Traditional Elites: Agricultural Productivity and the Persistence of Political Power

Sabrin Beg

University of Delaware

Annual Bank Conference on Development Economics, June 2016 Motivation

I Historic institutions matter for present economic outcomes

I One potential channel is through the identity and incentives of historic elites

I Influence policy and ultimately growth

I Elites are historically known to provide patronage to rural masses, capture the state

I Breaking down the paternalistic ties between elites and their clients led way for the formation of welfare states

I Empirical work is challenging

I How do elites acquire power and affect public goods? I What factors reinforce or undermine the power of existing political elites?

2 / 41 This Paper: Research Question

I Understand the relationship of historic land distribution with the distribution of political power, and how landowning elites affect electoral outcomes and public goods in context of

I Specifically, I ask three questions:

1. How do traditional landowning elites maintain political dominance? 2. How do they effect political competition and allocation of public goods? 3. How does the relationship between land and power change with development, vis. a permanent shift in agricultural technology and productivity?

3 / 41 Model

I Model land and electoral market to show how landowners can acquire political support I Derive testable predictions for tenure, electoral and public goods outcomes; what happens with technological change Empirics

I Exploit introduction of an election to test how landowning-politicians influence tenants I Exploit historic institutions and exogenous technological change to test the effect of a productivity shift on ag. tenure, electoral and public goods outcomes

This Paper: What I do Pakistan good example: unequal land ownership, landowning elites legacy of colonial times; able to retain political power and influence policy

4 / 41 Empirics

I Exploit introduction of an election to test how landowning-politicians influence tenants I Exploit historic institutions and exogenous technological change to test the effect of a productivity shift on ag. tenure, electoral and public goods outcomes

This Paper: What I do Pakistan good example: unequal land ownership, landowning elites legacy of colonial times; able to retain political power and influence policy Model

I Model land and electoral market to show how landowners can acquire political support I Derive testable predictions for tenure, electoral and public goods outcomes; what happens with technological change

4 / 41 This Paper: What I do Pakistan good example: unequal land ownership, landowning elites legacy of colonial times; able to retain political power and influence policy Model

I Model land and electoral market to show how landowners can acquire political support I Derive testable predictions for tenure, electoral and public goods outcomes; what happens with technological change Empirics

I Exploit introduction of an election to test how landowning-politicians influence tenants I Exploit historic institutions and exogenous technological change to test the effect of a productivity shift on ag. tenure, electoral and public goods outcomes

4 / 41 Related Literature

Institutional persistence and Elite Capture:

I Historic land tenures (Banerjee & Iyer 2001); property rights (Field 2005, 2007), colonial systems (Dell 2010) I Traditional elites (Acemoglu, Reed and Robinson 2014) Political Clientelism:

I Clientelism (Keefer & Valaicu 2008, Dixit & Londregan 1996) I Inefficiency of democracy in developing world (Persson & Tabellini 2000) Old political economy literature: Clientelist politics and reciprocity between traditional rural patrons and peasants (Powell 1970, Scott & Kerkvliet 1976, Alston & Ferrie 1999); Baland & Robinson (2008)

I Interlinked agrarian markets (Braverman & Srinivasan 1981, Braverman & Stiglitz 1982, Bell & Srinivasan 1989)

5 / 41 Road Map

1. Background 2. Model 3. Data 4. Empirical Strategy and Results 5. Discussion and Conclusion Road Map

Background

Model

Data

Empirical Strategy and Results

Discussion and Conclusion

6 / 41 I Within villages or clusters of villages, have monopsonist status

I Typically ≤3 large landowners per village (median village has one) controlling majority of village land

I 75% are small-holder (own ≤5 ac.) or landless I Tenants with same landlord for their entire farming career or over generations

Historic Landowning Elites

I Colonial institutions granted large estates to local landlords in some districts; led to the creation of a landowning class

7 / 41 I Typically ≤3 large landowners per village (median village has one) controlling majority of village land

I 75% are small-holder (own ≤5 ac.) or landless I Tenants with same landlord for their entire farming career or over generations

Historic Landowning Elites

I Colonial institutions granted large estates to local landlords in some districts; led to the creation of a landowning class

I Within villages or clusters of villages, have monopsonist status

7 / 41 Historic Landowning Elites

I Colonial institutions granted large estates to local landlords in some districts; led to the creation of a landowning class

I Within villages or clusters of villages, have monopsonist status

I Typically ≤3 large landowners per village (median village has one) controlling majority of village land

I 75% are small-holder (own ≤5 ac.) or landless I Tenants with same landlord for their entire farming career or over generations

7 / 41 Landowning Elites at Present

Economic Dominance

I An “oligarchy” of land owners interact with a large population of landless and small-holder households

I Sharecropping tenancy is high

Political Dominance

I 70% of members of Provincial Assembly are landowners (several owning over 1000’s of acres)

8 / 41 Landowning Elites and Politics

Mumtaz Ali Bhutto interview in TIME (2008)

Own 40,000 acres ... cultivated by thousands of sharecroppers dependent on [them] ... the Bhutto family can count on a large turnout of supporters at the polls, [in exchange] for a place to live, seeds and fertilizer. And patronage.

’If my tenants are happy with me, they work more efficiently on the lands. You help the people and they will help you.’

9 / 41 Landowning Elites - Common Interests, but Dwindling Influence

I Historically, the landowning families act in coordination to achieve common interests

’... they would decide between themselves, taking turns at being elected’ - villager (BBC 2008)

I More recently, politics is being re-shaped:

’vacuums formed as labour-intensive plantations decline, cotton farming modernises and old families lose clout’ (Economist 2013)

10 / 41 Landowning Elites

Source: Bloomberg.com

Results Overview

11 / 41 Road Map

Background

Model

Data

Empirical Strategy and Results

Discussion and Conclusion

12 / 41 Testable Predictions

Proposition Landlord politicians, who have incentive to offer private transfers, are more likely to: (a) Offer more sharecropping contracts (b) Offer contracts more favorable to tenants (by paying higher cost share for inputs)

13 / 41 Testable Predictions

Proposition A productivity shift causes (a) SC tenancy to fall and profits to rise The fall in tenancy: b) Reduces landlord’s vote share and winning probability; improves electoral competition c) Landlord “preferred” public goods lower relative to other public goods. and vice versa for the income effect

14 / 41 Road Map

Background

Model

Data

Empirical Strategy and Results

Discussion and Conclusion

15 / 41 Data

I Pakistan Rural Household Survey (2000, 2003)

I Election Commission (2002-2003)

I FAO Global Agro-Ecological Zones Data

I Decennial Agricultural Census (1960-2010), Annual Agricultural Statistics

I Quintennial Village Census (1993-2008)

I Pakistan Living Standards Measurement Surveys (2002-2006)

I Land settlement reports, district and province gazetteers from 1857-early1900s

16 / 41 Road Map

Background

Model

Data

Empirical Strategy and Results

Discussion and Conclusion

17 / 41 Testing Proposition 1

Landlords who have incentive to offer political transfers prefer sharecropping contracts, and offer contracts more favorable to tenants (by paying higher cost share)

I Empirical Strategy: compare LL politicians with electoral incentives to LL without electoral incentives using the introduction of election after a non-democratic regime

I Pakistan Rural Household Survey panel data

I 2000-01 during a military government; 2003-04 after the 2002 national elections

18 / 41 Testing Proposition 1 - Estimating Equation

yi,j,p,t = γ1LL winningpoliticiani,j,p + γ2PostElectiont + γ3ηp + γ4σj + γ5ςi,j + κi + εi,j,p,t,

19 / 41 Landlord Politician and Sharecropping

Dependent variables is dummy indicating plot is sharecropped (1) (2) (3) (4) (5) (6) (7) (8) LL is Politician PostElection 0.174** 0.169** 0.182** 14.70*** 0.169** 0.171** 0.166** 32.25*** (0.0726) (0.0741) (0.0827) (1.178) (0.0728) (0.0740) (0.0759) (4.785)

PostElection -0.0546*** -0.0546** -0.0598** -1.982*** -0.0496** -0.0485** -0.0384 0.836 (0.0202) (0.0222) (0.0244) (0.560) (0.0195) (0.0221) (0.0391) (1.672)

Observations 1034 1034 1034 1034 846 846 846 846 Plot Type Leasedin/out Leasedin/out Leasedin/out Leasedin/out Leasedin Leasedin Leasedin Leasedin Plot Level Controls No Yes Yes Yes No Yes Yes Yes LL/Tenant Level Controls No No Yes Yes No Yes Yes Yes Crop Composition No No No No No No Yes Yes Notes: Regressions are at plot level with household fixed effect. Robust standard errors clustered at household level. Plot level controls include plot soil, slope and area. Addiotional controls include landlord and tenant's landholdings. Leased out plots indicate that surveyed farmer is the lanlord, while leased in plots imply the surveyed farmer is the tenant. The composition of crops grown is provided when the surveyed farmer is the cultivator, i.e. only for leased in plots. Columns (4) and (8) report marginal effects from a probit regression.

20 / 41 Landlord Politician and Contracts

(1) (2) (3) (4) (5) (6) Plot is Landlord share of: Canal Dependent Variable: Sharecropped Fertilizer G-Water Thresher Output Access LL is Winning Politician x Sharecropped 0.855*** (0.0910)

LL is Winning Politician 0.166** 2.962 21.11* 17.09** 0.645 -0.827*** (0.0759) (2.985) (11.67) (8.166) (2.319) (0.0818)

PostElection -0.0384 -8.068*** -22.80*** -15.26** -1.092 -1.518** (0.0391) (2.205) (5.682) (5.949) (1.258) -0.078

Only Only Only Only All Leased Sample All Leased Plots Sharecropped Sharecropped Sharecropped Sharecropped Plots Observations 846 611 361 600 614 846 Notes: Regression are at plot level. Robust standard errors clustered at tenant level in brakets. All regressions include landlord's holdings, tenant fixed effects, plot controls (size, irrigation, soil/slope), crop controls, controls for landlord-tenant relation and length of relationship. The input share regressions are conditional on the use of the particular input on the plot

21 / 41 I I test for pre-trends using alternate data

I The results are robust to including placebo controls for other influential landlords (panchayat leader, religious head, large landlord)

I Also include control for LL politicians before election

Tables

Robustness

22 / 41 Robustness

I I test for pre-trends using alternate data

I The results are robust to including placebo controls for other influential landlords (panchayat leader, religious head, large landlord)

I Also include control for LL politicians before election

Tables

22 / 41 I Use variation in HYV penetration to construct a measure of technological change

I imported technology, increased productivity (Foster and Rosenzweig 1996, Bustos et al. 2014)

I area-specific differences in HYV suitability, amount of HYV varies over time and across crops

I need intensive inputs; fertilizer and irrigation (Shiva 1991)

Testing Proposition 2-Empirical Strategy

A productivity shift will cause a) Tenancy to fall and income to rise b) Lower (raise) landlords political incentives c) Improve(decrease) electoral competition d) Policy shifts

I Need plausibly exogenous shift in productivity and a measure for landlord dominance

23 / 41 Testing Proposition 2-Empirical Strategy

A productivity shift will cause a) Tenancy to fall and income to rise b) Lower (raise) landlords political incentives c) Improve(decrease) electoral competition d) Policy shifts

I Need plausibly exogenous shift in productivity and a measure for landlord dominance

I Use variation in HYV penetration to construct a measure of technological change

I imported technology, increased productivity (Foster and Rosenzweig 1996, Bustos et al. 2014)

I area-specific differences in HYV suitability, amount of HYV varies over time and across crops

I need intensive inputs; fertilizer and irrigation (Shiva 1991)

23 / 41 Empirical Strategy-measure of Land Productivity

FAO-GAEZ provides suitability indices for all crops for a world wide grid

I Suitability for each crop suiti,r,c under high and low technology

I suitH,Irr,c − suitL,R,c is a measure of area-specific suitability for HYV for any crop

I HYVp,t is the overall province level penetration of HYVs in any year (kg/ha)

I Prodc,j,t = (suitH,Irr,c,j − suitL,R,c,j ) × HYVpt

24 / 41 FAO Wheat Suitability (High, Low)

25 / 41 HYV Availability at Province Level

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) (per of Seeds 1970 1980 1990 2000 2010 1970 1980 1990 2000 2010 year Wheat Cotton Maize Rice

Graphs by province

26 / 41 Empirical Strategy - Measure Initial Landlord Dominance

I Use the prevalence of large estate-holders during British rule to proxy for initial landlord dominance

I Land assignments (“jagirs”) made by pre-colonial and colonial governments in British India

I Purpose was to maintain loyalty of local chiefs I Consisted of a village to several villages or an entire subdistrict

I Variation in the prevalence of these estates across districts:

27 / 41 Testing Proposition 2(a)-(c): Estimating Equation

yjt = β1+β2Prodjt +β3Prodjt ×LLDominatedj +νj +µpt +jt

I Outcome variables: Land holdings of winning politician, electoral competition, public goods

I Cluster standard errors at the geographic unit level (constituency, sub-district, district) level

28 / 41 Rate of sharecropping goes down

29 / 41 Likelihood of election of Landowners goes down

30 / 41 Electoral Competition Improves

31 / 41 Rate of sharecropping goes down

Perc of Tot Area Perc of Tot Area Perc of Leased Area Perc of Tot Area Perc of Tot Area Perc of Leased Area Dependant Variable: Sharecropped (SC) Self-cultivated (W) Sharecropped (SC) Sharecropped (SC) Self-cultivated (W) Sharecropped (SC) Suit Diff x HYV -0.128 0.114 -0.180 [-3.39]*** [2.81]*** [-4.87]*** Suit Diff x HYV x LL Dominated -0.0755 0.0518 0.0666 [-0.61] [0.47] [0.42] Predicted Yield -25.19 22.55 -33.44 [-3.33]*** [3.19]*** [-2.67]*** Predicted Yield x LL Dominated -2.257 1.564 0.537 [-0.65] [0.50] [0.17] Effect of 0.5ton/ha increase in Yield at LL Dominated=1 -7.12 5.80 -3.97 -13.72 12.06 -16.45 p-value 0.078 0.1 0.79 0.001 0.001 0.001 Mean of Dependant Variable (%) 25.0 69.3 78.7 25.0 69.3 78.7 Estimation OLS OLS OLS 2SLS 2SLS 2SLS

First Stage Regression Yield x LL Dependant Variable: Yield Dominated Suit Diff x HYV 0.01 -0.001 [2.99]*** [-1.41] Suit Diff x HYV x LL Dominated -0.002 0.04 32 / 41 [-0.85] [4.42]*** F-Stat 4.49 12.94 Observations 280 280 280 280 280 280 Notes: t statistics in brackets. Standard errors clustered at district level. All regression use district-level data from Agricultal Censuses of 1960 to 2010 and include district and province-year FE. The first stage in the bottom panel includes all controls from the seconds stage. 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 overall effect is calculated for a change in Suit Diff x HYV corresponding to a 0.5ton/ha increase in yield Likelihood of election of Landowners goes down

(1) (2) (3) (4) (5) Dependant Variable: Winner owns Winner owns Winner owns Agland Num of Ag AgLand AgLand AgLand (acres) Properties

Suit Diff x HYV 0.0000890 0.00244 -0.000241 0.400 0.00528 (0.000458) (0.00355) (0.000510) (0.705) (0.00498) Suit Diff x HYV x LL Dominated -0.00367 -0.0464 -0.00463 -4.698 -0.0369 (0.00182)** (0.0254)* (0.00221)** (2.582)* (0.0332) Effect of 0.5ton/ha increase in Yield at LL Dominated=1 -0.13 -1.54 -0.17 -150 -1.11 p-value 0.05 0.09 0.03 0.11 0.34

Mean of Dependant Variable 0.70 0.70 0.70 116.02 1.78 All Winning All Winning All Winning All Winning Politicians who Politicians who All Winning Sample includes:: Politicians Politicians declare exact declare exact Politicians acerage acerage Estimation OLS Probit OLS OLS OLS Observations 1568 1568 1402 1402 1568 Notes: All regressions include constituency and province-year FE and standard errors are clustered at constituency level. Marginal effects reported for probit model. 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. The overall effect is calculated for a change in the constructed measure of Suit Diff x HYV corresponding to an increase in yield of 0.5 ton/ha.

33 / 41 Electoral Competition Improves

(1) (2) (3) (4) (5) (6) Dependant Variable: H-dhal Index Winmargin (%) Low Competition Low Competition # of Candidates Turnout (%)

Suit Diff x HYV -0.000 -0.007 -0.000 -0.001 0.016 0.030 (0.000)* (0.015) (0.000) (0.006) (0.005)*** (0.007)*** Suit Diff x HYV x LL Dominated -0.001 -0.097 -0.005 -0.043 0.063 0.028 (0.001)* (0.068) (0.002)*** (0.015)*** (0.038) (0.016)* Effect of 0.5ton/ha increase in Yield at LL Dominated=1 -0.04 -3.64 -0.18 -1.54 2.77 2.03 p-value 0.0229 0.131 0.00224 0.00472 0.0395 0.00103

Mean of Dependant Variable 0.361 17.76 0.201 0.201 11.87 45.13 Estimation OLS OLS OLS Probit OLS OLS Observations 1,568 1,568 1,568 1,568 1,568 1,566 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). The results are robust to using other cut-off values. Turnout is unavailable for 2 constituencies where the election was uncontested. 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. The overall effect is calculated for a change in the constructed measure of Suit Diff x HYV corresponding to an increase in yield of 0.5 ton/ha.

34 / 41 I What does this mean for public goods provision?

35 / 41 Landlord preferred Public Goods

I Non-landlords are more likely to be elected → public goods desired by landowners lower relative to other types of public goods (PSLM 2004-2010)

(1) (2) (3) (4) (5) (6) Basic Health Fam Planning Dependent variable indicates use of: Unit Center Drinking Water Road Vet Police Station

Land owner -0.0582 -0.00813 0.00372 0.0156 0.0688 0.0360 (-3.18)*** (-1.73)* (1.25) (1.41) (5.60)*** (3.22)***

Mean of Dependent Variable 0.17 0.03 0.98 0.91 0.04 0.03

Observations 73950 73950 73950 73950 73950 73950 Notes: Landowner is a dummy indicating the respondent is in the top 5th percentile of the landowners in the sample. Data is obtained from the Pakistan Living Standards Measurement Surveys of 2006. Regressions are at individual level and control for district. Standard errors are clustered at district-region level. t- statistics reported in parentheses.

I Landlords may also prefer to invest in facilities in rural areas

I ‘Core voter’ view → landlords target their voter base (Beg 2016) → provide more to rural areas of the constituency

36 / 41 Resources targeted towards “broad” public goods

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

Dependent variable is the percentage of respondents reporting an improvement in the Fam Planning availability of: Basic Health Unit Center Drinking Water Road Vet Police Station

Suit Diff x HYV 0.0413 -0.0627 -0.0596 -0.0438 -0.0599 -0.0664 (1.67)* (-1.98)** (-2.51)** (-2.80)*** (-1.69)* (-1.65)*

Suit Diff x HYV x LL Dominated 0.0964 0.174 0.115 0.0475 0.0858 -0.00555 (1.84)* (1.97)** (2.96)*** (1.22) (0.98) (-0.06)

Observations 543 543 543 543 543 543 Notes: The point estimates are obtained using a Poisson regression model due to non-normally distributed observations. Data is obtained from the Pakistan Living Standards Measurement Surveys between 2006 and 2010. Regressions are at district-region-year level (region corresponds to rural/urban) and control for district, rural status, region-province-year fixed effects. Standard errors are clustered at district-region level. t- statistics reported in parentheses. 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.

37 / 41 Resources targeted away from landlord preferred goods

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

Dependent variable is the percentage of respondents reporting an improvement in the Fam Planning availability of: Basic Health Unit Center Drinking Water Road Vet Police Station

Suit Diff x HYV x LL Dominated 0.101 0.186 0.116 0.0526 0.0948 -0.00190 (2.68)*** (2.17)** (3.27)*** (2.19)** (2.15)** (-0.02)

Suit Diff x HYV x LL Dominated x Rural -0.0132 -0.0340 -0.00397 -0.0138 -0.0362 -0.00929 (-2.15)** (-1.60) (-0.70) (-3.77)*** (-3.49)*** (-0.62)

Overall Effect in Urban Areas ! ! ! ! ! " Overall Effect in Rural Areas ! ! ! " " "

Observations 543 543 543 543 543 543 Notes: The point estimates are obtained using a Poisson regression model due to non-normally distributed observations. Data is obtained from the Pakistan Living Standards Measurement Surveys between 2006 and 2010. Regressions are at district-region-year level (region corresponds to rural/urban) and control for district, rural status, and region-province-year fixed effects. Standard errors are clustered at district-region level. t- statistics reported in parentheses. 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.

38 / 41 Road Map

Background

Model

Data

Empirical Strategy and Results

Discussion and Conclusion

39 / 41 I Productivity shift → shift in distribution of political power as paternalistic ties become weak; electoral competition improves

I Transition from clientelism to more democratic

I Public goods desired by landlords go down; broad public goods go up, particularly in urban areas

I Initial asset distribution → economic and political inequality → growth → political inequality

Conclusion

I Micro-founded model, rigorous empirical analysis → landlord politicians offer private transfers to tenants

40 / 41 I Public goods desired by landlords go down; broad public goods go up, particularly in urban areas

I Initial asset distribution → economic and political inequality → growth → political inequality

Conclusion

I Micro-founded model, rigorous empirical analysis → landlord politicians offer private transfers to tenants

I Productivity shift → shift in distribution of political power as paternalistic ties become weak; electoral competition improves

I Transition from clientelism to more democratic

40 / 41 I Initial asset distribution → economic and political inequality → growth → political inequality

Conclusion

I Micro-founded model, rigorous empirical analysis → landlord politicians offer private transfers to tenants

I Productivity shift → shift in distribution of political power as paternalistic ties become weak; electoral competition improves

I Transition from clientelism to more democratic

I Public goods desired by landlords go down; broad public goods go up, particularly in urban areas

40 / 41 Conclusion

I Micro-founded model, rigorous empirical analysis → landlord politicians offer private transfers to tenants

I Productivity shift → shift in distribution of political power as paternalistic ties become weak; electoral competition improves

I Transition from clientelism to more democratic

I Public goods desired by landlords go down; broad public goods go up, particularly in urban areas

I Initial asset distribution → economic and political inequality → growth → political inequality

40 / 41 I Looking at other historical contexts I Testing model predictions about technology adoption I Testing the impact of productivity increase on landlords’ political participation and on partisan change

I Extensions:

Discussion

I Takes models of clientelism and traditional paternalistic relations to data

I Results speak to the rural-urban divide and rising rates of rural-urban migration

41 / 41 I Looking at other historical contexts I Testing model predictions about technology adoption I Testing the impact of productivity increase on landlords’ political participation and on partisan change

Discussion

I Takes models of clientelism and traditional paternalistic relations to data

I Results speak to the rural-urban divide and rising rates of rural-urban migration

I Extensions:

41 / 41 I Testing model predictions about technology adoption I Testing the impact of productivity increase on landlords’ political participation and on partisan change

Discussion

I Takes models of clientelism and traditional paternalistic relations to data

I Results speak to the rural-urban divide and rising rates of rural-urban migration

I Extensions:

I Looking at other historical contexts

41 / 41 I Testing the impact of productivity increase on landlords’ political participation and on partisan change

Discussion

I Takes models of clientelism and traditional paternalistic relations to data

I Results speak to the rural-urban divide and rising rates of rural-urban migration

I Extensions:

I Looking at other historical contexts I Testing model predictions about technology adoption

41 / 41 Discussion

I Takes models of clientelism and traditional paternalistic relations to data

I Results speak to the rural-urban divide and rising rates of rural-urban migration

I Extensions:

I Looking at other historical contexts I Testing model predictions about technology adoption I Testing the impact of productivity increase on landlords’ political participation and on partisan change

41 / 41 Summary Statistics

Estate=0 Estate=1 Difference Unit Source Observations Agriculture Perc. under sharecropping 23.2 31 7.8*** District Ag. Census 225 (17.0) (18.5) Land Concentration 39.0 44.8 5.8 District Ag. Census 47 (12.3) (11.2) Suitability Diff 2.00 2.22 0.22 9.25 × 9.25 km FAO (0.06) (0.14) Wheat Yield in 1965 (ton/ha) 0.84 0.68 -0.16 District Ag. Census 47 (0.27) (0.16) Wheat Yield in 1982 (ton/ha) 1.44 1.38 -0.06 District Ag. Census 47 (0.06) (0.14) Wheat Yield in 2002 (ton/ha) 2.50 2.55 0.05 District Ag. Census 47 (0.06) (0.14)

Notes: The agricultural data is at district level. The land concentration is calculated in the baseline period, 1972, when there are 47 districts. By 2008 there are a total of over 100 districts, but the later data is aggregated to the level of districts as demarked in 1972.

1 / 31 Summary Statistics

Estate=0 Estate=1 Difference Unit Source Observations

Electoral Ag.Land=1 0.65 0.79 0.14*** Constituency Elec. Comm 1402 (0.477) (0.410) Agr. Land Declared (Acres) 97.2 329 231.8** Constituency Elec. Comm 1402 (331) (1018) Win Margin 17.9 20.3 2.4** Constituency Elec. Comm 1635 (18.03) (19.10) Turnout 45.1 39.2 -5.9*** Constituency Elec. Comm 1635

Public Goods Police Station 86 79 -7*** District PSLM 543 (19) (22) Vet 82 77 -5*** District PSLM 543 (19) (22) BHU 49 43 -6*** District PSLM 543 (28) (28) Family Planning 69 62 -7*** District PSLM 543 (21) (25)

2 / 31 Productivity Increases with HYV

(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) 0.0145 0.00977 -0.1000 0.151 0.0145 0.0145 [4.46]*** [1.37] [-1.58] [3.08]*** [4.47]*** [4.58]***

Suit Diff x HYV x Estate 0.000557 [0.33]

Observations 1804 955 1110 1162 1804 1804 Notes: The regressions use FAO suitability indices for each crop. t statistics in brackets. District-level annual yields from 1979 onward. Standard errors clustered at province-year 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

3 / 31 I Use two natural experiments and a historic IV

I Introduction of election: look at landlords who are politicians before and after a general election (preceded by non-democratic regime)

I Construct a measure of productivity shift, using suitability for high yielding varieties of seeds

I Use a historic measure to proxy landlord political dominance - land grants given by colonial government

Preview: Empirical Strategy

I Challenges for identification: 1. Landlord politicians/tenants differ from other landlords/tenants in many ways 2. Political power and productivity are endogenous

4 / 31 Preview: Empirical Strategy

I Challenges for identification: 1. Landlord politicians/tenants differ from other landlords/tenants in many ways 2. Political power and productivity are endogenous

I Use two natural experiments and a historic IV

I Introduction of election: look at landlords who are politicians before and after a general election (preceded by non-democratic regime)

I Construct a measure of productivity shift, using suitability for high yielding varieties of seeds

I Use a historic measure to proxy landlord political dominance - land grants given by colonial government

4 / 31 Preview: Data

I Pakistan Rural Household Survey (2000, 2003)

I Election Commission (2002-2003)

I FAO Global Agro-Ecological Zones Data

I Decennial Agricultural Census (1960-2010), Annual Agricultural Statistics

I Quintennial Village Census (1993-2008)

I Pakistan Living Standards Measurement Surveys (2002-2006)

I Land settlement reports, district and province gazetteers from 1857-early1900s

5 / 31 Preview: Results

1. Political “private” transfers to tenants: Landlord who are politicians offer contracts favorable to tenants after an election (exploit introduction of an election)

2. An exogenous shift in land productivity (exploit high yielding variety seeds, and historic proxy for land distribution):

I reduces rate of tenancy;

I reduces likelihood that winning politician is land-owner;

I improves electoral competition;

I composition of public goods shifted; public goods that benefit landowners go down. Return

6 / 31 I Cost of tenancy higher with productivity

dΠS dΠL,N dΠL dτ = dτ > dτ

Productivity Shift and Tenancy

I If self-cultivating, the landlord choses higher inputs e, x

ΠS > ΠL

Details

7 / 31 Productivity Shift and Tenancy

I If self-cultivating, the landlord choses higher inputs e, x

ΠS > ΠL

I Cost of tenancy higher with productivity

dΠS dΠL,N dΠL dτ = dτ > dτ

Details

7 / 31 Secret Ballot not so secret

I (Chaudhry & Vyborny 2013) Rural households strongly convinced their vote was not secret from their patrons or officials

I Bloc voting: local leaders determine the candidate choice for an entire group of people.

I Field experiment (Gine & Mansuri 2010) aiming at educating female voters had to stress the significance of secret ballot

I FAFEN reports from last election (2013): government official, local influential persons and police in polling stations

8 / 31 Sharecropping Model

I Agent solves:

max EU(αgf (τ, e, x) − βxp, e) e,x

I Principal solves:

max E[(1 − α)gf (τ, e, x) − (1 − β)xq] = ΠL α,β

s.t U(αgf (τ, e, x) − βxp) ≥ U

and (e, x) ∈ argmaxU(αgf (τ, e, x) − βxp) e,x

9 / 31 Solution

I Agent choses e∗ and x∗ such that

1 EU2 I fe = − αρ EU1 βp I fx = − αρ

I Principal sets β s.t U(αgf (τ, e, x) − βxp) = U

I Principal chooses (α) to maximize profits ∂e ∂x ∂x ∂β I (1 − α)(fe ∂α + fx ∂α ) − f (e, x) − (1 − β)p ∂α + xp ∂α = 0 I If τ is labor saving technical change, as τ↑, x/e ↑,α and β go down → wage contract

Return

10 / 31 Agent

Principal/Agent beer off

Aer transfer: B U’ ✖ U ✖ A: Before transfer

Principal Π-γε Π

I

Landlord Advantage: LL can transfer utility by offering favorable contract

1. Fixed Transfer: LL wants to offer private transfer γε ; 2. Cost Share: Offer to lower cost share s.t. βεqx = γε.

11 / 31 Landlord Advantage: LL can transfer utility by offering favorable contract

1. Fixed Transfer: LL wants to offer private transfer γε ; 2. Cost Share: Offer to lower cost share s.t. βεqx = γε.

Agent

Principal/Agent beer off

Aer transfer: B U’ ✖ U ✖ A: Before transfer

Principal Π-γε Π

I 11 / 31 Fixed transfer vs. Cost Share

I Landlord is not worse off compared to (Π − γε) need following condition to hold

(1 − α){f (τ, e + eε, x + xε) − f (e, x)} − (1 − β)qxε − βεq(x + xε) ≥ −γε = −βεqx

∂Π∗ ⇒ ∂x > 0

I Lowering β cheaper than offering a lump sum transfer if ex post the principal wants agent to use more input qρ I β > β = p+qρ

Return

12 / 31 Productivity Shift

max (1 − α)gf (τ, e, x) − (1 − β)qx + EU(αgf (τ, e, x) − βpx, e) e,x,α,β,γ

s.t U(αgτf (e, x) − βpx) ≥ U

and f = − 1 EU2 , f = − βp e αρ EU1 x αρ

dΠSC dτ = (1 − α)fτ (τ, e, x) + αfτ (τ, e, x)E(U1.g)

I Note, if the agent is risk neutral we have dΠSC,N dΠS dτ = (1 − α)fτ + αfτ EU1Eg = dτ . I Since E(U1.g) = EU1.Eg + Cov(U1, g) < EU1Eg because Cov(U1, g) < 0 dΠW dΠSC,N dΠSC d(ΠW −ΠSC ) I dτ = dτ > dτ ⇒ dτ > 0. W SC I (Π − Π ) is increasing in τ

Return

13 / 31 Landlord politician and Sharecropping

14 / 31 Landlord politician and Input Shares Robustness

15 / 31 Landlord Politician and Loans

Return 16 / 31 Background: Land Concentration

Pakistan1 El Salvador2 India % of Holders Small 3 69 74 33 Av Size - Small (ha) 1.9 1.2 0.2 Av Size - Large (ha) 187 330 18 % land under large farms 41 64 30

1 Sources for Pakistan and India: Relevant Ag. Census 2 Sources: Jazairy et al 1992, Brockett 1992 3 Holding less than 5 ha 17 / 31 Land Concentration .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: (% of land with largeholders + % of land with smallholders/average size with smallholder) Source: Relevant Annual Census of Agriculture

18 / 31 Area under Sharecropping (%) .06 .04 Density .02 0 0 20 40 60 80 Perc. of Total Area under Sharecropping (%)

1960 1972 1980 2000 2010

Source: Relevant Annual Census of Agriculture

I Leasing Rates: India 0.08% (1995), Bangladesh 13% (2008), Pakistan 20% (2000)

19 / 31 Secret Ballot not so secret

I (Chaudhry & Vyborny 2013) Rural households strongly convinced their vote was not secret from their patrons or officials

I Bloc voting: local leaders determine the candidate choice for an entire group of people.

I Field experiment (Gine & Mansuri 2010) aiming at educating female voters had to stress the significance of secret ballot

I FAFEN reports from last election (2013): government official, local influential persons and police in polling stations ??

20 / 31 Background

Jagirs - An assignment of land, with or without conditions, to an individual for services rendered to the State.

I Alternately called feudatories, one holding lands by feudal tenure.

I Land grants made by the precolonial governments ( Dynasty, Kalhora Dynasty, Sikh Rulers, Mughals) as well as British. Usually consisted of a villages to several villages, or entire subdistrict

I Canal Colonies - land grants different from Jagirs

I Local chiefs - enjoyed administrative power and the right to land revenue

Return

21 / 31 Background

I Why and where

I Allegiance

I Turbulent and exposed tracts, like Hashtnagar and Mlranzai, were made over in jagir to the local chieftains, who enjoyed an almost complete independence I “All Jagirdars who offered their allegiance to the British Government within a specified time after the battle of Miani, would be confirmed in the possession of their estates.” (Gazetteer of Sindh)

I Military or other services I Gifts to friends/family of the ruling dynasty

I Later were classified in different classes according to the conditions on the rent and inheritance, and the length of the grant I Jagirdars enjoyed considerable influence in their jagir, and often were the sole Zamindar

Return

22 / 31 Analysis - FOCs of candidate

1 1 ψ M Pr(πA > ) = + nmφm[WmA − WmB ]dm 2 2 φm 0 ´ ´

n 1 ψ 0 o I G : r + nmφmH (G)dm = λ 2 φm ´ ´ n 1 ψ 0 o I fm : r + nmφmU (fm) = nmλ ∀m 2 φm ´ n 1 ψ B o I r : + nmφm[U(fm) + H(G) − W ]dm = λ 2 φm m ´ ´ I BC: R = r + G + nmfmdm ´

23 / 31 Analysis - FOCs of candidate

n 1 ψ 0 o I G : r + nmφmH (G)dm = λ 2 φm ´ ´ n 1 ψ 0 o I fm : r + nmφm(1 + η)U ((1 + η)fm) = nmλ ∀m ≤ T 2 φm ´ n 1 ψ B o I fm : + nmφm[U(fm) + H(G) − W ]dm = λ 2 φm m ´ ´ I BC: R = r + G + nmfmdm ´

24 / 31 Land Concentration 50 800 40 700 600 30 500 20 avgsize_over250 areashare_over250 400 10 300 0 0 1 2 3 4 percholdings_over250

25 / 31 Land Concentration 50 800 40 700 600 30 500 20 avgsize_over250 areashare_over250 400 10 300 0 0 1 2 3 4 percholdings_over250

26 / 31 Land Concentration 50 800 40 700 600 30 500 20 avgsize_over250 areashare_over250 400 10 300 0 0 1 2 3 4 percholdings_over250

27 / 31 Land Concentration 100 800 90 700 80 600 70 500 avgsize_over250 percholdings_under25 400 60 300 50 0 1 2 3 4 percholdings_over250

28 / 31 Land Concentration 100 25 80 20 60 15 40 10 # # of landowners 5 20 %age of land owned by landowners owned of land %age 0 0 0 200 400 600 800 avgholding_large

# of landowners %age of land owned by landowners

29 / 31 Distribution of Land Ownership of Landlords .003 .002 .001 0 0 1000 2000 3000

Density kdensity ll_ownland

30 / 31 Distribution of Land Ownership of Landlord Politician 4.0e-04 3.0e-04 2.0e-04 1.0e-04 0 0 2000 4000 6000 8000 10000

Density kdensity ll_ownland

31 / 31