Decoupling food insecurity from shocks: A global analysis of food system resilience

Significance

Our study provides a global assessment of food system resilience, based on panel data from high-, middle-, and low-income countries and using a resilience-as-capacity framework. By integrating multicountry indicators of shocks, governance, economy, system properties and social conditions, we offer an empirical multivariable analysis of the factors shaping countries’ ability to maintain food security during crises. This work advances resilience science beyond conceptual/theoretical models by offering data-driven evidence of the mechanisms underpinning responses to shocks at system level. These findings have broad relevance for researchers across disciplines, from sustainability science to public health and development studies, while offering direct policy-relevant insights for actors seeking to strengthen national preparedness and food security amid intensifying climate- and conflict-related shocks.

Abstract

Food systems worldwide are increasingly exposed to climate extremes, conflicts, pandemics, and economic shocks, raising concerns about their capacity to safeguard food security and nutrition. Although resilience is now central in global policy debates, empirical evidence on what enables national food systems to withstand shocks remains limited. Here, we analyze a unique multi-country dataset integrating measures of shock severity, governance, economic conditions, social capitals, and food-system characteristics. Using a causal resilience-as-capacity framework and a series of hybrid econometric models, we assess how structural factors and emergent system properties shape countries’ ability to limit rises in prevalence of undernourishment when affected by shocks. We show that experiencing more severe shocks does not systematically lead to worse food security: several highly affected countries—particularly in Asia—display encouraging outcomes despite severe disturbances. More generally, countries with stronger resilience capacities and more favorable food-system characteristics fare better during shocks. Especially, within-country improvements in government effectiveness, food-supply redundancy, or GDP per capita significantly reduce deterioration in undernourishment, while stronger social capital supports long-term improvement across countries. Together, these findings indicate that food-system resilience capacities can weaken the relationship between shock severity and food insecurity, with food-system resilience then understood as the ability to decouple food-security outcomes from shock severity. These findings provide one of the first global empirical assessments of the factors underpinning food-system resilience. Strengthening institutional effectiveness, enhancing redundancy, and investing in social capital emerge as critical pathways for building resilient food systems in an increasingly shock-prone world.

Around the globe, food systems face an increasing number of profound challenges, including rising population pressure, intensifying climate-related extreme events, pandemics, armed conflicts, and economic instability (1, 2). These shocks and stressors have revealed the fragility of current food systems and underline the urgency of strengthening their ability to withstand local and global disturbances (3). In particular, following the COVID-19 pandemic, numerous experts and institutions called for greater food-system resilience, rapidly elevating it to a political priority for governments and international organizations (4). More recently, Russia’s 2022 invasion of Ukraine underscored the world’s dependence on a small number of food-exporting countries but also how fragile prevailing systems are (5). At more local scales—and often outside international media attention—food systems are constantly disrupted by local shocks and adverse events such as droughts, floods and local armed conflicts (6, 7) regularly threatening the food security, nutrition and well-being of millions of people, particularly in low- and middle-incomes countries (LMICs).

Understanding the factors that contribute to the ability of (local and global) food systems to cope, adapt, or transform in the face of disruptions (that is, their resilience) is therefore critical for ensuring the food security and long-term well-being of human populations (8, 9). Yet, in the absence of a dominant explanatory model, the elements proposed in the literature to explain resilience are often presented as sets of relevant attributes rather than as components of a coherent explanatory framework. Two main issues severely restrict our ability to understand and address this critical issue: i) the paucity of conceptual frameworks providing a rigorous basis for measuring and assessing food systems’ resilience in a coherent, robust and consistent way (10, 11). Existing approaches tend to focus on specific components of the system—such as supply chains, production systems, or individual actors—without fully capturing the interactions, feedbacks, and cross-scale dynamics that underpin resilience processes in complex food systems (11); and ii) a general lack of data necessary for measuring resilience, including the absence of high frequency, longitudinal data documenting food system actors’ behavior, their adaptive responses to shocks, and the resulting effects on resilience outcomes (12). This lack of data constrains the ability to empirically test causal relationships and limits the development of robust metrics of resilience over time. As a consequence, while a rapidly growing number of modeling studies are exploring potential pathways to resilience, many of those remain essentially conjectural, e.g., ref. 13, often relying on stylized assumptions or simulated behaviors that are difficult to validate on the ground. Empirical studies that analyze food system resilience using real-world data are generally scarce and remain often local or at best regional, focusing on specific contexts such as conflict-affected areas or particular country groups, e.g., refs. 7 and 14, therefore offering limited insights into the generalizable drivers of resilience at a global scale.

This paper aims to address that gap by conducting a global assessment of national food systems’ resilience, combining a conceptually grounded causal framework and a series of econometric analyses performed on a multicountry, multiyear dataset. The dataset was built from the database made available by the Food System Countdown Initiative (FSCI) (15). A characteristic of the FSCI dataset is the transparent and rigorous protocol that was used to select the indicators—see table 3 in ref. 15 and figure 1 in ref. 16. Data for these indicators were identified through a multistage, multistakeholder process. They are mostly open access, generated through well-documented methodologies, and are clearly linked to specific food system dimensions and processes. As such, they offer a holistic overview of food systems, from drivers to outcomes (15). Most importantly, these indicators are available for a large number of countries across income groups, thus enabling a global multicountry analysis.

Informed by the recent global events that affected food systems, the following research questions structured the present exploratory work:

• ▪Do specific characteristics or properties of food systems play more decisive roles than others in building the resilience of these systems? While properties such as redundancy, diversity or good governance are generally considered central for the resilience of ecological, social or physical systems (17, 18), it is still unclear whether this is also the case for food systems.
• ▪Are food systems in LMICs more, or less, resilient than those in higher-income countries? While the literature tends to assume greater vulnerability of food systems in LMICs, e.g., refs. 19 and 20, the recent experience of COVID-19 and some regional and historical analyses suggest a more complex and nuanced reality from which important lessons may be derived (21, 22).
• ▪To what extent does reliance on external sources of food (import, aid, and trade) strengthen, or weaken, country’s food system resilience? Many experts contend that international connections are critically important for the stability of the whole system, e.g., refs. 23 and 24. Others challenge this view, arguing that overconnectivity (that is, a state of excessive, dense, and tightly coupled interdependencies between nations, firms, and financial markets) is a source of vulnerability and that more regional or even local systems would be preferable from a resilience perspective (25, 26).

Building on recent research on resilience measurement in the context of humanitarian and food security crises (27, 28), we draw on the “resilience-as-a-capacity” approach (29, 30) to structure the analysis. According to this approach, the resilience of a system mainly reflects the extent to which the actors of the system (individuals, communities, institutions) can rely on their various absorptive, adaptive and transformative capacities to develop specific strategies (responses) to mitigate the impact of the specific adverse event they are facing (31). These resilience capacities (such as individual or collective agency and assets; past investments/savings, levels of livelihood diversification; professional experience, etc.), combined with emergent properties at the system level (connectivity, redundancy, diversity, competitiveness), and mediated by the wider environment of the system (governance arrangements, level of infrastructure such as energy supply, communication and digital connectivity, etc.), are thought to be key in determining the ultimate outcome of the resilience process.

Under this approach, while resilience per se is not measured, its effect (or lack thereof) can be assessed by monitoring change (improvement, degradation) in the long-term outcomes of the system (32, 33). In the present case, this long-term outcome was assessed through the occurrences of fluctuations in the prevalence of undernourishment at the population level. This outcome-oriented perspective moves beyond measuring resilience through livelihood capitals (34) or treating resilience as a statistical property (35, 36) and explores instead whether resilience is realized in practice through specific strategies, thus offering valuable policy insights into which factors genuinely drive improvements in food system resilience and under what conditions.

Altogether these different elements can be organized into a logical framework comprising six components linked through cause–effect relationships: 1) shocks/disruption; 2) actors’ resilience capacities; 3) system emergent properties; 4) actors and system’s responses; 5) resilience outcomes and 6) systems’ general context (Fig. 1).

Fig. 1.

The six components of the “resilience-as-capacity” framework used in this analysis: (1) shocks; (2) resilience capacities; (3) system emergent properties; (4) system’s responses; (5) resilience outcomes and (6) food systems’ general context. Under the resilience-as-capacity approach, the resilience of a system mainly results from the strategies/responses (coping, adapting or transforming) put in place by the actors (in anticipation or in reaction) to a shock, by drawing on their resilience capacities and the emergent properties of the system, to mitigate the direct effect of the shock. The choice of the responses is itself directly influenced by the immediate impact (or its anticipation) and the resilience capacities at hand. These capacities include a wide range of individual and collective/institutional factors such as levels of assets/savings or infrastructure, prior investments, levels of livelihood diversification, social capital, access to knowledge/early warning systems/credits, etc. (11). Emergent properties (that is, properties that become apparent at system-level but which individual parts do not exhibit on their own) include factors such as connectivity, redundancy, diversity, competitiveness/rule of law, and inclusiveness (37, 38). The ultimate outcome of the resilience process is assessed through rapid changes in food security—where food security is measured by indicators such as the prevalence of undernutrition, Household Food Insecurity Access Scale (HFIAS) or Food Insecurity Experience Scale (FIES) (32).

Building on this framework, a multicountry, multiyear analysis was conducted using the FSCI dataset alongside three additional relevant indicators (Table 1 and Materials and Methods) where the dependent variable was the occurrence of annual fluctuations in the percentage of the population affected by undernourishment. Hybrid econometric models which combine the strengths of both fixed- and random-effect specifications were used for the analysis (39–41). These models are particularly well-suited as they capture both within-country changes over time (“level-one” variations) and between-country differences (“level-two” variations). This dual feature is crucial when studying dynamic processes such as resilience, which is shaped by both structural time-invariant factors (e.g., geographic location, access to seaports) and temporal changes (e.g., economic growth, social capital). Importantly, by leveraging this hybrid structure, the models move beyond purely descriptive analysis, helping us assess not only whether resilience capacities exist, but under what conditions they translate into improved food security outcomes. By combining these models, our approach not only aligns with established longitudinal research practices, but also generates more robust, actionable policy-relevant insights, offering a clearer understanding of the factors that drive resilience across diverse food system contexts.

Table 1.

List of indicators used in the analysis, organized into the six components of resilience analysis shown in Fig. 1

Resilience components	Indicators*	Rationale for selection	Data availability
Shocks	Shock severity derived from the EM-DAT global database normalized by the country’s √GDP	The original indicator proposed in the FSCI [the country’s level of exposure to shocks derived from the EM-DAT global database (42)] was amended to detect adverse events (shocks). Shocks (sudden increase in disaster-related damages, estimated in 103USD) were identified first by normalizing the EM-DAT annual disaster damages values by the country’s √GDP and then retaining the normalized values for years in which they exceed the values observed in the previous years, otherwise assigning zero. Rationale for inclusion: identifying and, when possible, assessing the severity of shocks affecting a particular system is an essential element of resilience analysis (29, 43).	Annual data were available for 193 countries over the period 2000–2024.
Resilience capacities	Social capital level estimated by the Social Capital component of the Legatum Prosperity Index	Social capital has long been recognized as a central element of resilience (17, 44, 45). In the present case, although the initial indicator (the Social Capital component of the Legatum Prosperity Index (46)) is computed for countries’ entire populations, it was assumed to be a reasonably good proxy for the level of social capital of the actors within the food system.	Annual data were available for 165 countries over the period 2007–2021.
	Mobile phone infrastructure	Level of infrastructure and level of communication are two key elements of resilience (10, 47). In the present case, although we recognize that mobile phone infrastructure is much more specific than infrastructure in general, the number of mobile cellular subscriptions offers the chance to capture both the level of infrastructure and the capacity for the food system actors to communicate/coordinate responses, making it particularly relevant for a resilience analysis. Like for social capital, although the initial indicator is estimated for countries’ entire populations, we assume it was a good proxy for the “sub-sample” of the food system actors’ population.	Annual data were available for 191 countries over the period 2000–2022.
	Government effectiveness estimated through the World Governance Indicators’ index	High level of government effectiveness has often been documented as an important pillar of resilience, at both local and higher/national levels (22, 37). Like for the social capital and the mobile cellular subscriptions indicators, although this index is initially estimated at country level, we assumed it is a reasonably good proxy for the sub-sample of the country’s food systems’ populations.	Annual data were available for 193 countries over the period 2002–2022.
Emergent properties	Food supply redundancy estimated through the dietary sourcing flexibility index	Diversification (of portfolios, livelihoods, income, source of food, etc.) is a risk strategy well established in both theoretical and empirical literatures (23, 45). In the present case, the dietary sourcing flexibility index computed by FAO is used to capture the diversification of sources of food supply at the system level. As such, it is a very relevant indicator for the resilience of food systems. It is presented here as an “emergent properties” of the systems because it results from the aggregation of multiple actors’ strategies across the systems.	Annual data were available for 142 countries over the period 2011–2019.
General context	GDP per capita	GDP per capita is not part of the original FSCI dataset. It was included however in our analysis because GDP per capita is usually seen as a good proxy for countries’ levels of economic development. As such, it is useful in our analysis for exploring the second research question and in particular testing whether, for a given shock, a higher level of economic development is associated with higher (or lower) change in the final outcomes.	We used the log-transformed series of the annual data available for 195 countries over the period 2000–2022.
	Cereal dependency ratio	The Cereal Dependency Ratio is the second indicator included in the analysis that was not built in the original FSCI. Including this indicator allows us to explore our third research question and in particular to investigate whether, for a given shock, higher cereal dependency ratios are generally associated with higher (or lower) changes in the final outcomes.	Annual data were available for 165 countries over the period 2001–2021.
	Geographical region	The following nine-category code was used: 1 = central Asia; 2 = eastern Asia; 3 = Latin America and the Caribbeans; 4 = North Africa and West Asia; 5 = North American and Europe; 6 = Oceania; 7 = Southeast Asia; 8 = Southern Asia; 9 = Sub-Sahara Africa. Rational for inclusion: Beyond the level of economic development, we are also interested in exploring whether particular regions of the world are associated with higher (lower) level of food system resilience.	Dummy variables available for all countries
	Food price volatility	Rational for inclusion: volatility in price has long been recognized as an indicator of short-term adjustment in economic systems, reflecting structural dynamics (48, 49). As such, it should not be seen as a final outcome (contrary to what many would assume) but instead as a proxy for food systems’ general context resulting from policies such as trade restrictions, taxes and subsidies, stockholding, and public long-term investment.	Annual data were available for 182 countries over the period 2002–2022.
	Food supply variability	Rational for inclusion: Variability in food supply is generally seen in the literature as resulting from a combination of instability and responses in production, trade, logistics and infrastructure, consumption, and storage (3, 48). As such it is a relevant proxy for food systems’ general context.	Annual data were available for 166 countries over the period 2002–2022.
Long-terms outcomes	Prevalence of undernourishment	The original indicator proposed in the FSCI (the country’s prevalence of undernourishment) was amended to detect marked fluctuations (positive = deteriorations, negative = improvements) in undernourishment. Relative deviations from the mean were used to identify these fluctuations. Rationale for inclusion: because the ultimate function of food systems is to ensure food security, deterioration in food security indicators in the face of shocks can be seen as indication of the food system’s lack of resilience (8, 29). On the contrary, absence of deterioration even after a shock would suggest a high level of resilience.	Annual data available for 161 countries over the period 2001–2021.

^*Indicators’ definitions detailed in SI Appendix, Table S1 in Materials and Methods section.

Results

The causal framework presented in Fig. 1 was used to identify relevant indicators among those originally included in the FSCI dataset. Eight of these FSCI indicators were considered particularly pertinent for the analysis: i) countries’ exposure to shocks; ii) national index of social capital; iii) level of infrastructure and communication; iv) national level food supply redundancy; v) government effectiveness; vi) food price volatility; vii) food supply variability; and viii) prevalence of undernourishment. These are listed in Table 1 along with the rationale for their inclusion. In addition, three supplementary indicators (not present in the original FSCI dataset) were included in the analysis: GDP per capita, cereal dependency ratio, and geographical regions—to capture important elements of the general context of country-specific food systems. The justification of their inclusion is also provided in Table 1. In total, 11 indicators were thus included in the analysis. The detailed definition for each of these 11 indicators is provided in SI Appendix, Table S1 (see also Materials and Methods section) while SI Appendix, Table S2 displays their general statistics and the pairwise correlation matrix. The time period and number of countries for which data were available across all these 11 indicators is nine years (from 2011 to 2019) and 87 countries.

Prior to conducting the multicountry multiyear analysis involving the full set of 11 indicators, we performed a preliminary descriptive exploration using two proxy indicators to visualize cross-country patterns. Specifically, we examined the relationship between the average severity of shocks experienced by countries (X-axis) and the ability of those countries to buffer the effects of those shocks on their food systems (Y-axis). It is important to note that these two proxy indicators were constructed for this exploratory visualization and are distinct from the variables subsequently used in the econometric analysis presented in Table 2.

Table 2.

Hybrids model estimations for the three best fitted models (2), (5) and (7)—see details in SI Appendix, Table S4

Models	(2)		(5)		(7)
Variables
Eastern Asia_Region	−0.13	(0.12)	−0.12	(0.13)	−0.13	(0.13)
Latin America and the Caribbeans_Region	−0.00	(0.11)	0.01	(0.11)	−0.00	(0.11)
North Africa and West Asia_Region	0.22*	(0.11)	0.22*	(0.10)	0.23*	(0.11)
North American and Europe_Region	−0.10	(0.14)	−0.08	(0.14)	−0.09	(0.14)
Oceania_Region	0.41***	(0.12)	0.44***	(0.12)	0.41***	(0.12)
Southeast Asia_Region	−0.09	(0.10)	−0.08	(0.10)	−0.08	(0.10)
Southern Asia_Region	0.04	(0.11)	0.05	(0.11)	0.04	(0.11)
Sub-Saharan Africa_Region	0.19*	(0.09)	0.19*	(0.09)	0.19*	(0.09)
Within-country effects
Shocks	0.02*	(0.01)	0.02**	(0.01)	0.02*	(0.01)
GDP_per_capita (Log)	−0.48***	(0.14)	−0.39*	(0.15)	−0.40*	(0.16)
Cereal_Dependency_ratio	−0.00	(0.00)
Cereal_Dependency_ratio_lag(1)			0.00	(0.00)	0.00	(0.00)
Government_effectiveness	-0.18*	(0.09)	−0.16*	(0.08)	−0.16*	(0.08)
Social_Capital_Index	-0.24	(0.30)	−0.19	(0.29)	−0.20	(0.29)
Food_Supply_Redundancy	−1.19**	(0.42)	−1.25**	(0.45)	−1.18**	(0.43)
Food_Supply_Variability	0.00	(0.00)			0.00	(0.00)
Food_Price_Volatility	0.00	(0.03)	0.01	(0.03)	0.00	(0.03)
Mobile_Infrastructure	−0.00	(0.00)	−0.00	(0.00)	−0.00	(0.00)
Between-country effects
Shocks	0.13	(0.11)	0.15	(0.11)	0.14	(0.11)
GDP_per_capita (Log)	−0.08	(0.08)	−0.10	(0.08)	−0.08	(0.08)
Cereal_Dependency_ratio	−0.00	(0.00)
Cereal_Dependency_ratio_lag(1)			−0.00	(0.00)	−0.00	(0.00)
Government_effectiveness	0.11	(0.06)	0.12*	(0.06)	0.11	(0.05)
Social_Capital_Index	−0.85**	(0.30)	−0.87**	(0.31)	−0.90**	(0.31)
Food_Supply_Redundancy	0.18	(0.29)	0.17	(0.29)	0.17	(0.29)
Food_Supply_Variability	−0.00	(0.00)			−0.00	(0.00)
Food_Price_Volatility	0.39*	(0.18)	0.38*	(0.19)	0.38*	(0.18)
Mobile_Infrastructure	−0.00	(0.00)	0.00	(0.00)	0.00	(0.00)
Constant	0.24	(0.54)	0.37	(0.54)	0.30	(0.53)
Observations (level 1)	678		678		676
Number of groups (level 2)	87		87		87

Dependent variable: fluctuations in prevalence of undernourishment. Note: Reported coefficients are unstandardized regression coefficients expressed in the original units of each variable. Consequently, coefficient magnitudes should not be directly compared across predictors as indicators of relative importance. Interpretation: the table shows that at country level, a 0.1-unit increase in government effectiveness is associated with a 0.016 to 0.018 percentage-point (pp) decrease in the prevalence of undernourishment, while a 0.1-unit increase in food-supply redundancy is associated with a 0.11 to 0.12 pp decrease in the prevalence of undernourishment. At the between-country level, a 0.1 increase in social capital is associated with a 0.09 pp decrease in the prevalence of undernourishment. Robust SE in parentheses. *P < 0.05; **P < 0.01; ***P < 0.001.

The first dimension (average shock severity) was constructed by computing the mean of the shock severity series (as defined in SI Appendix, Table S1) for each country over the period 2011–2019. The second dimension (ability to buffer shock effects) was proxied by computing the average deviation from the mean in prevalence of undernourishment (PoU) for each country over the same 2011–2019 period. The analysis was conducted for the 105 countries for which these two indicators are available (SI Appendix, Table S3).

Results are presented as a quadrant plot (Fig. 2). Countries above the horizontal dotted line display positive average deviations in undernourishment (i.e., an increase in their average prevalence of undernourishment over 2011–2019—thus signaling a degradation in their food security), whereas countries below the line display negative average deviations (i.e., decreases in PoU—improvement in long-term food security). The vertical dotted line represents the average shock severity computed across the 105 countries. Countries to the right of this line were exposed, on average, to more severe shocks than the global sample’s mean over 2011–2019, whereas those on the left experienced comparatively less severe shocks. The figure also incorporates countries’ population sizes and GDP per capita through marker size and color gradients.

Fig. 2.

Relationship between the average severity of shocks experienced by countries (X-axis) and changes in food insecurity (Y-axis). Average severity of shock is constructed by computing, for each country, the average value of the shock indicator (as defined in SI Appendix, Table S1) over the period 2011–2019 (note that for visualization purpose, Fig. 2 X-axis is log-transformed). Food insecurity is measured as the average deviation in the prevalence of undernourishment (PoU) over the same period. Importantly, these indicators are simplified proxies used for descriptive visualization and differ from the variables used in the econometric analysis presented in Table 2. The vertical dotted line indicates the average level of shock severity across the 105 countries. The horizontal dotted line separates countries with increasing food insecurity (positive average deviation in PoU) from those with improving food security (negative average deviation). Circle size represents country population and color represents GDP per capita. Country names are shown for outliers and for countries with populations above 80 million for illustrative purposes. The four quadrants provide a descriptive visualization of how countries differ in terms of shock exposure and their observed food-system responses to those shocks, proxied here by changes in the prevalence of undernourishment.

Fig. 2 reveals no clear correlation between countries’ average shock severity and countries’ average changes in PoU (R² = 0.0007, P > 0.05). This indicates that higher severity of shocks does not systematically translate into worse food security outcomes. The graph also reveals the absence of any clear color pattern in GDP per capita, suggesting that income levels alone do not shield populations from the impacts of shocks. A notable cluster of South and South-East Asian countries (Philippines, India, Nepal, Thailand, Vietnam, Japan) is visible in the lower-right quadrant. These countries are affected by above-average severity shocks but display favorable long-term outcomes in terms of PoU. Importantly, several of the most populous countries are concentrated in this group, meaning that large shares of the global population live in severe shock environments but do not necessarily experience worsening food security. Overall, the results highlight that shock severity alone is not a sufficient predictor of food insecurity outcomes. The absence of a clear relationship between the two provides empirical support for the concept of decoupling between food insecurity and shocks, suggesting that resilience capacities may enable some countries to buffer the effects of severe adverse events more effectively than others.

To explore further those preliminary observations, several econometric models were estimated, using the full set of 11 indicators, with the objective of identifying the relationships linking dependent and explanatory variables. In all models, the dependent variable was the occurrence of annual fluctuation (deteriorations, improvements) in PoU identified by computing the deviations from the mean in the 2011–2019 time series. For the explanatory variables, different combinations of the remaining ten variables were considered. SI Appendix, Table S4 details the composition of these different models along with their respective goodness of fit (see Methodology for variable selection and model identification).

Table 2 presents the results of the econometric analysis for the three top best-fitted models (models 2, 5 and 7 as per SI Appendix, Table S4) while the other best-fitted models initially explored for the analysis are presented in SI Appendix, Table S5. Robustness checks addressing potential endogeneity were conducted for all models using an instrumental-variable (IV) approach based on lagged values of the potentially endogenous variables (SI Appendix, Table S6).

Remarkably similar results are observed across the three best-fitted models, suggesting robustness of the structure identified through the econometric analysis. The first key result is that the same three regions (North Africa and West Asia; Oceania; and Sub-Sahara Africa) consistently exhibit statistically significant level-two effects across all three models (Table 2). This indicates that over the period considered (2011–2019), countries in these regions were likely to experience more severe fluctuations in undernourishment than those in other regions of the world, all else being equal.

Second key result: Among the within-country effects identified by the three models, the only variable positively associated with deviations in PoU from country-specific averages is the occurrence of episodes of increased disaster-related damages. This means that years characterized by larger-than-usual shock episodes are generally associated with more marked fluctuations in PoU, holding other factors constant. At the same time, three variables (GDP per capita, level of government effectiveness, and food supply redundancy) exhibit negative within-country effects, signaling that these factors have effective mitigation/buffer effects against undernourishment fluctuations. In other words, as individual countries see their level of economic development increase, their level of government effectiveness improves, and/or their food systems’ supply redundancy intensify, the severity of the fluctuations in undernourishment decreases, ceteris paribus.

Third key result: Country food price volatility systematically shows significant positive between-country effects across the three models. This implies that countries experiencing higher average price volatility over the period considered also show higher average undernourishment fluctuations, holding other factors constant. Conversely, countries with higher levels of social capital are, on average, less prone to severe fluctuations in undernourishment, everything else equal.

Discussion

This study develops a quantitative assessment of national food system resilience, using a conceptually grounded causal framework and a multicountry, multiyear dataset. Covering low-, middle-, and high-income countries, the analysis investigates whether some countries’ characteristics and/or system-specific factors contribute to build, or to undermine, food system resilience across heterogeneous shock contexts.

Resilience is understood here as the ability of the actors of the food system to maintain, protect, or recover the system’s long-term outcomes (food security) in the face of adversity, through their capacities to cope, adapt or transform. As such, resilience builds on actors’ agency and system’s emergent properties (38).

Using fluctuations in undernourishment prevalence as a proxy for long-term resilience outcomes, the analysis uncovers several important findings. First, the data reveal the absence of any clear correlation between average severity of shocks and average degradation in food security (Fig. 2). In other words, and contrary to conventional expectations, experiencing more severe shocks does not consistently translate into more severe food insecurity at national level. Some countries, despite being exposed to high levels of shocks, exhibit stable or improving trends in undernourishment. This suggests that these countries are able to buffer the effects of shocks more effectively and may therefore be interpreted as displaying stronger resilient food systems. This applies in particular to the countries located in the Bottom-Right Quadrant of Fig. 2. In contrast, countries in the Top-Left Quadrant of Fig. 2 experience on average less severe shock but still show high average degradations in undernourishment. This suggests they are less capable of buffering the effects of shocks and may therefore be considered “less resilient” than other countries. The only countries that follow the expected pattern “severe shocks –> high degradations in food security” are those located in the Top-Right Quadrant.

Fig. 2 also reveals the absence of any clear pattern related to GDP per capita. We will return to this issue below and reveal why this initial absence of pattern at global level does not necessarily contradict the result observed at country level. On the other hand, the graph indicates that many populous countries are among these exposed to higher-than-average shocks (larger bubbles on the right part of the graph). Among those, a cluster of South-(East) Asian countries is visible in the bottom right-hand side where larger-than-average shocks are recorded. This is in line with the fact that many Asian countries are recognized to be increasingly exposed to climate-related extreme events such as record-breaking monsoon (India), severe typhoons and frequent landslides (Philippines, Vietnam) but also occurrence of natural disasters such as earthquake (Japan, Nepal) (50). Yet, at the same time, these countries display positive trends in their average undernourishment (being below the 0-horizontal line). In other terms, being exposed to severe shocks did not stop these Asian countries from continuing on their already well-documented path toward effective undernourishment reduction (48). In sum, the relationship between shock severity and degradation of countries’ food security is not obvious, reinforcing the importance of better identifying what constitute(s) the ability—or inability—of some countries to mitigate the impact of these shocks—that is, better understanding their resilience and the interventions that may improve that resilience.

Taken together, these observations suggest that severe external shocks do not inevitably translate into deteriorating food security outcomes. Rather, resilience capacities can weaken this relationship by enabling food systems to absorb and buffer adverse events. In this sense, the resilience of food systems can be understood as their abilities to decouple food-security outcomes from shock severity.

The subsequent econometric analysis offers some initial insights into this apparent decoupling. First, the analysis reveals that three regions: North Africa and West Asia; Oceania; and Sub-Sahara Africa consistently exhibit higher fluctuations in undernourishment (level-two effects), suggesting greater vulnerability to food insecurity, irrespective of the severity or nature of the shocks. Several countries in these regions have been, or are still, affected by armed conflicts, which may be the reason for this result since protracted conflicts have been shown in the past to contribute to several of the worst food crises—in the Democratic Republic of Congo, Yemen, Afghanistan, the Syrian Arab Republic, Nigeria and South Sudan (51)—and more recently in Gaza (52).

The presence of one country from Oceania in this group suggests, however, that factors other than conflict and insecurity also play a critical role. Notably, most Oceania countries are Small Island Developing states (SIDS), often recognized as extremely vulnerable to various local or global adverse events—from economic disruptions, to sea level rise and saline intrusion, and climate-related extreme events (14, 44, 53). In this regard, it is worth recalling that the variable used to record the occurrence of disruptions in this analysis—derived from the EM-DAT International Disaster Database (42)—covers a wide range of shocks, including earthquakes, volcanic eruptions, storms, extreme temperatures, floods, landslides, droughts, epidemic, or insect infestation, among others. This broad range of shocks, combined with the large number of countries considered, represents a key strength of this analysis, as it reinforces the link between systemic disruptions and deteriorating food security, while the resilience framework (Fig. 1) provides a novel analytical framework to understand such vulnerabilities in the case of food systems.

In addition, our study provides a robust statistical framework to test the influence of factors often presented in the literature as critical for building or strengthening resilience capacities—such as social capital, infrastructure and communication, system redundancy and diversity, capacities of government, etc. (45). However, despite their prominence in theoretical and conceptual frameworks, the concrete evidence supporting the influence of these factors remains limited. Our results provide new, empirical insights into those factors. Our study suggests in particular that improvements in social capital or in infrastructure and communication (proxied here by the mobile phone subscriptions) do not have any noticeable effects on the resilience of food systems over the period considered. Two hypotheses could explain this: Either changes in these two variables over the study period are too small to generate measurable impacts—suggesting that these two variables are “slow variables”; or the proxies we used to capture these dimensions may reflect only partially the actual changes, thereby failing to detect their potential effects on the resilience outcomes. Further data analysis (SI Appendix, Fig. S1) indicates that the first hypothesis is credible: The two variables show only relatively small or no change over time, which is likely to explain their nonsignificance in the results.

In contrast, food supply redundancy, measured by the diversification of food supply sources (Table 1), shows a clear mitigating effect on the impacts of adverse events on undernourishment fluctuations. This is in line with broader system literature which generally identifies redundancy as a key component of system resilience (18, 54). The argument, however, has so far been explored predominantly in the context of “hard science” systems [e.g., power systems (55, 56)] or in theoretical frameworks derived from fields like ecology (57). This analysis provides one of the first empirical demonstrations of the buffering effect of redundancy in the context of national-level food systems.

This study also highlights the mitigating effect of two other factors at country level: government effectiveness—understood here as quality of public and civil services and processes (Table 1)—and economic development. Government effectiveness—and more generally government capacities—have long been recognized in the humanitarian and disaster-risk reduction literature as critical contributors to resilience (58, 59). In the context of food systems, its importance has been underscored during acute crises, such as the COVID-19 pandemic, when the effectiveness of governments played a key role in managing public health threats and safeguarding food security (60, 61). Our findings extend this argument by showing that the positive role of government effectiveness is not limited to pandemic scenarios; it generalizes across a broad range of types of shocks (geophysical, meteorological, hydrological, climatological and biological disasters), reinforcing its centrality in strengthening food system resilience.

The second mitigating factor is economic development, measured through GDP per capita. The fact that GDP per capita shows a mitigating within-country effect suggests that as individual countries advance economically over time, they conjointly enhance their abilities to mitigate the impacts of shocks on food systems and eventually on food security crises. This is in line with existing literature arguing that economic development often comes with a suite of increased capabilities such as improved adoption of new technologies, greater access to financial markets, etc. at both individual and institution/company’s levels, thus increasing their capacity to anticipate, absorb, adapt to, or even transform while facing adverse events (62). Importantly, it is not just the level of GDP per capita, but rather its growth (i.e., change over time) that matters. In that sense, rapidly growing LMICs may not necessarily have lower resilience potentials than slower-growing higher income countries. This last point may explain why no clear GDP-per-capita pattern was visible in Fig. 2.

The between-country component of the models reveals two other important findings. The first relates to the mitigating effects of social capital, indicating that countries with higher levels of social capital tend to experience lower average fluctuations in prevalence of undernourishment. Note that this result does not contradict but rather nuances the result discussed earlier where within-country changes in social capital were found to have no significant effect on undernourishment. Taken together, these findings suggest that while short-term or incremental improvements in social capital may have limited immediate effects, longer-term or structural differences (captured by cross-country differences) do contribute to mitigate the impact of shocks on undernourishment. In sum, continued investment in strengthening social capital remains important for enhancing food system resilience, even if the effects are not immediately visible.

The second between-country key finding relates to the association between food price volatility and fluctuations in undernourishment. The analysis shows that countries characterized by higher average price volatility also show higher average undernourishment fluctuations. A very rich empirical and theoretical literature is available that discusses the effect of shocks on price volatility (36, 63). The 2008–2009 food price crisis is for instance widely interpreted as resulting from a combination of disruptive factors, including adverse weather conditions (droughts and floods) that affected some of the major food-producing regions while, at a more global level, higher oil prices increased the cost of agricultural inputs, thus further pushing up food prices, and eventually driving more than 130 million people into poverty and an additional 75 million people into undernourishment (64). More recently, the major disruptions induced by COVID-19 (61, 65), and later by the Russian invasion of Ukraine (6) have also been shown to be associated with significant food price volatility and increase in food insecurity. This paper addresses a related but distinct question: It analyzes the impact of food price volatility on food system resilience, after holding the severity of the shocks constant. As such, it adds strong evidence that price volatility is an important factor to consider when protecting against food security crises.

The last result of note concerns the cereal dependency ratio, contributing to the current debate about the role of import and trade on food system resilience and the potential risk that overconnectiveness may induce: While trade is often presented in the literature as one potential strategy for buffering against food system disruptions (23, 24), at the same time, international trade and interdependencies are also often perceived as potential sources of vulnerability for these systems (25, 26, 66). Many of these analyses, however, are based on modeling, e.g., refs. 23 and 67, indicator-based analyses, e.g., ref. 47 or literature reviews of those same modeling and/or indicator-based analyses (11). Relying on empirical data, the models presented in this study did not find any statistically significant association (neither positive nor negative) between the cereal dependency ratio and fluctuation in undernourishment, thus highlighting the need for more empirical evidence on both sides of the narrative.

Several major methodological challenges may have affected the results of this study. First, capturing complex dynamic processes operating at different levels—from local to global—through a set of generic indicators can be difficult and problematic, as important nuances might be lost in the process. The strategy adopted to mitigate this issue has been to use one of the most carefully built, scientifically vetted open access databases available on food systems, established by a large group of international experts and validated through a series of regional consultations (15). Second, being a latent and dynamic variable, resilience remains inherently difficult to measure (29), with risks of endogeneity, possible misspecification and omitted variables. To reduce those risks, we built on one now well-established interpretation of resilience, combined with a rigorously constructed conceptual framework and relied on powerful econometric tools to test the framework. A third limitation concerns the normalization of disaster damages by the square root of the country’s GDP for the same year (i.e., contemporaneous GDP). Because GDP varies over time, changes in the normalized shock indicator could partly reflect changes in the denominator even when absolute disaster damages remain unchanged. We nevertheless retained contemporaneous √GDP as the denominator (as opposed to using a fixed—e.g., first year of the period—or a whole-period average GDP) because the indicator is intended to capture the relative economic severity of disaster losses with respect to the scale of the economy exposed at the time of the event; using a fixed or period-average GDP would disregard changes in that economic scale and introduce different distortions over time. Finally, the fact that many dimensions and processes of food systems are still not monitored by countries (including HICs) (15) remains probably the major challenge. Until indicators specifically dedicated to monitor food systems are routinely collected at local and national levels (ideally at a relatively high frequency), supporting food system resilience through appropriate and timely policies will remain a challenge.

That said, many of the findings revealed by the present analysis offer important initial policy insights. Variables which are associated with mitigating effects indicate areas where investments should be prioritized as they are likely to strengthen the resilience of national food systems. Those include the degree of food supply redundancy and the level of government effectiveness, while social capital also appears as a mitigating factor, albeit with effects that may take longer to materialize. Overall, it is critical to keep in mind that operational responses necessarily differ across shock types (e.g., drought adaptation, flood protection, or seismic preparedness).

Importantly, the results also suggest that the effectiveness and prioritization of these policy levers are likely to vary across country contexts. For instance, countries facing high levels of shock but maintaining stable or improving food security outcomes may benefit from consolidating existing resilience capacities, such as strengthening redundancy and institutional effectiveness (19, 47). In contrast, countries experiencing relatively lower levels of shocks but still showing deteriorating food security outcomes may require more fundamental investments in governance capacities and structural conditions (6). Similarly, in lower-income contexts, where economic and institutional capacities are still evolving, policies that support broader development trajectories may play a particularly important role in enhancing resilience, as improvements in economic development are associated with a greater ability to mitigate the impacts of shocks on food systems (59); whereas in higher-income settings, policy attention may need to focus more on managing systemic risks such as price volatility and interdependencies (48).

Materials and Methods

Definitions and Concepts.

In accordance with the recent literature, food system resilience is defined in this paper as “the ability of the different individual and institutional actors of a food system to maintain, protect, or successfully recover the key functions of that system despite the impacts of disturbances (38, 68). Under this definition, resilience is conceptualized as the ability of food system actors to cope, adapt, or transform in the face of shocks and stressors. This approach builds on the “resilience-as-a-capacity” body of literature that has emerged in the past 10 y, and which emphasizes actors’ agency as the main entry point of resilience analysis (29, 31, 69).

The key functions of food systems include i) ensuring the availability and affordability of sufficient, nutritious and safe food for all—considered as the core function of food systems, thus justifying the choice of a food security indicator (fluctuation in country undernourishment prevalence) as the long-term outcome of food system resilience. In addition, two other key functions are often considered in the literature: ii) the generation of decent livelihoods and viable incomes/profits for those who are economically engaged in food systems; and iii) the protection (or restoration/rehabilitation) of the environmental integrity of agro-ecosystems (38).

Throughout the paper, the terms “shocks” and “adverse events” are used interchangeably. They refer to a wide range of diverse “dramatic” events that are generally recognized to have impacts on the lives and livelihoods of people. These include geophysical events (earthquakes, landslides, tsunamis, volcanic eruptions), hydro-meteorological disasters (floods, seasonal droughts, tropical cyclones, severe storms, tornadoes, heatwaves), as well as other forms of human-made crises (armed conflicts, political or economic crises, pandemics). We acknowledge that “stressors” such as long-lasting multiyear droughts, sea level rise, temperature rising, environmental degradation or even decades of insecurity/low intensity conflicts are also (slow) “events” which can be sources of serious disruptions for food systems and their actors (43). As such, the standard approach is to include them in resilience analysis. Accounting for stressors and their potential impacts is, however, methodologically challenging. In particular, their start and end points (if any) are generally difficult to pinpoint with accuracy. Given these limitations, the current analysis only focused on acute events, measured by the occurrence of sudden increases in the total annual damage induced by all disasters recorded in individual country (cf. Table 1 and next section).

Data Sources and Selection of Relevant Indicators.

The primary data source for this analysis is the Food System Countdown Initiative (FSCI) database. Technically, the FSCI is an interdisciplinary collaboration of scientists and practitioners which was established in 2021 with the ambition to develop a robust food system monitoring framework to be used in complement of existing systems of indicators (for example, SDGs) when the latter are deemed insufficient for food systems (15). For that purpose, the FSCI is built around a monitoring architecture comprising five thematic areas related to food systems (Diets, nutrition and health; Environment, food production and natural resources; Livelihoods, poverty and equity; Governance; and Resilience—see table 1 in ref. 15). Series of regional workshops and global consultations were organized to vet, refine and finalize a pool of potential indicators for each of these thematic areas. Importantly, to be included, indicators had to satisfy a set of specific inclusion/exclusion criteria including clear conceptual relevance to food systems; existence of well-documented methodologies; grounded in accepted theory and practice; clear desirable direction of change, comparable across time and space; large geographic coverage (i.e. available in at least 70 countries); strong preference for indicators and data available in the public domain.

The result of this process is a database (15, 16) comprised of 50 globally relevant indicators, covering a wide range of countries across low-, middle-, and high-income regions. Depending on data availability, many indicators span two decades, dating back to the year 2000. Most of the data are open access and sourced from public domains (e.g., UN agencies, World Bank, OECD). The 50 indicators were structured to cover all the main components of the food system (15). As such, they offer a holistic overview of food systems, from drivers, to activities, to outcomes.

Among these 50 indicators, those relevant to the analysis were identified based on the resilience-as-capacity framework (Fig. 1) supported by the broader resilience literature, e.g., (30, 45, 69). Indicators were considered relevant if they provide direct measures—or reliable proxies—of one of the six components of the framework. Eight indicators meet this criterion: 1) exposure to shocks, directly linked to the shock component of the framework; 2) the national index of social capital; 3) level of infrastructure and communication; and 4) government effectiveness, all three serving as proxies for food system resilience capacities; 5) the dietary sourcing flexibility index, as a proxy for the level of redundancy of the system (an emergent property); 6) food price volatility and 7) food supply variability, reflecting the broader food system general contexts; and 8) prevalence of undernourishment, as a primary component to measure long-term food system resilience outcomes. Definitions for these indicators are provided in SI Appendix, Table S1 and the rationale for their inclusion in the resilience analysis was detailed in Table 1. Note that one other FSCI indicator—the percentage of population experiencing moderate or severe food insecurity, also known as the Food Insecurity Experience Scale (FIES)—could have been considered to assess long-term outcomes. However, this indicator is only available for 120 countries over 2015–2019, compared with 161 countries over 2001–2021 for the prevalence of undernourishment.

The decision to classify food price volatility and food supply variability under the category “food system general context” (rather than, for example, intermediary resilience outcomes) derives from the fact that both are widely recognized to reflect structural dynamics of food systems (3, 48). Indeed, in addition to a combination of instability and responses in production, trade, and storage, food supply variability and food price volatility result from changes in government policies, trade restrictions, taxes and subsidies, stockholding, and public long-term investment (22, 59). Likewise, the drivers of price volatility encompass natural shocks (e.g., extreme weather events such as droughts and floods, pests, crop diseases), but also market and policy factors, such as supply and demand (in)elasticities, speculative activities (e.g., biofuels), government interventions, and exchange rate fluctuations (48, 49). Eventually these two factors (food price volatility and food supply variability) become explanatory factors in our models, justifying their position as part of the general context of food systems.

In addition to these eight FSCI indicators, three additional country-level indicators were considered: 1) GDP per capita, 2) cereal dependency ratio, and 3) geographical region. These indicators further capture the broader context in which national food systems operate. Like the FSCI indicators, data for these variables are available in open access sources. Their definitions are provided in SI Appendix, Table S1 and the rationale for their inclusion is detailed in Table 1.

Altogether, 11 indicators were therefore selected for the analysis. Despite a thorough inspection, no FSCI indicators were found to be appropriate proxies for the food system actors’ responses component. The “reduced Coping Strategies Index” (rCSI) which monitors different strategies adopted by households when faced with shortages of food and/or financial resources (70) is included in the FSCI. It could have, in theory, been a suitable candidate. It is, however, currently reported only in 46 countries (mainly low-income countries) and only over the period 2020–2024.

Two of the 11 indicators—exposure to shocks and prevalence of undernourishment—were modified from their original FSCI format to better address resilience analysis requirements. The transformation of the exposure-to-shocks indicator was introduced to capture episodes of increasing disaster-related impacts—rather than countries’ structural exposure to hazards, which was the original purpose of the FSCI indicator. To do so, annual disaster damages from EM-DAT were first normalized by the square root of country GDP. This normalization had two purposes: first, to express disaster-related economic losses relative to the scale of the economy exposed to those losses. In other words, it intended to capture the relative severity of the economic shock, rather than its absolute monetary value; second, to ensure comparability across countries, avoiding in particular that wealthier countries appear disproportionately exposed simply because of their larger asset bases and more costly infrastructures [square-root transformations are commonly used in statistical analysis to reduce the influence of extreme values while preserving variation across observations (71, 72)]. Shocks were then identified whenever normalized damages in 1 y exceeded those observed in the previous year. In those cases, the normalized damage value for the current year was retained as the measure of shock severity; otherwise the value was set to zero. This procedure captures years characterized by increases in disaster-related damages while preserving information on the magnitude of the shock.

For undernourishment, rather than using the raw prevalence values, we computed a relative deviation measure by expressing annual departures from a country’s average prevalence as a proportion of that average. The resulting normalized series was used to identify rapid changes—both increases and decreases—in the prevalence of undernourishment level at the country’s population level.

Data availability eventually determined the final sample: 87 countries with complete data over a nine-year period (2011–2019). SI Appendix, Table S7 lists these 87 countries while their descriptive statistics can be found in SI Appendix, Table S2. This number, 87, somewhat limits the representativeness of the study. Nevertheless, the sample spans all major world regions and income categories and includes some of the world’s most populous countries, such as India, Indonesia, Pakistan, Bangladesh, Brazil, Nigeria, Mexico, and Japan. It also encompasses a wide range of food-system contexts, including food-importing and food-exporting countries, fragile and stable states, as well as island, landlocked, and continental economies.

Statistical Analysis.

Building on the multicountry multiyear database described above, the analysis was conducted using hybrid econometric models. This particular type of model was chosen because it explicitly decomposes each time-varying explanatory variable into level-one effects (how effects of specific variables such as GDP or social capital vary within-country and over time) and level-two effects (how these impacts vary between countries). This allows the two sources of variation (within- and between-country) to be estimated and interpreted simultaneously within a single model (39, 41). Compared with conventional fixed- or random-effects specifications, this provides a more transparent interpretation of the estimated coefficients and enables direct comparison of within-country and between-country relationships.

Model estimations were performed using the xthybrid command (40) available through the statistical software package Stata (version 18) (73). Because the objective was to explore empirical associations between the occurrence of shocks and outcomes and how these associations are influenced or mitigated by various factors, several models were tested.

The generic hybrid model used in this study was:

$$y_{\mathit{it}}={\beta }_W\left(x_{\mathit{it}}-{\overline{x}}_i\right)+{\beta }_B{\overline{x}}_i+\gamma c_i+u_i,$$

where $i$ denotes the level-two unit (country) and $t$ denotes the level-one unit (year/observation), $y_{\mathit{it}}$ is the dependent variable, $x_{\mathit{it}}$ is the vector of level one explanatory variables that vary within and between countries; $c_i$ is a vector of level-two variables that vary only between countries; $u_i$ is a random intercept; ${\overline{x}}_i$ is the vector of country-specific means; ${\beta }_W$ captures within-country effects and ${\beta }_B$ captures between-country effects. Such hybrid models include both the deviation from the country-specific mean $\left(x_{\mathit{it}}-{\overline{x}}_i\right)$ and the country-specific mean ${\overline{x}}_i$ among the model covariates. In all models, the dependent variable was the fluctuations (deviations from the mean) in the prevalence of undernourishment (PoU) series (used as a proxy for food system resilience’s long-term outcome) while the remaining ten indicators were used as explanatory variables. Significance tests (t tests/P-values) and diagnostic checks (multicollinearity test; residual analysis, heteroskedasticity) were implemented along with Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) to guide the variable selection process.

Hausman and F-tests were implemented as part of a broader robustness assessment. The Hausman test was used to evaluate whether the assumptions underlying random-effects models were compatible with the data by comparing fixed-effects and random-effects estimates. The F-test was used to compare fixed-effects models with pooled ordinary least squares (OLS) models. Results are presented in SI Appendix, Tables S8 and S9. Hausman test (SI Appendix, Table S8) reveals no systematic differences between the fixed-effects and random-effects estimates. This result is consistent with the use of a hybrid modeling framework combining features of both approaches while allowing separate estimation of within-country and between-country associations. In SI Appendix, Table S9, the significant P-value (Prob > F = 0.0000) indicates that fixed effect models would be preferred to pooled OLS specifications.

An additional test was conducted to assess the robustness of the transformation used for the shock variable. For this, the current models’ results were compared to estimations obtained with two other types of transformations: one where the annual disaster damages from EM-DAT were normalized by GDP (as opposed to square root of GDP) and one where the shocks were computed as total annual damages (as opposed to the increases in damages). All three models showed qualitatively comparable results—in particular a significant positive within-country effect for the shock variables, while the other variables show similar results across the models, confirming the stability of our results.

Lagged_0 and lagged_1 effects were systematically considered for all explanatory variables. Interactions between variables with the most likely chance of effect modification (i.e., GDPpc with government effectiveness or Food Supply Variability with Food Price Volatility) were also tested. The composition and structures of the best fitted models identified through the variable selection process are shown in SI Appendix, Table S4.

Complementary diagnostic tests were conducted to assess potential multicollinearity and nonstationarity issues. Inspection of the pairwise correlation matrix (SI Appendix, Table S2) revealed only moderate correlations among explanatory variables, with no evidence of excessively strong pairwise associations. Variance Inflation Factors (VIFs) were also computed for all explanatory variables and remained well below commonly used thresholds, indicating no problematic multicollinearity. In parallel, Augmented Dickey–Fuller (ADF) tests were performed on all variables and did not indicate nonstationarity, suggesting that the series used in the analysis are stationary.

Finally, while shocks are generally treated as exogenous to resilience capacities, potential endogeneity may still arise through feedback effects (e.g., past resilience influencing current resilience capacities and responses) or through omitted time-varying confounders (e.g., climate conditions) that may potentially affect both resilience capacities and food security outcomes. To assess the robustness of our analysis to these concerns, an instrumental variable (IV) approach was implemented, using lagged values of the potential endogenous variables as instruments for their current levels (SI Appendix, Table S6). This procedure was used as a robustness check rather than as a definitive solution to endogeneity. In addition, the hybrid specification controls for unobserved, time-invariant heterogeneity across countries.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Study data used in this analysis are available from the Food System Countdown Initiative metadataset (74). The FSCI metadataset contains the necessary links to access the underlying raw data.