Evaluating 20,000+ World Bank Development Projects: Can Machine Learning Help Predict which Projects Will Succeed or Fail?
WEDNESDAY, NOVEMBER 17, 2021
11:00–12:00 PM EST | Zoom Webinar
Presentation by Luke Jordan, MIT GOV/LAB Practitioner-in-Residence, with remarks by Dan Honig, Associate Professor, University College London and introductions by Lily L. Tsai, Professor, MIT, and Director and Founder of MIT GOV/LAB.
Free and open to the general public
Description
This event will explore the application of state-of-the-art machine learning (ML) methods to the evaluation of World Bank development projects. It will focus on predicting development outcomes conditional on projects’ core design features, the text used to propose and review the features, country and sector characteristics, and other features. The research seeks to understand which features of development projects, at their approval and after their completion, are most predictive of later outcomes, to inform both prior decision making and later review.
The research presented draws on existing datasets on aid flows and projects, combined with a new dataset of 20,000+ project documents and review documents, scraped from the World Bank’s public archives going back to 1947. That dataset will be described in the presentation and made easily available to researchers via public repositories.
The work-in-progress research finds that by processing the texts into high-dimensional “embeddings” and combining them with financial and other data, projects’ residual connection with development outcomes (i.e., controlling for other effects) can be predicted at both project inception and project conclusion with reasonable accuracy. Alongside projects’ degree of focus and the quality of host country institutions, important factors are the quality of project supervision and the content of projects’ abstracts and governing objectives.
The event will describe the data used, the embedding techniques, the models and their results, and most importantly explore the rich vein of possible research that this combination of data and methods may open up.
Speakers
Luke Jordan is Founder and Executive Director of Grassroot (South Africa) and currently the MIT GOV/LAB practitioner-in residence (intro Q+A). At MIT, Luke launched a guide for practitioners on building civic technology and is exploring different ways artificial intelligence and machine learning (AI/ML) can advance democracy. See links below to the guide and exploratory research.
Dan Honig is an Associate Professor of Public Policy at University College London’s School of Public Policy/Department of Political Science. His research focuses on the relationship between organizational structure, management practice, and performance in developing country governments and organizations that provide foreign aid (https://danhonig.info/).
Lily L. Tsai is the Director and Founder of the MIT GOV/LAB, Ford Professor of Political Science and Chair of the Faculty at the Massachusetts Institute of Technology (MIT). Her research focuses on accountability, governance, and political participation in developing contexts, particularly in Asia and Africa (https://mitgovlab.org/
This event is organized by MIT GOV/LAB as part of our practitioner-in-residence
Photo by Annie Spratt on Unsplash.