Researchers from Cornell University have pioneered a new machine-learning method for mapping poverty. This method could significantly enhance the ability of policymakers and non-governmental organisations to pinpoint and aid the most impoverished populations in developing countries. This novel approach merges national surveys, large-scale data, and machine learning to create a refined tool for resource allocation aimed at eradicating extreme poverty.

Extreme poverty is categorised as living on less than $2.15 per person daily. To effectively combat this, governments and agencies involved in development and humanitarian efforts must understand the scope of poverty—specifically, how many people fall below this economic threshold and their geographic locations. However, obtaining this data is a significant challenge in the regions that most require assistance, according to the researchers at Cornell.

Typically, poverty is measured using household surveys that track income or consumption. These surveys are considered the gold standard for establishing poverty thresholds. Still, they are often either unavailable or not current due to their high costs and the logistical challenges in administering them regularly. Conversely, there has been success in using satellite imagery and other Earth observation data to assess infrastructure, environmental conditions, and human activity. This data has been used to create asset-based poverty indexes, which, while applicable, do not align directly with the monetary metrics that are most valuable to decision-makers.

The innovative method developed by the Cornell team bridges this gap by converting extensive Earth observation data into more actionable metrics for policymakers. This approach was tested in a pilot project across four countries in southern and eastern Africa. It proved to be as precise as traditional asset index methods in mapping poverty. Still, it provided more relevant data, such as the proportion of the population living below the global poverty line. This structural poverty model performed better than previous monetary poverty estimation methods and offers a forward-looking perspective, making it particularly advantageous for planning and program development.

Chris Barrett, a professor of applied economics and management at Cornell and the study’s senior author, emphasised that while rapid advancements in data science have been made, their adoption has been limited because they have not yielded practical estimates for policy implementation. This research makes computational strides more applicable by correlating them with monetary poverty benchmarks.

The “Microlevel Structural Poverty Estimates for Southern and Eastern Africa” study has been published in the Proceedings of the National Academy of Sciences. It was part of a series of inaugural articles by new academy members, including Barrett, and focused on Ethiopia, Malawi, Tanzania, and Uganda. These countries were selected not only because of their significant poverty rates and the active presence of development agencies but also because they provided robust data on consumption and assets, enabling the researchers to model and understand their interconnections.

The research team utilised machine learning models trained on 13 national household surveys conducted between 2008 and 2020, linked with Earth observation data on various assets such as housing quality, land ownership, livestock, vehicles, and access to technology, including mobile phones. According to Elizabeth Tennant, the study’s first author and a research associate in economics, this integration of older survey data and recent satellite observations allows for generating localised “nowcasts” that reflect current conditions rather than past circumstances.

Barrett highlighted the dual benefit of their approach: achieving the precision offered by recent data science advances while maintaining the relevance of these findings for policy and programming. This method provides forward-looking insights, enabling policymakers and agencies to understand who will likely be poor in the present rather than relying on outdated data. This makes the Cornell team’s structural poverty models a powerful predictive tool in the ongoing fight against extreme poverty.

More information: Elizabeth Tennant et al, Microlevel structural poverty estimates for southern and eastern Africa, Proceedings of the National Academy of Sciences. DOI: 10.1073/pnas.2410350122

Journal information: Proceedings of the National Academy of Sciences Provided by Cornell University