Technological Change and Political Turnover FIGURE 1.Mean HYV Crop Adoption and Change in Congress Seat Share A.HYV Crop Adoption B.Congress Party Decline 回 0.000to0.126 國 -79.0to-33.39 口 0.128to0.170 ■ -33.39to-19.88 口 0.170to0.245 口 -19.88to-6.76 0.245to0.341 -6.76to8.25 0.341to0.727 8.25to52.0 Notes:Panel A:Districts shaded according to quintile of mean share of agricultural land under HYV crop cultivation between 1967- 1987.Panel B:Districts shaded according to quintile of pre/post-1967 percentage point change in the Congress party's seat share in state assembly elections. 91707.50109 the Indian government,based on agricultural sample adoption between 1967 and 1987 Panel B displays over- surveys conducted by local district-level officials.This time change in the Congress party's election perfor- data was compiled into a district-level panel dataset mance,computed by subtracting the average percent- covering 270 districts across India's major states be- age of seats won in elections after the introduction of tween 1957 and 1987 by Sanghi et al.(1998).Connect- HYV crops in 1967(inclusive)from the average per- ing the political and agricultural data yielded a panel centage of seats won before 1967 dataset linking district-level HYV crop adoption data A concern regarding a panel fixed effects OLS ap- to constituency-level electoral data for over 21,000 proach is that incumbent politicians may have strate- state assembly election races across India's major states gically manipulated the diffusion of HYV crops over between 1957 and 1987 Figure 1 displays district-level time,especially as agricultural policy became increas- maps of the data.Panel A displays mean HYV crop ingly politicized as a result of rural political mobiliza- tion.The direction of the bias implied by such dynamic targeting is unpredictable,depending upon whether 5 The districts covered in the dataset span all of India's major states the Congress party targeted supporters,opposition ar- except for Assam and Kerala,and account for over 85%of India's land area.To account for administrative splits of districts over time. eas,or swing voters (Dixit and Londregan 1996).To all data are aggregated to the level of 1961 district boundaries.During address endogeneity in the diffusion of HYV crops, the period under analysis,India experienced three rounds of redis- this paper utilizes an instrumental variables identifica- tricting of assembly constituency boundaries:in 1961,1967 and 1976. tion strategy.An instrumental variables identification This paper therefore utilizes the 1956,1961,1967,and 1976 reports of the Delimitation Commission to name match constituencies to con strategy also eliminates downward bias in coefficient temporaneous districts.It then utilizes the Administrative Atlas of estimates resulting from measurement error,which is India to match contemporaneous districts to 1961 boundaries. likely to be sizable,as a result of sampling error in the 923
Technological Change and Political Turnover FIGURE 1. Mean HYV Crop Adoption and Change in Congress Seat Share A. HYV Crop Adoption 0.000 to 0.126 0.128 to 0.170 0.170 to 0.245 0.245 to 0.341 0.341 to 0.727 B. Congress Party Decline −79.0 to −33.39 −33.39 to −19.88 −19.88 to −6.76 −6.76 to 8.25 8.25 to 52.0 Notes: Panel A: Districts shaded according to quintile of mean share of agricultural land under HYV crop cultivation between 1967– 1987. Panel B: Districts shaded according to quintile of pre/post-1967 percentage point change in the Congress party’s seat share in state assembly elections. the Indian government, based on agricultural sample surveys conducted by local district-level officials. This data was compiled into a district-level panel dataset covering 270 districts across India’s major states between 1957 and 1987 by Sanghi et al. (1998).5 Connecting the political and agricultural data yielded a panel dataset linking district-level HYV crop adoption data to constituency-level electoral data for over 21,000 state assembly election races across India’s major states between 1957 and 1987. Figure 1 displays district-level maps of the data. Panel A displays mean HYV crop 5 The districts covered in the dataset span all of India’s major states except for Assam and Kerala, and account for over 85% of India’s land area. To account for administrative splits of districts over time, all data are aggregated to the level of 1961 district boundaries.During the period under analysis, India experienced three rounds of redistricting of assembly constituency boundaries: in 1961, 1967, and 1976. This paper therefore utilizes the 1956, 1961, 1967, and 1976 reports of the Delimitation Commission to name match constituencies to contemporaneous districts. It then utilizes the Administrative Atlas of India to match contemporaneous districts to 1961 boundaries. adoption between 1967 and 1987. Panel B displays overtime change in the Congress party’s election performance, computed by subtracting the average percentage of seats won in elections after the introduction of HYV crops in 1967 (inclusive) from the average percentage of seats won before 1967. A concern regarding a panel fixed effects OLS approach is that incumbent politicians may have strategically manipulated the diffusion of HYV crops over time, especially as agricultural policy became increasingly politicized as a result of rural political mobilization. The direction of the bias implied by such dynamic targeting is unpredictable, depending upon whether the Congress party targeted supporters, opposition areas, or swing voters (Dixit and Londregan 1996). To address endogeneity in the diffusion of HYV crops, this paper utilizes an instrumental variables identification strategy. An instrumental variables identification strategy also eliminates downward bias in coefficient estimates resulting from measurement error, which is likely to be sizable, as a result of sampling error in the 923 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:06, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S000305541800031X
Aditya Dasgupta TABLE 1.Pretreatment Determinants of HYV Crop Adoption Dependent variable: HYV Intensity IADP (1) (2) (3) (4) (5) (6) Pre-INC 0.001 0.0000 (0.0004) (0.001) Pre-Irrigation 0.52* 0.24* (0.03) (0.06) Aquifer 0.15** 0.01 (0.02) (0.03) Constant 0.20* 0.13# 0.17** 0.05 -0.003 0.05* (0.03) (0.01) (0.01) (0.04) (0.02) (0.02) Observations 270 270 270 270 270 270 Notes:Unit of observation is district.Outcomes:HYV Intensity is mean district-level HYV crop adoption,1967-1987 IADP is an indicator for whether a district was selected for the Intensive Agricultural Districts Program.Explanatory variables:Pre-INC is pre-1967 average Congress seat share.Pre-irrigation is share of agricultural land with access to irrigation in 1966.Aquifer is the share of district land with a naturally occurring aquifer.Analysis estimated by OLS. p<0.1:*p<0.05;*p<0.01 4号 agricultural surveys from which the official statistics on via district and region-year fixed effects,which also ab- HYV crop adoption are obtained,as well as the aggre- sorb time invariant omitted variables as well as region- gated level,the district,at which HYV crop adoption is specific time shocks.The second stage regression is & measured. the main OLS specification.The exclusion restriction To construct an instrument for HYV crop adoption, requires that,conditional upon covariates,areas with this paper utilizes the fact that HYV crops,the cultiva- greater aquifer coverage saw a larger over-time decline tion of which required intensive and steady dosages of in the Congress party's election performance after 1967 water,delivered yield improvement only in conditions only as a result of differentially higher rates of HYV of access to controlled irrigation.As Table 1 illustrates, crop adoption over time and not for other reasons.The for this reason,after their introduction,HYV crops plausibility of this assumption is discussed and tested overwhelmingly disseminated to districts with preexist- following the main results. ing irrigation infrastructure or capacity for irrigation in To measure district-level aquifer coverage,this pa- the form of aquifers.The first stage regression takes the per utilized historical groundwater maps published in form: the National Atlas of India.based on information from the Geological Survey of India.The advantage of uti- HYVrdr yrd+in+BAquiferrd x Post967+vrdu. lizing historical maps,first published in 1977 is that they provide information about aquifers that was available where Aquiferrd x Post967 is the instrument,a cross- contemporaneously as the green revolution was occur- ring.The maps contain detailed local outlines of three sectional measure of the share of district land with a aquifer types in order of depth:(i)fairly extensive thick naturally occurring aquifer interacted with a dummy aquifers occurring beyond 150 meters,(ii)aquifers with variable that"switches on"'for all districts with the in- limited extent occurring between 100 meters and 150 troduction of HYV crops to India from 1967 onward. meters,and (iii)aquifers with restricted extent occur- This paper uses aquifer coverage as the cross-sectional ring up to 100 meters.To quantify district-level aquifer suitability measure since aquifers are naturally occur- coverage,the groundwater maps were first geocoded to ring and more clearly exogenous than preexisting irri- shapefiles of 1961 district boundaries.GIS software was gation infrastructure..Lower-order terms are absorbed utilized to generate polygons corresponding to aquifers and to compute the share of each district's surface area intersecting with an aquifer of any type.A snapshot of 6 As Table 1 illustrates,aquifer coverage was also less subject to ex- the coding process is depicted in Figure Al of the On- plicit government targeting by government programs,like the IADP. line Appendix. intended to disseminate HYV crops.Though less clearly exogenous, a possible advantage of preexisting irrigation coverage as the cross- sectional suitability measure is that it is more strongly predictive of REGRESSION RESULTS HYV crop adoption than is an instrument based on aquifer cover- age,suggesting a bias-variance tradeoff (Conley,Hansen,and Rossi A simplified version of the OLS regression analysis is 2012).Table A4 in the Online Appendix reports the main results us- /:sony ing preexisting irrigation coverage as an alternative cross-sectional visualized in Figure 2,which depicts a scatter plot of suitability measure in the instrument.The results are substantively change in Congress party vote share and seat share comparable. from the pre-to the post-1967 HYV crop introduction 924
Aditya Dasgupta TABLE 1. Pretreatment Determinants of HYV Crop Adoption Dependent variable: HYV Intensity IADP (1) (2) (3) (4) (5) (6) Pre-INC 0.001 0.0000 (0.0004) (0.001) Pre-Irrigation 0.52∗∗∗ 0.24∗∗∗ (0.03) (0.06) Aquifer 0.15∗∗∗ 0.01 (0.02) (0.03) Constant 0.20∗∗∗ 0.13∗∗∗ 0.17∗∗∗ 0.05 − 0.003 0.05∗∗ (0.03) (0.01) (0.01) (0.04) (0.02) (0.02) Observations 270 270 270 270 270 270 Notes: Unit of observation is district. Outcomes: HYV Intensity is mean district-level HYV crop adoption, 1967–1987. IADP is an indicator for whether a district was selected for the Intensive Agricultural Districts Program. Explanatory variables: Pre-INC is pre-1967 average Congress seat share. Pre-irrigation is share of agricultural land with access to irrigation in 1966. Aquifer is the share of district land with a naturally occurring aquifer. Analysis estimated by OLS. ∗ p < 0.1; ∗∗ p < 0.05; ∗∗∗ p < 0.01 agricultural surveys from which the official statistics on HYV crop adoption are obtained, as well as the aggregated level, the district, at which HYV crop adoption is measured. To construct an instrument for HYV crop adoption, this paper utilizes the fact that HYV crops, the cultivation of which required intensive and steady dosages of water, delivered yield improvement only in conditions of access to controlled irrigation. As Table 1 illustrates, for this reason, after their introduction, HYV crops overwhelmingly disseminated to districts with preexisting irrigation infrastructure or capacity for irrigation in the form of aquifers. The first stage regression takes the form: HYVrdt = γrd + τrt + βAqui f errd × Post1967 t + νrdit, where Aqui f errd × Post1967 t is the instrument, a crosssectional measure of the share of district land with a naturally occurring aquifer interacted with a dummy variable that “switches on”’ for all districts with the introduction of HYV crops to India from 1967 onward. This paper uses aquifer coverage as the cross-sectional suitability measure since aquifers are naturally occurring and more clearly exogenous than preexisting irrigation infrastructure.6 Lower-order terms are absorbed 6 As Table 1 illustrates, aquifer coverage was also less subject to explicit government targeting by government programs, like the IADP, intended to disseminate HYV crops. Though less clearly exogenous, a possible advantage of preexisting irrigation coverage as the crosssectional suitability measure is that it is more strongly predictive of HYV crop adoption than is an instrument based on aquifer coverage, suggesting a bias-variance tradeoff (Conley, Hansen, and Rossi 2012). Table A4 in the Online Appendix reports the main results using preexisting irrigation coverage as an alternative cross-sectional suitability measure in the instrument. The results are substantively comparable. via district and region-year fixed effects, which also absorb time invariant omitted variables as well as regionspecific time shocks. The second stage regression is the main OLS specification. The exclusion restriction requires that, conditional upon covariates, areas with greater aquifer coverage saw a larger over-time decline in the Congress party’s election performance after 1967 only as a result of differentially higher rates of HYV crop adoption over time and not for other reasons. The plausibility of this assumption is discussed and tested following the main results. To measure district-level aquifer coverage, this paper utilized historical groundwater maps published in the National Atlas of India, based on information from the Geological Survey of India. The advantage of utilizing historical maps, first published in 1977,is that they provide information about aquifers that was available contemporaneously as the green revolution was occurring. The maps contain detailed local outlines of three aquifer types in order of depth: (i) fairly extensive thick aquifers occurring beyond 150 meters, (ii) aquifers with limited extent occurring between 100 meters and 150 meters, and (iii) aquifers with restricted extent occurring up to 100 meters. To quantify district-level aquifer coverage, the groundwater maps were first geocoded to shapefiles of 1961 district boundaries.GIS software was utilized to generate polygons corresponding to aquifers and to compute the share of each district’s surface area intersecting with an aquifer of any type. A snapshot of the coding process is depicted in Figure A1 of the Online Appendix. REGRESSION RESULTS A simplified version of the OLS regression analysis is visualized in Figure 2, which depicts a scatter plot of change in Congress party vote share and seat share from the pre- to the post-1967 HYV crop introduction 924 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:06, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S000305541800031X