Environmental Risk Emerges as the Most Consistent Driver of Global Hunger in New Study of 120 Countries

Hunger is not distributed evenly across the world, and neither are the risks that produce it. A new open-access study published in Discover Sustainability by Emna Essadik of the Research Laboratory International Finance Group Tunisia at University Tunis El Manar tackles one of the most stubborn questions in development economics: why do the same global pressures—political instability, demographic stress, environmental degradation, and economic shocks—produce wildly different levels of food insecurity in different countries? The research, which examines 120 countries over the period from 2001 to 2022, argues that the answer lies in moving beyond averages. By combining machine learning clustering with a quantile-based econometric framework, the study maps how four dimensions of systemic risk are associated with hunger across distinct vulnerability profiles, offering a more granular evidence base for the second Sustainable Development Goal, which aims to end hunger and ensure access to safe, nutritious, and sufficient food for all.

The centerpiece of the analysis is the prevalence of undernourishment, the standard indicator used to track progress toward SDG 2. Rather than treating all countries as a single homogeneous sample, the study first builds composite indices that capture four risk dimensions: institutional risk, demographic risk, environmental risk, and economic risk. These indices aggregate multiple underlying variables into coherent measures, allowing the researcher to test whether broad systemic pressures—not just isolated factors such as conflict or drought—are statistically associated with hunger. This multidimensional framing reflects a growing consensus in the food security literature that undernourishment is rarely the product of a single cause. Instead, it emerges from the interaction of fragile governance, population pressure, climate exposure, and volatile economic conditions, each capable of amplifying the others.

The methodological innovation lies in a two-stage empirical strategy. In the first stage, a K-means clustering algorithm—an unsupervised machine learning technique that partitions observations into groups based on similarity—classifies the 120 countries into three food insecurity profiles: high, moderate, and low. The high-insecurity cluster includes countries such as Afghanistan, the Democratic Republic of the Congo, Kenya, Liberia, Madagascar, Sierra Leone, Uganda, and Zambia. The moderate cluster contains a broad group including Bangladesh, Ethiopia, India, Nigeria, and Pakistan, among others. The low-insecurity cluster spans most of Europe, North America, parts of Asia and Latin America, and several African countries such as Ghana, Senegal, and South Africa. This data-driven grouping avoids imposing arbitrary regional or income-based categories and lets the underlying risk structure define the comparison groups.

In the second stage, the study applies pooled quantile regression within each cluster. Unlike ordinary least squares regression, which estimates the effect of a variable on the mean of the outcome, quantile regression estimates effects at different points of the outcome distribution—the 10th, 50th, and 90th percentiles, for example. This distinction matters enormously for food insecurity research. A country sitting at the 90th percentile of undernourishment is in a fundamentally different situation from one at the median, and the factors pushing it deeper into crisis may differ from those that matter for countries closer to the average. Mean-based studies, the paper argues, obscure exactly this kind of heterogeneity, potentially leading policymakers to design uniform interventions that fail where they are needed most.

The headline finding is stark: environmental risk is the most consistent factor associated with undernourishment across all three clusters. In the quantile estimates, the environmental composite index shows positive and statistically significant coefficients across most quantiles in the moderate and low-insecurity groups, with particularly strong associations in the low-risk cluster at the median and upper quantiles. When the composite is decomposed into its components, carbon dioxide emissions show positive and significant associations with undernourishment across nearly all quantiles in the high and moderate clusters, and across the upper half of the distribution in the low-insecurity cluster, where the coefficient at the 90th percentile reaches 1.20. Agricultural land use shows negative associations in the moderate cluster but positive ones in the low cluster, suggesting that the environmental dimension operates through different channels at different stages of development.

Demographic risk tells a more complicated, and arguably more surprising, story. In countries with high and moderate food insecurity, the demographic composite index is positively associated with undernourishment at the lower quantiles, indicating that population pressure adds to hunger risk where food systems are already fragile. But the sign flips at the upper quantiles and in the more secure cluster, where demographic risk is negatively associated with food insecurity. Life expectancy, one of the demographic components, shows negative and significant coefficients at the upper quantiles in both the high and moderate clusters, consistent with the idea that longer-lived, healthier populations are better positioned to secure adequate nutrition. The lesson is that demographic dynamics are not inherently good or bad for food security; their effect depends on the vulnerability context in which they unfold.

Institutional quality, captured by the institutional composite index, emerges as a protective factor—but only in certain settings. In the moderate and low-insecurity clusters, the index shows negative and significant coefficients at the median and upper quantiles, with the strongest effects in the low-risk cluster, where the coefficient reaches −0.63 at the 90th percentile. In other words, better institutions are most strongly associated with lower undernourishment precisely in the countries that are already relatively secure, while in the most vulnerable group the institutional variable shows no significant association at any quantile. This asymmetry carries a sobering implication: governance improvements alone may not deliver measurable hunger reductions in the most fragile states, where environmental and demographic pressures may overwhelm institutional gains, or where institutions may be too weak for the index to register meaningful variation.

Economic risk, by contrast, is the least consistent of the four dimensions. Across most clusters and quantiles, the economic composite index fails to reach statistical significance. The exception is telling: in the most vulnerable group, at the 90th percentile, the index shows a negative and significant coefficient. This means that economic risk matters only at the extreme upper tail of undernourishment in the hardest-hit countries—exactly the situations of acute crisis that mean-based analyses tend to smooth away. The finding underscores the paper’s central methodological claim that quantile-specific, asymmetric associations are invisible to average-effect estimation, and that policies calibrated to the average country may miss the dynamics driving the worst outcomes.

What does this mean for the pursuit of SDG 2? The study’s answer is that interventions should be severity-specific rather than uniform. For countries in the high-insecurity cluster, environmental pressures and demographic stress dominate the picture, suggesting that climate adaptation, emissions mitigation, and investments in health and human capital are inseparable from food policy. For moderate-risk countries, strengthening institutions appears to be a viable pathway to reducing hunger, particularly for populations at the upper end of the undernourishment distribution. For relatively secure countries, the strong environmental signal—especially the carbon emissions component—suggests that even wealthy nations’ food security trajectories are tied to their environmental footprints. The clustering approach itself offers a practical template: by grouping countries according to their risk profiles rather than their geography or income alone, policymakers can identify peer countries facing similar systemic pressures and adapt strategies that have worked elsewhere in the same cluster.

The research also carries broader lessons for how the development community measures progress. Global hunger statistics are typically reported as averages, and international targets are set against those averages. But if the drivers of undernourishment differ systematically across the outcome distribution—as this study demonstrates—then average-based monitoring can mask deterioration among the most vulnerable even as headline numbers improve. The combination of unsupervised clustering and pooled quantile regression offers a replicable framework for other SDG indicators, from poverty to health access, wherever heterogeneity across vulnerability contexts is suspected. As climate change intensifies and geopolitical instability persists, the study’s core message grows more urgent: ending hunger will require understanding not just which risks matter, but for whom, and at what point of the crisis spectrum they bite hardest.

Subject of Research: Heterogeneous associations between multidimensional systemic risks and food insecurity across 120 countries

Article Title: Assessing food insecurity under global risks using a clustered pooled quantile framework for sustainable development goal 2

Article References: Essadik, E. (2026). Assessing food insecurity under global risks using a clustered pooled quantile framework for sustainable development goal 2. Discover Sustainability. https://doi.org/10.1007/s43621-026-04811-4

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04811-4

Keywords: food insecurity, Sustainable Development Goal 2, undernourishment, systemic risk, quantile regression, K-means clustering, environmental risk, institutional quality, demographic risk, economic risk, food security policy, Discover Sustainability