How Mechanistic Explanations Reshape Learning and Behavior: Evidence from a Fertilizer Choice Experiment in Eastern Uganda, with Anirudh Sankar, Ben Davies, Vesall Nourani, Jess Rudder, Godfrey Taulya, and Abraham Salomon
Funding from IGC, the King Center on Global Development, and the Weiss Fund. In collaboration with Agriworks Uganda.
Abstract: Mechanistic explanations—descriptions of a system through the causal interactions of its parts—play a key role in human cognition and scientific progress. Despite their importance, we lack systematic evidence on whether and how mechanistic explanations help lay decision-makers interpret information in complex economic environments. We evaluate the causal impact of including mechanistic explanations in an information intervention: public demonstrations of fertilizer use for smallholder tomato farmers in Eastern Uganda. In all demonstrations, extension officers showcased the impact of a recommended fertilizer recipe. In the treatment group, officers also explained the mechanisms underlying the recipe’s effect — introducing the language of macronutrients and the causal processes linking nutrients, soil features, and plant growth. We collected detailed data on beliefs and behaviors from 797 farmers in a lab-in-the-field experiment conducted at the demonstration site and followed up with them over two growing seasons. In the lab-in-the-field, treated farmers generalized more effectively—making better substitution and arbitrage decisions among fertilizers and achieving 9% higher simulated profits in an incentivized fertilizer-application task. At endline, treated farmers’ real fertilizer choices reflected improved nutrient timing and balance, and their yields were 14% higher.
Towards a Culture of Learning at Scale through Teacher Professional Development in Uganda, with Vesall Nourani and Sara Restrepo Tamayo
Funding from J-PAL Learning for All Initiative. In collaboration with the Kimanya-Ngeyo Foundation for Science and Technology. AEA RCT Registry Entry. Baseline data collection ongoing.
Abstract: Worldwide, pedagogical culture in schools is didactic, emphasizing one-way knowledge flows, limiting student learning. We study innovations at teacher training institutions in Uganda to reverse this towards a “culture of learning.” Teachers are trained to apply scientific approaches to improve their practice. A coordination structure generates spillovers to transform system-level practice. With these ingredients in place, change takes place through three steps. First, the initial training empowers teachers to use action-research to continuously improve their pedagogy, fostering dynamic, learner-centered, environments both inside and outside the classroom. Second, teachers form communities of practice, where collective reflection leads to a shared analysis of effective practices, enabling continuous improvement and adaptation. Finally, as communities grow and are coordinated by key actors, the program catalyzes a systemic shift, attracting untrained schools and teachers to participate. This embeds a culture of continuous learning that improves and grows independently, benefiting students on a larger scale. Our research seeks to understand how effects of this intervention, dubbed the "Learning to Learn" (LTL) approach, spread and sustain change over time. In this light, we ask: What density of teachers must be trained for effects to spread? How do the extent and speed of spillovers depend on a “coordinator”? What conditions cause a culture of learning to emerge, with dynamically increasing effects on learning? Our design involves randomizing characteristics of the implementation process at three levels: clusters (groups of schools), schools, and teachers. Answers will inform future scale-up plans beyond a single context, including how many teachers to target, what coordinating structures support impact, and conditions of long-lasting change.
Information Resolution and Generalizable Learning: Evidence from Soil Tests and Agronomic Training in Kenya, with Anirudh Sankar, Jess Rudder, and Vesall Nourani
Abstract: Learning about technologies in heterogeneous environments is difficult—what works for others may not work for oneself. We evaluate two approaches to overcoming heterogeneity: delivering high-resolution tailored information and providing generalizable knowledge of the production function that is applicable across contexts. These approaches may substitute for or complement each other. We cross-randomize soil-testing resolution (number of tests per village) and a generalizable soil fertility training to measure effects on smallholders’ beliefs, fertilizer customization, and farming outcomes. By randomizing test resolution and giving each farmer recommendations from the nearest soil sample, we estimate how the “signal translation function” changes with contextual distance. By randomizing the training, we test whether stronger mental models enhance or substitute for data relevance, identifying the value of models in signal translation. We also inform policy: how many soil tests should be conducted at scale, and can farmer training complete the “last mile” of fertilizer customization?
Using a Digital Client Feedback Platform to Improve Health Care Quality: Evidence from Tanzania, with Pascaline Dupas, Dylan Groves, and John Marshall
Funding from FID and the J-PAL Governance Initiative. In collaboration with Wezesha Tanzania. Pilot ongoing.
Abstract: Improvements in health outcomes in the Global South are limited by underutilized and low-quality public health services. Under-utilitzation and low-quality often stem from weak accountability structures, limited information flows between citizens, providers, and bureaucrats, and inadequate methods for gathering citizen feedback. This study evaluates a mobile citizen feedback system designed to improve service quality and accountability in Tanzanian public health facilities. The intervention we study enables patients to anonymously report healthcare experiences via SMS surveys, providing real-time, facility-specific insights on service quality across multiple domains. Feedback is aggregated and shared with facilities and government facility monitoring teams. Using a randomized controlled trial, we assess the intervention's impact on service quality, utilization, community satisfaction, and health outcomes.