Development Through the Looking Glass: Predictive Modeling Development Project Performance

In 2022, $210.6 billion was donated as official development assistance through bilateral or multilateral channels. Yet this development aid has been notoriously ineffective. While many development scholars have investigated this problem through data analysis and case studies, there has been little to no consensus on what makes development projects fail. This thesis probes whether governance data about aid recipient countries could help make that aid more effective.

The rise of machine learning has provided new tools to find patterns between successes and failures within development projects. In this thesis, using machine learning and an experimental survey against human judgment, I built a model to predict the performance of development projects. I used project information spanning over 60 years combined with governance indices spanning from public sector performance to the rule of law to understand whether project characteristics, governance levels, or both influence the performance of development projects. I found that a model that incorporates governance indicators can predict project performance significantly more accurately compared to a model that does not incorporate governance measures, demonstrating the difference a holistic understanding of governance can make for projecting development outcomes.

This rudimentary predictive modeling exercise shows that machine learning can point out patterns of success and failure among a vast corpus of development projects, pointing out policy and project design levers that can prove useful to development professionals as aid is modernized. Ultimately, this thesis shows how these new data science tools can be translated into meaningful contributions to development policy.