Bundling Weather Forecasts and Agronomic Advisory for Farmers in Pakistan

How to cite

Rudder, Jess; Viviano, Davide. 2022. "Bundling Weather Forecasts and Agronomic Advisory for Farmers in Pakistan." AEA RCT Registry. AEARCTR-0009945. https://doi.org/10.1257/rct.9945-2.1

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Jess Rudder* , Davide Viviano

* Primary investigator

Registry ID
AEARCTR-0009945
DOI
https://doi.org/10.1257/rct.9945-2.1
First registered
Last updated
Status
Completed
Trial dates
2022-05-01 – 2023-04-01
Countries
Pakistan ()
Registry record
http://www.socialscienceregistry.org/trials/9945

Abstract

Agricultural production is inherently risky and subject to weather shocks that threaten livelihoods for small-scale farmers worldwide. Short-range weather forecasts can help farmers plan immediate agricultural activities and take preventative measures that lower the likelihood of losses due to unforeseen weather. Advisory services coupled with weather forecasts can speed up learning about how forecasted weather patterns can affect crop production. We propose to test an optimal design strategy to detect treatment and spillover effects in a large population of potential program participants. We do so by developing a research design that accounts for both spillovers and adaptive assignments. We propose to assign clusters to have variation in treatment intensity of the share of users per cluster assigned to any treatment. This will generate exogenous variation in the share of farmers in a community that receive the information. After the first round of assignments, we propose to use information on uptake and engagement to reassign users in an adaptive experiment to maximize benefits to users.

Intervention

Precision Development (PxD) is rolling out a new service that provides short-run weather forecasts to cotton farmers in Punjab province Pakistan using a voice message system that automatically calls users and delivers a pre-recorded message. PxD’s core service provides agronomic advisory recommendations via the same voice message system. The intervention will compare outcomes for users randomly assigned to one of two groups: 1. A control group that receives advisory messages only, and 2. A treatment group that receives advisory messages plus weather forecasts. The core user base includes about 500,000 cotton farmers located in 42 tehsils in the cotton-growing belt of Pakistan.

Experimental design

The design consists of at least three consecutive waves of randomization. Each wave assigns treatments with different probabilities across different tehsils (equivalent of a district unit).

We consider three groups of tehsils denoted as high/medium/low saturation, each corresponding to larger treatment probability within each tehsil (approximately 60-40-12% of treated individuals in the first experimentation wave).

Throughout each wave, we increase the treatment probability within each tehsil. By the last wave, the high saturation teshils will present the largest share of treated individuals (approximately 90%).

In the first wave, we induce small local perturbation to treatment probability across pairs of tehsils. In the consecutive waves we assign treatments also using engagement data, and induce local perturbation to treatment probabilities based on individual engagement.

Engagement data is readly available for all experimental participants. For phone surveys, we sample individuals at random in multiple waves (up to 1,800 individuals/month). In each wave, sampling is stratified by treatment status, and tehsil.

Design details

Randomization method
randomization done in office by a computer
Randomization unit
Clusters and units within clusters
Clusters
42
Observations
Approximately 500 000 participants in the experiment with engagement outcomes and up to 10,000 individuals with outcomes from survey data. Follow-up survey number of observations: 2,000
Arms
Intervention starts with 30% of participants assigned to treatment and increases each wave
Primary outcomes
Agricultural input use, timing of input applications, agricultural yields, sharing of information, information spillovers, individual weather forecast/belief, engagement Outcomes for follow-up survey launched in September 2023: perceptions of climate change, perceptions of risks from climate change, 2022 input expenditure and yields, 2023 short-run adaptation (crop, seed choices, conservation agriculture practices), spillovers on network sample
Secondary outcomes
Sources of heterogeneity: variation in climate trends at the tehsil level, variation in 2022 yield expectations compared to realized yields, variation in forecast accuracy, growing zone

Cited by

No citation data available for this registration yet.

Keywords