Using ML-Generated Data to Study Development

The use of machine learning for causal inference has become increasingly popular in the social sciences. But relatively less attention has been paid to how machine learning (ML) algorithms can be used to generate novel measures in data-sparse environments like those that prevail in many developing countries, particularly those in sub-Saharan Africa.

Here, I present results from a suite of projects that utilize high-resolution measures of economic development generated by a convolutional neural net trained on satellite imagery. I show that, in addition to superior spatial and temporal coverage, this ML-generated data resolves serious inferential shortcomings in existing national and sub-national estimates of wealth, alters influential findings in African political economy, and opens up several promising avenues of research. I demonstrate one such avenue by discussing ongoing work that investigates the impact of climate change on political behavior, an emerging area of scholarship that demands accurate, high-resolution data in places where such data rarely exist. 

ABOUT THE SPEAKER

Brandon Miller-de la Cuesta is a postdoctoral fellow at CDDRL and the Center on Food Security and the Environment (FSE). He received his PhD in Politics from Princeton University in 2020. He has a strong regional focus in sub-Saharan Africa with a special interest in applied methods and political accountability. His current work utilizes machine learning in both dataset generation and causal inference to estimate the impact of infrastructure investments on economic well-being and to investigate how climate change is altering the strength and substance of accountability demands.

Virtual to Public. Only those with an active Stanford ID with access to E008 in Encina Hall may attend in person.