Yale University, Economic Growth Center, New Haven, CT
I show that malaria misdiagnosis, common in resource-poor settings, decreases the expected effectiveness of an important new therapy - since only a fraction of treated individuals have malaria - and reduces the rate of learning via increased noise. Using pilot program data from Tanzania, I exploit variation in the location and timing of survey enumeration to construct reference groups composed of randomly chosen, geographically and temporally proximate acutely ill individuals. I show that learning is stronger and adoption rates are higher in villages with more misdiagnosis. Subsidizing diagnostic tools or improving initial targeting of new technologies may thus accelerate uptake through learning.