Main

Global food security, the access to sufficient, safe and nutritious food, has declined slowly but steadily since 20151. Recently, this decline has accelerated owing to the coronavirus disease 2019 (COVID-19) pandemic and measures taken to mitigate its impact on human health. Millions of people were pushed into immediate hunger, and in the medium to long term, existing inequalities are expected to exacerbate further2. To address food security challenges, we cannot rely exclusively on market mechanisms to deliver sufficient quantities of nutritious food at affordable prices. This was illustrated all too well by the world food price crises of 2007–2008 and 2010, which were largely a result of speculative activity in global commodity markets3 and increasing demands for grains as feed and fuel4. We need intervention in the form of governance5: public (for example, government), private (for example, food retailers) or communal (for example, farmer cooperatives) entities that implement measures to improve food security6.

To govern effectively, we need to measure the impact of governance implementation ex post or simulate the impact of governance implementation ex ante. The literature on ex-post assessments of food security governance is too limited to provide empirically founded guidance with regard to choice of governance measures to improve food security6,7. The literature on ex-ante assessment, however, is vast and growing. Ex-ante assessments are made using simulation models that compare scenarios where food systems are governed differently to projected business as usual scenarios8. Food security governance simulation studies have the following characteristics: they use simulation models (1) to assess the impact of governance measures ex ante (2) on food security (3) within a food value chain context (4).

There are many types of simulation models, including agent-based, system dynamic, optimization and equilibrium models (for full list, see Supplementary Note 1), which have different capabilities, making them fit for different governance cases and contexts9. Models may be coupled to overcome weaknesses associated with certain model types10. Governance measures are the tools that governing entities use to affect society11. These can be categorized into: nodality (information dissemination), authority (laws and regulation), treasure (financial incentives) or organization (capacity building, punishment and crisis management)11. Within these categories, measures vary in how socially and spatially targeted they are (Extended Data Table 1). Social and spatial targeting is increasingly adopted around the world in the hope of maximizing impact with limited funds12,13. Socially targeted measures, such as social protection policies14,15,16, can reach vulnerable groups more effectively13. Spatially targeted measures can reach geographically vulnerable groups that, for example, live in drought-prone regions17,18,19 or food deserts20,21,22 more effectively23. In addition, spatially targeted governance can, through more efficient use of natural resources, contribute to enhanced food production24,25,26. Assessment of socially and spatially targeted measures does, however, require socially and spatially disaggregated models and data such as household surveys and gridded data, respectively. Most simulation models are not spatially disaggregated by default, and some model types, such as equilibrium models, traditionally rely on socially aggregated data such as trade balance sheets27.

Impacts of governance on food security are assessed with indicators of food availability, access, utilization or stability (for indicators, see Supplementary Dataset 1). Impacts may occur within or outside the jurisdictions within which measures are implemented. A trade moratorium on grains from Russia and Ukraine may, for instance, reduce grain prices in Russia and Ukraine, but increase prices globally, especially in net importing countries28. As both the COVID-19 pandemic and the war in Ukraine have illustrated, capturing such spillover effects is increasingly important in our highly interconnected world29. Simulated changes in food security occur within food value chain contexts, which differ from setting to setting. In subsistence settings, they may consist of production and consumption echelons only, but more often than not, middle echelons also play an important role through distribution, processing, storage and packaging, trade and wholesale, and/or retail of food30. Spurred on by widespread urbanization and associated dietary changes, these middle echelons have grown rapidly and continue doing so throughout the Majority World (that is, low-income countries)30. Rapid urbanization without accompanied development of road infrastructure may, for instance, leave urban consumers unable to obtain or afford produce from rural areas31. Despite this, the middle echelons tend to be missing in both development initiatives32 and simulation studies33.

Previous reviews have discussed subsets of above-mentioned characteristics (for example, how to bridge micro–macro scales in food security models10,34) or pointed out the shortcomings of the ex-post literature on food security governance6,7. Yet, no studies have systematically reviewed food security governance simulation studies, providing no community-wide understanding of common modelling practices, potential blind spots and promising developments. In this Article, to facilitate targeted and relevant future model development, we aim to summarize and critically appraise existing food security governance simulation studies, which, in contrast to ex-post studies, have not been systematically reviewed. In particular, we ask how food security governance is simulated. This question, in turn, is broken down into four subquestions: (1) Which modelling approaches are used? (2) How do simulation studies represent the food system, and to what extent are they (capable of) capturing up- and downstream value chain dynamics? (3) Which governance measures are simulated, and are simulation models (capable of) assessing socially and spatially targeted governance measures? And lastly: (4) How is food security measured, and to what extent are models (capable of) assessing spillover effects?

Results

A total of 1,953 potentially relevant studies were identified through a database search in Scopus and Web of Science (for full list, see Supplementary Dataset 2). Among these, 110 remained after title and abstract screening, full-text reading and qualitative content analysis (Table 1). The Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) flow diagram (Fig. 1) illustrates the selection process of the simulation studies for systematic review (for more information, see Supplementary Note 2, Supplementary Figs. 1–5 and Supplementary Dataset 3).

Table 1 Data extracted from eligible simulation studies during qualitative content analysis
Full size table
Fig. 1: PRISMA flow diagram illustrating the identification and screening process.
figure 1

Search terms were compiled on the basis of related systematic reviews, and studies were identified through a systematic search and checked against exclusion criteria.

Full size image

Modelling approaches

Food security governance was simulated by means of: cellular automata (CA, n = 6), agent-based models (ABM, n = 25), system dynamics models (SDM, n = 19), optimization models (n = 11), partial equilibrium (PE, n = 11) and computable general equilibrium (CGE, n = 24) models, micro-simulation models (n = 7), econometric models (n = 15) and/or other mathematical models (n = 13) (Fig. 2). Thematically, equilibrium and micro-simulation models were mostly economic, while ABM, SDM, optimization and econometric models were more diverse, often covering both the bio-physical and/or the social domain in addition to the economic (Fig. 2a and Extended Data Fig. 1).

Fig. 2: Domains, scales, geographic focus and food system representation of different types of simulation models.
figure 2

a–d, Number of studies using models with different modelling domains (a), spatial scales (b), simulated regions (c) and value chain echelons (d). Sizes of numbers and bubbles are proportional to number of studies. If a study covers multiple domains, is multi-scale, simulates or spans multiple regions, or covers multiple value chain echelons, it is counted multiple times.

Source data

Full size image

While most studies simulated governance using stand-alone models, a minority used coupled models (n = 21). Coupling to specialist, typically bio-physical, mathematical models to simulate, for example, crop growth18,35,36, hydrology37 or transportation20 occurred among ABM, SDM, optimization, CGE and PE studies. Further, CGE models were frequently coupled to micro-simulation models to scale down findings, enabling the assessment of governance impacts on different socio-economic groups, for example31,38,39 (Table 2).

Table 2 Use of coupled model types by food security governance simulation studies
Full size table

The spatial scale of simulation studies varied from village to global (Fig. 2b). With the notable exceptions of ABM, CA, and CGE and PE studies, most studies had a national scale. ABMs and CA tended to have subnational scales, simulating governance interventions within provincial or municipal jurisdictions. CGE and PE studies, contrastingly, were often global (Fig. 2b). In terms of geographic focus, there were no obvious patterns distinguishing model types (Fig. 2c).

Overall, two out of the three most food-insecure regions globally, East Asia and the Pacific (n = 43) and Sub-Saharan Africa (n = 40), were the regions covered most by the studies reviewed. South Asia (n = 21), however, received little attention considering the region’s high population and food insecurity status (Fig. 2c). Considering individual countries, China (n = 30) and the United States (n = 26) dominated. Here, studies reflected national policy, with models covering the United States typically aiming to improve access to healthy food in ‘food deserts’, poor city districts, through various pricing, safety net or city planning policies14,20,22, and models covering China typically aiming to obtain or maintain food self-sufficiency through, for instance, land protection schemes40,41,42.

Value chain coverage

No study covered all food value chain echelons, and only two studies43,44 covered all value chain echelons but one: retail. Most studies covered food production only, consumption only, or a combination of production, consumption and trade (Fig. 3). Food distribution, processing or storage, and retail were rarely simulated. Value chain coverage by CA, ABMs, SDMs and optimization models was overwhelmingly production centric (Fig. 2d). CGE and PE models, contrastingly, were centred around the trade echelon, often combined with production and/or consumption echelons.

Fig. 3: Food value chain coverage by food security governance simulation studies.
figure 3

Row numbers and bubble sizes refer to the number of studies that cover the specific combinations of value chain echelons. Column numbers refer to the total number of studies that cover each specific value chain echelon.

Source data

Full size image

Studies that covered the distribution, processing, storage and retail echelons came in three fashions. Firstly, some ABM studies assessed policies that aimed to improve urban food access by transporting food to the consumers22, or the consumers to the food20,21, or by relocating or creating new supermarkets20,21. Secondly, some optimization and SDM studies simulated measures that aimed to improve the availability and stability of (perishable) foods by, for instance, setting up a bottom-up cooperative for dairy distribution, processing and retail45, increasing the shelf life of bread46 or meat47 at retail, using import quotas to strengthen the domestic supply chains48, or improving sourcing strategies of public distribution programmes49. Lastly, some CGE and PE model studies simulated measures that aimed to reduce poverty and improve food access through investment in transportation31,50, processing51 or marketing50,51 infrastructure.

Governance measures

The governance measures assessed most through food security governance simulation modelling were treasure or authority based (Table 3). Among treasure-based measures, bearer-directed payments (n = 31) dominated, and among authority-based measures, standard constraints dominated (n = 38). Frequently simulated authoritarian measures were land protection24,40,52, tax policies53,54,55 and trade regulations such as tariffs15,54,56, quotas28,55,57 or bans28,43,58. These measures have in common that they affect or are meant to affect the wider public (first row, Table 3). Infrequent or absent authoritarian measures were conditional tokens59 and enablements16, and certificates and directed constraints, respectively, which are customized for individuals (second and third row, Table 3). Treasure-based measures were more frequently targeted towards groups or individuals than authority-based measures. Social protection policies such as conditional cash transfers14,15,16 or food stamps, for example60,61,62, were specifically targeted towards the poorest, most vulnerable citizens.

Table 3 Classification of governance measures
Full size table

Organization and nodality were simulated much less than treasure and authority. Among organization-based measures, at-large treatment (typically, big infrastructure projects such as construction of dams63,64,65 or irrigation canals17,65,66, roads31,50 or market facilities20,21,50) was simulated most. Additionally, some studies assessed the impact of governance measures aiming to improve food access through better distribution systems39,57,67. The most common nodal measures assessed were group-targeted messages. These were usually policies through which the government educated68,69,70 or tried to convince farmers of the benefit of certain products70,71 or practices72,73,74 (Table 3).

Studies’ choices of governance measures sometimes coincided with choices of model types. At-large trade policies relying on authority such as import tariffs on food commodities were, for instance, almost exclusively simulated by CGE15,54,56 and PE models55,70. Contrastingly, infrastructure projects and other organizational governance measures were simulated by SDM66,75,76, ABM20,63,65 and optimization49,50,64 models. Nodal measures were mostly simulated by ABMs and SDMs74,77,78, as well. Treasure-based measures such as subsidies on farm inputs were, however, simulated by the full range of model types. Socially targeted governance measures, too, were simulated by the full range of simulation models (Fig. 4a and Extended Data Fig. 2a). Spatially targeted governance measures were rarer than socially targeted governance measures, but were also simulated by most model types (Fig. 4b and Extended Data Fig. 2b).

Fig. 4: Social and spatial targeting of governance measures.
figure 4

a,b, Dark-green bubbles show the number of studies that simulated socially (a) or spatially (b) targeted governance measures with socially (a) or spatially (b) disaggregated data. Turquoise bubbles show the number of studies that simulated untargeted measures with or without socially (a) or spatially (b) disaggregated data. Pink bubbles show the number of studies that simulated socially (a) or spatially (b) targeted governance measures without using socially (a) or spatially (b) disaggregated data. Bubble sizes and numbers are proportional to number of studies.

Source data

Full size image

To be able to assess the impact of socially and/or spatially targeted governance measures, models need to be socially and/or spatially disaggregated. We found that most studies (n = 65) were able to identify food-insecure socio-economic groups within the wider study population, and most of these studies also simulated governance measures that targeted these people directly (n = 41; Fig. 4a). Most of these were treasure based. Socially targeted governance measures relying on organization, and especially nodality and authority, were much rarer (Table 3). Most studies did not use spatially disaggregated data (n = 68), rendering them unable to identify vulnerable geographic groups (for example, inhabitants of drought-prone regions; Fig. 4b). Among the spatially disaggregated studies, few (n = 14) explored the potential benefits of spatially targeted measures that could potentially reach these vulnerable groups17,18,19 or enhance production through more efficient natural resource management17,25,26 (Table 3 and Fig. 4b). Further, our review showed that some studies simulated targeted measures with models calibrated using socially (n = 14) and/or spatially (n = 12) aggregated data only (Fig. 4). Bazzana et. al.65, for instance, assessed the impact of the construction of hydro-electric dams, electric grids and water canals, and payments as compensation to those who lost resources because of these infrastructure projects. Despite the spatially targeted nature of these governance measures, the assessment was made without using any spatial data, without reporting the spatial extent or resolution of the model and without providing a spatially disaggregated visualization of the impact of the governance measures.

Governance impacts

Governance impacts on food security were most frequently assessed using indicators for food access (44%) or availability (27%). Few studies assessed the impact of governance on food utilization (n = 39), and even fewer on the nutritional qualities of crops (n = 1)68 and diets (n = 3)37,72,79. The number of studies that assessed the impact of governance on stability was also low (n = 29), but as many studies (also) provided indicator values over time (n = 41); in total, more than half of the studies assessed stability. However, most models had too low temporal resolution (≥annual) to account for seasonal differences. In total, 123 different food security indicators were used (Supplementary Dataset 1). The majority of these (60%) were unique to single studies. The price of food commodities (n = 38) and income of citizens (n = 37) were most frequently used to assess the impact of governance on food access. The production of food (n = 33) and the area used to grow crops or keep livestock (n = 26) were most frequently used to assess the impact of governance on food availability. The consumption (n = 21) or purchase (n = 6) of food were most frequently used to assess the impact of governance on food utilization. Lastly, self-sufficiency (n = 8) and stocks (n = 8) of food crops were most frequently used to assess the impact of governance on stability. Only a minority of the studies (n = 23) captured changes in the distribution of food security within a population through indicators such as poverty incidence (n = 13), or income inequality (n = 9), though a few additional studies gave insight into inequality (n = 8), for example, by providing the incomes for different socio-economic groups31,80,81. Most indicators were assessed only within the jurisdiction where the governance measures were implemented (n = 90), that is, potential spillover effects were seldom captured. When the impacts were assessed globally (n = 14), or locally outside a jurisdiction (n = 12), this was done by PE28,37,55 or CGE54,82,83 simulation studies.

Discussion

A wide variety of governance measures have been simulated in the reviewed studies, though authority- and treasure-based measures were much more frequent than organization-based, and especially nodality-based, measures. Concerning food security governance, there has not been a comprehensive overview of governance measures implemented in practice. Country responses to the 2008–2009 food price crisis were, however, mapped by Demeke et. al.13. In line with this systematic review, they found that a lot of authority- and treasure-based at-large trade policies were implemented. They also registered that many countries implemented treasure-based food price regulation policies and social protection measures such as cash and food transfers to vulnerable people, and subsidy schemes aiming to increase food production. However, they did not register any nodality or organizational measures. This might be because, in contrast to the implemented authority- and treasure-based measures, they do not provide immediate help in times of crises. Nodality and organizational measures typically require longer time horizons, as they often—though not always—require actors to adopt new knowledge54,70,73 or norms20,60,84, or require the construction of large-scale infrastructure24,66,76 or logistic networks21,49,85.

As research and practice have thus far focused on treasure- and authority-based measures, the potential of many organizational, and especially nodal, measures remains unexplored. In the field of public health, nodal measures such as mass radio campaigns have been shown to be highly cost-effective, mostly owing to the fact that they are very cheap compared with other health interventions86. In the field of food security governance, we found only one realistically simulated example of nodal governance implementation and its effectiveness. This study, performed by Williams et al.18, found the impact of sharing climate forecasts with farmers (a nodal measure) to be similar to the impact of providing 20% of farmers with jobs (an organizational measure). However, they expected the nodal measure to be much cheaper and easier to realize18. The rarity of simulated nodal measures could be due to the fact that this requires models with high degrees of social disaggregation, and the ability to simulate interacting agents, which requires approaches such as ABMs. As increased cost-effectiveness in food security governance could speed up the alleviation of food security, we see the development of such models as an important research priority.

Shifting the focus from spatially untargeted at-large measures based on treasure and authority towards spatially and socially targeted measures based on organization, authority and nodality would be helpful to expand our knowledge base, potentially increase the cost-effectiveness of food security governance and support the decision-making process of policymakers already interested in implementing such governance measures but who often fail to do so effectively12,13. This can be achieved through the use of spatial models such as ABMs20,63,87 or CA25,40,52, or by disaggregating traditionally aspatial models such as CGE or optimization models by either parameterizing them differently for different geographic regions39,88 or by coupling them to spatial models73. In terms of the assessment of the impact of governance on food security, simulation models tend to assess governance impacts on availability, access and extra-seasonal stability, but rarely intra-seasonal stability, nutrition and social inequality. As these aspects are prerequisites to achieving food security5,89, and their assessment is technically feasible18,67,68, we call researchers to include them in future studies.

Steering away from unwanted, negative spillover effects, that is, reducing food security within a region at the expense of food security outside that region, is another prerequisite to achieving net food security improvement. To avoid spillover effects, studies need to assess the impacts of governance measures not only within the jurisdiction where they are implemented, but also in regions connected to this jurisdiction29. However, only a small minority of studies did this, all using CGE or PE models. Further, only three of the studies capturing spillover effects used socially disaggregated data70,90,91, and only one reported the impacts of governance implementation in a socially disaggregated way90. This illustrates that the ability to assess and target governance in a socially disaggregated way and the ability to assess potential spillover effects are rarely combined. Yet, this is crucial for achieving social justice outcomes5. To combine these two abilities, we need to bridge the gap between macro- and micro-scale models10,92, by downscaling macro-models or by upscaling10,34 or by telecoupling29 micro-models.

Scaling down is a technical development that has opened up for equilibrium models through coupling with micro-simulation models39,88. Scaling up micro-scale models is less common than scaling down macro-scale models93, and none of the models eligible for this review did this. Nevertheless, this is technically feasible and has been illustrated by Niamir et al.94, who simulated the energy consumption choices of households with an ABM and upscaled their impact on total EU-26 energy use with a CGE model. Telecoupling, which accounts for socioeconomic–environmental interactions between distant places29, has been illustrated in various food system simulation models. Dou et. al.95, for instance, used a telecoupled ABM to assess the impact of increased demand for soybeans in China on land use and farmers’ welfare in Brazil. However, none of our identified food security governance simulation studies use telecoupling.

Lastly, we found that simulation models typically either have a macro-economic focus and simulate the inter-linked impact of governance on food production and consumption through trade, or focus solely on either the production or consumption end of the value chain (Fig. 3). The governance measures simulated reflect this trend: measures attempting to improve transportation20,22,49 and retail20,47,67, and especially storage57,62,96 and processing45,51, were rare. The neglect of the value chain echelons between production and consumption is in line with previous reviews—which have termed them the ‘missing’32 or ‘hidden’30 middle—except that the ‘productivist paradigm’7,33 was less striking. We observed a stronger consumption focus. This may be due to a production bias present in previous reviews7, which we also observed when compiling the search string used for this review (Supplementary Note 2). Though not all value chain echelons are relevant for all food systems and commodities, it is important for the field of governance simulation to advance the ability to simulate the role of the missing middle in up- and downstream value chain dynamics. This can be done with any type of simulation model, but requires hardly accessible data on missing middle value chain actors, especially in the Majority World97.

Analyses of the quantitative impacts of governance measures on food security are important to inform decision makers about the effectiveness of different governance measures and, through this, speed up the alleviation of food insecurity. To facilitate such analyses, we explored which studies and governance measures within the reviewed pool of literature could be compared as they used the same food security indicators, measured with the same spatial and temporal precision, and were implemented either inside a jurisdiction, outside it or globally. Using these criteria for comparability, we found 26 groups with three to six studies, exploring up to 13 governance measures, that could potentially be compared in meta-analyses. Multiple comparisons could be made between ABM, SDM and optimization modelling studies, SDM and PE modelling studies, or CGE and PE modelling studies. The first group of studies assessed the impacts of nodal, treasure-based and organizational measures on land use and farm income. The second two groups assessed the impacts of treasure-based and authority-based measures on the consumption, import, production and price of cereals. For the full list of comparable simulation studies, their simulated governance measures and their impacts on food security, see Supplementary Dataset 4. Owing to the absence of reporting guidelines within the field, researchers may need to request additional information from the authors of the studies.

We note that our findings are subject to limitations. Firstly, studies were identified using a search string compiled from existing reviews on simulation modelling, governance, food security and food systems. Biases present in these reviews may be only partly addressed through the compiled search string used in our research. We noticed, for instance, that we initially lacked terms for the consumption echelon, which we therefore supplemented. Other biases may have gone unnoticed. Secondly, we did not include specific governance programmes or food security indicators in our search string. Studies that assess the impact of specific governance measures (for example, the Indian public food distribution programme49) on specific food security indicators (for example, dietary income differential67) without framing their studies or outcomes in the context of governance and food security might have been overlooked. Thirdly, we did not include grey literature, but the grey literature we did identify did not contain any approaches or topics that were not covered by the peer-reviewed studies (Supplementary Note 2). Lastly, our findings, resulting from a qualitative content analysis, involved interpretation.

Nevertheless, by this systematic review, we connected the fragmented landscapes of studies that simulate food security governance implementation and provided a comprehensive overview of the state of the field, mapping dominant modelling approaches (ABMs, SDMs and CGE models), governance measures (treasure- and authority-based) and value chain echelons (production, trade and consumption). To be able to generate useful knowledge for the community (for example, for meta-analyses similar to those in public health98 or ecology99), however, future research could benefit from harmonization of food security indicators. We also identified blind spots regarding studies’ choices of governance measures, food security indicators and value chain coverage. We recommend the development of ABMs that simulate the implementation of nodal governance measures (for example, provision of seasonal weather forecasts for farmers18); socially and spatially disaggregated models that consider socially vulnerable groups and bio-physical heterogeneity and simulate the implementation of socially and spatially targeted governance measures21,22,65; macro–micro coupling or telecoupling to capture potential spillover effects (for example, trade moratorium on grains from Russia and Ukraine28); and a re-orientation moving beyond availability and access-focused production, trade and consumption studies, towards utilization-, nutrition- and social justice-focused value chain studies.

Methods

Studies sought for review

Food security governance simulation studies sought for this review meet four criteria: they use simulation models (1) to assess the impact of governance measures ex ante (2) on food security (3) within a food system context (4). A paper was considered a simulation study if it used a model that simulates alternative scenarios with future projections or alternative (future) realities. Models that simulated alternative historical realities (≤1990) and empirical models that assessed the impact of governance that was already implemented were excluded. Additionally, our scope was limited to peer-reviewed English-language journal papers published after 2000. Governance was approached from a tool-kit perspective, looking at the tools that governance entities may use to affect society11. We distinguished four main categories of governance tools: nodality (information dissemination), authority (laws and regulation), treasure (financial incentives) and organization (capacity building, punishment and crisis management)11. Any study that described a governance measure that was described clearly enough to be categorized in this way is eligible for this review (Table 3 and Extended Data Table 1).

With regard to food security, we considered any study that assessed the impact of governance on an indicator for any of the four dimensions of food security: availability, access, utilization or stability (for full list of indicators, see Extended Data Table 1). Indicators related to food safety and overconsumption, though important aspects of the wider field of food governance, were beyond the scope of this review, and have been studied by others100,101,102. Food systems could potentially encompass all aspects of our lives, but for this review we confined them to the three elements identified by the High-Level Panel of Experts (2017) (ref. 103): (1) food value chains (including food production, storage, distribution, processing, packaging and retail), (2) food environments (the physical, economic, political or socio-cultural context within which people interact with their food) and (3) consumer behaviour. Aquatic food production, consumer behaviour outside of food environments, and hunting, gathering or fishing were beyond the scope of this review.

Search term selection

To identify food security governance simulation studies, we collected search terms from existing systematic literature reviews on the topics of (1) food systems, (2) food security, (3) simulation modelling and (4) governance (for list of reviews, see Supplementary Table 1). Different combinations of search terms were tested on Web of Science and Scopus. For each combination, we went through the first 40 studies. If fewer than two studies were eligible on the basis of abstract and title scanning, the term responsible for those results was dropped (for search terms, see Supplementary Table 2). To test the sensitivity of the selected search terms, the collection of identified studies was compared with the collection of studies identified for the review by Utomo et al.33. Terms used in studies from their review that met the inclusion criteria of our review were added. The final search string used to identify food security governance simulation studies on 21 April 2021 was:

  1. 1.

    ((‘food system’ OR agricultur* OR farm* OR ((food OR agricultural) AND (‘value chain*’ OR ‘supply chain*’ OR value chain* OR supply-chain*)) OR agri-food OR ‘food processing’ OR ‘food production’ OR agri-business OR ‘food transfer*’ OR livestock OR pasture OR ‘food consumption’ OR ‘food purchase’) AND

  2. 2.

    (nutrition* OR diet* OR ‘food secur*’ OR ‘food insecur*’ OR ‘food access*’ OR ‘food availab*’ OR ‘food demand’ OR ‘food supply’ OR ‘food sovereign*’ OR ‘food sufficien*’ OR ‘food insufficien*’ OR ‘food utili[sz]ation’ OR hunger OR malnutrition OR poverty OR livelihood) AND

  3. 3.

    ((agent-based AND model*) OR (multi-agent) OR (individual-based AND model*) OR (simulation AND model*)) AND

  4. 4.

    (govern* OR stewardship OR regime OR politic* OR polic* OR accountability OR incentiv*)).

Screening process

A total of 1,953 potentially relevant studies were identified through a database search in Scopus and Web of Science (for full list, see Extended Dataset 2). In addition to 624 duplicate studies, a total of 154 studies were excluded before screening as they did not meet the criteria: English-language peer-reviewed journal papers published after 2000 (for screening criteria, see Supplementary Figs 1–5). The remaining 1,175 studies were screened, first on the basis of titles, abstracts and keywords, then, if deemed eligible, on the basis of full text. The first round of screening was performed in Ryyan QCRI, a webtool that helps track and expedite the screening process104. The second round was performed in ATLAS.ti v9.1.7 (ref. 105). The PRISMA flow diagram (Fig. 1) illustrates the process of the selection of simulation studies for the systematic review.

Data extraction and analysis

Eligible studies were imported to ATLAS.ti for a directed, qualitative content analysis106 where descriptive codes were used to extract text fragments containing information about the simulation approaches (for the ATLAS.ti project, see Data Repository107). Coded information relating to characteristics 1–5 is specified in Table 1. The text fragments, along with the descriptive codes attached to them, and meta-data concerning the document they were extracted from, were exported to Excel (for raw data, see Data Repository107), then imported to R v4.1.0 for data processing and analysis. For each of the reported results, the number of studies in which (a combination of) codes occurred in was counted. If a study contained multiple code categories per variable (for example, multiple model domains), each code category was counted.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.