wells, water, and welfare: the impact of access to groundwater … · 2017. 3. 16. · wells,...

76

American Economic Journal: Applied Economics 2014, 6(3): 76–102 http://dx.doi.org/10.1257/app.6.3.76

Wells, Water, and Welfare: The Impact of Access to Groundwater

on Rural Poverty and Conflict †

By Sheetal Sekhri *

This paper evaluates the impact of access to groundwater on poverty using data from rural India. The estimation exploits the fact that the technology required to access groundwater changes exogenously due to constraints imposed by laws of physics at a depth of eight meters. I find that rural poverty in areas where depth from surface is below the cutoff is 9 to10 percent higher. Using survey data for a subsample of villages, I also show that disputes over irrigation water increase by 25 percent around the cutoff. Historical endowments of groundwater facilitate adoption of yield enhancing technologies over the long-run. (JEL D74, I32, O13, O15, O18, Q15, Q16)

Groundwater irrigation sustains more than 50 percent of the world’s food pro-duction. Groundwater access has the potential to raise agricultural productivity

by providing farmers with a more reliable source of crop irrigation than rainfall. By increasing agricultural productivity, groundwater access can address enduring goals in poverty alleviation: stabilizing rural incomes, raising living standards, and enhanc-ing food security. Amid declining groundwater supplies, policymakers the world over debate policy interventions to achieve the optimal allocation of finite groundwater resources and encourage sustainable use. Absent from these debates is a precise assessment of how groundwater access reduces poverty, especially in rural areas.

I identify the causal effect of groundwater access on rural poverty, based on variation in exogenous technological constraints to groundwater extraction. When groundwater is within eight meters below ground, it can be extracted with a simple, low-cost centrifugal pump that utilizes atmospheric pressure to raise water above ground. Pumps using atmospheric pressure cannot support the weight of groundwa-ter deeper than eight meters. Costlier submersible pumps are necessary to extract deeper groundwater. I exploit the random variation in the feasibility of technol-ogy used to extract groundwater to compare poverty rates in villages with low and high access to groundwater across five Indian districts. Villages whose depth to

* University of Virginia, PO Box 400182, Department of Economics, Monroe Hall, Charlottesville, VA 22904-4182, (e-mail: [email protected]). I thank the Ministry of Water Resources, Government of India, for providing the groundwater data and Ajay Gupta for technical assistance. The paper has benefited from suggestions by Tania Barham, Andrew Foster, Jessica Goldberg, Anne Harrison, Seema Jayachandran, and Michael Kremer. Thanks to Petia Topalova for sharing Indian districts poverty data. Ting Wang and Sisir Debnath provided excellent research assistance. Funding from Bankard Fund for Political Economy is greatly acknowledged.

† Go to http://dx.doi.org/10.1257/app.6.3.76 to visit the article page for additional materials and author disclosure statement(s) or to comment in the online discussion forum.

VoL. 6 No. 3 77Sekhri: Groundwater and Poverty

groundwater is greater than 8 meters have 9 to 10 percent higher poverty rates than villages with groundwater depth within 8 meters. These estimates are robust across several empirical specifications and estimation techniques. Survey evidence from a subsample of villages shows that the percentage of farmers that own submers-ible pumps rises sharply for villages whose groundwater depth from the surface is greater than eight meters. This finding further validates the identification strategy I employ. In other settings, water shortages have been shown to result in escalated conflict (Devoto et al. 2012). Using survey data, I find that self-reported disputes over irrigation water increase by 25 percent around this cutoff.

I use the discontinuity as an instrument for submersible pumps in an instrumental variable framework and establish that an increase in submersible pumps is associ-ated with a significant rise in poverty. I show that a change in agricultural yield mediates this effect. I also use historical endowment of groundwater and the tim-ing of the introduction of high-yielding varieties in India to show that groundwater endowments facilitated adoption of yield enhancing varieties that ushered in the Green Revolution in India.

This study evaluates the impact of groundwater access on rural poverty in India. India exemplifies developing countries’ dependence on groundwater and its promise as a poverty-alleviation tool. Nearly 54 percent of India’s population is employed in agriculture. Sixty percent of Indian agriculture depends on groundwater crop irriga-tion. Almost 92 percent of the groundwater that India extracts is used for irrigation (Jha and Sinha 2009). India also exhibits challenges to sustainable groundwater sup-ply management. It is the world’s single largest user of groundwater. Between 1960 and 2010, the area irrigated by groundwater increased by 500 percent (Garduño, and Foster 2010). Across India, nonrenewable aquifers are rapidly depleting (Rodell, Velicogna, and Famiglietti 2009; Sekhri 2012). Protecting groundwater for future use has taken a center stage at policy discussions. But market or nonmarket solu-tions to addressing overextraction can be meaningfully designed only if the benefits of access to groundwater are well known.

This paper contributes to the literature on groundwater irrigation in the develop-ing world. Most of the existing literature on groundwater in the United States has focused on studying externalities arising from groundwater use (Brill and Burness 1994; Gisser and Sánchez 1980). A small but growing body of work focuses on examination of institutions for groundwater allocation or inter-sectoral effects of groundwater use in developing countries (Aggarwal and Narayan 2004; Anderson 2011; Banerji, Meenakshi, and Khanna 2012; Foster and Rosenzweig 2008; Foster and Sekhri 2007; Jacoby, Murgai, and Rehman 2004; Keskin 2009; Sekhri 2011, 2012). A few studies focus on estimating the effect of groundwater irrigation on agricultural output. Sekhri (2013) evaluates the impact of groundwater irrigation on agricultural production in India and investigates coping mechanisms to ground-water distress. Hornbeck and Keskin (2013) also focus on agricultural output in the midwestern United States and look at the role groundwater plays in adaptation to climate stress. The impact of electricity subsidies on groundwater extraction and agricultural production has been investigated by Badiani and Jessoe (2011). This paper is the first study to provide credible estimates of the effect of groundwater access on rural poverty that are vital for policy design.

78 AmErIcAN EcoNomIc JoUrNAL: AppLIEd EcoNomIcs JULy 2014

This paper also complements the literature on the welfare effects of irrigation. The most credible estimates on the effects of irrigation have been established by Duflo and Pande (2007). Their study assesses the effect of dam irrigation on agri-cultural output and poverty in Indian districts. By contrast, this paper focuses on the impact of groundwater irrigation on poverty using village-level data. Because Indian agriculture relies heavily on groundwater, understanding the welfare implications of access to groundwater irrigation is important.

The findings of the paper have important policy implications. In light of rapid declines in groundwater tables around the world, there is a significant policy debate about whether groundwater irrigation should be constrained completely or whether innovative policies should be implemented to encourage sustainable use. The esti-mates of the impact on welfare of access to groundwater are extremely crucial for making informed policy choices. This paper indicates that groundwater irrigation leads to a significant reduction in poverty.

The rest of the paper is organized as follows. Section II discusses the naturally occur-ring variation in the cost of groundwater access that I exploit in the empirical analysis. Section III describes the data. Section IV outlines the estimation strategy. Section V reports the results and discusses the robustness tests. Section VI provides insights into mechanisms that mediate this effect. Sections VII and VIII discuss further concerns about the validity of the estimation strategy and interpretation of results. Section IX provides a discussion of the caveats and Section X provides concluding remarks.

I. Spatial Variation in Access to Groundwater

The technology suitable for extracting groundwater depends on the depth from which groundwater is extracted. Mechanized pumps are used to extract water for irrigation and use electricity or diesel as fuel in India. Centrifugal pumps are the most prominently used mechanized pumps (Raghunath 1982, 363) and are installed on the surface. Centrifugal pumps create a low pressure in the tube of a tube well by suction. As a result, the atmospheric pressure pushing down on water outside the well causes the water level in the tube to rise. The weight of the water inside the tube exerts a downward pressure and the water outside exerts an upward pressure. This process continues until the pressure inside and outside the tube well equalizes. In a perfect vacuum, the water would rise to a height of 34 feet (10.36 meters) because the weight of a column of this height exerts pressure equal to atmospheric pressure (Spellman 2004, 122). However, since suction cannot create a perfect vacuum, the accepted practical standard for extracting water using centrifugal pumps is approxi-mately eight meters (Gibson and Singer 1969, 116).

Therefore, if the depth to groundwater is more than eight meters, the centrifugal surface pumps cannot be used to access water. In such a scenario, a submersible pump that is placed inside the well tube is used to extract water (Gibson and Singer 1969, 116 and 124). The submersible pumps cost three times as much as centrifugal pumps.1 One concern might be that technology has improved since these estimates

1 These details have been outlined in Sekhri (2011), which uses this variation in cost of the pumps to estimate the effect of public wells on groundwater in a triple difference approach.


were published and submersible pumps may have become cheaper. To allay this con-cern, I obtained recent estimates of the cost of centrifugal and submersible pumps of various horsepower from an established pump vendor in this area. The difference in the cost of the pumps persists. For example, the cost of a 1 hp centrifugal pump is as large as Rs 6,830, whereas the cost of 1 hp submersible pump is Rs 16,750. 2

The difference in technology required is induced by one of the laws of physics, and this law generates an exogenous variation in cost as a function of water table depth (changing at eight meters). In the subsequent analysis, I estimate a reduced-form effect of access to groundwater on poverty, exploiting the cost of access that changes at eight meters as a source of spatial variation.

II. Data

A. main sample

The paper employs three main sources of data. The groundwater data at the vil-lage-level are from two waves (1993 and 2000) of the Minor Irrigation Census (MI Census) conducted by the Ministry of Water Resources, Government of India, on a quinquennial basis. The paper uses the MI data for five districts in Uttar Pradesh. Poverty data are cross-sectional, and are from a poverty census conducted by the government of Uttar Pradesh in 2002. These data provide the list of village residents determined to be “Below Poverty Line” (BPL status) using the criterion established by the government of India.3 A number of social benefit programs use this definition of poverty to identify beneficiaries. For example, the Targeted Public Distribution System (TPDS) provides higher subsidy for grains to households with BPL status. The flagship “Janani Suraksha Yojna” which provides cash incentives for institu-tional births, has provisions for higher benefits for BPL status mothers.

The Primary Census Abstract of the Population Census of India 2001, provides the village-level demographic variables including literacy rate, percentage of sched-uled caste population, number of households, female population, female literacy rate, and total population. A comprehensive set of village infrastructure variables are available in the village directory of the Census of India 2001. These variables include whether a village is electrified, and whether it has a school, a medical facil-ity, a bank, and bus service. The village directory also reports distance to the nearest town and expenditure of the village council on public goods. These four datasets have been matched at the village level.

Geographical controls, including rainfall and temperature, are from the University of Delaware climate data (see Matsuura and Willmott (2009a, b) for details). These data have been interpolated from station data to 0.5 degree grid cells.4 The village level annual rainfall and temperature data have been extracted using spatial data for the respective villages. Elevation and slope data have been extracted from the

2 These receipts are available on request.3 This methodology is based on asset based “proxy means testing.”4 Available at http://climate.geog.udel.edu/ climate/html_pages/download.html.

http://climate.geog.udel.edu/climate/html_pages/download.html


Digital Elevation Model of India (SRTM at 1 km resolution)5 by superimposing vil-lage spatial coordinates on this raster data.

The main sample is restricted to villages where the depth to groundwater is either under or over eight meters since 1993.6 The sample has 1,171 revenue villages. Table 1 provides the summary statistics. I use headcount as a measure of poverty. Headcount is the fraction of individuals below the poverty line in the total population of the vil-lage. The average headcount in this sample is 0.128 and the standard deviation is 0.11. By 2001, most of the villages had been electrified, but few had access to a banking facility or a dispensary within the village. Villages with groundwater above the cutoff that determines the feasibility of surface pumps are different from the ones below the cutoff on many dimensions. The most notable differences are scheduled caste popu-lation, number of households, rainfall, slope, elevation, access to schools, and total expenditure of the village council on public goods. However, one of the concerns about the identification strategy employed in the paper is whether these variables are smooth around the cutoff of eight meters below ground level (mgbl). More rigorous tests of differences in the two groups around the cutoff are presented in a later section.

5 The source for these data is the Global Land Cover Facility, www.landcover.org.6 The villages where depth to groundwater was less than eight meters in 1993, and then exceeded eight meters

in 2000, have been excluded. Also, villages where depth to groundwater was greater than eight meters in 1993, and then became less than eight meters in year 2000, have also been excluded. In a robustness test, I show that including these villages that cross-over does not influence the results, which are discussed later in Section VF.

Table 1—Summary Statistics Full Sample

Variable MeanStandarddeviation

Head count 0.128 0.11

demographyPercentage literate 0.28 0.12Percentage scheduled caste 0.3 0.24Number of households 141 160Female population 0.47 0.03Female literate population 0.14 0.1

climate and geographyRainfall 71.66 8.13Temperature 25.6 0.3Slope 0.12 0.2Elevation 107 48.6

Village infrastructureSchool 0.66 0.47Medical facility 0.13 0.33Distance from nearest town 11.4 10.44Power supply 0.8 0.4Bank facility 0.05 0.2Total panchayat expenditure 29,813 114,890

Observations 1,171


B. survey sample

As mentioned before, the primary data does not have information regarding sub-mersible pumps in the villages. Hence, the analysis is limited in that we cannot examine the first stage.7 In order to address this concern, I conducted a very short survey in 400 villages around the cutoff of 8 meters. The villages were chosen from depths between 3 and 12 meters.8 After restricting to this depth, 400 villages were randomly chosen.9 Ten farmers were randomly chosen in these villages. Table 2 shows the summary statistics for this subsample of villages. The characteristics of these villages look similar to those in the main sample reported in Table 1.

C. mechanisms

regression discontinuity sample—Survey of Living Conditions.—In order to address underlying mechanisms and bolster the findings from the main sample, I use the Survey of Living Conditions (SLC) 1997–1998, Uttar Pradesh and Bihar col-lected by the World Bank. The SLC dataset is representative of the area from which the main sample is drawn and provides comprehensive economic and social data collected through individual household surveys.10 The dataset has only 120 villages,

7 I wish to thank Seema Jayachnadran for her suggestions on this issue.8 These depths balanced the possible villages on either side of the cutoff.9 The sample size was chosen in light of the budget available for the survey.10 Anderson (2011) uses this dataset to study bilateral groundwater markets in this region.

Table 2—Summary Statistics Survey Sample


Head count 0.13 0.1

demographyPercentage literate 0.26 0.12Percentage scheduled caste 0.32 0.25Number of households 147.6 180.7Female population 0.47 0.03Female literate population 0.13 0.1

climate and geographyRainfall 70.3 4.6Temperature 25.6 0.3Slope 0.13 0.18Elevation 115.78 55.4

Village infrastructureSchool 0.65 0.47Medical facility 0.135 0.34Distance from nearest town 11.5 11.4Power supply 0.81 0.38Bank facility 0.05 0.22Total panchayat expenditure 24,889.34 66,659.45

Observations 400


but has comprehensive coverage of 2,250 households. Out of these villages, I was able to match 103 villages to rainfall, elevation, and the Minor Irrigation (MI) sur-vey. 11 The SLC data report farm-level inputs and outputs for households, agricul-tural yield per acre in value terms, wage income of individual household members, and migration information. I use the entire sample in the analysis around the discon-tinuity due to the limited number of villages in the sample. The summary statistics are reported in Table 3.

difference-in-differences sample—World Bank India Agricultural and Climate Data.—I examine the long-term benefits of access to groundwater, using the district level Agriculture and Climate in India (IAC) data collected by the World Bank. These data have comprehensive information about agricultural production, inputs, area under high-yielding varieties, and farm gate prices from 1957 through1987.12 The data also contain information about soil characteristics and geographical fea-tures, such as altitude and distance from the sea. These data include 271 districts in 13 major states in the country.13 I matched average annual rainfall and temperature from University of Delaware data to this dataset. These data also report the extent (thickness) of the aquifers underneath districts. The 3 reported binary variables indi-cate extent (thickness) greater than 150 meters, between 100 and 150 meters, and less than 100 meters. These 3 indicators are reported for 124 districts. The IAC data do not provide information about the extent of aquifers in the remaining districts. I used the Water Resources plates from the National Atlas of India that are the source

11 Spatial methods were used for matching, and due to same village name within the same districts, I was not able to isolate the coordinates of 17 villages. Three villages have some missing variables. Hence, I use data form 100 villages.

12 These data were compiled by Sanghi et al. (1998)(World Bank) and draw on data assembled by James McKinsey and Robert Evenson of the Yale Growth Center (Yale University).

13 All splits in districts have been accounted for in the data, and the boundaries correspond to 1961 boundaries.

Table 3—Summary Statistics Survey of Living Conditions


Annual wage earnings Individual 12,517.63 34,939.78Yield per acre (value) Household 763.12 1,597.18Households with seasonal out-migrants Household 0.31 0.46Fertilizer applied per acre Household 80.07 105.3Cultivated land Household 5.75 9.13Percentage irrigated land Household 85.16 27.27Fraction of SC households Village 0.25 0.2Rainfall shock Village −9.5 4.2Elevation Village 74.6 38.5Power supply Village 0.51 0.5School Village 0.8 0.38Bank Village 0.06 0.24Medical facility Village 0.41 0.45Caste homogeneity index Village 0.5 0.2

Villages 100Villages with 1993 groundwater depth > 8 meters

54


of the data to identify the remaining districts as areas with sporadic aquifers and verify this classification. I use the IAC aquifer classification for classifying the aqui-fer thickness of Indian districts in these four categories: thickest (aquifer greater than 150 meters), fairly thick (aquifer thickness between 100 and 150 meters), thick (aquifer thickness up to 100 meters), and sporadic (sporadic aquifers with little water). The thickest aquifers are under various districts of Punjab, Haryana, and Tamil Nadu—the model Green Revolution states in India. Fairly thick aquifers are under other major food-grain-producing states including West Bengal, and Uttar Pradesh. The thickness of aquifers was determined prehistorically. Jain, Agarwal, and Singh (2007) document the age of various aquifers in India. The youngest aqui-fers are from the Pleistocene age. I restrict the sample to the first three categories. I do not compare areas with and without aquifers, because these might differ sig-nificantly from each other in geological make-up. Rather, the identification relies on comparing areas with different water endowment. Table 4 reports the summary statistics.

III. Estimation Strategy

A. main Estimation strategy

One concern with comparing places that are on either side of the threshold is that they may differ in other unobserved correlates of groundwater that influence pov-erty. This would result in omitted variable bias. While this could be true in the tails, such unobserved variables are likely to be similar close to the threshold. Therefore, I estimate the impact of access to groundwater in very narrow intervals around the threshold. I use the threshold of eight meters, at which the technology required to extract groundwater changes, to isolate the causal estimates of the effect of ground-water access on poverty. The reduced-form empirical model can be characterized as follows:

(1) y i = α + β G i + δ X i + f ( w i ) + ϵ i ,

where y i is the outcome of interest in village i, G i is an indicator variable equal to 1 if the depth to groundwater exceeds eight meters below ground level, and β is the

Table 4—Summary Statistics Agriculture and Climate Dataset


Area under high yielding varieties (’000 hectares) 93,780.04 124,065.2Quantity of fertilizer applied (’000 tonnes) 13,189.72 19,201.93Average annual rainfall 85.66 43.6Average annual temperature 25.55 1.2Altitude 355.71 153.70Distance from sea 353.98 210.58

Number of districts 124Number of years 32


parameter of interest. X i is the vector of village-specific characteristics, f ( w i ) is a control function (a function of depth to groundwater in village i), and ϵ i denotes the error term. The design allays selection concerns by comparing otherwise similar observations just above and below a cutoff (the threshold that determines feasibility of cheaper surface pumps in this case).

Because the feasibility of the pumps changes technically at 10.36 meters (under perfect vacuum conditions), the industry standard of 8 meters provides a “fuzzy” discontinuity. Hence, I use this discontinuity in an instrumental variable approach. I show that the use of high-cost submersible pumps increases at the threshold using nonparametric estimation procedure. I then use the discontinuous change in feasibil-ity of low-cost surface pumps as an instrument for submersible pump ownership to examine the impact on poverty.

First, I use an regression discontinuity (RD) approach to examine the reduced-form effect of the change in feasibility of pumps on poverty. I follow Hahn, Todd, and Van der Klaauw (2001), and use local linear regressions to estimate the left and right limits of the discontinuity, where the difference between the two is the estimated treatment effect. The estimation is done in one step using a simple rectan-gular kernel. Although a triangular kernel puts more weight on observations closer to the cutoff and is boundary optimal (Cheng, Fan, and Marron 1997), I follow Lee and Lemieux (2010), who argue that a more transparent way of putting more weight on observations close to the cutoff is to re-estimate a model with a rectan-gular kernel using smaller bandwidths.14 The choice of bandwidth can also have a significant bearing on the results. I present results for a broad range of bandwidths. I show results for bandwidths five and two. In addition, I also carry out the estima-tion in the optimal bandwidth proposed by Imbens and Kalyanaraman (henceforth, IK bandwidth or IK optimal bandwidth) (Imbens and Kalyanaraman 2012).15I use this approach to show that there is a discontinuous increase in poverty, submersible pumps, and dispute over irrigation water around this threshold. I use a parametric approach to examine underlying mechanisms. I show that agricultural yield-per-acre and wage earnings from casual labor fall to the right of the cutoff. The nonparametric estimation of the change in submersible pumps around the cutoff offers the first stage of the two-stage least square procedure.

Because groundwater access status is unchanged since 1993, concerns about reverse causality are mitigated to some extent. If poverty affects depth to groundwa-ter, we would expect less groundwater to be extracted in poorer areas. This implies that such areas would be groundwater abundant. Therefore, I would expect to see a negative correlation between headcount and depth to groundwater. Figure 1 shows the prima facie evidence from the preliminary results that this is not the case. The figure shows the regression function of a local regression of head-count on depth to groundwater using the IK bandwidth. We observe a positive jump at the threshold of eight meters below ground level. 16

14 The results are not sensitive to the choice of the kernel.15 The IK bandwidth is designed to minimize MSE.16 Online Appendix Figure A1 (panel A) shows the local polynomial regression function with confidence

intervals.


B. Ancillary Estimation strategy for Long-Term mechanisms

High-yielding varieties (HYV) are sensitive to watering. Therefore, I hypoth-esize that groundwater abundance, which would influence cost of production, would result in significantly higher adoption of these varieties.

The thickness of the aquifers under Indian districts is used as a proxy for groundwa-ter abundance. Hence, adoption of HYV is compared across the districts with different endowments of groundwater. The high-yielding varieties arrived in India in 1966. I calculate the area under high-yielding varieties by year using the IAC data, and docu-ment it in online Appendix Figure A5. Before 1966, no HYV were planted in India. However, there was a sharp increase in the area in the post-1966 years. This rise in area under high-yielding varieties was also accompanied by a sharp rise in groundwa-ter irrigation, as shown in online Appendix Figure A6. Use of groundwater, which is popularly termed water by demand, as a source of irrigation increased significantly after 1966, and it quickly took over surface water as the main source of irrigation.17

I use the temporal variation in introduction of the HYV in India and spatial varia-tion in factor (groundwater) endowments to examine how groundwater abundant areas differed in their adoption of HYV. The approach of comparing areas with dif-ferential factor endowments has been used by Michaels (2011) to estimate returns to

17 This figure is based on data collated by the Directorate of Economics, Ministry of Agriculture in India.

Figure 1. Headcount on Depth to Groundwater

Notes: This figure plots the regression functions from a local linear regression of headcount on deviation of depth to groundwater from the cutoff that determines feasibility of surface pumps on either side of the cutoff. The optimal bandwidth proposed by Imbens and Kalyanaraman (2012) designed to minimize MSE has been used. The bandwidth is 2.64.

source: Author’s calculations

0

0.2

0.4

0.6

0.8

1

Hea

dcou

nt

‒10 0 10 20

Deviation of depth to groundwater (in m.b.g.l) from the cutoff


oil production in the United States and by Hornbeck and Keskin (2013) to examine adaptation of agriculture to climate shocks.

In the empirical approach, I formally estimate the following specification:

(2) y it = α 0 + α 1 A i T + α 2 A i FT + T t + α 3 A i T × post

+ α 4 A i FT × post + α 5 X it + d i × T t + ε it ,

where y it is the outcome of district i in year t. A i T is an indicator that takes the value one if the district is in the thickest aquifer category, and zero otherwise. A i FT is an indicator that takes the value one if the district is in the fairly thick aquifer category, and zero otherwise. These indicators control for aquifer category-specific, time-invariant differences in districts. T t are year indicators to capture year specific shocks. post is an indicator that takes the value one after 1966, and zero otherwise. X it are time varying characteristics of the districts. d i × T t is a district-specific trend to account for time-varying unobservable differences in districts. The districts in the thick aquifer category are the excluded group. α 3 and α 4 are the parameters of interest. Note that the sample excludes districts with sporadic aquifers and compares outcomes within gradation of aquifer thickness. The standard errors are clustered at the level of districts.

I also estimate the time-varying effects of the introduction of the Green Revolution across districts in different aquifer categories. The empirical model is as follows.

(3) y it = β 0 + β 1 A i T + β 2 A i FT + T t + Σ l=1957 1987 α l T A i T × T l

+ Σ l=1957 1987 α l FT A i FT × T l + ε it .

In this empirical specification, α l T and α l FT are streams of coefficients estimated for years 1957 to 1987. Year 1956 and its interactions are excluded. Standard errors are clustered at the district level.

IV. Results and Robustness Tests

A. Full-sample Estimation

I report the preliminary results of parametric specifications of equation (1) using the entire sample in Table 5. Column 1 reports the result of a regression of headcount on the indicator for depth to groundwater greater than eight without specifying a control function. The coefficient is positive and statistically significant at 1 percent. This indicates that a switch from groundwater depth less than 8 to greater than 8 leads to a 6.8 percent increase in poverty. In column 2, I control for linear depth to groundwater; and in column 3, I control for squared depth to groundwater. The result holds across these specifications and continues to be statistically significant at 1 percent. Adding a cubic control function in column 4 does not affect the coefficient or the standard error. Thus, I use quadratic control function in the specifications that follow. In online Appendix Table 1, I add a very comprehensive set of controls and


show that the results reported in column (1) are robust to including these controls. State might use discretion to classify below-poverty households across blocks. I show that results are robust to including block fixed effects. The details are provided in the online Appendix I: Robustness Tests and Ancillary Findings (1.1 and 1.2). If 8 mbgl is indeed the cutoff at which the feasibility of the surface pumps changes, and if there is a relationship between poverty and access to groundwater, then we should see an effect at 8 mbgl. We should not observe an effect at cutoff values lower than eight. I discuss this in online Appendix I: Robustness Tests and Ancillary Findings (1.3). The results are in online Appendix Figure A2 and online Appendix Table 2. Lateral velocity of groundwater is low. Depending on the velocity, it can be a few centimeters to a meter a year (Todd 1980). Hence, spatial externalities due to interconnectedness of aquifers may not arise over short durations of time. But to allay concerns over this, I allow for spatial correlation across villages in the empiri-cal model. The sensitivity test is reported in online Appendix Table 3 and discussion is provided in online Appendix I: Robustness Tests and Ancillary Findings (1.4).

B. regression discontinuity Estimates

The results from the nonparametric RD analysis are reported in Table 6. Columns 1 and 2 show the results for bandwidths five and two, respectively. Column 3 shows the results for the estimation carried out using the optimal bandwidth as proposed by Imbens and Kalyanaraman (2012).18 The results are similar to the full sample parametric estimates reported in Table 5 and are statistically significant across the various specifications. The preferred specification is column 3, which uses the opti-mal bandwidth. This implies that going from below the cutoff to above increases

18 The code is available at http://www.economics.harvard.edu/faculty/imbens/files/rdob.ado.

Table 5—Impact of Access to Groundwater on Poverty

Dependent variable: headcount

(1) (2) (3) (4)

Indicator for depth to water > 8 0.068*** 0.075*** 0.10*** 0.11***(0.0065) (0.007) (0.01) (0.01)

Water level linear No Yes Yes YesWater level squared No No Yes YesWater level cubed No No No Yes

Observations 1,171 1,171 1,171 1,171r2 0.08 0.08 0.09 0.1

Notes: Indicator for depth to water > 8 is an indicator variable that takes the value 1 if ground-water level since 1993 is at a depth greater than 8 meters below ground level. Robust standard errors are reported in parentheses.

*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.



headcount by 9.7 percent. This is a 1 standard deviation increase in head-count rate and is large in magnitude.

In the absence of other credible estimates of the impact of access to groundwa-ter on poverty, I rely on contextual information to ascertain if these magnitudes are plausible. Duflo and Pande (2007) estimate a 0.15 percent decline in poverty in districts that are downstream from a dam and a 0.77 percent increase in pov-erty in districts with dams. Groundwater irrigation is used much more extensively in India than dam irrigation. Groundwater is readily accessible relative to surface water. More than 60 percent of Indian irrigation is supported by groundwater and there were close to 20 million wells by 2000 (Ministry of Water Resources 2001). Out of a total irrigated area of 54.27 million hectares in India, 32.7 million hectares are irrigated using groundwater (Agriculture Census of India 2005). Sekhri (2013) evaluates the link between groundwater and agricultural production using data from Indian districts. Conditional of year fixed effects, district fixed effects, and district specific trends, this study shows that a 1 meter deviation of groundwater around the long-term mean in Indian districts leads to an 8 percent decrease in food grain production, a 5 percent decrease in cash crops, and a 9 percent decrease in water-intensive crops, whereas dam irrigation increases crop production by 0.34 percent (Duflo and Pande 2007). Given this large effect on crop production in contrast to dam irrigation, it is not surprising that the welfare effects are also large.

Online Appendix Figures A3 and A4 address the concerns about the smooth-ness of other variables around the discontinuity and the discussion is provided in online Appendix I: Robustness Tests and Ancillary Findings (1.10). I do not directly observe agricultural yield or agricultural production at the village level in the main sample. Such data for India is not available below district level from government sources. However, in order to address whether the effect is being mediated through agriculture, I evaluate how the groundwater irrigated area, as a fraction of total sown area, changes around the threshold. I also examine the surface water irrigated area. I find that the groundwater irrigated area reduces significantly in villages where sur-face pumps become infeasible, whereas there is no change in the surface irrigated area (Discussion is in online Appendix I: Robustness Tests and Ancillary Findings (1.5) and results are in online Appendix Table 4). In a later section, I use the SLC sample to estimate the affect on yield per acre and casual wage earnings.

Table 6—Nonparametric RDD Estimates of the Impact of Access to Groundwater on Poverty

Dependent variable: headcount

Bandwidth 5 Bandwidth 2Optimal

bandwidth

(1) (2) (3)

Indicator for depth to groundwater > 8 0.11*** 0.10** 0.097**(0.02) (0.049) (0.04)

Notes: Each column reports the estimated coefficient from a regression of headcount on indi-cator for depth to groundwater greater than 8 mbgl. The nonparametric specifications with dif-ferent bandwidths are reported in columns 1–3. Optimal bandwidth proposed by Imbens and Kalyanaraman (2012) is used in column 3.



C. survey-sample Estimation

I estimate the change in submersible pumps, headcount, and reported conflict over irrigation water in the survey sample using nonparametric RD. Figure 2 shows the regression functions from a local polynomial regression of the percentage of farmers with submersible pumps at the village level on depth to groundwater. The percent-age of farmers who report owning a submersible pump rises sharply around the cutoff. Table 7 shows the results from nonparametric regressions. In the first two columns, a bandwidth of five has been used. In columns 3 and 4, I use the IK optimal bandwidth. Columns 2 and 4 include covariates in the regressions. My most pre-ferred specification from column 3 yields a 27 percent increase in submersible pump owners, and this is highly statistically significant.19 The results for reported conflict are shown in Figure 3 and Table 8. Figure 3 shows the regression function from a local polynomial regression on either side of the cutoff. Table 8 reports the results of specifications similar to those in Table 7. The estimate using IK optimal band-width is reported in column 3. I observe that a jump in the depth escalates disputes over irrigation water by 25.22 percent. In an urban setting in Morocco, Devoto et al. (2012) also find very large reductions of 69 percent in self-reported conflict over drinking water. Their study shows that access to piped water completely eliminates such feuds in the treated area. Groundwater used in irrigation is a vital resource

19 Online Appendix Figure A1 (panel B) replicates Figure 1 and graphically shows the result for headcount in the survey sample.

0

20

40

60

80

Per

cent

age

with

sub

mer

sibl

e pu

mps

‒5 0 5 10

Normalized groundwater depth (mbgl)

Figure 2. Survey Sample

Notes: This figure graphs the regression functions from local polynomial regression of percent-age of farmers with submersible pumps on deviation of depth to groundwater from the cutoff that determines feasibility of surface pumps on either side of the cutoff.



influencing rural livelihood. Consistent with this other study, I find that water scar-city escalates conflict over water in rural settings as well.

D. Instrumental Variable Estimates

Since the setting indicates a fuzzy rather than a sharp design, I implement an instrumental variable procedure using the discontinuity in feasibility of cheaper sur-face pumps as an instrument. The results are reported in Table 9. Using the survey sample of 400 villages, I find that the change in groundwater depth at 8 increases the percentage of farmers with submersible pumps by 34.3 percent. T and F-statistics from this stage are reported. The instrument is strong. The second stage indicates that a 1 percent increase in submersible pumps increases headcount by 0.12 percent. This coefficient is highly statistically significant at 1 percent.

V. Mechanisms

A. yields, Labor Earnings, and other Inputs

I use the SLC data to examine the mediating channels. Figure 4 shows that yield-per-acre in value terms for villages where surface pumps are feasible is much higher, and it drops substantially at the cutoff. 20 I also examine how the wage income of casual labor is affected. The SLC dataset reports the days per month worked for every

20 This is reported by households that sell their output. As in Anderson (2011), the estimate is computed con-ditional on selling.

Table 7—Nonparametric RDD Estimates of the Impact on Percentage of Submersible Pumps

Survey sample

Dependent variable: percentage of submersible pumps

Bandwidth 5 Optimal bandwidth (3.6)

(1) (2) (3) (4)

Indicator for depth 29.5*** 22*** 27*** 22.6*** to groundwater > 8 (4.8) (6.1) (5.16) (6.6)

Covariates No Yes No Yes

Notes: Each column reports the estimated coefficient from a regression of the percentage of reported disputes on the indicator for depth to groundwater > 8 mbgl. The nonparametric spec-ifications with different bandwidths are reported in columns 1–4. Optimal bandwidth proposed by Imbens and Kalyanaraman (2012) is used in columns 3 and 4. Columns 2 and 4 include demographic, geographical, and infrastructure controls. Demographic controls include number of households, fraction of literate population, scheduled caste population, fraction of females in the population, and fraction of literate females. Village infrastructure includes availability of banking facilities, medical facilities, schools, electrification, distance to nearest town, and total expenditure of the village panchayat council. Geographical controls include annual rainfall, temperature, slope, and elevation. Sample includes 400 villages.




0

10

20

30

40

Per

cent

age—

disp

ute

over

irrig

atio

n w

ater

−5 0 5 10

Normalized groundwater depth (mbgl)

Figure 3. Disputes over Water—Survey Sample

Notes: This figure graphs the regression functions from local polynomial regression of per-centage of farmers reporting disputes over irrigation water on deviation of depth to ground-water from the cutoff that determines feasibility of surface pumps on either side of the cutoff.


Table 8—Nonparametric RDD Estimates of the Impact on Disputes over Irrigation Water

Dependent variable: percentage of farmers in a dispute over irrigation water

Survey sample

Bandwidth 5 Optimal bandwidth

(1) (2) (3) (4)

Indicator for depth 29.1*** 23.5*** 25.22*** 24.6*** to groundwater > 8 (5.4) (6.5) (6.19) (7.2)

Covariates No Yes No Yes

Notes: Each column reports the estimated coefficient from a regression of percentage of reported disputes on indicator for depth to groundwater > 8 mbgl. The nonparametric specifi-cations with different bandwidths are reported in columns 1–4. Optimal bandwidth proposed by Imbens and Kalyanaraman (2012) is used in columns 3 and 4. Columns 2 and 4 include demographic, geographical, and infrastructure controls. Demographic controls include number of households, fraction of literate population, fraction of scheduled caste population, fraction of females in the population, and fraction of literate females. Village infrastructure includes availability of banking facilities, medical facilities, schools, electrification, distance to near-est town and total expenditure of the village panchayat council. Geographical controls include annual rainfall, temperature, slope, and elevation. Sample includes 400 villages.




Table 9—Instrumental Variable Estimates of the Impact of Groundwater Access on Poverty

Survey sample

First stage Second stage

Dependent variable:percentage with

submersible pumps

Dependent variable:

headcount(1) (2)

Indicator for depth to 34.3*** Predicted percentage 0.0012*** groundwater > 8 (2.84) with submersible pumps (0.0003)

T-statistic 12.07 T-statistic 3.8F-statistic 25.34 F-statistic 15.28

Covariates Yes Yes

Notes: The discontinuity in the depth to groundwater that determines the feasibility of surface pumps is used as an instrument for submersible pumps. The controls include demographic, geographical, and infrastructure characteristics. Demographic controls include number of households, fraction of literate population, fraction of scheduled caste population, facilities, fraction of females in the population, and fraction of literate females. Village infrastructure includes availability of banking, medical facilities, schools, electrification, distance to nearest town, and total expenditure of the village panchayat council. Geographical controls include annual rainfall, temperature, slope, and elevation. Sample includes 400 villages.



650

700

750

800

850

900

Yie

ld p

er a

cre

in v

alue

‒10 ‒5 0 5 10

Normalized depth to groundwater

Figure 4

Notes: This figure graphs the regression functions from local polynomial regression of yield per acre (value terms) on deviation of depth to groundwater from the cutoff that determines feasibility of surface pumps on either side of the cutoff.



month, hours per day worked, and cash wage received. I compute the individual-level wage income from these data. Figure 5 shows the decrease in casual annual wage earnings around the cutoff. Table 10 reports the results from the regression analy-sis. Panel A reports the estimates for wage earnings and panel B for the yield-per-acre. Column 1 controls for village characteristics including availability of banks,

10,000

12,000

14,000

16,000

18,000

Ann

ual w

age

earn

ings

—ca

sual

uns

kille

d la

bor

−10 −5 0 5 10 15

Normalized depth to groundwater (mbgl)

Figure 5

Notes: This figure graphs the regression functions from local polynomial regression of annual wage income for unskilled labor on deviation of depth to groundwater from the cut-off that determines feasibility of surface pumps on either side of the cutoff.


Table 10—Estimates of Impact of Access to Groundwater on Yields and Wage Earnings

(1) (2) (3)

panel A. dependent variable: annual individual wage earnings (rupees)Indicator for depth −6,557.032*** −6,590.318*** −6,697.061*** to groundwater > 8 (1,207.6) (1,256.6) (1,268.5)

panel B. dependent variable: yield per acre in value terms (rupees/acre)Indicator for depth −600.55*** −525.7** −500** to groundwater > 8 (225.58) (226.16) (228.05)

Infrastructure and controls Yes Yes YesElevation and rain shocks No Yes YesCaste homogeneity No No Yes

Notes: Indicator for depth to groundwater > 8 is an indicator variable that takes value 1 if groundwater depth in 1993 is greater than 8 meters below ground. Infrastructure and demographic controls include village electrification, availability of banks, schools, medical facilities, and the share of scheduled caste population in the village. Robust standard errors are reported in parentheses. The household survey data from the World Bank Survey of Living Standards in Uttar Pradesh and Bihar (1997) are used for the analysis.




schools, medical facilities and electrification, and the fraction of scheduled caste households.21 In column 2, I control for deviation of rainfall from long-term means and elevation. Using the SLC data, Anderson (2011) shows that caste homogeneity in the village affects groundwater markets. Since groundwater markets can influ-ence depth to groundwater and poverty, I control for the caste homogeneity in the village in column 3 in addition to the previous controls.22 Yield-per-acre falls by 500 rupees in villages where surface pumps are infeasible. This is statistically significant at 5 percent and 1 _ 3 of a standard deviation. Annual wage earnings for casual labor fall by 6,697 rupees. This coefficient is statistically significant at 1 percent and is 0.2 of a standard deviation. In online Appendix Table 5, I show that use of other inputs, such as fertilizers, also falls to the right of the discontinuity, and individual farmers also report a significant reduction in the irrigated areas. One concern might be that agriculturally enterprising farmers may have sorted to the left of the discontinuity, in which case, we should find an increase in cultivated area to the left.23

B. Adoption of High-yielding Varieties

I estimate equation (2) for area under high-yielding varieties and report the coef-ficients on the interaction terms in Table 11. The first row reports the effect on the districts with the thickest aquifers post the arrival of the HYV in India. The second row reports the effect on districts with fairly thick aquifers. The excluded group is

21 Scheduled castes are historically marginalized population groups in India.22 Caste homogeneity index is calculated as the probability that two randomly selected individuals in the village

are from the same caste.23 Cultivated area is the farmed area that is used for agriculture. Farmers may make sowing decisions from

year to year depending on annual conditions of weather and other macro shocks. Hence, sown area is time variant (Mandal 1979). But I do not find evidence of this type of sorting. I discuss this test in online Appendix I: Robustness Tests and Ancillary Findings (1.6) and panel A of online Appendix Table 5.

Table 11—Effect of Groundwater Abundance on Area under High Yielding Varieties

Dependent variable: area under high yielding varieties in ’000 hectare

(1) (2) (3) (4)

Post × thickest aquifers (> 150m ) 114.7*** 70.8** 71.5** 76.7***(23.56) (28.15) (28.3) (28)

Post × thick aquifers (100–150m) 64.37*** 43.35** 43.5** 45.7**(16.4) (21.4) (21.4) (21.8)

Controls No Yes Yes YesYear fixed effects No No Yes YesDistrict-specific trends No No No Yes

Observations 3,936 3,936 3,936 3,936r2 0.35 0.47 0.63 0.81

Notes: Post is an indicator that takes value 1 for post-1966 years. Controls include geographical and edaphic char-acteristics including average annual rainfall and temperature, type of soil, altitude, and distance from sea interacted with post. Robust standard errors are clustered at the district level and reported in parentheses. Historical District-Level data are from the India Agricultural and Climate dataset.



the districts with thick aquifers. In the first column, I report the coefficients from a basic model with no additional controls. In column 2, I add annual average rainfall, temperature, soil type, altitude, and distance from sea.24 In column 3, I add year fixed effects to control for year-specific macro shocks to the economy. In column 4, I add district specific trends. These trends control for any underlying changes in the demographic and economic environment in the district over time.

As expected, the effects are larger for the districts with the thickest aquifers. The coefficient in the first column is large and positive for both interactions. After I add the controls, these coefficients drop, but subsequently remain stable upon adding year fixed effects or even district-specific trends. The results indicate an increase of 76.7 thousand hectares in area under HYV in the districts with the thickest aquifers (0.6 of a standard deviation) and 45.7 thousand hectares in area under HYV in dis-tricts with fairly thick aquifers (0.36 of a standard deviation). These estimates are highly statistically significant at the 1 and 5 percent significance levels, respectively, and large in magnitude. Table 12 shows a corresponding increase in the quantity of fertilizers applied. The application of fertilizers increased by 13.8 thousand tons in districts with the thickest aquifers and by 7 thousand tons in districts with fairly thick aquifers relative to districts with thick aquifers. These coefficients are similarly significant and large. I estimate equation (3) and plot the year-by-year estimates for the districts with the thickest aquifers and those with fairly thick aquifers relative to the districts with thick aquifers. Figure 6 shows the coefficients for the area under HYV. Online Appendix Table 7 reports the standard errors and t-statistics. The coef-ficients for the districts with the thickest aquifers and those with fairly thick aquifers are zero before 1966. The coefficients start diverging right after 1966, and then the

24 The latter three are interacted with the post indicator to allow for the controls to vary over time.

Table 12—Effect of Groundwater Abundance on Fertilizer Use

Dependent variable: quantity of total fertilizers used (’000 tonnes)(1) (2) (3) (4)

Post × thickest aquifers (> 150m) 22.3*** 13*** 13.4*** 13.8***(1.47) (3.8) (3.8) (3.7)

Post × thick aquifers (100–150m) 9.8*** 7.3*** 7.46*** 7**(0.7) (2.4) (2.4) (3.3)

Controls No Yes Yes YesYear fixed effects No No Yes YesDistrict-specific trends No No No Yes

Observations 3,936 3,936 3,936 3,936r2 0.3 0.41 0.63 0.78

Notes: Post is an indicator that takes the value 1 for post-1966 years. Controls include geographical and edaphic characteristics including average annual rainfall and temperature, type of soil, altitude, and distance from sea inter-acted with post. Robust standard errors are clustered at district level and reported in parentheses. Historical district level data are from the India Agricultural and Climate dataset.




divergence grows over time. The coefficients are highly statistically significant after 1966. Figure 7 plots the coefficients for quantity of fertilizer applied. This figure too shows similar patterns.

These patterns highlight that access to groundwater played an instrumental role in adoption of HYV in Indian agriculture. 25 Foster and Rosenzweig (1995) docu-ment high positive returns to HYVs in India. They show that profits per hectare were 1,100 rupees from use of conventional varieties, whereas they were 2,400 rupees

25 The earliest available estimates of poverty and the data that can be used to construct such measures are avail-able in 1983. Due to lack of predata, I cannot estimate the long-term effects on poverty or other welfare measures.

0.00

20000.00

40000.00

60000.00

80000.00

100000.00

120000.00

140000.00

160000.00

180000.00

200000.00

1957

19

58

1959

19

60

1961

19

62

1963

19

64

1965

19

66

1967

19

68

1969

19

70

1971

19

72

1973

19

74

1975

19

76

1977

19

78

1979

19

80

1981

19

82

1983

19

84

1985

19

86

1987

Thickest aquifers >150m

Thick aquifers 100–150m

Figure 6. Year-by-Year Regression Coefficients for Area under High Yielding Varieties


Figure 7. Year-by-Year Regression Coefficients for Quantity of Fertilizer Used


0.00

5,000.00

10,000.00

15,000.00

20,000.00

25,000.00

30,000.00

35,000.00

40,000.00

45,000.00

50,000.00

1957

19

58

1959

19

60

1961

19

62

1963

19

64

1965

19

66

1967

19

68

1969

19

70

1971

19

72

1973

19

74

1975

19

76

1977

19

78

1979

19

80

1981

19

82

1983

19

84

1985

19

86

1987

Thickest aquifer >150m

Thick aquifers 100-150m


per hectare from HYV. 26 Thus, the growth rate of area under high-yielding varieties suggests that the districts’ income profile changed differentially over the long-term horizon after 1966.

VI. Reverse Causality

Poverty rate can have an effect on groundwater use.27 In many contexts, poverty has been found to be positively correlated with resource abundance.28 The data does not bear out evidence for this resource curse hypothesis. In Figure 1 and online Appendix Figure A1, we clearly observe an opposite relation from that hypoth-esized by the resource curse literature. We see that poverty rates are higher in areas where water is relatively inaccessible or is scarce. If poverty rates precluded water access, then poverty rates would have been associated with water abundance, or in other words, groundwater would be found closer to the surface due to less deple-tion. This argument would be undermined if households to the left of the threshold were richer due to opting out of agriculture and, hence, extracted less groundwater. I address this in online Appendix I: Robustness Tests and Ancillary Findings and online Appendix Table 8, and do not find evidence that rich areas extract less. 29

VII. Establishment of Rural Settlements, Population Sorting, and Migration

Historical rural settlement patterns and their evolution over time have been exam-ined by geographers. Prior to the thirteenth century, independent tribes and septs comprised numerous small kingdoms and principalities in Eastern Uttar Pradesh. There was a significant movement of the Hindi-speaking Rajputs into this region in the eleventh through thirteenth centuries. Under the Rajputs, clan-based system of organizing settlements emerged (Singh 1968; Mandal 1979). A dominant clan formed the raj or the pargana, which was divided into subclan areas called tappas, and these were subdivided into smaller parcels called gaons or grams. A pargana was geographically contiguous and was based on loose ties. Wider kindred subclans lived in tappas and closer ones comprised the gaon. Marriages were formalized within kindred and migration of elites and Brahmins and other service providers within the original settlement was allowed (Singh 1968).

The original pargana in a region was settled by clearing forests and had a cow shelter called ghosa around which activity sprung. Over time, both pargana and tap-pas centers emerged as urban centers with “haats” or markets for the clan members.

26 Foster and Rosenzweig (1995) also document that irrigation assets have a significant influence on adoption of high-yielding varieties. Since groundwater endowments influence the fixed cost of owning irrigation assets, I consistently find that water abundance influenced adoption of HYV.

27 See Alix-Garcia et al. (2013) for a review of this literature and an application showing the effect of poverty alleviation on forest cover.

28 Several studies posit this is on account of extractive institutions that emerge in these areas.29 Note that the villages are classified on the basis of their 1993 depth and are water scarce or not since 1993.

Hence, the estimates are yielding medium-term effects of water accessibility. I do not have village-specific poverty rates at an earlier time period. Therefore, I cannot carry out a panel estimation. I estimated the impact of access to groundwater on poverty for the sample that includes villages where depth to groundwater crossed over the thresh-old since 1993. The estimated effects on headcount are similar to the ones reported in Table 5. Online Appendix Table 10 shows the results, which are similar to ones reported in Table 5.


Gaons emerged as residential communities. A system of Jajmani was adopted in which services were rendered to the land owners within a gaon and the gaon was self-contained for basic needs. The communities grew over generations and new gaons were carved when an original gaon became large. This system persisted through the subsequent muslim rule of the area. Under the British, the heads of the tappas became more formalized as “zamindaars,” but the basic structure persisted. After independence, tappas became the model for subdistricts called “tehsils.”30 The gaons formed the villages.

Different settlement structures evolved in these villages and varied depending on water-availability (Ahmad 1952). Masonry wells and pond water were the main sources of drinking water in these villages. In places where masonry wells were expensive, such as in areas with deep water tables, the population settled around wells or other water bodies, such as ponds. Thus, higher depth to groundwater resulted in more compact settlements grouped closer to each other.31 On the other hand, water abundance resulted in a scattered clustering of residential settlements with several hamlets scattered over a larger area (Ahmad 1952). The area in the sample is characterized by this type of clustering with scattered hamlets. The soil is uniformly fertile with interspersed pockets of sand.

Within villages, agricultural land use was characterized by fragmentation. The par-cels around the settlements were the most irrigated and heavily manured. The outer ring was used to produce fodder and staples. All farmers typically had parcels in these different rings. These patterns of residential and agricultural land use persisted even after laying of the canal systems (Ahmad 1952). Under the Zamindari system, a signifi-cant part the of pargana was under absentee landlords. Tenant farmers from backward castes cultivated these farms. When the Zamindari system was abolished, the ten-ant farmers were endowed ownership of lands. Consequently, backward agricultural castes became one of the most dominant cultivator-owners in this region. Scheduled castes typically resided in marginal areas of the villages around the settlements and cultivated in these marginal areas (Singh 1968). Abolition of Zamindari and land ceil-ing improved the holdings of the marginalized castes. But social norms and kinship ties aided the persistence of this structure of village organization. Land is not typically transacted. Divisions happen when households split or through inheritance. Anderson (2011) has documented the stability of this population over time.

These records of history of settlements in this area indicate that areas with abundant groundwater and scarce groundwater were settled alike according to the clan-based system. Water abundance determined the structure of residential organization within the clans and the comprising villages. Irrigation was not reliant on groundwater his-torically. In addition, only in the late 1960s did groundwater become a main source of irrigation—several hundred years after the settlements of the villages took root. Using the SLC data in Section VIA, I demonstrated that groundwater accessibility directly increased agricultural yields and wage earnings of casual labor, which may reduce poverty. Given that water accessibility influenced residential compactness of

30 These are the administrative units for collecting land revenue in modern India. The subclans of the tappas form the “up-jaatis” under the Hindu system of caste.

31 Insecurity from raids of other clans, such as Jats and Marathas, also contributed to compactness.


a village, it may undermine the ability to coordinate and demand public services due to geographical dispersion of residents. This can adversely affect poverty and other measures of welfare. Since overall poverty is lower in areas where groundwater is accessible, and public services, such as schools and medical facilities, do not look any different at the threshold, the second effect is potentially small and offset by an increase in yield and agricultural productivity.

A related issue is the interplay of water scarcity and migration. Water inacces-sibility could induce out-migration, which can affect poverty rates. Alternatively, the poor might have difficulty in migrating (Bryan, Chowdhury, and Mobarak 2011). Historically, migration rates out of villages have been low (Anderson 2011). In post-independence India, land markets are thin. Hence permanent migration is limited. However, households do resort to temporary migration. The SLC village sample allows me to examine seasonal migration patterns. Online Appendix Figure A7 shows that the percentage of households reporting that at least one member of the household seasonally migrated is higher in the villages where groundwater is accessible. This percentage drops in villages where surface pumps are infeasible, and hence groundwater is less accessible. Online Appendix Table 9 reports the estimates from the regression analysis. A discussion of the results is provided in online Appendix I: Robustness Tests and Ancillary Findings.

VIII. Caveats

Conflict and discrimination can also result in residential sorting of households into different types of villages. Poorer residents might be forced to areas with poor groundwater conditions. Historical records described in the last section indi-cate that such sorting occurs only within and not across villages. The population shares of different castes in villages have been shown to be stable in these areas (Anderson 2011). I also include percentage of scheduled caste, who are histori-cally marginalized poor population groups in India, in the regression analysis. But to the extent that this is only a crude measure to capture this, the data do not allow me to explicitly address this in more detail.

I do not have village-level consumption or income data by household to con-struct other measures of poverty. The data used are based on a survey that the government conducts. This survey assesses a household’s income potential and asset holding to assign a score to it. Then the government uses a cutoff score that isolates the poverty threshold. In this sense, the data are a proxy for poverty. This assigned poverty status is used in the allocations of benefits in all social benefit programs in India. Thus, although the analysis is limited in that I cannot look at other dimensions of poverty, it is informative because most policies are targeted based on this measure.

The data uses a cross section to compare the villages above and below the thresh-old. The “above or below the threshold” status in this cross section has been sta-ble since 1993. But given this is only seven years, I cannot identify if this is the short-run or long-run equilibrium. It is possible that this is short run, so that house-holds have not fully adapted yet, in which case if households can adapt by changing crop mixes or exiting agriculture in the long run, the poverty may reduce over longer


horizons. I demonstrate that groundwater endowments induce long-term invest-ments in yield-enhancing technologies that can have persistent effects on poverty. My results support the policy recommendation that sustainable access to groundwa-ter for rural poor is better from a welfare perspective than cutting access altogether, at least over medium-term periods such as a decade.

IX. Conclusion

This paper provides estimates of the impact of access to groundwater on rural pov-erty and conflict. A transition from below the cutoff of 8 meters below ground level to above the cutoff results in a 9–10 percent increase in the poverty rate. These results are robust to a wide array of functional form assumptions and specifications. I estimate the local average treatment effect (LATE), but the estimates do provide insights for policy design. The results imply that depleting groundwater reserves in areas that rely heavily on groundwater irrigation will lead to a substantial increase in poverty. From a policy perspective, shutting off access to groundwater in response to rapid depletion will have perverse effects on welfare. Hence, policies that promote sustainable use of groundwater should be at the center stage of discussions to protect groundwater reserves. Compared to other levers of poverty reduction, providing unchecked free access to groundwater results in immediate benefits in terms of increasing welfare, but could result in long-term costs as the reserves begin to deplete.

REFERENCES

Aggarwal, Rimjhim M., and Tulika A. Narayan. 2004. “Does inequality lead to greater efficiency in the use of local commons? The role of strategic investments in capacity.” Journal of Environmental Economics and management 47 (1): 163–82.

Agriculture Census of India. 2005. “Agriculture Census 2005-06.” http://agcensus.dacnet.nic.in/nationalcharacterstic.aspx (accessed March 2012).

Ahmad, Enayat. 1952. “Rural Settlement Types in the Uttar Pradesh.” Annals of the Association of American Geographers 42 (3): 223–46.

Alix-Garcia, Jennifer, Craig McIntosh, Katherine R. E. Sims, and Jarrod R. Welch. 2013. “The Eco-logical Footprint of Poverty Alleviation: Evidence from Mexico’s Oportunidades Program.” review of Economics and statistics 95 (2): 417–35.

Anderson, Siwan. 2011. “Caste as an Impediment to Trade.” American Economic Journal: Applied Economics 3 (1): 239–63.

Badiani, Reena, and Katrina K. Jessoe. 2011. “Electricity subsidies for agriculture: Evaluat-ing the impact and persistence of these subsidies in India.” http://econweb.ucsd.edu/cee/papers/Jessoe_4april.pdf.

Banerji, A., J. V. Meenakshi, and Gauri Khanna. 2012. “Social Contracts, Markets and Efficiency: Groundwater Irrigation in North India.” Journal of development Economics 98 (2): 228–37.

Bhattarai, Madhusudan, and A. Narayanamoorthy. 2003. “Irrigation and other Factors Contribu-tion to the Agricultural Growth and Development in India: A Cross-State Panel Data Analysis for 1970 to 94.” http://www.iwmi.cgiar.org/assessment/files_new/publications/Workshop%20Papers/TATA_2003_Bhattarai2.pdf.

Brill, Thomas C., and H. Stuart Burness. 1994. “Planning versus Competitive Rates of Groundwater Pumping.” Water resources research 30 (6): 1873–80.

Bryan, Gharad, Shyamal Chowdhury, and A. Mushfiq Mobarak. 2011. “Seasonal Migration and Risk Aversion: Experimental Evidence from Bangladesh.” Bureau for Research and Economic Analysis of Development (BREA) Working Paper 319.

Cheng, Ming-Yen, Jianqing Fan, and J. S. Marron. 1997. “On automatic boundary corrections.” Annals of statistics 25 (4): 1371–1803.

http://agcensus.dacnet.nic.in/nationalcharacterstic.aspx

http://econweb.ucsd.edu/cee/papers/Jessoe_4april.pdf

http://www.iwmi.cgiar.org/assessment/files_new/publications/Workshop%20Papers/TATA_2003_Bhattarai2.pdf

http://pubs.aeaweb.org/action/showLinks?crossref=10.1016%2FS0095-0696%2803%2900034-2

http://pubs.aeaweb.org/action/showLinks?system=10.1257%2Fapp.3.1.239

http://pubs.aeaweb.org/action/showLinks?crossref=10.1029%2F94WR00535

http://pubs.aeaweb.org/action/showLinks?crossref=10.1080%2F00045605209352092

http://pubs.aeaweb.org/action/showLinks?crossref=10.1016%2Fj.jdeveco.2011.07.005

http://pubs.aeaweb.org/action/showLinks?crossref=10.1214%2Faos%2F1031594728


Conley, T. G. 1999. “GMM estimation with Cross Sectional Dependence.” Journal of Econometrics 92 (1): 1–45.

Devoto, Florencia, Esther Duflo, Pascaline Dupas, William Parienté, and Vincent Pons. 2012. “Happi-ness on Tap: Piped Water Adoption in Urban Morocco.” American Economic Journal: Economic policy 4 (4): 68–99.

Duflo, Esther, and Rohini Pande. 2007. “Dams.” Quarterly Journal of Economics 122 (2): 601–46.Foster, Andrew D., and Mark R. Rosenzweig. 1995. “Learning by Doing and Learning from Oth-

ers: Human Capital and Technical Change in Agriculture.” Journal of political Economy 103 (6): 1176–1209.

Foster, Andrew D., and Mark R. Rosenzweig. 1996. “Technical Change and Human-Capital Returns and Investments: Evidence from the Green Revolution.” American Economic review 86 (4): 931–53.

Foster, Andrew, and Mark Rosenzweig. 2008. “Inequality and the Sustainability of Agricultural Pro-ductivity Growth: Groundwater and the Green Revolution in Rural India.” http://poppov.org/Por-tals/1/documents/papers/Foster_Rosenzweig.pdf.

Foster, Andrew, and Sheetal Sekhri. 2007. “Can Expansion of Markets for Groundwater Decelerate the Depletion of Groundwater Resource in Rural India?” Unpublished.

Garduño, Héctor, and Stephen Foster. 2010. sustainable Groundwater Irrigation: approaches to rec-onciling demand with resources. World Bank.

Gibson, Ulrich P., and Rexford D. Singer. 1969. small Wells manual: Location, design, construction, Use and maintenance. Washington DC: Department of State Agency for International Development.

Gisser, Micha, and David A. Sánchez. 1980. “Competition versus Optimal Control in Groundwater Pumping.” Water resources research 16 (4): 638–42.

Global Land Cover Facility. 2004. “Shuttle Radar Topography Mission, 1 Arc Second scene SRTM Unfilled Unfinished 2.0.” http://staff.glcf.umd.edu/sns/branch/htdocs.sns/data/srtm/.

Hahn, Jinyong, Petra Todd, and Wilbert Van der Klaauw. 2001. “Identification and Estimation of Treatment Effects with a Regression Discontinuity Design.” Econometrica 69 (1): 201–09.

Hornbeck, Richard, and Pinar Keskin. 2014. “The Historically Evolving Impact of the Ogallala Aqui-fer: Agricultural Adaptation to Groundwater and Drought.” American Economic Journal: Applied Economics 6 (1): 190–219.

Imbens, Guido, and Karthik Kalyanaraman. 2012. “Optimal Bandwidth Choice for the Regression Discontinuity Estimator.” review of Economic studies 79 (3): 933–59.

Imbens, Guido W., and Thomas Lemieux. 2008. “Regression Discontinuity Designs: A Guide to Prac-tice.” Journal of Econometrics 142 (2): 615–35.

Jacoby, Hanan G., Rinku Murgai, and Saeed Ur Rehman. 2004. “Monopoly Power and Distribution in Fragmented Markets: The Case of Groundwater.” review of Economic studies 71 (3): 783–808.

Jain, Sharad K., Pushpendra K. Agarwal, and Vijay P. Singh. 2007. Hydrology and Water resources of India. Netherlands: Springer.

Jha, B. M., and S. K. Sinha. 2009. “Towards Better Management of Ground Water Resources in India.” http://www.cgwb.gov.in/documents/papers/incidpapers/Paper%201-B.M.Jha.pdf.

Jin, Songqing, Hans G. P. Jansen, Rie Muraoka, and Winston Yu. 2012. “The Impact of Irrigation on Agricultural Productivity: Evidence from India.” Paper presented at the International Association of Agricultural Economists (IAEE) Triennial Conference, Foz do Iguacu, August 18–24.

Keskin, Pinar. 2009. “Thirsty Factories, Hungry Farmers: Intersectoral Impacts of Industrial Water Demand.” http://www.pinarkeskin.com/Keskin%20--%20Thirsty%20Factories.pdf.

Lee, David S., and Thomas Lemieux. 2010. “Regression Discontinuity Designs in Economics.” Journal of Economic Literature 48 (2): 281–355.

Mandal, R. B. 1979. Introduction to rural settlement. New Delhi: Concept Publishing Company.Matsuura, Kenji, and Cort J. Willmott. 2009a. “Terrestrial Air Temperature: 1900-2008 Gridded

Monthly Time Series (V 2.01).” http://climate.geog.udel.edu/~climate/html_pages/Global2_Ts_2009/README.global_t_ts_2009.html.

Matsuura, Kenji, and Cort J. Willmott. 2009b. “Terrestrial Precipitation: 1900-2008 Gridded Monthly Time Series (V 2.01).” http://climate.geog.udel.edu/~climate/html_pages/Global2_Ts_2009/README.global_p_ts_2009.html.

Michaels, Guy. 2011. “The Long Term Consequences of Resource-Based Specialisation.” Economic Journal 121 (551): 31–57.

Ministry of Water Resources. 2001. report on census of minor Irrigation schemes 1993-94. Govern-ment of India. New Delhi, March.

Minor Irrigation Census. 2000. Third minor Irrigation census of India. Government of India, Minis-try of Water Resources. March.

http://climate.geog.udel.edu/~climate/html_pages/Global2_Ts_2009/README.global_t_ts_2009.html

http://climate.geog.udel.edu/~climate/html_pages/Global2_Ts_2009/README.global_p_ts_2009.html

http://pubs.aeaweb.org/action/showLinks?crossref=10.1086%2F601447

http://pubs.aeaweb.org/action/showLinks?crossref=10.1111%2F1468-0262.00183

http://pubs.aeaweb.org/action/showLinks?crossref=10.1016%2FS0304-4076%2898%2900084-0

http://pubs.aeaweb.org/action/showLinks?crossref=10.1111%2Fj.1467-937X.2004.00304.x


http://pubs.aeaweb.org/action/showLinks?system=10.1257%2Fpol.4.4.68

http://pubs.aeaweb.org/action/showLinks?system=10.1257%2Fjel.48.2.281

http://pubs.aeaweb.org/action/showLinks?crossref=10.1029%2FWR016i004p00638

http://pubs.aeaweb.org/action/showLinks?crossref=10.1111%2Fj.1468-0297.2010.02402.x

http://pubs.aeaweb.org/action/showLinks?crossref=10.1093%2Frestud%2Frdr043

http://pubs.aeaweb.org/action/showLinks?crossref=10.1162%2Fqjec.122.2.601

http://pubs.aeaweb.org/action/showLinks?crossref=10.1016%2Fj.jeconom.2007.05.001


Population Census of India. 2001. Registrar of India, Government of India (http://censusindia.gov.in/).Raghunath, H. M. 1982. Ground Water: Hydrogeology, Ground Water survey and pumping Test, rural

Water supply and Irrigation systems. New Delhi: Wiley Eastern Limited.Rodell, Matthew, Isabella Velicogna, and James S. Famiglietti. 2009. “Satellite-based estimates of

groundwater depletion in India.” Nature 460: 999–1002.Sanghi, Apurva, K. S. Kavi Kumar, and James W. McKinsey, Jr. 1998. “India Agriculture and Climate

Data Set.” http://ipl.econ.duke.edu/dthomas/dev_data/datafiles/india_agric_climate.htm.Sekhri, Sheetal. 2011. “Public Provision and Protection of Natural Resources: Groundwater Irrigation

in Rural India.” American Economic Journal: Applied Economics 3 (4): 29–55.Sekhri, Sheetal. 2012. “Sustaining Groundwater: Role of Policy Reforms in Promoting Conservation

in India.” In India policy Forum, edited by Shekhar Shah, Barry Bosworth, and Arvind Panagariya, 149–86, Vol. 9. Thousand Oaks: Sage Publications.

Sekhri, Sheetal. 2013. “Missing Water: Agricultural Stress and Adaptation Strategies in Response to Groundwater Depletion in India.” http://www.virginia.edu/economics/RePEc/vir/virpap/papers/virpap406.pdf.

Sekhri, Sheetal. 2014. “Wells, Water, and Welfare: The Impact of Access to Groundwater on Rural Poverty and Conflict: Dataset.” American Economic Journal: Applied Economics. http://dx.doi.org/10.1257/app.6.3.76.

Shah, Shekhar, Barry Bosworth, and Arvind Panagariya, eds. 2012. India policy Forum, 2012–13. Vol. 9. Thousand Oaks: Sage Publications.

Singh, Kashi N. 1968. “The Territorial Basis of Medieval Town and Village Settlement in Eastern Uttar Pradesh, India.” Annals of the Association of American Geographers 58 (2): 203–20.

Spellman, Frank R. 2004. mathematics manual for Water and Wastewater Treatment plant operators. Boca Raton: CRC Press.

Todd, David Keith. 1980. Groundwater Hydrology. 2nd ed. New York: John Wiley and Sons.World Bank. 1998. “Survey of Living Conditions, Uttar Pradesh and Bihar, 1997–1998.” Unpub-

lished.

http://www.virginia.edu/economics/RePEc/vir/virpap/papers/virpap406.pdf

http://dx.doi.org/10.1257/app.6.3.76


http://pubs.aeaweb.org/action/showLinks?crossref=10.1111%2Fj.1467-8306.1968.tb00640.x

http://pubs.aeaweb.org/action/showLinks?crossref=10.1038%2Fnature08238

wells, water, and welfare: the impact of access to groundwater … · 2017. 3. 16. · wells,...

Documents