Leveraging Integrated E-extension Services to Reduce Sri Lankan Farmers' Vulnerability

The randomized controlled trial (RCT) aims to evaluate the impact of the S.E.E.D. platform on Sri Lankan paddy farmers' vulnerability. The study samples 220 farmers’ organization (2,200 farmers) stratified across agro-ecological zones and irrigation schemes to improve the representativeness. The treatment group includes 110 FOs that will receive early access to the hub. Half of them, 55 FOs, will also receive hands-on training on the usage of the platform. The remaining 110 FOs will serve as the control group until March 2024, after which the Government of Sri Lanka may consider scaling up the intervention. The impact of the S.E.E.D. platform will be assessed against different dimensions such us farmers' access to information and advices, their vulnerability (measured as harvest losses), as primary outcomes. Other secondary outcomes include paddy farmers' productivity (rice yield), farming management decisions (e.g., use of fertilizer, crop and sales diversification), and household food security (e.g., diet diversity and food consumption).

2023-07-06

2024-03-31

- Access to weather and agricultural information
- Harvest losses

- Rice yield
- Quantity of inorganic and organic fertilizers
- Crop diversification
- Sale diversification
- Sale prices
- Food security indicators (HDDS, and FCS)

The RCT includes 2,200 paddy farmers enrolled in 220 FOs. The sample is nationally representative of the paddy farmers enrolled in the FOs (more than 90% of the total paddy farmers in the country). The sampling strategy is a multi-stage stratified cluster sampling. In the first stage, districts are stratified based on agro-ecological zones and randomly sampled with a probability proportional to the share of land devoted to paddy fields over the total extent of the district. Agro-ecological zones, other than defining all districts exhaustively and mutually exclusively, identify different agricultural conditions that are likely correlated with outcomes. In the second stage, within each sampled district, farmer organizations were stratified based on irrigation scheme, and then 220 were randomly sampled. Sri Lanka has three irrigation schemes: the major irrigation (45%), rainfed (29%), and minor irrigation (26%). Irrigation defines all farmer organizations exhaustively and mutually exclusively. Finally, for each farmer organization, farmers are randomly selected with a probability proportional to the size of the organization.
The treatment assignment follows a two-stage cluster stratified randomized design. FOs in the treatment group will receive early access to the S.E.E.D hub (login credential). Prior to treatment assignment, experimental units (FOs) were classified into agroecological zones (blocks/strata). This allows grouping units that are likely to be homogeneous, reducing uncertainty about outcomes. In the first stage, units within each stratum are then assigned to the treatment and control groups, each consisting of 110 FOs. The treatment assignment has been selected from 5,000 random draws in order to minimize the total number of treated and control pairs whose distance is smaller than 5 km. Such an assignment ensures complete randomness within each stratum while reducing possible contamination across the treatment and control arms. In the second stage, half of the treated (55 FOs) were randomly assigned to an additional training on the usage of the platform ('access plus training' arm). According to the first stage randomization, the final training assignment has been selected from 5,000 random draws to minimize the total number of FO pairs across treated arms whose distance is smaller than 5 km. Farmers enrolled in the FOs included in the treatment arms will be encouraged to attend the "hub deployment meeting" that will take place in each farmer organization prior to the start of the main agricultural season. During this visit, the login credential will be shared with the participants. The training on the usage of the platform delivered to the FOs in the 'access plus training' arm consists of a lecture session, a practical demonstration on how to use the application, and a Q&A session. The training as a whole will last approximately 2.5 hours. The process to distribute the credential and the training will follow standardized protocols to reduce possible bias due to the trainers. Farmers' participation in the intervention is voluntary

Not available

Cluster Stratified Randomization has been implemented at FO level to address the risk of contamination within the FO as well as to take into account the fact that many management decisions in the Sri Lankan rice sector are coordinated at FO level.
In the first stage, units within each stratum are then assigned to the treatment and control groups, each consisting of 110 FOs. The treatment assignment has been selected from 5,000 random draws in order to minimize the total number of treated and control pairs whose distance is smaller than 5 km. Such an assignment ensures complete randomness within each stratum while reducing possible contamination across the treatment and control arms. In the second stage, half of the treated (55 FOs) were randomly assigned to an additional training on the usage of the platform ('access plus training' arm). According to the first stage randomization, the final training assignment has been selected from 5,000 random draws to minimize the total number of FO pairs across treated arms whose distance is smaller than 5 km. Farmers enrolled in the FOs included in the treatment arms will be encouraged to attend the "hub deployment meeting" that will take place in each farmer organization prior to the start of the main agricultural season.

In the first stage, FOs within each stratum are assigned to the treatment and control groups, each consisting of 110 units. The treatment assignment has been selected through 5,000 random draws from each stratum (agroecological zones) in order to minimize the total number of treated and control pairs whose distance is smaller than 5 km. Such an assignment is expected to ensure the complete randomness within each stratum while reducing possible contamination across the treatment and control arms. In the second stage, half of the treated units (55 FOs) were randomly assigned to an additional training on the usage of the platform ('access plus training' arm). The final training assignment has been selected from 5,000 random draws to minimize the total number of FO pairs across treated arms whose distance is smaller than 5 km.

Yes

220 Farmers organizations (FOs)

2,200 household farms (10 units from each FOs).

55 FO clusters in “access plus training” arm, 55 FO cluster in ‘access without training’ arm, 110 FO clusters in control arm.

The total sample size (2200 farmer households) was defined following multi-stage stratified cluster sampling and to be nationally representative of paddy farmers in Sri Lanka (it includes 0.3% of the paddy farming population). Considering the entire treatment and control arms and setting a two tailed test and the significance level at 5 per cent (α=0.05), the experiment has 80% power to detect a 0.003 increase in harvest losses measured as normalized difference between area planted and harvested in square meters (effect size of 0.122), and 0.034 increase in the access of agricultural and weather information through application or text (effect size of 0.132). Considering the Bernoulli correction for two primary outcome variables and the covariates (household head education, sex, and age, household size, number of children, household income derived from off-farm activities, and total area of plots), the minimum detectable effects become 0.001 (effect size of 0.079) and 0.033 (effect size of 0.134) respectively. Considering the ‘access with and without training’ arms (i.e. 55 FOs and 550 farmer households in each arm), and setting a two tailed test and the significance level at 5 per cent (α=0.05), the experiment has 80% power to detect a 0.004 increase in harvest losses (effect size of 0.172), and 0.049 increase in the access of agricultural and weather information through application or text (effect size of 0.194). Considering the Bernoulli correction for two primary outcome variables and the covariates, the minimum detectable effects become 0.003 (effect size of 0.333) and 0.047 (effect size of 0.189) respectively.