About the Author

Decision Support Systems to Enhance Food Security in the U.K.

By Martine J. Barons

Data-driven decision support systems help governing bodies, business leaders, and other policymakers compare the likely outcomes of different combinations of actions. Such support is particularly valuable in dynamic systems that incorporate a lot of uncertainty. Uncertainty stems from the limits of our knowledge of the external world, the modeling and analysis process, or internal doubt, and generally pertains to an inability to answer questions in a precise manner. Scientific approaches must therefore address and faithfully communicate this uncertainty to the people who design, implement, and feel the impact of response measures.

Food Security and Food Banks

After the 2008 financial crisis, food banks began to appear in increasingly large numbers throughout the U.K. [1]. Unlike countries such as the U.S., Canada, and Australia, the U.K. was relatively unfamiliar with the concept of food banks; instead, it has a welfare system based on the notion of shared risk that was developed after World War II as a last line of defense against hunger and destitution.

Figure 1. Food-insecure households struggle to regularly access, afford, and utilize nutritious food. Figure courtesy of Martine Barons.
Organizations in the U.K. have vocalized an intent to eliminate the need for food banks. Many adhere to the philosophy that food banks are an emergency measure and do not tackle the underlying causes of household food insecurity; some non-food interventions have proven more successful in that regard [8]. Food security exists when all people, at all times, have physical and economic access to a sufficient, safe, and nutritious food supply that meets their dietary needs and preferences for an active and healthy lifestyle (see Figure 1) [4]. Under the International Covenant on Economic, Social and Cultural Rights, the U.K. government has a legal duty to take appropriate steps to realize the right of all people to be free of hunger.

According to the Global Food Security Index, the U.K. is among the most food secure countries in the world; in 2017, it ranked third out of 113, just after Ireland and the U.S. By 2022, however, it had declined to ninth place. For many years, the exact scale of household-level food insecurity in the U.K. was unknown because the country lacked systematic national assessments on this topic. But beginning in 2016, the U.K. Food Standards Agency began to utilize the Household Food Security Survey Module that originated from the U.S. Department of Agriculture. This module divides households into four categories: 

  1. High food security: no reported indications of food access problems or limitations
  2. Marginal food security: one or two reported indications that typically mention anxiety about food sufficiency or shortage in the house, but little to no indication of changes in diet or food intake 
  3. Low food security: reports of reduced quality, variety, or desirability of diet, with little to no indication of reduced food intake
  4. Very low food security/hunger: reduced food intake of household members at certain times throughout the year and disruption of normal eating patterns due to lack of money and other resources. 

The latest data in the U.K. indicates that 92 percent of households are food secure, four percent have low food security, and four percent have very low food security [10]. Food insecurity is not evenly dispersed across society; age, disability, ethnicity, and geographical location all play a role. Missing meals and changing one’s diet—especially over extended periods of time—can lead to adverse health effects, especially in children. Possible consequences include wasting, diabetes, cardiovascular disease, certain cancers, food-borne illness, mental health impacts due to social exclusion, weakened responses to infection, and increased risk of illness or death from stunted growth [6]. Food insecurity can also be a dimension of wider insecurity [3].

Integrating Decision Support Systems

An integrating decision support system (IDSS) seeks to provide an unambiguous, complete framework through which to evaluate the efficacy of different policies in complex, evolving scenarios [9]. Researchers partner with relevant decision-makers and expert panels to develop an IDSS in an iterative manner [2]; the IDSS then aids governing bodies by providing a clear evaluation of the candidate policies. More specifically, it combines expert judgement with data for each subsystem, ultimately yielding a full inferential procedure that can represent complex systems. The decision-making panel (e.g., a board of directors) receives guidance from domain experts who are involved throughout the process. The key development stages are as follows: soft elicitation [5], model building, model quantification, model evaluation and feedback, software engineering, and launch and adoption. Communication with specialists from other expert domains sometimes poses a challenge but is absolutely necessary to render effective decision support [11].

Figure 2. The required expert panels for the integrating decision support system (IDSS) on food insecurity in the U.K. Each node represents an expert panel that uses its models and data to provide summaries of expected values and relevant moments for each policy decision under consideration. Figure courtesy of [1].
Probabilistic graphical models are particularly well suited for decision support, as they represent the world as a set of variables and use a graph to model the probabilistic dependencies between these variables. We can build these graphs based on the knowledge of domain experts, which provides a narrative for the system that can be transparently and coherently revised as the domain changes.

Consider a vector of random variables that are relevant to the system \(\mathbf{Y}=(Y_1,\ldots, Y_n)\). Typically, different expert panels have expertise in particular aspects of the multivariate problem. After identifying the most appropriate expert panels for every subsystem, each subpanel should defer to the others and adopt their models, reasoning, and evaluations in the relevant domains. Each expert panel \(G_i\) is responsible for a subvector \(\mathbf{Y}_{B_i}\) of \(\mathbf{Y}\), with \(B_1,\ldots,B_m\) as a partition of \(\{1,\ldots,n\}\). We then decompose the multivariate problem into submodels. The joint model accommodates the diversity of information from the different component models and robustly handles the intrinsic uncertainty in these submodels. In large problems, the decision-maker (DM) is often a council that comprises several individuals. A DM makes decisions \(d\in \mathcal{D}\), where \(\mathcal{D}\) represents the set of all policy options that the DM plans to consider. It receives information from each panel and reaches a conclusion that depends on a reward (utility) function \(R(\mathbf{Y},d)\), \(\mathbf{Y}\in R_{\mathbf{Y}}\), \(d\in \mathbf{D}\). Any valid IDSS must respect a set of common assumptions that all panels share, as well as the union of the utility, policy, and structural consensus [9].

When the utility function is in an appropriate polynomial form for a distributive IDSS, each panel must deliver only a short vector of conditional moments rather than entire distributions; this is because the embedded conditional independence statements allow the use of tower rule recurrences [7] for fast calculations and the embedding of propagation algorithms within the customized IDSS. When relevant new information emerges, individual panels can quickly and easily perform prior and posterior analyses to update both the information that they donate and the expected utility scores. This aspect is especially useful for emergency decision support and represents a huge efficiency gain over the reconstruction and reparameterization of an entire large model. A number of frameworks satisfy the IDSS property requirements, including staged trees, Bayesian networks, chain graphs, multiregression dynamic models (MDMs), and uncoupled dynamic Bayesian networks.

Figure 3. A standardized utility posterior median that has been decomposed for health and education variables. Figure courtesy of [1].

Decision Support for Household Food Security in the U.K.

Recent increases in food poverty within the U.K. have raised pertinent questions about the main drivers of food insecurity, their changes over time, and the use of evidence for the evaluation of decision support policies. In this context, we developed an IDSS for household food security that helps decision-makers compare several candidate policies that pertain to food insecurity at the household level (see Figure 2). This proof-of-concept IDSS application for government policy uses an MDM as the overarching framework.

Policymakers have a limited number of decisions at their disposal (i.e., a small decision space), and they can compare the effects of various combinations of decisions on their outcome of interest. Because household food security was not directly measurable, the IDSS utilized a proxy of health and educational attainment. Decision-makers can compare the scores for this multi-attribute utility under every candidate policy against both each other and the baseline values to identify the optimal policy. This outcome may then serve as the basis of a business case for investment. Because policymakers are usually not mathematicians, the results must be displayed in a straightforward, understandable way; as such, an overall score and a score for each utility’s attributes often benefits decision-making discussions (see Figure 3). In some circumstances, comparison with an independent standard—e.g., the national average or an ideal situation—can offer useful context.

[1] Barons, M.J., Fonseca, T.C.O., Davis, A., & Smith, J.Q. (2022). A decision support system for addressing food security in the United Kingdom. J. R. Stat. Soc. Ser. A Stat. Soc., 185(2), 447-470.
[2] Barons, M.J., Wright, S.K., & Smith, J.Q. (2018). Eliciting probabilistic judgements for integrating decision support systems. In L.C. Dias, A. Morton, & J. Quigley (Eds.), Elicitation: The science and art of structuring judgement (pp. 445-478). Cham, Switzerland: Springer.
[3] Chilton, M., Rabinowich, J., Council, C., & Breaux, J. (2009). Witnesses to hunger: Participation through photovoice to ensure the right to food. Health Hum. Rights, 11(1), 73-85.
[4] Food and Agriculture Organization of the United Nations. (1996). Rome declaration on world food security (Technical report). Rome, Italy: World Food Summit. Retrieved from https://www.fao.org/3/w3613e/w3613e00.htm.
[5] French, S. (2022). From soft to hard elicitation. J. Oper. Res. Soc., 73(6), 1181-1197.
[6] Friel, S., & Ford, L. (2015). Systems, food security and human health. Food Secur., 7(2), 437-451.
[7] Leonelli, M., & Smith, J.Q. (2015). Bayesian decision support for complex systems with many distributed experts. Ann. Oper. Res., 235(1), 517-542.
[8] Phojanakong, P., Welles, S., Dugan, J., Booshehri, L., Weida, E.B., & Chilton, M. (2020). Trauma-informed financial empowerment programming improves food security among families with young children. J. Nutr. Educ. Behav., 52(5), 465-473.
[9] Smith, J.Q., Barons, M.J., & Leonelli, M. (2017). Coherent inference for statistical inference serving integrating decision support systems. Preprint, arXiv:1507.07394.
[10] U.K. Department for Environment, Food & Rural Affairs (2021). Theme 4: Food security at household level. In United Kingdom food security report 2021. Retrieved from https://www.gov.uk/government/statistics/united-kingdom-food-security-report-2021/united-kingdom-food-security-report-2021-theme-4-food-security-at-household-level.
[11] Walton, J.L., Levontin, P., Barons, M.J., Workman, M., Mackie, E., & Kleineberg, J. (2022). Communicating climate risk: A toolkit (Version 3). Preprint, Cambridge Open Engage.

Martine J. Barons is director of the Applied Statistics and Risk Unit at the University of Warwick, which serves as the Knowledge Exchange Hub for the Department of Statistics. She works with nonacademic partners from business, industry, and government to enhance their business needs, with a particular focus on evidence-informed decision support across a range of applications. Prior to her becoming an academic, Barons had a career in accountancy and auditing.