1. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
The Value of Advice: Evidence from Agricultural
Production Practices
Shawn Cole (Harvard) and Nilesh Fernando (Harvard)
IFPRI
August 2, 2012
2. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Agricultural Extension Services
Widely credited for speeding the green revolution
But...
“often fail due to inadequate consultation of farmers about
their information needs” (Babu et al., 2012)
are costly, reach few, and su¤er from limited accountability
(Anderson and Feder, 2007)
Arrival of low-cost ICT may provide improved way to convey
information
3. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Today
Evaluate a mobile-phone, voice-based agricultural advice and
information service
“Avaaj Otalo,” or AO, a for-pro…t startup in Gujarat
RCT with 1,200 cotton farmers in 40 villages, randomized at
individual level
400 get AO & Physical Agricultural Extension
400 get AO only
400 serve as pure controls
[No control group of Extension only]
Impact on:
Sources of information
Agricultural knowledge
Real outcomes
Peer e¤ects and learning:
Information sharing
Learning by observation
Peer agricultural behavior
4. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Motivation
Modern growth theory attempts to explain productivity
di¤erences within and across countries through varying
technology use
Large productivity di¤erences in crop yields exist within and
across countries. To what extent are these di¤erences
explained by ine¢ cient agricultural practices?
Does a lack of awareness and technical know how explain the
limited adoption of pro…table agricultural investments in the
context of Gujarat? Through what mechanisms do such
’informational ine¢ ciencies’limit technology adoption?
How do farmers share information? Does informing some
farmers lead to positive spillovers, or less observation and
information exchange? What factors serve to promote or limit
the di¤usion of agricultural technologies?
5. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Contributions
Rigorous evaluation of agricultural extension
E¢ cacy of training (Financial literacy, management
consulting)
Test of validity of rural surveys via mobile phone
6. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Speci…c Open Questions
Impact: Can e¤ective agricultural extension be delivered via
mobile phone?
Inform general debate on delivery of public services via ICTs.
Is ICT education a substitute or complement to traditional
(in-person) extension?
Demand-Driven Extension: What is the importance of
’top-down’information provided by experts versus ’bottom-up’
information generated by users?
Predictors of Adoption: What demographic factors explain
di¤erences in technology use?
Technology Di¤usion: Is valuable agricultural information
shared among peers?
7. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Location in Literature
Identifying the Impact of Agricultural Extension : Bardhan &
Mookherjee (2011), Gandhi et al. (2009), Evanson et al.
(1990)
Explaining Technology Adoption in Agriculture : Du‡o,
Kremer and Robinson (2011), Suri (2011), Udry & Conley
(2010)
Markets for Advice : Anagol & Kim (2012), others????
Productivity and Technology (Banerjee-Du‡o 2005,
Hsieh-Klenow 2009)
Entire …eld of agricultural economics
8. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Feedback sought
What are the most pressing questions in agricultural
extension?
Particular mechanisms worth testing
Alternative applications in our setting
9. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Avaaj Otalo
Avaaj Otalo (“Voice Stoop”)
Based on open source software (hence scalable)
Mobile and voice-based touch-tone platform
Good for low literacy environments
Facilitates consistent delivery and reception of information
Easy to monitor information delivery, evaluate services
Enables farmers to receive, solicit, and share agricultural
information
Bottom-up and top-down agricultural information
Gujarat-based startup, founded by Stanford and Berkeley
computer scientists
About 8 implementations in 6 states of India (information
network for sex workers, ag info, etc.)
10. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Motivation for AO
Training & Visit method of agricultural extension considered
unsustainable (IFPRI 2010)
Up to 97 percent of Ag Extension budgets pay salaries,
leaving little resources for …eld visits
Caste and gender limits use of e-Choupal kiosks (Kumar,
2004)
Provides information on information needs quickly and cheaply
AO provides ongoing ‡ow of information rather than one-shot
training
Agri-input dealers provide advice, but may have incentives to
recommend incorrect quantities or even products
Local NGO, DSC, has provided radio program with farming
practices for 5 years in study area
11. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Avaaj Otalo Service: Main Features
Push calls
Question and answer service
Experience Sharing
Farmers can volunteer agricultural practice information,
perspectives, etc.; respond to others
Radio Program
Normal implementation: farmer pays airtime
Our implementation: computer provides free callbacks in
response to a missed call
12. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Push Calls
Delivered every Wednesday to all 800 treatment respondents
Average Length: ~5 minutes
Each message is based on:
Weekly calls to ten randomly selected farmers about their
information needs for the following week
Questions from incoming calls the week before
Weather information from Indian Meteorological Dept.
Agronomists’knowledge of crop phase, and agricultural best
practices
13. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Incoming Call Features
QnA: Farmers can record their own questions as well as listen
to or respond to existing questions and answers
Typical response time ~24 hours
Announcements: This forum contains all the push messages
that are sent out weekly by DSC and CMF
Radio: Many episodes on agriculture are available from a
radio show run by DSC over the previous …ve years.
Experience-sharing: This forum encourages farmers to share
their own innovative practices.
Personal Inbox: Gives users access to their own questions and
messages.
14. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Value Proposition
Provides customized, timely, regular and relevant agricultural
information
Mitigates failures of traditional extension systems
Addresses spatial failures by providing geography-speci…c
information, and mobile-based delivery decreases cost of
delivery and thus has higher reach
Addresses temporal failures by delivering information that is
sensitive to local weather conditions, and is available 24/7
Addresses institutional failures by delivering information that is
demand-driven, and the web-based moderation platform allows
for centralized monitoring of extension delivery
Demand-driven
Builds on trust and expertise established by DSC, a local NGO
15. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Physical Extension
Two hours long; one session in Kharif (monsoon), one in Rabi
(winter)
Run by NGO (AKRSP), not government
Invited 400 people from treatment group, 168 came
Provided free transportation and a meal, no other
compensation
At NGO site, ca. 10-50 km from respondents households
Rabi session focuses on:
Wheat and cumin variety selection
Cotton pesticide usage
Based on time in season and informed by AO questions
20-30 people per session
16. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
AO vs. Physical Extension Cost
AO: Assume 20 minutes of push calls and 18 minutes of
incoming usage each month
Per Farmer Cost Monthly Cost (USD) Monthly Cost (USD)
(N=800) (N=2000)
Airtime .60 .60
Agronomist .90 .36
AO System .40 .17
Total Monthly 1.90 1.13
One agronomist can handle 2,000 farmers on a regular basis
Physical Extension Cost: $8.50 per farmer
17. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Sample Selection
Focus on cotton farmers
Important crop for millions of farmers
Similar varieties and irrigation methods over large area
Well-settled science, but uncertainty about practice remains
Identify 40 villages in which DSC has a good presence
Identify all cotton farmers in village (NGO workers created
lists)
Selected 30 from these lists at random, strati…ed by subcaste
Assigned 10 to Control, 10 to AO, and 10 to AO & Physical
Extension
18. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Project Timeline
Surveying and Intervention Timeline
Date Event
May/June 2011 Cotton planting decisions begin
May 2011 Listing for Baseline Survey
July 2011 Baseline (Paper) Survey
August 2011 AO training for treatment respondents
August 2011 AO missed call service activated, first treatment message delivered
October 2011 Encouragement reminder calls
November 1-6, 2011 Physical Extension
November 10, 2011 Round 1 of phone survey
November 21, 2011 Peer / General Reminder Calls Begin
November 2011 Cumin planting decisions for Rabi 2012
December, 2011 Round 2 of phone survey
March, 2012 Peer survey
July, 2012 Midline survey
July, 2012 2nd AO training session starts for treatment respondents
19. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Avaaz Otalo Administrative Data
All calls, duration of call, features access
Whether individual listens to push call or not
Whether control group calls in to non toll-free line
Linked by mobile phone number
20. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Results Map
Randomization Check and Summary Statistics
Impact Evaluation
Randomization Check
Sources of Information
Knowledge Index
Pesticide Purchase and Usage
Sowing Decisions
Peer E¤ects and Learning
Information Sharing Behavior
Spillovers within study sample
Spillovers outside of study sample
Attrition and other concerns
21. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Empirical Approach
Standard simple di¤erence estimator, village …xed-e¤ects,
robust standard errors
Comparing AO and AO+Extension group to controls
yi ,v = αv + β (AO or AO&Extension ) + ε i
Sample size with paper survey 1,200=398 control + 399 AO +
403 AO&E
Sample size with phone survey 737=369+184+184
Comparing AO to control
yi ,v = αv + β (AO ) + ε i
Sample size with paper survey: 797=398+399
Sample size with phone survey: 553=369+184
22. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Sample Characteristics and Randomization Check
Balance Information Sources (Paper)
Cotton Fertilizer Cotton Pesticide Cumin Planting
Control (AO+AOE)-C Control (AO+AOE)-C Control (AO+AOE)-C
Cell contents: Mean ITT Mean ITT Mean ITT
(3) (4) (5) (6) (9) (10)
Asked for or received advice 0.265 -0.005 0.594 0.035 0.131 -0.022
(0.442) (0.027) (0.492) (0.030) (0.337) (0.020)
N 392 1180 392 1180 398 1200
Importance of source consulted
Past experience 0.048 -0.014 0.030 -0.016 - 0.023
(0.215) (0.025) (0.171) (0.012) (0.016)
Gov't extension 0.010 -0.010 0.004 0.002 0.019 -0.019
(0.098) (0.010) (0.066) (0.006) (0.139) (0.019)
NGO 0.019 -0.014 0.004 0.002 - 0.011
(0.138) (0.014) (0.066) (0.006) (0.012)
Other farmers 0.731 -0.009 0.421 -0.005 0.635 0.101
(0.446) (0.054) (0.495) (0.039) (0.486) (0.082)
Input shops 0.144 0.027 0.515 0.019 0.269 -0.062
(0.353) (0.043) (0.501) (0.040) (0.448) (0.076)
N for Mean 104 233 52
N for ITT Regression 309 729 139
No di¤erence for Cotton and Wheat Planting, or by treatment
23. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Sample Characteristics and Randomization Check
Randomization mostly successful, except Cotton 2010
Balance Planting (Paper)
Control Group AO Only AO+Extension AO-C AOE-C (AO+AOE)-C
Cell contents: Mean Mean Mean ITT ITT ITT
(1) (2) (3) (4) (5) (6)
A. Sample Size
Entire Sample 398 399 403 797 801 1200
B. Planting in Kharif '10
Planted Cotton 0.98 0.98 0.99 -0.01 0.00 0.00
(0.12) (0.15) (0.11) (0.01) (0.01) (0.01)
Area Cotton Planted 4.45 5.01 4.74 0.57** 0.29 0.43*
(3.62) (4.05) (4.43) (0.27) (0.29) (0.24)
Planted Wheat 0.78 0.72 0.72 -0.05* -0.05 -0.05**
(0.42) (0.45) (0.45) (0.03) (0.03) (0.03)
Area Wheat Planted 1.17 1.35 1.07 0.18 -0.10 0.04
(1.35) (2.30) (1.25) (0.13) (0.09) (0.09)
Planted Cumin 0.42 0.40 0.41 -0.02 -0.01 -0.02
(0.49) (0.49) (0.49) (0.03) (0.03) (0.03)
Area Cumin Planted 0.76 0.79 0.70 0.03 -0.06 -0.02
(1.41) (1.50) (1.34) (0.10) (0.10) (0.09)
Imbalance in reported area cotton planted in Kharif 2010
(No cotton imbalance in Kharif 2011)
24. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Usage (First Stage)
Initially, some concern about usage / take-up
AO service …rst provided free by IBM/DSC, high usage
Subsequently required farmers to pay own airtime, usage
dropped
Maximize research power with free service for treatment group
DSC started to charge farmers nominal fee (ca. $4/year)
Could be …nancially self-sustaining even without subscription
(airtime charges)
25. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Usage (First Stage)
First Stage: Effect of Assignment on AO Usage and Physical Extension Visits
Control (AO+AOE)-C AO-C AOE-C
Mean ITT ITT ITT
(1) (2) (3) (4)
Listened to more than 50% of push calls 0 0.755*** 0.745*** 0.766***
(0.017) (0.024) (0.024)
Called AO 0 0.603*** 0.555*** 0.650***
(0.017) (0.023) (0.025)
Duration of usage 0 72.548*** 53.866*** 90.828***
(10.644) (8.902) (18.589)
Winsorized duration of usage (90%) 0 45.341*** 40.106*** 50.540***
(4.140) (4.534) (5.783)
Total no. of questions asked 0 1.441*** 1.22*** 1.656***
(0.201) (0.205) (0.272)
Total no. of times Q&A accessed 0 3.416*** 2.28*** 4.496***
(0.571) (0.338) (0.957)
Attended Physical Extension 0.005 0.214*** 0.013* 0.413***
(0.071) (0.017) (0.007) (0.033)
26. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Question Topics
Thematic breakdown from Start until Round 1 Phone Survey:
Topic Percent of Calls
Pest Management 59
Fertilizers 8
Seeds 1
Crop Planning 5
Others 24
27. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Sources of Information: Planting Decisions
AO may complement or substitute information collection
Impact: Information Sources
Cotton Planting Wheat Planting Cumin Planting
Control (AO+AOE)-C Control (AO+AOE)-C Control (AO+AOE)-C
Cell contents: Mean ITT Mean ITT Mean ITT
(1) (2) (7) (8) (9) (10)
Importance of source consulted
Past experience 0.612 0.078** 0.138 -0.040* 0.179 -0.035
(0.488) (0.035) (0.346) (0.024) (0.384) (0.027)
Gov't extension 0.008 -0.003 0.000 0 0.005 -0.005
(0.090) (0.006) (0.074) (0.004)
NGO 0.043 -0.008 0.014 -0.003 0.005 0.005
(0.204) (0.014) (0.116) (0.008) (0.074) (0.007)
Mobile phone-based 0.003 0.087*** 0 0.052*** 0 0.125***
(0.052) (0.015) (0.012) (0.017)
Other farmers 0.230 -0.127*** 0.033 -0.019* 0.070 -0.046***
(0.422) (0.027) (0.178) (0.011) (0.256) (0.016)
Input shops 0.070 -0.016 0.000 0.005 0.011 0.000
(0.256) (0.018) (0.000) (0.004) (0.104) (0.008)
35
28. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Sources of Information: Agricultural Inputs
Impact: Information Sources
Cotton Fertilizer Cotton Pesticide
Control (AO+AOE)-C Control (AO+AOE)-C
Cell contents: Mean ITT Mean ITT
(3) (4) (5) (6)
Importance of source consulted
Past experience 0.496 0.029 0.291 -0.018
(0.501) (0.037) (0.455) (0.033)
Gov't extension 0.011 0.005 0.008 0.003
(0.104) (0.009) (0.091) (0.007)
NGO 0.051 -0.013 0.044 -0.011
(0.221) (0.015) (0.206) (0.014)
Mobile phone-based 0.003 0.223*** 0.006 0.297***
(0.052) (0.022) (0.074) (0.024)
Other farmers 0.252 -0.149*** 0.177 -0.073***
(0.435) (0.028) (0.382) (0.026)
Input shops 0.146 -0.081*** 0.446 -0.190***
(0.354) (0.022)
35 (0.498) (0.035)
29. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Impact on Agricultural Knowledge
Measure agricultural knowledge with a series of ten questions
related to agricultural practices
Asked once at baseline paper survey, once in round 1 phone
survey
Example questions
“If money were not a constraint, what is the best pesticide for
cotton white‡y?”
“Which fungicide should be applied to control wilt in cotton?”
“Which variety of cumin is recommended as wilt-resistant?”
30. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Impact on Agricultural Knowledge
Impact: Agricultural Knowledge
Baseline Survey Round 1 Phone Survey
Control (AO+AOE)-C Control (AO+AOE)-C
Cell contents: Mean ITT Mean ITT
(1) (2) (3) (4)
Total 0.289 -0.001 0.350 0.008
(0.212) (0.014) (0.173) (0.011)
Cotton-related 0.585 0.024 0.576 0.025
0.493 0.034 (0.380) (0.022)
Fertilizer-related (0.162) -(0.004) 0.321 -0.015
0.279 0.016 (0.200) (0.014)
Pesticide-related 0.284 0.003 0.202 -0.008
(0.451) (0.028) (0.257) (0.014)
Cumin-related 0.254 -0.024 0.340 0.123***
(0.436) (0.025) (0.474) (0.035)
Limited e¤ect, only on cumin-related question
31. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Cumin Adoption
Cash crop with potential for high-return, but risky
Important risks: wilt, frost
AO provides timely weather forecasting and planting
suggestions
Nearly half of push messages (20) discuss Cumin, and physical
extension covers cumin cultivation
32. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Impact on Adoption of Cumin
Impact: Cumin Adoption
Control (AO+AOE)-C AO-C AOE-C
Mean ITT ITT ITT
(1) (2) (3) (4)
Planted cumin in R'11 but not in R'12 0.183 -0.016 0.007 -0.036
(0.387) (0.030) (0.039) (0.033)
Planted cumin in R'12 but not in R'11 0.138 0.064** 0.054 0.064*
(0.345) (0.028) (0.041) (0.035)
Planted cumin both in R'11 and in R'12 0.233 -0.032 -0.016 -0.045
(0.423) (0.029) (0.040) (0.034)
33. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Cumin: Acreage Planted (No Control)
Control (AO+AOE)-C AO-C AOE-C
Mean ITT ITT ITT
Baseline (N=1200) (1) (2) (3) (4)
Did you plant cumin in Rabi 2011? 0.425 -0.017 -0.023 -0.012
(0.495) (0.030) (0.034) (0.035)
Total area of cumin planted in Rabi 2011? 0.762 -0.019 0.018 -0.055
(1.406) (0.078) (0.085) (0.106)
Baseline Phone Respondents (N=798)
Did you plant cumin in Rabi 2011? 0.425 -0.018 -0.035 -0.001
(0.495) (0.032) (0.039) (0.039)
Total area of cumin planted in Rabi 2011? 0.762 0.000 0.036 -0.037
(1.406) (0.081) (0.110) (0.119)
Round 1 Phone Survey (N=737)
Did you plant cumin this Rabi 2012? 0.274 0.046 0.070* 0.022
(0.446) (0.033) (0.042) (0.039)
Total area of cumin planted in Rabi 2012? 0.522 0.243** 0.240* 0.255*
(1.174) (0.112) (0.127) (0.154)
34. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Cumin: Acreage Planted (Control for Baseline Cotton
Area)
Cumin Sowing Decisions
Control (AO+AOE)-C AO-C AOE-C
Mean ITT ITT ITT
Baseline (N=1200) (1) (2) (3) (4)
Did you plant cumin in Rabi 2011? 0.425 -0.025 -0.032 -0.019
(0.495) (0.030) (0.035) (0.035)
Total area of cumin planted in Rabi 2011? 0.762 -0.057 -0.036 -0.080
(1.406) (0.079) (0.085) (0.105)
Baseline Phone Respondents (N=798)
Did you plant cumin in Rabi 2011? 0.425 -0.034 -0.052 -0.014
(0.495) (0.033) (0.039) (0.038)
Total area of cumin planted in Rabi 2011? 0.762 -0.086 -0.058 -0.099
(1.406) (0.084) (0.102) (0.115)
Round 1 Phone Survey (N=737)
Did you plant cumin this Rabi 2012? 0.274 0.038 0.051 0.019
(0.446) (0.035) (0.045) (0.040)
Total area of cumin planted in Rabi 2012? 0.522 0.225 * 0.202 0.246
(1.174) (0.117) (0.138) (0.156)
35. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Why Pest Management?
Indian cotton yields are one third of Chinese yields (NCC,
2012)
Cotton production accounts of 54% of all pesticide usage in
India
Pesticide accounts for roughly 15% of input costs in cotton
production
Pilot studies of AO in rural Gujarat suggest large demand for
pest management information (Patel at al. 2010)
Types of pests a- icting cotton vary by season and develop
resistance to pesticides and varieties of cotton putting a
premium on learning and timely information
36. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Pesticide Choice for Cotton
Monocrotophos Imidachlororprid
Price per Liter $8 $20
Dose for 1 acre 1.5L 300 ml
Cost/acre $12 $6
Introduced 1980 2000
Kills Bollworm Yes No
Many sucking pests
Kills sucking pests have developed Yes
immunity to this
Toxicity High Medium-Low
Key fact: 95% of farmers grow BT Cotton, which is resistant
to bollworm
37. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Control Actions on Monocrotophos
Indonesia: Banned for use in rice in 1986
Kuwait: Severely Restricted
Germany: May not be handled by adolescents, pregnant and
nursing women
Malaysia: Registered for speci…c use
Philippines: Severely restricted
Sri Lanka: Severely restricted. Banned since 1995.
US: Use is prohibited
38. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Use of Cotton Pesticides: More of the Good
Control (AO+AOE)-C AO-C AOE-C
Mean ITT ITT ITT
Baseline (N=1200) (1) (2) (3) (4)
Purchased imidacloprid in K'10 0.428 0.028 -0.018 0.074
(0.496) (0.038) (0.041) (0.050)
Amount of imidacloprid used in K'10 0.435 0.034 0.066 0.019
(0.837) (0.068) (0.093) (0.083)
Round 1 Phone Survey (N=737)
Purchased imidacloprid in K'11 0.388 0.128 *** 0.136 *** 0.122
(0.488) (0.037) (0.049) (0.043)
Used imidacloprid in K'11 0.388 0.125 *** 0.136 *** 0.116
(0.488) (0.036) (0.049) (0.044)
Amount of imidacloprid used in K'11 0.492 0.105 0.149 0.050
(1.254) (0.082) (0.128) (0.071)
Intensity of imidacloprid used in K'11 0.110 0.037 ** 0.039 0.032
(0.214) (0.018) (0.027) (0.019)
39. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Use of Cotton Pesticides: Less of the Bad
Usage of Monocrotophos
Control (AO+AOE)-C AO-C AOE-C
Mean ITT ITT ITT
Baseline (N=1200) (1) (2) (3) (4)
Purchased monocrotophos in K'10 0.962 0.000 -0.001 -0.001
(0.191) (0.012) (0.010) (0.016)
Amount of monocrotophos used in K'10 2.328 0.254* 0.346* 0.219
(1.866) (0.137) (0.189) (0.169)
Baseline Phone Respondents (N=798)
Purchased monocrotophos in K'10 0.962 0.002 -0.011 0.013
(0.191) (0.013) (0.015) (0.017)
Amount of monocrotophos used in K'10 2.328 0.306* 0.412 0.226
(1.866) (0.180) (0.287) (0.213)
Round 1 Phone Survey (N=737)
Purchased monocrotophos in K'11 0.945 -0.013 -0.024 -0.002
(0.229) (0.019) (0.020) (0.026)
Used monocrotophos in K'11 0.942 -0.010 -0.021 0.001
(0.234) (0.019) (0.021) (0.026)
Amount of monocrotophos used in K'11 3.870 -0.486** -0.307 -0.684**
(4.005) (0.210) (0.251) (0.278)
40. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
At baseline, each farmer was asked to identify top agricultural
contacts with whom they exchange information
Can imagine two models for e¤ect of AO
AO dramatically increases sharing of knowledge, because
quality of knowledge increases, so returns to sharing are higher
AO reduces sharing of knowledge, because farmers avail of AO
41. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
How does AO a¤ect farmers propensity to share
information?
Impact: Information Sharing
Control (AO+AOE)-C AO-C AOE-C
Mean ITT ITT ITT
(1) (3) (5) (7)
Shared agricultural information with top contacts 0.693 -0.019 0.003 -0.046
(0.462) (0.034) (0.038) (0.043)
Topics of information shared:
Crop decision 0.122 -0.037* -0.033 -0.041
(0.328) (0.020) (0.025) (0.027)
Fertilizers 0.313 -0.053 -0.042 -0.067*
(0.464) (0.033) (0.042) (0.040)
Pests and diseases 0.043 0.027* 0.010 0.040*
(0.204) (0.016) (0.016) (0.024)
Pesticides and fungicides 0.454 -0.005 0.025 -0.038
0.499 0.036 0.040 0.046
Harvesting 0.014 -0.014** -0.014** -0.014**
(0.116) (0.006) (0.006) (0.006)
42. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
E¤ect on Information Collection
Less likely to receive information from top contacts
Impact: Information Reception and Observation
Control (AO+AOE)-C AO-C AOE-C
Mean ITT ITT ITT
(1) (2) (3) (4)
Received information from top contacts 0.563 -0.076 *** -0.086 ** -0.066
(0.497) (0.029) (0.040) (0.037)
Topic of information received:
Crop decision 0.114 -0.054 *** -0.046 ** -0.061
(0.318) (0.019) (0.022) (0.021)
Field Preparation 0.038 -0.020 * -0.016 -0.026
(0.192) (0.011) (0.015) (0.013)
Seeds 0.179 -0.045 * -0.057 * -0.033
(0.384) (0.026) (0.029) (0.030)
Fertilizers 0.204 -0.052 ** -0.049 -0.058
(0.403) (0.025) (0.032) (0.033)
Pests and diseases 0.035 -0.002 -0.003 -0.002
(0.185) (0.012) (0.016) (0.015)
43. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Less Learning from Neighbors
Control (AO+AOE)-C AO-C AOE-C
Mean ITT ITT ITT
(1) (2) (3) (4)
Learned information from observing top 0.239 -0.107 *** -0.115 *** -0.102
contacts' fields (0.427) (0.028) (0.035) (0.029)
Topic of information learned:
Crop decision 0.049 -0.040 *** -0.034 *** -0.043
(0.216) (0.013) (0.012) (0.013)
Field Preparation 0.041 -0.024 ** -0.025 * -0.023
(0.198) (0.011) (0.015) (0.012)
Seeds 0.035 -0.021 ** -0.034 *** -0.008
(0.185) (0.010) (0.010) (0.012)
Fertilizers 0.041 -0.020 -0.037 ** -0.004
(0.198) (0.015) (0.016) (0.018)
Pests and diseases 0.014 -0.011 -0.008 -0.014
(0.116) (0.006) (0.008) (0.006)
44. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Peer E¤ects: Spillovers on Imida Use, Control Group
Peer Effects on Imidacloprid Adoption within Study Group
1=Purchased Imidacloprid in K'11
(1) (2) (3) (4) (5)
At least one top contact 0.123 0.304*
received treatment (0.118) (0.160)
0.165 0.185 0.283**
Proportion of top contacts
(0.130) (0.133) (0.138)
who received treatment
Controls
Age -0.008 -0.006 -0.003
(0.006) (0.006) (0.006)
Years of education -0.003 0.004 0.012
(0.017) (0.015) (0.014)
Land holdings (ac) 0.018 0.034 0.0419*
(0.022) (0.021) (0.024)
Area Cotton Planted (K'10) -0.016 -0.0495* -0.048
(0.031) (0.029) (0.034)
Top contacts' average land holdings 0.003 0.010 0.010
(0.012) (0.013) (0.017)
Number of references received -(0.048) (0.033)
(0.039) (0.043)
Village Fixed Effects Yes Yes Yes Yes Yes
N 120 120 120 102 90
No. of villages 38 38 38 38 36
45. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Spillovers, Imida Use Among Peers
1=Purchased Imidacloprid in K'11
(1) (2) (3) (4) (5)
At least one top contact 0.033 0.033
received treatment (0.046) (0.051)
Proportion of top 0.029 0.026 0.065
contacts who received (0.051) (0.051) (0.190)
treatment
Controls
Age -0.001 -0.002 0.001
(0.002) (0.003) (0.003)
Years of education 0.006 0.007 0.007
(0.005) (0.007) (0.005)
Land holdings (ac) 0.005** 0.005** 0.0051**
(0.003) (0.002) (0.002)
Top contacts' average land holdings 0.002** 0.001*** 0.002***
(0.001) (0.000) (0.001)
Number of references received 0.003 0.005
(0.028) (0.051)
Village Fixed Effects Yes Yes Yes Yes Yes
N 687 687 651 360 528
No. of villages 40 40 40 40 40
46. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Caveats
Attrition
Number of attritors equal in treatment and control group
Imbalance in attrition in round 1 phone: treatment group
attritors more likely to have planted cumin in ’10
Demand e¤ects
55% of treatment group reports having called into AO to ask
question
Administrative data indicates 53% actually did so
Continue knowledge tests
47. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Proof of Concept
Take-up of AO is high, among randomly selected sample of
poor farmers
75% listen to more than half of push calls
60% call into system,
Each person asks 1.4 questions on average
Young, technophiles more likely to use
Telephone surveys appear to work well
Midline: Section ‘
Z’randomly assigned phone or paper
administration
48. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Impact Evaluation: Information
Between 10-30% report AO as main information source on
various topics
Reduction in reliance on agro-dealers for pesticide (from 45%
to 25%) and fertilizer (from 15% to 7%)
No dramatic change in measured agricultural knowledge
49. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Impact Evaluation: Agricultural Practices
Cumin
Approximately 6 percentage point increase in cumin adoption
(from base of 12 pp)
Average area planted in cumin increase .22 acres, o¤ base of
.52 acres
Pesticide
Dramatic increase in use of imidacloprid (from 40% to 55%)
Modest reduction in intensity of monocrotophos (from 3.9 to
3.5 L)
E¤ect on Yields
Collecting data currently
50. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Conclusion
Farmers will listen to advice
Treatment has important e¤ects:
pesticide choice
cumin sewing
Behavior may change without generic change in knowledge
New technology may a¤ect information-sharing behavior
Less reported sharing, but likely better quality information
51. Intro Context Usage Impact: Info Impact: Practices Peer E¤ects Discussion
Future work
Audit study of agri-dealers
Peer survey
Peer_Outcome=a+β Peer _Treated + ε i
Health outcomes
Measuring willingness to pay
Role of trust in learning (DSC well-known)
Education vs. persuasion