Download StudyGeneral

Study Overview

Picture-based crop insurance: A randomized control trial evaluating the impacts of using smartphone camera data for claims verification in India
Study ID:
Initial Registration Date:
Last Update Date:
Study Status:
In Development
Climate change is exposing smallholder farmers increasingly to extreme weather events, causing financial hardship and discouraging profitable yet risky agricultural investments even before a disaster occurs. In this study, we will therefore evaluate the impacts of Picture-Based Crop Insurance (PBI) on the livelihoods of farming households in Haryana, India. PBI compensates farmers for individual crop damage visible from a stream of pre- and post-damage pictures of insured fields, uploaded by farmers throughout the season using a tamper-proof smartphone app. The proposed evaluation will study the long-run sustainability and social impacts of offering a range of PBI products over three years, targeting 2,000 farmers from 100 villages in Haryana. We will randomize, at the village level, whether or not farmers are offered the opportunity to purchase PBI. The study will quantify impacts of this intervention on how farmers affected by a natural disaster cope with agricultural income losses, and we will assess whether PBI, by reducing exposure to risk for a wide range of crops, improves agricultural investments, management practices, production diversity, and agricultural incomes.
Agriculture and Rural Development
Information and Communications Technology
Additional Keywords:
Risk and insurance, Resilience, Basis risk, Randomized control trial, Picture-based insurance, Climate change
Secondary ID Number(s):

Principal Investigator(s)

Name of First PI:
Berber Kramer
International Food Policy Research Institute (IFPRI)
Name of Second PI:
Francisco Ceballos
International Food Policy Research Institute (IFPRI)

Study Sponsor

International Initiative for Impact Evaluation (3ie)
Study Sponsor Location:
Funding Proposal:

Research Partner

Name of Partner Institution:
Borlaug Institute for South Asia (BISA)
Type of Organization:
Research institute/University

Intervention Overview

Farmers will be offered a range of picture-based crop insurance products that cover visible damage in a stream of smartphone pictures, taken from sowing to harvest. We will start offering products for wheat and tomatoes in the first season, adding more crops in future seasons. To enroll, farmers download a smartphone app from the Google Play Store, register personal details and plots to be enrolled, submit initial georeferenced pictures for these plots, and pay the insurance premium. Throughout the season, built-in reminders ask farmers to take georeferenced pictures of enrolled sites. The app interface shows—as a ‘ghost’ image—the initial picture for that site in order to ensure repeat pictures are always taken from the same location with the same view frame. To prevent tampering, farmers can only take and upload pictures within the app. At the end of each season, experts will inspect incoming pictures for visible damage to inform insurance payments. Over time, as ground truth data availability improves, we will use vegetation indices derived from the crop pictures, weather data, crop models and machine learning algorithms to automate loss assessment.
Theory of Change:
On the supply side, damage assessments based on visible crop characteristics in pictures can be done at low marginal cost, reducing the costs of timely loss verification compared to yield-based products such as the PMFBY. We assume that poor management practices are visible in smartphone pictures (or at least farmers perceive this to be the case) and having ‘eyes on the ground’ through uploaded pictures can limit moral hazard and adverse selection. On the demand side, assuming that damage is visible in the pictures, the product will help minimize basis risk compared to standard index-based insurance products. Moreover, taking a stream of plot-level pictures will improve awareness by increasing farmer engagement and product tangibility, which will improve the willingness to pay for crop insurance. Moreover, having to take pictures might improve monitoring of crop health, potentially reducing farmers’ exposure to production risk, bringing down insurance premiums further. Combined, improved supply of affordable insurance and a higher willingness to pay can generate higher insurance uptake, with two effects. Ex-post, if farmers experience production shocks, they will receive timely insurance payouts. Anticipating this will encourage risk-averse farmers, for whom these risks are a barrier to invest, to invest in high-return activities, which includes diversifying into high-risk horticultural crops covered by insurance. Lenders, expecting improved repayment rates from risk-rationed farmers, will be more likely to provide credit for productivity-enhancing yet risky investments. In the long run, we expect improved incomes and risk coping (both directly, by providing timely payouts when farmers suffer damage, and indirectly, by raising agricultural incomes). Improvements in income and coping, combined with improved production diversity, can also increase dietary diversity and food security, ultimately contributing to sustainable human capital formation.
Multiple Treatment Arms Evaluated?

Implementing Agency

Name of Organization:
HDFC ERGO General Insurance, Ltd.
Type of Organization:
Private for profit organization

Program Funder

Name of Organization:
CGIAR research program for Policies, Institutions and Markets (PIM)
Type of Organization:
Research Institution/University

Intervention Timing

Intervention or Program Started at time of Registration?
Start Date:
End Date:
Evaluation Method

Evaluation Method Overview

Primary (or First) Evaluation Method:
Randomized control trial
Other (not Listed) Method:
Additional Evaluation Method (If Any):
Other (not Listed) Method:

Method Details

Details of Evaluation Approach:
The quantitative study will test the PBI theory of change using a cluster randomized trial in which 50 villages are randomly selected to be exposed to the intervention above, and another 50 villages (again randomly selected) will serve as control group, meaning that they will not be exposed to the PBI intervention. Because this is a randomized experiment, treatment status is determined at random. Hence, we can compare outcome and impact variables in follow-up survey rounds between these two sets of villages and interpret any statistically significant difference between the two groups as a causal effect of the intervention. Further, by randomizing at the village level, we minimize the risk of jealousy between farmers within a village, and of contamination and spillovers within the village. Randomization will be done in private, by the researchers, for a list of study villages stratified by block or district.
Outcomes (Endpoints):
Primary outcomes: - Increases insurance take-up and retention - Improved crop management practices - Reduced exposure to production risk - Diversification into high-risk horticultural crops - Improved access to credit for high-risk agricultural investments - Reduced indebtedness Secondary outcomes - Agricultural incomes - Risk coping strategies - Dietary diversity and food security - Human capital formation, particularly among women
Unit of Analysis:
The main unit of analysis will be household i in season s (with 2000 households and 6-9 seasons)
Unit of Intervention or Assignment:
The intervention will be randomized at the village level (if villages are too small to sample at least 20 farmers, we will cluster two neighboring villages into one).
Number of Clusters in Sample:
Number of Individuals in Sample:
2000 (20 per cluster)
Size of Treatment, Control, or Comparison Subsamples:
50 clusters in the treatment, 50 clusters in the control

Supplementary Files

Other Documents:

Outcomes Data

We will measure these outcome variables through repeated household surveys with all targeted households. Short surveys will be administered with the full sample at the end of each season, while longer face-to-face surveys have been planned for the fall of 2019 (after two-three seasons of implementation) and the fall of 2021 (after six to nine seasons of implementation). These surveys will be administered using computer-assisted personal interviewing.
Data Already Collected?
Survey Name:
Data Previously Used?
Data Access:
Data Obtained by the Study Researchers?
Data Approval Process:
Approval Status:

Treatment Assignment Data

Participation or Assignment Information:
While the survey will collect self-reported enrollment data, we will also use actual enrollment data generated through the administrative insurance data. This data will also include information on whether farmers reported damage and whether any insurance payouts were made.
Data Obtained by the Study Researchers?
Data Previously Used?
Data Access:
Data Obtained by the Study Researchers?
Data Approval Process:
Approval Status:

Data Analysis

Data Analysis Status:

Study Materials

Upload Study Materials:

Registration Category

Registration Category:
Prospective, Category 1: Data for measuring impacts have not been collected

Completion Overview

Intervention Completion Date:
Data Collection Completion Date:
Unit of Analysis:
Clusters in Final Sample:
Total Observations in Final Sample:
Size of Treatment, Control, or Comparison Subsamples:


Preliminary Report:
Preliminary Report URL:
Summary of Findings:
Paper Summary:
Paper Citation:

Data Availability

Data Availability (Primary Data):
Date of Data Availability:
Data URL or Contact:
Access procedure:

Other Materials

Survey Instrument Links or Contact:
Program Files:
Program Files Links or Contact:
External Link:
External Link Description:
Description of Changes:

Study Stopped