Experimental Design
We use a randomized controlled trial (RCT) to investigate the impact of a digital agricultural advisory service on smallholder farmers’ yield. We recruited eligible households cultivating rice across approximately 903 villages in 19 blocks using a random-walk method and randomly selected half of the households to receive a free digital advisory service.
We selected 19 blocks in five districts in the state for inclusion in the study. All the included blocks had low (<18.5 percent) “penetration” of the information service, as measured by the estimated share of farmers that were enrolled prior to the study, and were covered by the same satellite path. Within these 19 blocks, we selected all villages that met the following criteria during the 2011 Census of India: (i) classified as “rural”; (ii) had more than 50 households; (iii) located in a panchayat with service “penetration” below 10 percent. A total of 1,313 or 39 percent of the villages met these conditions and constituted our study sites. In each village, we selected 5 - 10 percent of households using a random-walk sampling method. Households are eligible for inclusion in the study if they meet a set of criteria and are willing to allow the research team to collect the GPS coordinates of their primary rice plot.
The eligible households (one person per household) are invited to participate in a baseline survey, including questions on demographics, farm characteristics, agricultural knowledge and practice, yield, and GPS coordinates of the primary rice plot.
We then randomize those farmers who have completed the baseline survey and have not registered for this study’s digital extension service before into control and treatment groups. Randomization is conducted at the individual farmer level, and stratified by panchayat group, baseline survey verison, an indicator for participants’ self-reported productivity at baseline being above the median or not, and (for cohort 2 only) an indicator for participants’ gps measured plot size at baseline being above the median or not.
Farmers in the treatment group start to receive free digital advisory service in the coming agriculture season post their baseline survey. Moreover, those farmers are administered a phone-based compensation and service enrollment survey, which collects necessary information to send farmer compensation for their baseline survey participation and explains the digital advisory service to farmers.
The study began implementation of the random walk sampling method in March 2021, but suspended all field operations on 23rd April following a steep rise in COVID-19 cases in the state. 5,204 households were enrolled in the study before the fieldwork suspension (aka “initial” sample, or cohort 1), of which half were randomized to the treatment group and invited to enroll in the information service via phone. 2,571 (99.5%) members of the treatment group (2,602 farmers) opted-in to the service and began receiving advisory messages from 9th June. As the COVID-19 situation alleviated in the study location, we resumed our field activities in October, 2021. We used the same random-walk method in the same locations as previously and sampled a further 8,471 households (aka. cohort 2).
After the 2021-22 and 2022-23 Kharif seasons, we will conduct a midline survey and an endline survey with cohort 1 farmers, allowing us to measure their knowledge, practice, yield and profit.