|
Field
Last Published
|
Before
August 14, 2024 02:36 PM
|
After
August 14, 2025 12:29 PM
|
|
Field
Intervention (Public)
|
Before
In the flyer arm, we distribute flyers throughout the village providing farmers with information about PMFBY and guidance on how to enroll.
In the meeting arm, we distribute flyers and additionally host an information session in the village to inform people about the scheme and answer questions. We also offer transportation to the local government office where farmers can enroll.
In the diffusion arm, we seed a random subset of farmers within each village with information about the scheme. Among this group, we also randomly select a subset of farmers to whom we offer a small incentive to be distributed conditional on their successful application to the program.
For each of the three arms listed above, we compare outcomes against people in control villages where information on the scheme was not distributed.
In a final set of villages, we elicit farmers’ willingness-to-pay for transportation to the local government office where farmers can enroll. We randomize the nature of this offer. First, we randomize the identity of a partner for each respondent. In some cases, the partner is in the farmers’ risk-sharing network, and we randomize whether this person has a perceived high- or low-covariance of agricultural risk with the respondent. In other cases, the partner is another person in the village who is not in the respondents’ risk-sharing network. Within these partner types, we randomize two more dimensions. First, we randomize whether we tell the respondent that we have also offered the transportation to the partner. Second, we randomize whether we tell the respondent that the partner may be informed if the respondent takes up the transportation.
|
After
In the flyer arm, we distribute flyers throughout the village providing farmers with information about PMFBY and guidance on how to enroll.
In the meeting arm, we distribute flyers and additionally host an information session in the village to inform people about the scheme and answer questions. We also offer transportation to the local government office where farmers can enroll.
In the diffusion arm, we seed a random subset of farmers within each village with information about the scheme. Among this group, we also randomly select a subset of farmers to whom we offer a small incentive to be distributed conditional on their successful application to the program.
For each of the three arms listed above, we compare outcomes against people in control villages where information on the scheme was not distributed.
In a final set of villages, we elicit farmers’ willingness-to-pay for transportation to the local government office where farmers can enroll. We randomize the nature of this offer. First, we randomize the identity of a partner for each respondent. In some cases, the partner is in the farmers’ risk-sharing network, and we randomize whether this person has a perceived high- or low-covariance of agricultural risk with the respondent. In other cases, the partner is another person in the village who is not in the respondents’ risk-sharing network. Within these partner types, we randomize two more dimensions. First, we randomize whether we tell the respondent that we have also offered the transportation to the partner. Second, we randomize whether we tell the respondent that the partner may be informed if the respondent takes up the transportation.
In a follow-up of the treatment in summer 2025, we returned to villages in the meetings arm and conducted a new set of meetings as well as also offering mobile reminders about sign-up.
In a separate set of villages in 2025, we offer people a series of choices about spreading information about PMFBY. We approach random households using a right-hand rule, explain PMFBY to them, and then elicit their decisions about six scenarios. The order of the scenarios is randomized. In half of the scenarios, people are choosing about distributing information about PMFBY, and in the other half, they are choosing about distributing information about the weather. Crossed with these arms about the content of the information, there are three additional variations. In one, the respondent themselves is being offered a prize to go and distribute the information. In two others, the enumerator team is distributing the information. In the first enumerator-distribution, the identity of the respondent is kept private. In the second, the identity of the respondent is told to the recipient of the information. In each scenario, the respondent makes multipe choices among pairs of people, some of whom are randomly selected from the respondent's risk sharing network and some of whom are random other people in the village. One of these six scenarios and one of the choices will be implemented with a random probability. Additionally, there are four hypothetical scenarios involving a third person who will always be told the information. The arms in the hypotheticals are similarly placebo weather information vs. PMFBY information and visibility of the respondent.
|
|
Field
Primary Outcomes (End Points)
|
Before
Our main outcomes for the flyer, meeting, and diffusion arms require a combination of follow-up surveys and administrative take-up data, which is publicly available on a government portal at the village level.
In the flyer and meeting arms, we focus on enrollment and take-up of the program and impacts on agricultural production decisions, consumption, and risk sharing behavior. In the diffusion arm, we focus on the spread of information about the scheme and subsequent enrollment throughout the village social network both during the current and future agricultural seasons.
We hypothesize that overall insurance take-up will be affected by ease of enrollment (e.g., distance to the government office), prior exposure to natural disasters, prior exposure to the policy (by caste), the occupational mix of the village (e.g., fishing vs. agriculture, land distribution across households), and eligibility status of households (e.g., loanees vs. sharecroppers vs. non-loanee landholders). At the household level, we are also collecting information on caste and land-holdings. We also hypothesize that take-up will vary by village network characteristics and, in the diffusion experiment, the treatment assignment of other nodes.
We anticipate having natural variation in weather outcomes across our sample, which will permit us to study whether insurance payouts help households mitigate negative shocks and the extent to which weather and payout realizations impact take-up in subsequent years. Floods may also vary in their incidence across the village.
We plan to collect data at several points in time. First, we will conduct a short follow-up survey in the weeks following the enrollment deadline to measure take-up, beliefs about the policy, and social network information.
We will return to respondents following the harvest of the Kharif crop to measure farm inputs and yields along with detailed information on social networks and transfers between households. We will return to households again following planting in summer 2025 to measure agricultural and labor allocation decisions for the subsequent season along with insurance take-up.
In the willingness-to-pay experiment, we focus on demand for the transportation to the government office as the main outcome. In addition to the dimensions of potential heterogeneity mentioned above, we will additionally have information on respondents’ beliefs about the quality of the program, access to other transportation options, and network composition, which may each affect demand for take-up and demand for privacy.
|
After
Our main outcomes for the flyer, meeting, and diffusion arms require a combination of follow-up surveys and administrative take-up data, which is publicly available on a government portal at the village level.
In the flyer and meeting arms, we focus on enrollment and take-up of the program and impacts on agricultural production decisions, consumption, and risk sharing behavior. In the diffusion arm, we focus on the spread of information about the scheme and subsequent enrollment throughout the village social network both during the current and future agricultural seasons.
We hypothesize that overall insurance take-up will be affected by ease of enrollment (e.g., distance to the government office), prior exposure to natural disasters, prior exposure to the policy (by caste), the occupational mix of the village (e.g., fishing vs. agriculture, land distribution across households), and eligibility status of households (e.g., loanees vs. sharecroppers vs. non-loanee landholders). At the household level, we are also collecting information on caste and land-holdings. We also hypothesize that take-up will vary by village network characteristics and, in the diffusion experiment, the treatment assignment of other nodes.
We anticipate having natural variation in weather outcomes across our sample, which will permit us to study whether insurance payouts help households mitigate negative shocks and the extent to which weather and payout realizations impact take-up in subsequent years. Floods may also vary in their incidence across the village.
We plan to collect data at several points in time. First, we will conduct a short follow-up survey in the weeks following the enrollment deadline to measure take-up, beliefs about the policy, and social network information.
We will return to respondents to measure farm inputs and yields along with detailed information on social networks and transfers between households. We had initially planned to return to households again following planting in summer 2025 to measure agricultural and labor allocation decisions for the subsequent season along with insurance take-up. We will return in the middle of the Kharif season of 2025.
In the willingness-to-pay experiment, we focus on demand for the transportation to the government office as the main outcome. In addition to the dimensions of potential heterogeneity mentioned above, we will additionally have information on respondents’ beliefs about the quality of the program, access to other transportation options, and network composition, which may each affect demand for take-up and demand for privacy.
In the information distribution choices experiment, we focus on the outcome of who the respondent chooses to receive the information, and specifically their relative risk sharing relationship to the respondent.
|
|
Field
Experimental Design (Public)
|
Before
Sampling disproportionately from areas at risk of flooding, we randomly assign villages to pure control arms (where no information is distributed) and to flyer and meeting arms. For the diffusion arm, we randomly select villages to be seeded, and then within village, randomly seed households. For the willingness-to-pay arm, we randomly draw partners for the respondent from a list of names elicited in previous surveys.
|
After
Sampling disproportionately from areas at risk of flooding, we randomly assign villages to pure control arms (where no information is distributed) and to flyer and meeting arms. For the diffusion arm, we randomly select villages to be seeded, and then within village, randomly seed households. For the willingness-to-pay arm, we randomly draw partners for the respondent from a list of names elicited in previous surveys. We follow the same procedure for the information diffusion choices experiment.
|
|
Field
Randomization Method
|
Before
The village selection for the flyer, meeting, and diffusion arms is done via computer. The individual selection for the diffusion arm is done via a right-hand rule in which enumerators begin the survey at a randomly selected household by counting houses according to a number generated by the computer, and then proceed subsequently by skipping houses. The randomization for the willingness-to-pay arm is done via computer.
|
After
The village selection for the flyer, meeting, and diffusion arms is done via computer. The individual selection for the diffusion arm is done via a right-hand rule in which enumerators begin the survey at a randomly selected household by counting houses according to a number generated by the computer, and then proceed subsequently by skipping houses. The randomization for the willingness-to-pay arm and the information diffusion choices arm is done via computer.
|
|
Field
Randomization Unit
|
Before
The randomization unit for the flyer and meeting arms is at the village-level. For the diffusion rm, we randomize across villages whether we seed anyone, and within villages, whom we seed. The randomization unit for the willingness-to-pay experiment is at the farmer level.
The meetings, flyer, diffusion and control conditions are clustered, in the sense that the unit of randomization is at the village level.
The willingness-to-pay treatment is not clustered.
|
After
The randomization unit for the flyer and meeting arms is at the village-level. For the diffusion arm, we randomize across villages whether we seed anyone, and within villages, whom we seed. The randomization unit for the willingness-to-pay experiment is at the farmer level. The randomization unit for the information diffusion choices arm is at the farmer level.
The meetings, flyer, diffusion and control conditions are clustered, in the sense that the unit of randomization is at the village level.
The willingness-to-pay treatment is not clustered.
|
|
Field
Sample size (or number of clusters) by treatment arms
|
Before
We expect approximately 80 villages in the flyer arm, 110 villages in the meeting arm, 30 villages in the diffusion arm with between five and 10 respondents per village (depending on village size, household eligibility and door-locks), and 120 people in the willingness-to-pay experiment allocated approximately evenly across the different partner types.
|
After
We expect approximately 80 villages in the flyer arm, 110 villages in the meeting arm, 30 villages in the diffusion arm with between five and 10 respondents per village (depending on village size, household eligibility and door-locks), and 120 people in the willingness-to-pay experiment allocated approximately evenly across the different partner types. We expect about 30 villages with 10 respondents each for a total of 300 respondents in the choices information diffusion experiment.
|