who held a political office before the coup. In the second round tenants are asked: ”Has any member of the landlord’s household held any political, religious or hereditary office or other official position?”31 They are also asked what kind of office the landlord’s household member holds. I code P lothasP oliticianLandlord as 1 if the tenant answers ”Yes” to the former question, and the office held by the landlord’s household member is an elected position.

B.2 Though experiment on landlords’ control To understand the electoral advantage of the landowners, I do a brief thought experi- ment here: The election I study is a provincial election held in 577 constituencies across the country, each of which constitutes on average 120,000 voters. Given an average 44% turnout, a candidate needs on average 26,000 votes to win for sure. The landlords de- scribed as large in the village survey have about 100 tenants each. This is a reasonable number, given the large holder category in the census have average of 500 acres each; leasing out 5 acre plots would get us the same figure. Using household size of 4 adults, we attain that every landowner described as large, would be employing over 400 voters. An average constituency has 30 large landholders; a collaboration among them, some of who could be in the same family, would constitute 12,000 voters, which is over 40% of the votes needed for absolute majority.

B.3 FAO and High Yielding Seeds Data Food and Agriculture Organization of the United Nations (FAO) and the International Institute for Applied Systems Analysis (IIASA) have developed the Agro-Ecological Zones (AEZ) methodology, which provides crop suitability and potential yield across the world under various combinations of input-levels and water supply (IIASA/FAO 2012). From FAO-GAEZ database I obtain the suitability for any crop for each grid point under two extreme levels of technological inputs used in production (low and high) and two extreme levels water availability (rain-fed and irrigated). When the level of technology is assumed to be low, agriculture is not mechanized; it uses traditional cultivation and does not use nutrients or chemicals for pest and control. When the level of technology is high instead, production is fully mechanized, it uses improved or high yielding varieties and ”optimum” application of nutrients and chemical pest, disease and weed control. For any grid point in the FAO data, Suit Hc is suitability for crop c with high technology (mechanized inputs fertilizer, and irrigation), while Suit Lc is the same with low tech- nology (traditional inputs and rain fed). For any region j I take the average for all grid points within the region, and take the difference (Suit Hcj − Suit Lcj) as a measure of suitability for HYV seeds for crop c in geographic region j.

31The survey organization confirmed that the version of this question asks if the landlord’s household member currently holds this position. If the variable identifies landlord households who ever held a political position, then the point estimates provide a lower bound for the effect of active politicians.

55 HYVpt is the total volume of seeds provided in any province, in any year as reported by government statistical authorities. These seeds are imported both by government agencies as well as private companies and distributed to farmers. HYVpt is a measure of the availability of the technology. The interaction of the exogenous suitability for HYV and the availability of HYV (Suit Hcj − Suit Lcj) × HY V pt is a measure of exogenous technological change. From district level data on crop shares in 1980 I take the most commonly grown crop for j and use the constructed measure for that crop.

C Mathematical Appendix

I set up the basic land market and election market, allowing landlords to run in an election. Landlords first choose whether or not to run - if not running, landlords makes choices about agricultural contracts. If running, landlords make simultaneous choices about agricultural contracts and their electoral platform. After elections happen, winning candidates deliver promises and everyone receives their payoff.

C.1 Set up of Electoral Market. Following Persson and Tabellini (2008), I assume two candidates denoted by j = {A, B} and a continuum of voters, mass N. Voters get utility from private income and public goods offered by a candidate, as well as from their ideological affinity for the candidate. j j Each candidate j offers two types of public goods G1 and G2, from which the entire population benefits, and private transfers f j which can only benefit one voter at a time. G2 is the public good prioritized by landlords, thus distinguished from G1. Voters get utility U(·) from private income and H1(·) and H2(·) from each of the public goods. σi the ideological bias of voter i towards candidate B, is assumed to be uniformly distributed 1 1 such that σi ∼ U(− 2φ , 2φ ). Aggregate uncertainty is caused by an ideological shock 1 1 δ ∼ U(− 2ψ , 2ψ ), which shifts the votes in favor of B. Candidates know voters’ ideological distribution but do not observe any individual voter’s ideological preference, and hence, do not make voter-specific transfers. In the analysis below, I assume the candidates make the same private transfer promise to all voters. Allowing the candidate to target some voters does not change the analysis in any way; this is precisely because the transfers are not truly targetable since voters cannot reveal their type before the election. I assume credibility (voters believe the candidates will fulfill their promises) and truthful voting (voters vote for the candidate whose win results in the highest utility for the voter), with no strategic behavior on part of the voters. Voter i will vote for candidate A if: A A A B B B U(f ) + H1(G1 ) + H2(G2 ) > U(f ) + H1(G1 ) + H2(G2 ) + σi + δ (7) Given the distribution of σ and δ, we get the expression for the total vote share of A, 1 A B 1 1 A B πA = 2 + φ[W − W − δ] and A’s probability of win P r(πA > 2 ) = 2 + ψ(W − W ), j j j j where W = U(f ) + H(G1) + H(G2). These expressions are symmetric for candidate B. Note that the winning probability is simply a function of the total utility that any candidate offers to the voters relative to the competitor. All candidates pay a fixed cost of running C. The winning candidate gets non-pecuniary rents from office denoted by χ and funding from central government of R, which is used to fulfill the candidates promises of G1,G2 and f conditional on winning. χ is interpreted as the bureaucratic connections (the benefits are large but not immediate) available to an office holder as opposed to monetary benefits that could be used in combination with R to fulfill promises.32 However, I allow candidates to choose some P from their private

32While it is possible to use bureaucratic connections to benefit voters, e.g. through offering public sector 56 income Π to pay for f and G conditional on winning.33 Candidates maximize expected pay-off, subject to the feasibility of the payments f and G. Thus, candidate j’s problem is defined by: Ä j jä 1 j 1 max χ + Π (Γ) − P P r(πj > ) + Π (Γ)P r(πj < ) − C (8) P,G1,G2,f,Γ 2 2

j j j j s.t. R + P = G1 + G2 + Nf

and 0 ≤ P j ≤ Πj Πj(Γ) is the private income of candidate j of which she chooses P j to spend on election promises. Γ is the set of choices which determine Πj. When the profit maximizing choices are independent of the policy choices (P, f, G1,G2), these can be made separately. The total money available to fulfill campaign promises is thus R + P j. If she wins, the politician’s payoff is the office rents χ and the profits left after paying for f and G. If she loses, she does not have to pay anything to voters and gets no political rents, so the payoff in the case of electoral loss is just private profits Πj. C is the cost of running. Candidates maximize expected pay-off, subject to the feasibility of the payments f and G.

C.2 Set up of Land Market Normalizing plot size to 1, the production function for land as is θf(e, x, τ) where θ is a random variable with E(θ) = 1, e and x are per acre labor and physical inputs, τis a measure of productivity. I assume landlords are risk neutral, profits dented by Π, while tenants are risk averse and have utility U, which is a function of income and effort. Tenant’s reservation utility is denoted U. Following the literature on contractual choice in agriculture I consider three kinds of rental contracts: fixed wage contract, a fixed rent contract, or a sharecropping contract. 1. Fixed Wage Contract (W): The risk neutral land owner chooses optimal e and x at given prices to maximize profits. Since workers offered a fixed wage have an incentive to shirk, this contracts entails a supervision cost denoted c. The landlord’s profits ΠW are given by solving: ΠW = max[θf(e, x, τ) − px − w − c] x,e,w

st. U(w, e) = U

where w is payment to the worker, pis the price of x, and c is the landlord’s cost of supervision, which can alternately be interpreted as the opportunity cost of being present on the farm to prevent shirking. The landowner takes up all the risk in this case, and since she is risk neutral the inputs are applied until the marginal product equals the price.

employment (Robinson and Verdier 2002), I abstract from that dimension of office rents and restrict the ability of the politician from using χ towards voters; this makes the problem tractable, although including this ability will not change results in any substantial way. 33More formally, candidates should use their wealth, which consists of accumulated wealth and profits; since I am not interested in the effect of candidates’ initial wealth, and always assume candidates are equally wealthy, I restrict my focus on just candidates’ profits as they may differ for landlord candidates. Including assets in the budget does not change the marginal problem, so having assets or not having them is effectively equivalent as long as they are the same for both candidates 57 2. Fixed Rent Contract (R): The land owner offers her land to a tenant and allows him to farm it in return for a fixed fee r. In this case, the tenant solves:34

max[EU(θf(e, x, τ) − px − r, e)] x,e

Given a fixed rental rate r tenant chooses (e(r), x(r)); the landlord sets the max- imum r such that the incentive compatibility constraint is satisfied i.e. r∗ is such that

EU(θf(e(r), x(r), τ) − px(r) − r∗, e(r)) = U¯

where U¯ is the reservation utility of tenants. The landlords’ profits are given by ΠR = r∗

Monitoring is unnecessary since the tenant has incentives to supply optimal inputs. However, the risk averse tenant also assumes all the risk in this case; thus if uncer- tainty or risk aversion is high the optimal rent in an incentive compatible contract may be too low. The sharecropping contract deals with this problem, by allowing the landlord assume part of the risk associated with the farm output.

3. Sharecropping Contract (SC): This contract is typically given by (α, β) where α is the output share and β is the cost share of the tenant. Conversely, the landlord receives 1 − α of the output produced by the tenant, and pays 1 − β of the cost of the physical input x, which the tenant supplies. The tenants problem can be written as:

maxEU(αθf(e, x, τ) − βpx, e) e,x

The landowner chooses (α, β) to get optimal profits ΠSC

ΠSC = max(1 − α)θf(e, x, τ) − (1 − β)px α,β

st. EU(αθf(e, x, τ) − βpx, e) ≥ U¯ and (e, x) ∈ arg maxEU(αθf(e, x, τ) − βxp) e,x

In this case the tenant supplies the inputs,35 but doesn’t need to bear all the risk. This contract may dominate the fixed rent one in the absence of insurance markets. It dominates the fixed wage contract if supervision costs are high. However, the agency problem poses a tradeoff; since the tenant consumes only a fraction of the output, he has lesser incentive to exert the optimal inputs.36

34For simplification, the tenancy contracts assume no limited liability. 35In theory, as well as in practice, the landlord may pay some cost to prevent the sharecropper from shirking. I assume away the monitoring cost in the sharecropping case, or analogously assume that it is cheaper to monitor the sharecropper relative to the wage worker; this will be the case given that the sharecropper is payed partly in terms of the output and his incentives to shirk are thus lower. 36Another rationale for having the sharecropping contract is the absence or imperfection of some factor markets, e.g. markets for management, supervision or family labor (Bell and Zusman, 1979, Eswaran and Kotwal 1985). If the tenants’ competitive advantage is in supervision (family labor) and landlords’ is in managerial ability, the sharecropping contract allows pooling of skills while providing incentives. 58 The contractual literature studies the conditions under which any of the above contracts may be optimal (Cheung 1969, Eswaran and Kotwal 1985, Stiglitz 1974). In each case the tenant gets at least his reservation utility U¯; the landlord chooses the contract which maximizes her payoff subject to this participation constraint.37 The tradeoffs between incentives, monitoring costs and risk sharing can lead to one contractual arrangement dominating the other. The contract choice can be summarized as follows:

1. If monitoring is costly and tenants are risk averse, landlord prefers sharecropping, SC

2. If monitoring is costly, but risk aversion is low, the landlord prefers fixed rent, R

3. If monitoring is cheap landlord chooses wage contract, W

At any level of productivity τ, there is an optimal contract which yields profits Π?:

Π?(τ) = max{ΠW (τ), ΠSC (τ), ΠR(τ)}

ΠW ,ΠF and ΠSC are the maximized profits from the wage contracts, fixed rent con- tracts and sharecropping contract, respectively, for a unit plot size (given above). For a landlord with total land L,38 the total profits are given by LΠ?. For simplicity I assume homogeneous plots, the optimal contract is one of the three, and is used to farm all plots by a profit maximizing landowner. All the intuitions follow through if the assumption is not imposed.39

C.3 Technical Change and Optimal Contract Consider a generalized contract of the form (α, J), where α is the tenant’s output share and J is a fixed payment to the tenant. α > 0 and J = 0 correspond to a pure share- cropping contract, α = 1 and J < 0 correspond to a fixed rent contract, while the case with α = 0 and J > 0 represents a pure wage contract. Given any such contract the tenant chooses e to solve (I am abstracting from the physical input here for simplicity, and denote the production function as θq(τ, e)):

maxEU(αθq(τ, e) + J, e) e We get the following choices for any α > 0 1 EU2 qe = − (9) αρ EU1 where ρ = EU1g . The landlord solves: EU1 max E[(1 − α)θq(τ, e) − J] α,J

s.t U(αθq(τ, e) + J, e) ≥ U

and e ∈ argmaxU(αθq(τ, e) + J, e) e

37 Different plots may have different optimal contracts depending on factors like the type of crop, the level of technology, the development of markets, social factors, as well as a combination of these factors. 38Equivalent to L plots 39With heterogeneous plots, the choice for contract will not be discreet. Instead, the landlord will choose the share of her land to cultivate under each type of contract. In other words, for each level of τ the landlord will choose the optimal (T ∗,R∗) to maximize profits given by Π?(τ) = max{(L−T −R)ΠW (τ)+T ΠSC (τ)+RΠR(τ)}. T,R 59 The solution is given by setting J ∗ such that the worker gets exactly U. Then the rental ∗ share is given by: α∗ = 1 − θq+Jα . θqeα Suppose I write the production function more specifically as Q(τL, τE) where L is land and E is total labor. The per acre production function is given by τq(e), where increase in τ represents a Hicks-neutral technical change and q is concave.40 Similarly, writing the production function as Q(L, τE) or Q(τL, E) represents a labor and land augmenting technical change parameter, respectively. I discuss how contracts and landlords profits shift as τ rises under each kind of technical change. With a factor neutral change, the production function is τq(e), the rental share is not sensitive to changes in τ, while J is decreasing in τ. ∗ Proof. Note α∗ = 1 − θτq+Jα and J ∗ is defined by U(αθτq(e) + J ∗, e) = U. Then θτqeeα ∗ ∗ ∗ Jα = −ρτq, can be substituted into the expression for α to show that ατ = 0. Similarly ∗ Jτ = −αρq < 0 It can also be shown that in the case of a land saving change, the rental share must go up. Proof. Writing the production function as Q(L, τE) and using CRS gives the per acre 1 EU2 production function q(τe). The input choice of the tenant is then defined by qe = − ατρ EU1 and output share is given by α∗ = 1 − q+Jα = 1 − q−ρq and is increasing in τ. τqeeα τqeeα Using similar argument it can be shown that a labor saving change will decrease the optimal rental share. In other words a technical change that increases labor per acre, will cause the land owner to provide greater incentives by increasing the output share and shifting towards fixed rental, or providing closer supervision, shifting to the wage contract. dΠW,R,SC To see how the optimal contract shifts with technical change, I calculate dτ . First note, that due to the incentives and risk aversion, the input provision are highest in the wage contract, followed by fixed rent, and lowest in the sharecropping contract. Thus: qW > qR > qSC dΠW W Consider first the Hicks neutral change. Using envelope theorem we have dτ = q . To dΠSC see how this compares with dτ I write the landlord’s problem as maximizing the sum of utilities max (1 − α)θτq(e) − J + EU(αθτq(e) + J, e) e,x,α,J s.t U(αθτq(e) + J, e) ≥ U 1 EU2 and qe = − ατρ EU1 Invoking the envelope theory again, we can ignore the effect of τ through (α, J, e), and dΠSC SC SC SC dΠW we have dτ = (1 − α)q + αq E(U1.θ) < q < dτ since E(U1.θ) = EU1.Eθ + Cov(U1, θ) < EU1Eθ < 1. The first inequality in the last expression is due to risk aversion, which has Cov(U1, θ) < 0; the second inequality is also because of risk aversion dΠR R dΠW (for risk neutral agent EU1 = 1). Similarly, dτ = q E(U1.θ) < dτ , if the tenant R SC SC R dΠ R dΠ q −q EU1θ is risk averse. However, = q E(U1.θ) if α SC . Thus, if the dτ R dτ R q (1−EU1θ) optimal contract is sharecropping, technical change (when land saving) will increase α, and eventually shift the optimal contract to fixed rent. With fixed cost of monitoring c, further technical progress shifts the optimal contract to wage, W . c can be imaginably proportional to the change in technology, particularly if the new technology requires more careful monitoring, the landlord will need to provide better dΠW W incentives to the worker. Suppose cτ > 0, then dτ = q − cτ , then the optimal contract is not necessarily wage one after technical progress.

40This generalizes to two inputs.

60 C.4 Timing The timing of the model is as follows: 1. Landlord chooses whether or not to run

2. If running, landlord chooses (f, G1,G2,P ) and tenure contract type; competing candidate chooses (f,G1,G2,P ) 3. If not running, landlord only makes the contract decision for farming; candidates choose( f, G1,G2,P ) 4. Election happens

5. Production happens

6. Winner delivers promises, payoffs are realized I consider two cases: 1) Case 1: Baseline case where two non-landlords compete. 2) Case 2: I allow landlord to run in election. In this case, I assume the members of the landowning oligarchy act as a single entity, which I will refer to as the landlord candidate. I assume log functional form for U(f) and H(G) to get closed form expressions for the policy platforms. I am interested in analyzing how the policy platform and electoral outcome is different in these cases; and also how these change in response to shifts in land productivity τ. Before solving the electoral equilibrium, I show how landowners can make private transfers to sharecropping tenants.

C.5 Landlord can make electoral transfers to sharecroppers cheaply Suppose a candidate, who owns land, wants to offer a private transfer to a voter who is a sharecropping tenant. She can offer a direct lump-sum transfer γε or alternately offer to lower the tenant’s cost share by βε, such that it is monetarily equivalent to γε. However, lowering the cost share induces the tenant to apply higher inputs and effort on the farm, which is partly internalized by the landlord who shares the output. I show that:

Lemma 1 It will be cheaper to lower cost share of the tenant than offering a lump sum transfer if ex-post the landlord would want the tenant to apply more input. Proof. Rewriting the equations from the SC contract above, the tenant’s problem is: maxEU(αgτf(e, x) − βpx, e) x,e The land owner’s problem is: max(1 − α)gf(τ, e, x) − (1 − β)px α,β st. U(αgf(τ, e, x) − βpx, e) ≥ U¯ and (e, x) ∈ argmaxU(αgf(τ, e, x) − βpx), e,x Suppose the landlord wants to offer a private transfer γε. She can alternately offer to lower the cost share by βε such that it is monetarily equivalent (at existing level of inputs (e, x)), i.e βεpx = γε. Thus βε is equivalent to γε if inputs stay unchanged with the change in β. Now, if the tenant can change inputs then by revealed preference, he is indifferent or better off than receiving γε. If landlord is weakly better off relative to (Π − γε) with cost share=β − βε, then lowering the cost share is cheaper for landlord relative to directly offering γε. 61 To ensure the the landlord is not worse off compared to (Π − γε) we must need the following condition to hold: (1 − α)τ{f(e + eε, x + xε) − f(e, x)} − (1 − β)qxε − βεq(x + xε) ≥ −γε = −βεqx or (1 − α)τ{f(e + eε, x + xε) − f(e, x)} ≥ {(1 − β)(xε) + βεxε}q ≥ 0 That is, the net cost to landlord of changing the cost share, after the tenant alters the effort e and input x exerted in response to the new cost share, is lower than the lump sum transfer γε. eε and xε represent the change in inputs applied by tenant when the cost share is changed. Ignoring the double differential the LHS can be reorganised and written as: (1 − α)τ{f(e + eε, x + xε) − (1 − β)(xε + x)q} − {(1 − α)τf(e, x)} − (1 − β)xq} ≥ ∂Π∗ (1 − α)τ{f(e, x + xε) − (1 − β)(xε + x)q} − {(1 − α)τf(e, x)} − (1 − β)xq} = ∂x ∂Π∗ Where ∂x is the rate of change of the profits Π with respect to x evaluated at the optimal x∗ chosen by the tenant. It will be cheaper to lower than offering a lump sum transfer if ∂Π∗ ∂x ≥ 0. i.e. ex post the principal would want the agent to use more input. L 1−β We must check if that is the case. The landlord’s optimal x is given by fx = 1−α q, but the tenant chooses x such that f SC = βp , where ρ = E((U1g) < 1 for a risk averse tenant. x αρ EU1 L SC 1−β βp Now, fx < fx ⇔ 1−α q < αρ . Substituting α = 1/2, the most common rate for output qρ share observed in the data, the last condition is true if β > β = p . Given an optimal chosen contract (α∗, β∗), I define η as the cost incurred by landlord for lowering the tenant’s cost share to β1 such that it raises his utility equivalent to a unit ∗ ∗ monetary transfer. Thus, β satisfies: R β dU dβ = R 1 U . Then η = R β − dΠS dβ. From 1 β1 dβ 0 1 β1 dβ Lemma 1, η < 1 for a range of β values.

C.6 Equilibrium I solve for the equilibrium by backward induction. The payoffs (steps 4-6) to landlord and tenants are described in the setup of the contracts, while the voters’ and candidates’ payoff is described in the election model setup.

C.6.1 Solving the candidate problem in the case of non-landlord candidate (Step 3) Consider the problem of a non-landlord politician j given by (8), which can be simplified to:

Ç å 1 −j max (χ − P ) + ψ(U(f) + H1(G1) + H2(G2) − W + Π − C (10) P,G1,G2,f 2

st. R + P = G1 + G2 + Nf

and 0 ≤ P ≤ Π

where W −j is the voter welfare promised by the competitor. Π denotes the maximized profits, where the optimizing decisions can be made independently of the electoral deci- sions. Assuming log functional form for U, H1 andH2, the first order conditions are the following (λ is the Lagrange multiplier on the budget constraint)

f:(χ − P )ψU 0(f) = λN

62 0 G1:(χ − P )ψH1(G1) = λ

0 G2:(χ − P )ψH2(G2) = λ

G We get G1 = G2 = G and f = N ; substituting the platforms into the budget constraint R+P gives G = 3 , which can be use to write the winning probability, w, in terms of P .

1 R+P comp w = 2 + ψ(3log( 3 ) − logN − W ). To get the optimal P , I just take first order conditions with respect to P : i.e. solve max(χ − P )w + Π − C, to get: P

dw (χ − P ) dP + w(−1) = 0 The cost of increasing P is higher expenditure on election promises, conditional on win- ning. Thus the marginal cost of P is just unity times the probability of win, w, and the marginal benefit is the increased chance of getting the office rents, (χ−P ). Thus, the can- 3ψχ−wR didate equates the marginal benefit to the marginal cost, to get the optimal P = 3ψ+w , where w is given above.41 In case 1, with identical candidates the problem is symmetric, both candidates have the same platform and winning probability is equal to one half.

C.6.2 Solving the candidate problem in the case of landlord candidate (Step 2) I account for the landlord’s ability to efficiently transfer utility to a tenants (as shown in section C.5) by assuming that offering an extra dollar as private transfer to a tenant costs the landlord η < 1. I denote transfers specific to tenants by ft and to non-tenants f−t. Denote the mass of sharecroppers by T , which the landlord chooses.42Additionally, the landlord’s payoff consists of direct utility from G1, which she chooses directly if elected and is given by K(G1). In the case of a land owner with land L, the private pay-off contract contract is given by LΠ + K(G1), where LΠ are total farm profits depending on the choice of contract and K represents the land owners private benefit function from

G1. Non-landlord candidate’s offers are denoted (ft, f−t,G1,G2). Assuming the tenants’ ideological preferences are distributed in the same way as the total voters, the landlord politician’s problem can be simplified to: max (χ − P + K(G ))w + Π + (1 − w)K(G ) − C (11) ? 1 1 T,Π ,P,G1,G2,ft,f−t

R + P = G1 + G2 + (N − T )f−t + T ηft (12) 1 ψ w = + ÄTU(f ) + (N − T )U(f ) + NH (G ) + NH (G ) − NW −jä (13) 2 N t −t 1 1 2 2 0 ≤ T ≤ L (14) 0 ≤ P ≤ Π (15) Π = Π?(L − T ) + T ΠSC (16) (12) is the budget constraint, w is the probability of win. (14) and (15) show that the landlord can hire only up to L (total land-size) tenants, and cannot spend more than her

41With sufficiently large χ the candidate sets P = Π. 42T = L if Π∗ = ΠSC . Given that sharecropping allows landlord to make transfers efficiently, the landlord may choose to have T > 0 plots under sharecropping, even when sharecropping is not the optimal contract. 63 private income on campaign promises. (16) gives the total farm profits of the landowner, which will depend on the total acreage under each kind of contract. The non-landlord

competitor’s problem is as in section C.6.1; she only chooses {P , ft, f−t,G1,G2}. First order conditions for {ft, f−t,G1,G2} leads to proposition 1:

Proposition 1 If landlord runs in the election:

(a) Landlord selects policies such that G1 > G2 and G2 < G2 = G, i.e. she over provides her preferred public good. If T > 0 , (b) Landlord offers higher private transfers to sharecropping tenants, ft > f−t. Thus at any level of ideological preference, a tenant is more likely to vote for the landlord candidate relative to another voter with the same ideological preference. (c) The landlord’s vote share and probability of winning exceeds the competitor’s, all else equal. Proof. The first order conditions with respect to the electoral platform ft, f−t,G1,G2 gives 0 0 0 0 0 U (ft) = ηU (f−t) < U (f−t) and U (f−t) = NH2(G2). The former expression leads G2 to the conclusion that ft > f−t. In the log utility case this leads to: f−t = N and f−t 1 1 0 ft = . The first order condition with respect to G1 gives − = wK (G1), where η G2 G1 0 G2 K (G1) > 0 is the marginal benefit for landlord of G1. Thus, G1 = > G2. 1−wG2KG The optimal choice of P is given by: dw dG (χ − P + K(G )) + w(−1) + wK 1 = 0 (17) 1 dP G dP Because the landlord politician also gains from the additional benefit of imposing her preferred public good if she wins, the marginal benefit of P is higher for the landlord. With sufficiently high office rents, both the landlord and non-landlord candidate set P = Π. In this case, the total budget of the landlord candidate and an equally wealthy competitor is the same. R+Π G From the solution of step 3 in the timing, we have G = 3 and f = N for the non- landlord competitor. From the budget constraint and the expressions for the landlord’s platforms, it follows that G1 + 2G2 = R + Π = 3G. Since G1 > G2, it must be the case that G2 < G2 = G. Suppose the landlord offers ft = f−t = f and G1 = G2 = G. The landlords cost of the electoral platform will be:

1 R+Π 1 R+Π 2 (N − T ) 3 N + ηT 3 N + 3 (R + Π)

< R + Π Thus when T > 0, the landlord can always promise more and have a higher vote share. We need to check if this will also be true in equilibrium. The landlord winning probability will be greater than the competitors iff

1 2 + ψ{(T ln ft + (N − T ) ln f−t + ln G1 + ln G2) − (N ln f + ln G + ln G)} > 1 2 + ψ{(N ln f + ln G + log G) − ((T ln ft + (N − T ) ln f−t + ln G1 + ln G2))}

⇔ N ln G2 + ln G1 + ln G2 − T ln η > N ln G + ln G + ln G (18) 0 Taking logs of the expression of G1 in terms of G2 gives ln G1 = ln G2−ln(1−wG2K (G1)). Thus the above condition is true if 64 0 (N + 2) ln G2 − T ln η − ln(1 − wG2K (G1)) > (N + 2) ln G (19) 0 R+Π 1−wG2K (G1) Using the budget constraint G1 + 2G2 = R + Π gives G2 = 3 2 0 ; 19 is true 1− 3 wG2K (G1) iff: T N + 2 2 − ln η − ln(1 − wG K0(G )) > − ln(1 − wG K0(G )) N + 1 N + 1 3 2 1 2 1 Since η < 1, landlord offers more than her competitor and will have a higher vote share in equilibrium when T > 0. Landlord also chooses T . The first order condition for T is given by:43 ψ (χ − P + K(G ) − K(G))(U(f ) − U(f )) = Π? − ΠSC (20) N 1 t −t That is, the candidate chooses T so the marginal benefit of each tenant, (which the increased chance of getting the payoff from winning) equates the marginal cost, which is the extra farm profits per plot she could make if she chose the optimal contract. If Π? = ΠSC 44, the solution is a corner one given by T = L. If Π? 6= ΠSC , the landlord candidate may choose T > 0 as given by (20). Using this and the result from Lemma 1, I propose:

Proposition 2 Landlord politicians, who have incentive to offer private transfers, are more likely to offer sharecropping contracts, and offer contracts more favorable to tenants (by paying higher cost share).

The RHS of (20) is increasing in U(ft) − U(f−t), which is increasing in η (the higher the η the larger the difference between the transfers made to tenants versus non tenants). Thus, T is increasing in η. The RHS is also increasing in K(G1) − K(G), i.e. a larger marginal benefit of G1 to the landlord implies she would want more tenants. Similarly, the larger Π? − ΠSC , the lower the optimal T . This leads to proposition 3:

Proposition 3 When sharecropping is optimal the landlord sets T = L. Otherwise, the landlord candidate sets 0 < T ≤ L as long as Π? − ΠSC is small. In this case, T is larger: (a) the cheaper it is to make transfers to sharecroppers (b) larger the marginal payoff of ? SC G1 (c) smaller the Π − Π . When Π? − ΠSC is large T = 0

C.6.3 Landlord choice for running (Step 1) The landlord will run as long as the expected pay off from running exceed that of not running. i.e. ∗ E[(χ − P + K(G1))w + Π + (1 − w)K(G) − C] > LΠ

C.6.4 Effect of Technological Progress - τ It can be shown that ΠSC is lower relative to ΠF and ΠW with technological change (Eswaran and Kotwal 1985, Stiglitz 1974, Bardhan and Srinivasan 1971). In other words technological change causes a shift away from share tenancy to either fixed rent or to fixed wage contracts, depending upon the type of technological change. Labor augmenting technical change increases the capital to effective labor ratio making supervision relatively

43Assuming optimal T is interior 44The optimal contract for any landowner is SC

65 less costly, leading to wage contracts. On the other hand a land augmenting technological change increases the effective labor per acre and the need to provide greater incentives, shifting the optimal contract to fixed rent. What does the shift in optimal agricultural contract imply for the political economy? It d(Π?−ΠSC ) ? SC can be shown that dτ > 0 if Π 6= Π , so as τ goes up the RHS of (20) goes up. At the original choices of the landlord candidate, the cost of sharecropping tenants exceeds the benefit, so the landlord candidate must lower T . Hence,

Proposition 4 A rise in τ that shifts the optimal contract away from sharecropping will cause the landlord candidate to lower T

The intuition is that as land productivity increases, it is increasingly costly to have share- cropping tenants on one’s land. Moreover, profits are higher regardless of the contact, so the landlord doesn’t need tenants any more to increase her vote share. So landlord farms more of her land under the optimal contract (wage or fixed rent).

C.7 Model Discussion There are two things to consider - firstly, the commitment problem and secondly vote- buying.45 I base the enforcement assumption based on the re-election incentives and informal/uninstitutionalised reciprocity, which are not explicitly modeled. However, as long as there is some incentive to prevent deviation, politicians will fulfill promises with some positive probability. Both voters and politicians realize this, and there will be partial enforcement, which is enough for the results of the model to go through. All transfers are made after the election, as driven by the timing assumption. Particu- larly, the landlord’s offers to tenants do not play the role of vote-buying, and are in fact post-election transfers conditional on winning. Vote-buying and turnout buying are also not explicitly modeled for two reasons; the first is for sake of simplicity and succinctness. Since the purpose of the model is to derive testable predictions, lack of data on elec- toral transfers has deterred incorporating vote-buying into the framework. Additionally, specifically in the case of landlords transfers to tenants, evidence from tenant-level data shows that in-fact the landlords’ offer higher inputs after the election rather than before. For both these reasons I believe it is reasonable to restrict focus to only post-election transfers.46

45Another point to consider is the electoral competition among landlords. Even though I argue that in an environment with high land concentration we can treat the oligarchy of landowners as a single entity, one might wonder about a third case, where there is competition between landlords. I make a case in the background that landlords represent a single class, have common interests and often belong to the same extended family (through inter marriage etc), it is reasonable to model them as cooperating, without modeling the cooperation game amongst them. While anecdotally rare, it is possible to consider the case with two landlords who can run against each other. It can be shown (see appendix) that if the optimal T for any landlord candidate is interior, she weakly prefers to cooperate with the other landlord. And moreover, if the optimal T leads to a corner solution, she strictly prefers to cooperate. Thus it suffices to consider the case with the monopsonist landlord. 46Empirical violations of these assumptions would bias my results against finding an effect of landlord prefer- ences, since I will be comparing landlord politicians after they win in an election to other landlords. If vote-buying occurs, and other landlords also change their contracts, I should not find an effect. 66