DEEP Dives | November 2025 | New evidence, insights and policy takeaways on extreme poverty

Welcome to the second issue of DEEP Dives! The bulletin from the Data and Evidence to End Extreme Poverty (DEEP) research programme.

This is what we’ve been reading over the last few months. Our round-up includes a number of flagship global reports released to coincide with International Day for the Eradication of Poverty on the 17th October.

Child Poverty: Global, Regional and Select National Trends, produced by the World Bank and UNICEF, highlights how children are overrepresented among people living in extreme poverty and that the rate of poverty reduction among children has been slower than that of the general population.

An update on the global status of multidimensional poverty from UNDP and OPHI, Global MPI 2025, Overlapping Hardships: Poverty and Climate Hazards, also highlights the prevalence of child poverty, reporting that around 27.8% of all children live in multidimensional poverty – more than double the rate among adults (13.5%). The main focus though is on the links between multidimensional poverty and climate hazards; with around 90% of multidimensionally poor people living in areas exposed to one or more climate hazards, including high heat, drought, flooding and air pollution.

The World Bank has also produced an update to its poverty assessment in Poverty & Inequality Update. And, plugging a DEEP-supported piece, the Behind the Numbers: How We Measure Global Poverty blog provides a useful overview on global poverty measurement.

We’ve also enjoyed a couple of pieces looking at how effectively machine learning approaches can predict food security and poverty. How Robust are Machine Learning Approaches for Improving Food Security Amid Crises? – Evidence from COVID-19 in Uganda finds that models trained on data from before the pandemic were quite robust in identifying food-insecure households during the pandemic, suggesting that machine learning models can capture underlying vulnerabilities that persist even during acute shocks better than simple regressions. The main variables were household-level demographic and asset indicators. District-level geospatial data also contributed, although more granular spatial data (county or sub-county level) led to overfitting. These are important methodological contributions, although given that the best model correctly identified 80% of food insecure households at the cost of misclassifying 40% of food secure households, there is still work to be done before these approaches can be relied upon to plan aid interventions.

Equitable AI: Exploring the Role of Gender in Poverty Estimation Models using Geospatial Data compares the performance of machine learning models trained using geospatial data to predict the Wealth Index from the Ghana Demographic and Health Survey for male- and female-headed households. Both models showed reasonable predictive accuracy, confirming that machine learning models can be used to fill the temporal and spatial gaps in survey-derived poverty data. However, the machine learning models trained on male-headed households were noticeably more accurate than those trained on female-headed households. The authors show that perhaps half of the performance difference is attributable to the lower number of female-headed households in the sample. The real puzzle is the other half: they speculate this could be because the way the Wealth Index itself is composed is a better measure of the wealth of male-headed households.

More on gender and poverty; in Intra-household Resource Allocation and Changes in Child and Gender Poverty between 2011–12 and 2023–24 (full pre-print here: Intra-Household Poverty Reduction March 2025), Ravi and Kapoor interrogate the intra-household distribution of poverty in India by analysing trends in expenditure on clothing (which, unlike food or general household goods, can be assigned to individuals) to understand how resources are allocated. They find a shift in household resources from adult males to children and adult females over the period. This further resulted in dramatic shifts in the incidence of poverty for these three groups, with poverty among children reducing the most, followed by adult women. Poverty among adult men marginally decreased.

Poverty-Conflict Nexus in sub-Saharan Africa: A Scoping Literature Review synthesises evidence on how poverty fuels conflict and vice versa – a relationship that is more and more important to understand as poverty is increasingly focused in fragile and conflict-affected states. (See also an earlier DEEP thematic paper that unpacks this: Ending Extreme Poverty amidst Fragility, Conflict, and Violence.)

And finally, onto another global driver of extreme poverty – climate change. The authors of Heat Stress Causes Economic and Welfare Disparities across Agroecological Zones in Burkina Faso undertake simulations to understand the impacts of heat stress on poverty and inequality through the channel of reduced labour capacity. The authors highlight how the impacts of heat stress will not be uniform and that certain sectors (agriculture), geographic zones and types of household (poor households) will be affected more and differentially, with impacts for both poverty and inequality.