Experimental Design
This study aims to understand how farmers interpret and respond to monsoon onset forecasts delivered through voice calls (IVRS) during the 2025 Kharif season in Odisha, India. The study will evaluate whether access to these forecast messages influences farmers' early-season agricultural decisions. Forecast messages will be disseminated via the GoO’s KS platform.
GoO is implementing IVRS monsoon onset forecast message dissemination via the KS system across a subset of 118 blocks in Odisha. Given the KS system call handling capacity constraints, GoO randomly selected 57 of these 117 blocks to begin receiving monsoon onset forecast messages starting May 22, 2025. The remaining 60 blocks did not receive these messages. Farmers in all blocks received text messages with the forecasts sent by the Ministry of Agriculture and Farmers’ Welfare, Government of India. The list of farmers in the state government’s list may be slightly different from the list of farmers receiving the text messages.
The study team intends to leverage this randomized rollout - where 57 of the 117 eligible blocks (treatment group) are receiving the IVRS messages and 60 are not (comparison group) - to evaluate the impact of forecast dissemination on farmer message recall, message comprehension, decision-making, and other agricultural outcomes. A first round of phone-based surveys was conducted in August 2025, focusing on early-season agricultural decisions, including total area of cultivation, crop and variety choices, and planting date. Each survey round is expected to last approximately five weeks.
Round 2 edit: A second phone-based survey is planned for February 2026, focused on eliciting end-of-season agricultural outcomes such as yields, input choices and expenditure, sales, rabi season crop choices, off-farm investments, and migration.
The anticipated sample includes up to 2000 farmers per survey round selected from administrative databases. The study sample will be drawn from all 117 blocks that were suitable for the monsoon forecast dissemination - 57 treatment and 60 comparison blocks. Within each block-engagement stratum, simple random sampling was used to select respondents.
Round 2 edit: We successfully completed 2,058 surveys in round 1. For round 2, we plan to call back the same round 1 participants to build a panel survey. We will administer questions on yields, sales, input choices, and expenditures to one half of the sample and questions on yields, sales, damages, and off-farm investments and migration to the other half of the sample.
Subject to expanding sample: We will survey an additional 2000 farmers from the same blocks. The additional sample will also be split in two and will receive one of the two questionnaires.