Predicting poverty and malnutrition for targeting, mapping, monitoring, and early warning

Abstract

More frequent and severe shocks combined with more plentiful data and increasingly powerful predictive algorithms heighten the promise of data science in support of humanitarian and development programming. We advocate for embrace of, and investment in, machine learning methods for poverty and malnutrition targeting, mapping, monitoring, and early warning while also cautioning that distinct tasks require different data inputs and methods. In particular, we highlight the differences between poverty and malnutrition targeting and mapping, identification of those in a state of structural versus stochastic deprivation, and the modeling and data challenges of 1 We thank Nathan Jensen, Varun Kshirsagar, participants and presenters at the AAEA ASSA 2021 Big Data and Near-Real-Time Monitoring of Food Emergencies session for feedback and insights on earlier drafts of this work. Any remaining errors are our own. We thank the United States Agency for International Development for financial support under cooperative agreement # 7200AA18CA00014, "Innovations in Feed the Future Monitoring and Evaluation -Harnessing Big Data and Machine Learning to Feed the Future". The contents are solely the authors' responsibility and do not necessarily reflect the views of USAID or the United States Government.

Food security analysis and forecasting: A machine learning case study in southern Malawi

“…We contribute to a rapidly growing literature that focuses on using new, systematized approaches to improve predictions of food security. McBride et al (2021) provide an excellent review of the existing evidence, emphasizing that these predictive models should be built for purpose.…”

Satellite Image and Machine Learning based Knowledge Extraction in the Poverty and Welfare Domain

“…Baulch and Hoddinott [5] found that 20-65% from 13 panel studies were classified as sometimes poor which were more numerous than the always poor category. McBride [31] means that being able to distinguish between structurally poor and stochastically poor is crucial for well-targeted interventions and would require models (for targeting, mapping, monitoring or early warning) to account for the structural determinants of impoverishment. This relates to the potential to observe if change occurs in either positive or negative direction.…”

“…Targeting is then directed at identifying the households and individuals at risk. McBride [31] discusses the trade-offs between predictive models with hundreds of features and one that includes a few well-known easily observed variables. The latter being useful for poverty targeting, evaluation and monitoring and the former more suited for poverty mapping.…”

Small Area Estimation of Monetary Poverty in Mexico Using Satellite Imagery and Machine Learning

“…Traditionally, small area estimation has been carried out by utilizing survey data to estimate a model and simulating that model in household-level census data (e.g., Elbers, Lanjouw, and Lanjouw, 2003). In the last five years, however, a growing body of innovative research has used geospatial or other big data to predict poverty and generate purely synthetic predictions of poverty or welfare, often generated by convolutional neural networks (Jean et al, 2017;Yeh et al, 2020;Engstrom et al, 2022;McBride et al 2021). These estimates are typically validated against survey data at the village level, using sample data withheld from the prediction model.…”