Making impact with agricultural development projects: The use of innovative machine learning methodology to understand the development aid field

Lindsey Moore, Mindel van de Laar, Pui-hang Wong & Cathal O'Donoghue


This paper introduces a novel methodology aimed at addressing a critical knowledge gap related to the lack of a systematic understanding of agriculture projects across spatial and temporal dimensions. This gap has impeded efforts to enhance learning and accountability, thereby reducing the overall effectiveness of foreign assistance to the agriculture sector. To address this gap, deductive and inductive methodologies are applied to develop a standardized taxonomy for benchmarking United States Agency for International Development (USAID) agricultural projects. By applying this taxonomy to code all available final evaluations of USAID projects, a large qualitative dataset was generated. This dataset facilitates the analysis of the rich qualitative information available within public project evaluations and covers ninety countries over a span of six decades. The result of this research is a new dataset on the multi-layer composition of development projects, forming the foundation for a machine learning algorithm that expedites the process of synthesizing qualitative evidence and measuring the impact of development aid projects at a systems level. The overarching objective of this research is to contribute to the improvement of project and policy implementation in the field of agriculture development.

JEL Classification: C40, F35, O13

Keywords: agricultural projects, development aid, interventions, machine learning, USAID

Download the working paper