Arjunan Subramanian
University of Glasgow, Glasgow, UK Parmod Kumar
Institute of Social and Economic Change, Bangalore, India
Innovation and welfare effects of extension services: Experimental evidence from India
India Agricultural outlook forum 2019 Ministry of Agriculture & Farmers Welfare
Government of India
Based on the project
“Information, Market Creation and Agricultural Growth”
funded by
3
Motivation
Research Questions
• Can farmers’ income be doubled?
• Can providing agricultural information improve farm productivity among small holder agriculture?
• What are the labour market consequence of increased farm incomes?
• What are the consequence of increased farm income
Farming practices below optimal
• New pests and diseases
• Development of resistance by old pests
• New seed varieties with better traits
• Change in chemical composition of soil Climate Change!
Huge potential exists for yield increase & reduction in cost of cultivation
• Better sprays
• Choice of appropriate variety of seeds
• Application of fertilizer at the right time and quantity
5
How information is delivered?
Traditional agriculture extension
• Considerable resources spent by government
• Few farmers report contact/limited evidence of impact
• Serious governance issues
• Concerns about two-way information flow
• Agro-dealers mainly provide information – perverse
Novelty of the study
General equilibrium effects – doubling farmers’
income
Relaxing multiple constraints with extension
information – credit; inputs; water; soil quality
Quantifying spillover from information dissemination
7
Matrix of effects in yield (quintals/ acre)
Intervention
Effect size
Estimation strategy
9
𝑂
𝑖𝑡= 𝛽
0+ 𝛽
1𝑇𝑟𝑒𝑎𝑡
𝑖+ 𝛽
2𝑆𝑝𝑖𝑙𝑙 + 𝛽
3𝑂
𝑖0+ 𝑌
𝑡+ 𝛿
𝑣+ 𝜀
𝑖𝑡𝑂
𝑖𝑡- outcome of interest in crop plot i in period t
𝑇𝑟𝑒𝑎𝑡
𝑖- dichotomous variable equal to 1 if household received treatment
𝑆𝑝𝑖𝑙𝑙 – dichotomous equal to 1 if spillover farmers 𝑂
𝑖0- value of the dependent variable at the baseline 𝑌
𝑡- year fixed effects
𝛿
𝑣- group fixed effects
𝜀
𝑖𝑡- error term
11
Intervention–relax multiple constraints
• Providing real-time, comprehensive and
contextual agricultural information to treated farmers
Crop production – soil testing, fertilizer, pesticide
Livestock production – feed fodder, diseases control
Regular updates of input and output price
Eligibility on agricultural credit
Crop insurance
Cattle insurance
13
Intervention–relax multiple constraints
• We hired scientists from UAS Bangalore and Raichur
• Disciplines: agronomy; entomology; pathology;
biotechnology; genetics; agri economics
• Tablet with information and real-time link with experts at Agri Universities
• Meet treated farmers every 12th day in their farm
• Provide information on every aspect throughout
the crop-cycle for 3 seasons
Focus crops
Gubbi Siriguppa
Paddy Paddy
Red Gram Bengal Gram
Ragi Sunflower
Cotton
15
There are 30 other crops grown that includes horsegram;
maize; sugarcane; cowpea; barley; groundnut; castor; and
green gram
Samples split by treatment group
Statistics Number of
samples Farmers total
Treatment farmers Spillover farmers Control farmers
1320 600 120 600
Survey timeline – panel data
Survey timeline Reference year
Baseline round 0 Follow-up round 1 Follow-up round 2 Follow-up round 3
June 2012- May 2013 July 2013 – May 2014 June 2015 – May 2016 June 2016 – May 2017
17
Household incomes
Dep. Var. :
Household income from sources:
Log crop
income Log livestock
income
Log off- farm income
Log nonfarm income
Treated Spillover
0.3567**
(0.1554) 0.2641*
(0.1551)
-0.1456 (0.3283)
-0.4740 (0.3313)
0.0912***
(0.0267) 0.0458 (0.0637)
-0.3521***
(0.0851) -0.3805**
(0.1269)
Control mean Dep. Var. 216,213 -3793 12,993 29,697 Village FE
Year FE
YES
YES YES
YES YES
YES YES
YES
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Labour market – extensive margin
19
Dep. Var. :
Number of family and hired labour Crop
cultivation Livestock Off-farm
work Nonfarm work
Treated Spillover
9.2727***
(2.2913) 9.6129***
(1.8131)
-0.0675 (0.2476)
0.0335 (0.2645)
0.0908***
(0.0344) 0.2322***
(0.0519)
-0.1636*
(0.0995) -0.1636 (0.1122)
Control mean Dep. Var. 53.4825 1.0122 2.5017 1.3265 Village FE
Year FE
Clustered SE
YES YES YES
YES YES YES
YES YES YES
YES YES YES
R-squared Observations
0.2542
4,250 0.3416
1,983 0.4741
2,084 0.3630 2,084
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Standard errors in parentheses clustered by GP code (12 clusters) * p < 0.1, ** p < 0.05, *** p < 0.01
Labour use – intensive margin
Dep. Var. :
Hours /days worked to total employed Crop
cultivation Livestock Off-farm
work Nonfarm work
Treated Spillover
-0.4201**
(0.1905) -0.3365**
(0.1660)
0.1044 (0.1431) 0.4911**
(0.1770)
3.8965 (3.1539) 11.0604*
(6.5880)
1.3263 (9.9914)
7.9254 (15.0317)
Control mean Dep. Var. 1.7351 3.3320 103 293 Village FE
Year FE
YES
YES YES
YES YES
YES YES
YES
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Household crop income - post treatment
21
Dep. Var. :
Crop income by program round
One round Two rounds Three rounds Treated
Spillover
0.3085**
(0.1594) 0.2284 (0.1650)
0.2953 (0.1865)
0.1709 (0.1944)
0.3567**
(0.1554) 0.2641*
(0.1551) Control mean Dep.
Var. 197,613 187,027 216,213
GP FE Year FE
Clustered FE
YES YES YES
YES YES YES
YES YES YES R-squared
Observations 0.6719
811 0.5067
1,171 0.4502
1,599
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Standard errors in parentheses clustered by GP code (12 clusters) * p < 0.1, ** p < 0.05, *** p < 0.01
Crop yield – all crops
Dep. Var. :
Crop yield by program round
One round Two rounds Three rounds Treated
Spillover
0.2648***
(0.0606) 0.3072***
(0.0736)
0.4818***
(0.0769) 0.4723***
(0.0891)
0.5072***
(0.0789) 0.4763***
(0.0922) Control mean Dep.
Var. 18.3985 17.8756 18.0919
GP FE
Year FE YES
YES YES
YES YES
YES
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Crop yield – cotton
23
Dep. Var. :
Cotton yield by program round
One round Two rounds Three rounds Treated
Spillover
0.1168 (0.0782)
0.1175 (0.1090)
0.1310*
(0.0700) 0.1534 (0.0981)
0.1240**
(0.0546) 0.1002 (0.0856) Control mean Dep.
Var. 7.9717 6.6782 6.6017
GP FE Year FE
Clustered FE
YES YES YES
YES YES YES
YES YES YES R-squared
Observations 0.4950
601 0.4572
852 0.4070
1,088
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Standard errors in parentheses clustered by GP code (12 clusters) * p < 0.1, ** p < 0.05, *** p < 0.01
Crop yield – paddy
Dep. Var. :
paddy yield by program round
One round Two rounds Three rounds Treated
Spillover
0.1737***
(0.0425) 0.2186***
(0.0474)
0.1796***
(0.0253) 0.2162***
(0.0281)
0.1659***
(0.0280) 0.2020***
(0.0296) Control mean Dep.
Var. 24.6525 25.1372 25.8278
GP FE
Year FE YES
YES YES
YES YES
YES
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Crop profit - cotton
25
Dep. Var. : Profits per acre -yearly
2013-
2014 2014-
2016 2016-
2017 All years Treated
Spillover
-0.2359*
(0.1366) -0.4206 (0.3585)
-0.3507 (0.3314) -1.0326*
(0.6155)
-0.0104 (0.2704)
-0.4816 (0.3496)
-0.1240 (0.1391)
-0.3997 (0.2466) Control mean Dep. Var. 190.4224 627.6106 6934.918 3287 GP FE
Year FE
Clustered FE
YES YES YES
YES YES YES
YES YES YES
YES YES YES R-squared
Observations 0.4049
218 0.3036
127 0.4061
146 0.3718 364
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Standard errors in parentheses clustered by GP code (12 clusters) * p < 0.1, ** p < 0.05, *** p < 0.01
Crop profit - paddy
Dep. Var. :
Profits per acre -yearly
2013-2014 2014-2016 2016-2017 All years
Treated Spillover
0.3938***
(0.1043) 0.3294**
(0.1096)
0.7690***
(0.0534) 0.6876***
(0.0705)
0.9006***
(0.2586) 0.9202***
(0.2604)
0.4789***
(0.0234) 0.4408***
(0.0389)
Control mean Dep. Var. 17,581 21,579 27,648 21,311 GP FE
Year FE
Clustered FE
YES YES YES
YES YES YES
YES YES YES
YES YES YES
R-squared 0.2887 0.0818 0.2570 0.2531
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Price received from crop sold
27
Dep. Var. : Price per quintal
Cotton Paddy
Treated Spillover
0.0264**
(0.0138) 0.0235*
(0.0179)
0.0690***
(0.0150) 0.0676***
(0.0157)
Control mean Dep. Var. 4370 1521
GP FE Year FE
Clustered FE
YES YES YES
YES YES YES R-squared
Observations 0.4860
1,088 0.4292
2,295
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Standard errors in parentheses clustered by GP code (12 clusters) * p < 0.1, ** p < 0.05, *** p < 0.01
Returns to intervention for all years- cotton
Dep. Var. :
Crop-wise plot level per acre: cotton Log Yield
(quintal) Log profits
(Rs)
revenue Log (Rs)
Log cost (Rs) Treated
Spillover
0.1240**
(0.0546) 0.1002 (0.0856)
-0.1240 (0.1391)
-0.3997 (0.2466)
0.1480**
(0.0545) 0.1209 (0.0925)
0.1392***
(0.0370) 0.0744*
(0.0414) Control mean Dep. Var. 6.6017 3287 28736 25449 GP FE
Year FE YES
YES YES
YES YES
YES YES
YES
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Returns to intervention for all years- paddy
29
Dep. Var. : Crop-wise plot level per acre: paddy Log Yield
(quintal) Log profits
(Rs) Log revenue
(Rs)
Log cost (Rs)
Treated Spillover
0.1659***
(0.0280) 0.2020***
(0.0296)
0.4789***
(0.0234) 0.4408***
(0.0389)
0.2009***
(0.0300) 0.2343***
(0.0313)
0.1281***
(0.0275) 0.1347***
(0.0326)
Control mean Dep. Var. 25.8278 21,311 40,665 19,354 GP FE
Year FE
Clustered FE
YES YES YES
YES YES YES
YES YES YES
YES YES YES
R-squared Observations
0.2225
2,295 0.2531
1,978 0.2905
2,295 0.2069 2,426
𝑂 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑇𝑟𝑒𝑎𝑡 𝑖 + 𝛽 2 𝑆𝑝𝑖𝑙𝑙 + 𝛽 3 𝑂 𝑖0 + 𝑌 𝑡 + 𝛿 𝒗 + 𝜀 𝑖𝑡
Standard errors in parentheses clustered by GP code (12 clusters) * p < 0.1, ** p < 0.05, *** p < 0.01
Crop profit and household welfare
Consumption per capita per day Cereals Pulses Fruit&
Vegetable Diary Meat Sugar Elasticity,
log(profit) 0.001
(0.048) 0.418**
(0.149) -0.018
(0.063) 0.353*
(0.214) 0.518*
(0.309) 0.206**
(0.090) Control mean 0.324 0.040 0.193 0.112 0.092 0.039 GP FE
Year FE
Clustered FE
Yes Yes Yes
Yes Yes Yes
Yes Yes Yes
Yes Yes Yes
Yes Yes Yes
Yes Yes Yes R-squared 0.494 0.197 0.634 0.129 0.369 0.320
𝑪 𝑖𝑡 = 𝛽 0 + 𝛽 1 𝑃𝑟𝑜𝑓𝑖𝑡 𝑖 + 𝛽 2 𝐶 𝑖0 + 𝑌 𝑡 + 𝛿 𝑔𝑝 + 𝜀 𝑖𝑡
Results summary
• Crop income for treated relative to control increased by 35%
• Combined crop income (treated+spillover) increased by
56% as a result of treatment while for paddy returns almost doubled
• Off-farm income also increases but only by 9% from labour reallocation from non-farm to farm
• Revenues for cotton and paddy were higher while greater cotton cultivation costs reduced profits
• Farmers received higher prices for cotton and paddy
relative to control farmers
31Thank you very much
Research design – randomized control trials
Information dissemination experiment
• Sample selection at household/village/gram panchayat?
• Strong spill over effects smaller the area
• Two stage randomization procedure – (1) GP (2) HH
• Random selection of spill over group within the village
• 50 treatment households + 10 spill over households
• 50 control households
Components of input costs
Dep. Var. :
Input use Crop-wise plot level
All crops Cotton Paddy Plowing
Sowing
0.100**
(0.047) -0.015 (0.033)
0.149**
(0.065) 0.088***
(0.032)
0.091 (0.065)
-0.063 (0.041)
Interculture 0.134*
(0.073) 0.181***
(0.077) Weeding
Ferilizer application
0.102***
(0.039) 0.062**
(0.027)
0.279***
(0.076) 0.102**
(0.052)
0.067 (0.048) 0.082**
(0.036) Insecticide application 0.056* 0.103** 0.019