UC Berkeley

Big data promises to bring new opportunities of cost saving and data-driven decision making into the field of international development. This dissertation illustrates different ways of utilizing the ever growing repository of satellite imagery and crowd-sourced data to evaluate effectiveness of development programs, while achieving cost saving and timely delivery of insights. Chapter 2 leverages high-resolution daytime satellite imagery and deep learning methods to evaluate development aid programs entirely remotely, using an unconditional cash transfer program in western Kenya as a proof of concept. Chapter 3 shows how we compile crowd-sourced policy data to deliver timely insights on the effectiveness of international COVID-19 containment policies. Chapter 4 explores a new idea to quickly and relatively inexpensively gather more evidence on the long-term impacts of development programs - instead of implementing a new program and tracking the participants for decades to come, we revisit existing randomized control trials that were conducted in the past few decades and identify those that can be followed up for evaluation.