Impact Evaluations in Data Poor Settings: The Case of Stress-Tolerant Rice Varieties in Bangladesh

New technologies are sometimes introduced at times or in places that lack the necessary data to conduct a well-identified impact evaluation. We develop a methodology that combines Earth observation (EO) data and advances in machine learning with administrative and survey data so as to allow researchers to conduct impact evaluations when traditional economic data is missing. To demonstrate our method, we study stress tolerant rice varieties (STRVs) first introduced to Bangladesh 15 years ago. Using EO data on rice production and flooding for the entire country, spanning two decades, we find evidence of STRV effectiveness. We highlight how the nature of the technology, which is only effective under a specific set of circumstances, creates a Goldilocks Problem that EO data is particularly well suited to addressing. Our findings speak to the promises and challenges of using EO data to conduct impact evaluations in data poor settings.