Global measures consistently underestimate food insecurity; one in five who suffer from hunger may go uncounted
International humanitarian aid organizations rely on analyses from the Integrated Food Security Phase Classification (IPC) system, a global partnership that monitors and classifies the severity of food insecurity to help target assistance where and when it is most needed. Those analyses are multifaceted and complex – often taking place in regions where data is scarce and conditions are deteriorating – and stakeholders tend to assume they overestimate need. However, a new study from the University of Illinois Urbana-Champaign and collaborators, published in Nature Food, finds the opposite is the case: global food insecurity analyses systematically underestimate hunger.
“Evaluating the accuracy of these analyses is difficult because IPC are trying to identify crises that are coming in the near term. If they are effective and the humanitarian community responds to their analyses, those crises will be averted or at least lessened. This means that if they are correct, they are in a sense always wrong,” said lead author Hope Michelson, professor in the Department of Agricultural and Consumer Economics (ACE), part of the College of Agricultural, Consumer and Environmental Sciences at U. of I.
In 2023, about 765 million people around the world lacked sufficient food to meet their basic needs, and almost one-third of those experienced acute food insecurity that put their lives in danger. Accurately identifying hunger crises is crucial to directing international humanitarian responses.
IPC was established in 2004 as a consortium of 21 partner organizations, and – as of 2024 – it was used to allocate more than $6 billion in humanitarian aid annually.
“The IPC conducts subnational analyses of the food security situation in about 30 countries across the world. They focus on places that face, or are likely to face, tough circumstances whether due to chronic deprivation, an acute climate shock, or geopolitical issues,” Michelson said.
She conducted the research with Chungmann Kim, a doctoral student in ACE at Illinois; Kathy Baylis, professor in the Department of Geography at the University of California, Santa Barbara; and Erin Lentz, associate professor of public affairs at the University of Texas at Austin.
Michelson and her colleagues had previously researched food insecurity outcomes, including the role of machine learning applications in crisis prediction. In 2021, the IPC approached them to conduct an evaluation of their system; the main findings of this report are presented in the Nature Food paper.
IPC conducts consensus-based analyses based on data from different external sources like national statistical agencies or third-party NGOs, measuring dietary quality and diversity, food prices, weather information, and more.
Technical working groups of trained analysts gather to evaluate this information, discussing the aggregated data and taking into consideration the local context. They follow a set of protocols guided by a technical manual drafted by the IPC. Analysts are highly trained in the protocols and processes. Based on their analysis, they assign classifications for each subnational zone, ranging from phase 1 (none/minimal), 2 (stressed), 3 (crisis), 4 (emergency), to 5 (catastrophe/famine).
“We started out by conducting approximately 20 interviews with different humanitarian agencies and organizations that use the IPC system in their decision-making. The results from that qualitative work showed that these users tend to assume the IPC overstates the severity of crises,” she said.
The researchers looked at the same food security data the IPC working groups use in order to assess the process and results. They analyzed nearly 10,000 food security subnational analyses covering 917 million individuals (adding up to 2.8 billion people with multiple rounds of counting) in 33 countries between 2017 and 2023.
To evaluate accuracy in the analyses, they looked at the distribution of population percentages at the threshold between phase 2 and phase 3. When at least 20% of the population moves to phase 3, it signals an urgent need for assistance, so this number serves as a cutoff determining whether a location is in crisis. This is a point in the distribution where under- or overcounting is most likely to show up.
The researchers found clear evidence of “bunching” just below the phase 3 threshold, and this effect occurred for multiple countries with different levels of overall food insecurity.
They furthermore conducted their own estimates based on the available data and compared their results to IPC’s analysis. They identified 293.1 million people in phase 3 or higher, compared to IPC’s estimations of 226.9 million people. That means 66.2 million people, or one in five, who are in urgent need, could go uncounted.
The team found when the food security data that IPC working groups have access to about a given area provide conflicting information about the severity of the situation on the ground, they are more likely to classify that area to be just under the threshold.
“The food security indicators that are available to the IPC analysis teams don’t always agree with each other. The working groups will have different information about the same region over the same amount of time. And we found that they tend to take a more conservative approach in their analysis, especially when indicators are contradictory,” Michelson stated.
The researchers note that the IPC process continues to provide a very important measure of global food insecurity. Working to improve data collection and decision-making can help to improve confidence in the system. For example, machine learning techniques can enhance data collection and modeling, complementing the existing consensus process. The authors stress that automated systems should not replace the current system but might supplement or support it.
There is already a large gap between addressed needs and available funding, and the need is expected to grow while international humanitarian support is shrinking, Michelson said.
“Understanding that the current figures are likely to underestimate the actual global population of food-insecure people further underscores the scale and the scope of need, and the importance of allocating more resources to alleviating hunger worldwide,” she concluded.
The paper, “Global Estimates Systematically Undercount Acute Hunger,” is published in Nature Food [DOI:10.1038/s43016-025-01267-z]
Research in the College of ACES is made possible in part by Hatch funding from USDA’s National Institute of Food and Agriculture.