Measuring technology adoption is central to evaluating development policies, yet standard approaches face trade-offs between accuracy, cost, and scalability. Using data from a randomized controlled trial in Niger, we compare four methods for measuring adoption of demi-lunes, an agricultural technology: direct observation, survey self-reports, manual satellite-based observation, and satellite imagery with machine learning (SIML). Treating direct observation as ground truth, we quantify measurement error on extensive and intensive margins and assess trade-offs between costs and error at project and policy scales. We find that self-reports perform well at both scales, while satellite-based methods scale cheaply but generate significant measurement error.
Aker et al. (Fri,) studied this question.